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.
10 #include "dm-uevent.h"
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/mutex.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/signal.h>
18 #include <linux/blkpg.h>
19 #include <linux/bio.h>
20 #include <linux/mempool.h>
21 #include <linux/dax.h>
22 #include <linux/slab.h>
23 #include <linux/idr.h>
24 #include <linux/uio.h>
25 #include <linux/hdreg.h>
26 #include <linux/delay.h>
27 #include <linux/wait.h>
29 #include <linux/refcount.h>
30 #include <linux/part_stat.h>
31 #include <linux/blk-crypto.h>
32 #include <linux/blk-crypto-profile.h>
34 #define DM_MSG_PREFIX "core"
37 * Cookies are numeric values sent with CHANGE and REMOVE
38 * uevents while resuming, removing or renaming the device.
40 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41 #define DM_COOKIE_LENGTH 24
44 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
45 * dm_io into one list, and reuse bio->bi_private as the list head. Before
46 * ending this fs bio, we will recover its ->bi_private.
48 #define REQ_DM_POLL_LIST REQ_DRV
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;
65 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
66 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
68 void dm_issue_global_event(void)
70 atomic_inc(&dm_global_event_nr);
71 wake_up(&dm_global_eventq);
74 DEFINE_STATIC_KEY_FALSE(stats_enabled);
75 DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
76 DEFINE_STATIC_KEY_FALSE(zoned_enabled);
79 * One of these is allocated (on-stack) per original bio.
86 unsigned sector_count;
87 bool is_abnormal_io:1;
88 bool submit_as_polled:1;
91 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
93 return container_of(clone, struct dm_target_io, clone);
96 void *dm_per_bio_data(struct bio *bio, size_t data_size)
98 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
99 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
100 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
102 EXPORT_SYMBOL_GPL(dm_per_bio_data);
104 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
106 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
107 if (io->magic == DM_IO_MAGIC)
108 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
109 BUG_ON(io->magic != DM_TIO_MAGIC);
110 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
112 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
114 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
116 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
118 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
120 #define MINOR_ALLOCED ((void *)-1)
122 #define DM_NUMA_NODE NUMA_NO_NODE
123 static int dm_numa_node = DM_NUMA_NODE;
125 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
126 static int swap_bios = DEFAULT_SWAP_BIOS;
127 static int get_swap_bios(void)
129 int latch = READ_ONCE(swap_bios);
130 if (unlikely(latch <= 0))
131 latch = DEFAULT_SWAP_BIOS;
135 struct table_device {
136 struct list_head list;
138 struct dm_dev dm_dev;
142 * Bio-based DM's mempools' reserved IOs set by the user.
144 #define RESERVED_BIO_BASED_IOS 16
145 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
147 static int __dm_get_module_param_int(int *module_param, int min, int max)
149 int param = READ_ONCE(*module_param);
150 int modified_param = 0;
151 bool modified = true;
154 modified_param = min;
155 else if (param > max)
156 modified_param = max;
161 (void)cmpxchg(module_param, param, modified_param);
162 param = modified_param;
168 unsigned __dm_get_module_param(unsigned *module_param,
169 unsigned def, unsigned max)
171 unsigned param = READ_ONCE(*module_param);
172 unsigned modified_param = 0;
175 modified_param = def;
176 else if (param > max)
177 modified_param = max;
179 if (modified_param) {
180 (void)cmpxchg(module_param, param, modified_param);
181 param = modified_param;
187 unsigned dm_get_reserved_bio_based_ios(void)
189 return __dm_get_module_param(&reserved_bio_based_ios,
190 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
192 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
194 static unsigned dm_get_numa_node(void)
196 return __dm_get_module_param_int(&dm_numa_node,
197 DM_NUMA_NODE, num_online_nodes() - 1);
200 static int __init local_init(void)
204 r = dm_uevent_init();
208 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
209 if (!deferred_remove_workqueue) {
211 goto out_uevent_exit;
215 r = register_blkdev(_major, _name);
217 goto out_free_workqueue;
225 destroy_workqueue(deferred_remove_workqueue);
232 static void local_exit(void)
234 flush_scheduled_work();
235 destroy_workqueue(deferred_remove_workqueue);
237 unregister_blkdev(_major, _name);
242 DMINFO("cleaned up");
245 static int (*_inits[])(void) __initdata = {
256 static void (*_exits[])(void) = {
267 static int __init dm_init(void)
269 const int count = ARRAY_SIZE(_inits);
272 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
273 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
274 " Duplicate IMA measurements will not be recorded in the IMA log.");
277 for (i = 0; i < count; i++) {
291 static void __exit dm_exit(void)
293 int i = ARRAY_SIZE(_exits);
299 * Should be empty by this point.
301 idr_destroy(&_minor_idr);
305 * Block device functions
307 int dm_deleting_md(struct mapped_device *md)
309 return test_bit(DMF_DELETING, &md->flags);
312 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
314 struct mapped_device *md;
316 spin_lock(&_minor_lock);
318 md = bdev->bd_disk->private_data;
322 if (test_bit(DMF_FREEING, &md->flags) ||
323 dm_deleting_md(md)) {
329 atomic_inc(&md->open_count);
331 spin_unlock(&_minor_lock);
333 return md ? 0 : -ENXIO;
336 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
338 struct mapped_device *md;
340 spin_lock(&_minor_lock);
342 md = disk->private_data;
346 if (atomic_dec_and_test(&md->open_count) &&
347 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
348 queue_work(deferred_remove_workqueue, &deferred_remove_work);
352 spin_unlock(&_minor_lock);
355 int dm_open_count(struct mapped_device *md)
357 return atomic_read(&md->open_count);
361 * Guarantees nothing is using the device before it's deleted.
363 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
367 spin_lock(&_minor_lock);
369 if (dm_open_count(md)) {
372 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
373 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
376 set_bit(DMF_DELETING, &md->flags);
378 spin_unlock(&_minor_lock);
383 int dm_cancel_deferred_remove(struct mapped_device *md)
387 spin_lock(&_minor_lock);
389 if (test_bit(DMF_DELETING, &md->flags))
392 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
394 spin_unlock(&_minor_lock);
399 static void do_deferred_remove(struct work_struct *w)
401 dm_deferred_remove();
404 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
406 struct mapped_device *md = bdev->bd_disk->private_data;
408 return dm_get_geometry(md, geo);
411 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
412 struct block_device **bdev)
414 struct dm_target *tgt;
415 struct dm_table *map;
420 map = dm_get_live_table(md, srcu_idx);
421 if (!map || !dm_table_get_size(map))
424 /* We only support devices that have a single target */
425 if (dm_table_get_num_targets(map) != 1)
428 tgt = dm_table_get_target(map, 0);
429 if (!tgt->type->prepare_ioctl)
432 if (dm_suspended_md(md))
435 r = tgt->type->prepare_ioctl(tgt, bdev);
436 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
437 dm_put_live_table(md, *srcu_idx);
445 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
447 dm_put_live_table(md, srcu_idx);
450 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
451 unsigned int cmd, unsigned long arg)
453 struct mapped_device *md = bdev->bd_disk->private_data;
456 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
462 * Target determined this ioctl is being issued against a
463 * subset of the parent bdev; require extra privileges.
465 if (!capable(CAP_SYS_RAWIO)) {
467 "%s: sending ioctl %x to DM device without required privilege.",
474 if (!bdev->bd_disk->fops->ioctl)
477 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
479 dm_unprepare_ioctl(md, srcu_idx);
483 u64 dm_start_time_ns_from_clone(struct bio *bio)
485 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
487 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
489 static bool bio_is_flush_with_data(struct bio *bio)
491 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
494 static void dm_io_acct(struct dm_io *io, bool end)
496 struct dm_stats_aux *stats_aux = &io->stats_aux;
497 unsigned long start_time = io->start_time;
498 struct mapped_device *md = io->md;
499 struct bio *bio = io->orig_bio;
500 unsigned int sectors;
503 * If REQ_PREFLUSH set, don't account payload, it will be
504 * submitted (and accounted) after this flush completes.
506 if (bio_is_flush_with_data(bio))
508 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
509 sectors = bio_sectors(bio);
511 sectors = io->sectors;
514 bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
519 if (static_branch_unlikely(&stats_enabled) &&
520 unlikely(dm_stats_used(&md->stats))) {
523 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
524 sector = bio->bi_iter.bi_sector;
526 sector = bio_end_sector(bio) - io->sector_offset;
528 dm_stats_account_io(&md->stats, bio_data_dir(bio),
530 end, start_time, stats_aux);
534 static void __dm_start_io_acct(struct dm_io *io)
536 dm_io_acct(io, false);
539 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
542 * Ensure IO accounting is only ever started once.
544 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
547 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
548 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
549 dm_io_set_flag(io, DM_IO_ACCOUNTED);
552 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
553 spin_lock_irqsave(&io->lock, flags);
554 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
555 spin_unlock_irqrestore(&io->lock, flags);
558 dm_io_set_flag(io, DM_IO_ACCOUNTED);
559 spin_unlock_irqrestore(&io->lock, flags);
562 __dm_start_io_acct(io);
565 static void dm_end_io_acct(struct dm_io *io)
567 dm_io_acct(io, true);
570 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
573 struct dm_target_io *tio;
576 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
577 /* Set default bdev, but target must bio_set_dev() before issuing IO */
578 clone->bi_bdev = md->disk->part0;
580 tio = clone_to_tio(clone);
582 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
585 io = container_of(tio, struct dm_io, tio);
586 io->magic = DM_IO_MAGIC;
587 io->status = BLK_STS_OK;
589 /* one ref is for submission, the other is for completion */
590 atomic_set(&io->io_count, 2);
591 this_cpu_inc(*md->pending_io);
593 io->split_bio = NULL;
595 spin_lock_init(&io->lock);
596 io->start_time = jiffies;
599 if (static_branch_unlikely(&stats_enabled))
600 dm_stats_record_start(&md->stats, &io->stats_aux);
605 static void free_io(struct dm_io *io)
607 bio_put(&io->tio.clone);
610 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
611 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
613 struct dm_target_io *tio;
616 if (!ci->io->tio.io) {
617 /* the dm_target_io embedded in ci->io is available */
619 /* alloc_io() already initialized embedded clone */
622 struct mapped_device *md = ci->io->md;
624 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
628 /* Set default bdev, but target must bio_set_dev() before issuing IO */
629 clone->bi_bdev = md->disk->part0;
631 /* REQ_DM_POLL_LIST shouldn't be inherited */
632 clone->bi_opf &= ~REQ_DM_POLL_LIST;
634 tio = clone_to_tio(clone);
635 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
638 tio->magic = DM_TIO_MAGIC;
641 tio->target_bio_nr = target_bio_nr;
646 clone->bi_iter.bi_size = to_bytes(*len);
647 if (bio_integrity(clone))
648 bio_integrity_trim(clone);
654 static void free_tio(struct bio *clone)
656 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
662 * Add the bio to the list of deferred io.
664 static void queue_io(struct mapped_device *md, struct bio *bio)
668 spin_lock_irqsave(&md->deferred_lock, flags);
669 bio_list_add(&md->deferred, bio);
670 spin_unlock_irqrestore(&md->deferred_lock, flags);
671 queue_work(md->wq, &md->work);
675 * Everyone (including functions in this file), should use this
676 * function to access the md->map field, and make sure they call
677 * dm_put_live_table() when finished.
679 struct dm_table *dm_get_live_table(struct mapped_device *md,
680 int *srcu_idx) __acquires(md->io_barrier)
682 *srcu_idx = srcu_read_lock(&md->io_barrier);
684 return srcu_dereference(md->map, &md->io_barrier);
687 void dm_put_live_table(struct mapped_device *md,
688 int srcu_idx) __releases(md->io_barrier)
690 srcu_read_unlock(&md->io_barrier, srcu_idx);
693 void dm_sync_table(struct mapped_device *md)
695 synchronize_srcu(&md->io_barrier);
696 synchronize_rcu_expedited();
700 * A fast alternative to dm_get_live_table/dm_put_live_table.
701 * The caller must not block between these two functions.
703 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
706 return rcu_dereference(md->map);
709 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
714 static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
715 int *srcu_idx, unsigned bio_opf)
717 if (bio_opf & REQ_NOWAIT)
718 return dm_get_live_table_fast(md);
720 return dm_get_live_table(md, srcu_idx);
723 static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
726 if (bio_opf & REQ_NOWAIT)
727 dm_put_live_table_fast(md);
729 dm_put_live_table(md, srcu_idx);
732 static char *_dm_claim_ptr = "I belong to device-mapper";
735 * Open a table device so we can use it as a map destination.
737 static int open_table_device(struct table_device *td, dev_t dev,
738 struct mapped_device *md)
740 struct block_device *bdev;
744 BUG_ON(td->dm_dev.bdev);
746 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
748 return PTR_ERR(bdev);
750 r = bd_link_disk_holder(bdev, dm_disk(md));
752 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
756 td->dm_dev.bdev = bdev;
757 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off);
762 * Close a table device that we've been using.
764 static void close_table_device(struct table_device *td, struct mapped_device *md)
766 if (!td->dm_dev.bdev)
769 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
770 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
771 put_dax(td->dm_dev.dax_dev);
772 td->dm_dev.bdev = NULL;
773 td->dm_dev.dax_dev = NULL;
776 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
779 struct table_device *td;
781 list_for_each_entry(td, l, list)
782 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
788 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
789 struct dm_dev **result)
792 struct table_device *td;
794 mutex_lock(&md->table_devices_lock);
795 td = find_table_device(&md->table_devices, dev, mode);
797 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
799 mutex_unlock(&md->table_devices_lock);
803 td->dm_dev.mode = mode;
804 td->dm_dev.bdev = NULL;
806 if ((r = open_table_device(td, dev, md))) {
807 mutex_unlock(&md->table_devices_lock);
812 format_dev_t(td->dm_dev.name, dev);
814 refcount_set(&td->count, 1);
815 list_add(&td->list, &md->table_devices);
817 refcount_inc(&td->count);
819 mutex_unlock(&md->table_devices_lock);
821 *result = &td->dm_dev;
825 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827 struct table_device *td = container_of(d, struct table_device, dm_dev);
829 mutex_lock(&md->table_devices_lock);
830 if (refcount_dec_and_test(&td->count)) {
831 close_table_device(td, md);
835 mutex_unlock(&md->table_devices_lock);
838 static void free_table_devices(struct list_head *devices)
840 struct list_head *tmp, *next;
842 list_for_each_safe(tmp, next, devices) {
843 struct table_device *td = list_entry(tmp, struct table_device, list);
845 DMWARN("dm_destroy: %s still exists with %d references",
846 td->dm_dev.name, refcount_read(&td->count));
852 * Get the geometry associated with a dm device
854 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
862 * Set the geometry of a device.
864 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
866 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
868 if (geo->start > sz) {
869 DMWARN("Start sector is beyond the geometry limits.");
878 static int __noflush_suspending(struct mapped_device *md)
880 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
884 * Return true if the dm_io's original bio is requeued.
885 * io->status is updated with error if requeue disallowed.
887 static bool dm_handle_requeue(struct dm_io *io)
889 struct bio *bio = io->split_bio ? io->split_bio : io->orig_bio;
890 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
891 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
892 (bio->bi_opf & REQ_POLLED));
893 struct mapped_device *md = io->md;
894 bool requeued = false;
896 if (handle_requeue || handle_polled_eagain) {
899 if (bio->bi_opf & REQ_POLLED) {
901 * Upper layer won't help us poll split bio
902 * (io->orig_bio may only reflect a subset of the
903 * pre-split original) so clear REQ_POLLED.
905 bio_clear_polled(bio);
909 * Target requested pushing back the I/O or
910 * polled IO hit BLK_STS_AGAIN.
912 spin_lock_irqsave(&md->deferred_lock, flags);
913 if ((__noflush_suspending(md) &&
914 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
915 handle_polled_eagain) {
916 bio_list_add_head(&md->deferred, bio);
920 * noflush suspend was interrupted or this is
921 * a write to a zoned target.
923 io->status = BLK_STS_IOERR;
925 spin_unlock_irqrestore(&md->deferred_lock, flags);
929 queue_work(md->wq, &md->work);
934 static void dm_io_complete(struct dm_io *io)
936 struct bio *bio = io->split_bio ? io->split_bio : io->orig_bio;
937 struct mapped_device *md = io->md;
938 blk_status_t io_error;
941 requeued = dm_handle_requeue(io);
943 io_error = io->status;
944 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
946 else if (!io_error) {
948 * Must handle target that DM_MAPIO_SUBMITTED only to
949 * then bio_endio() rather than dm_submit_bio_remap()
951 __dm_start_io_acct(io);
956 this_cpu_dec(*md->pending_io);
958 /* nudge anyone waiting on suspend queue */
959 if (unlikely(wq_has_sleeper(&md->wait)))
962 /* Return early if the original bio was requeued */
966 if (bio_is_flush_with_data(bio)) {
968 * Preflush done for flush with data, reissue
969 * without REQ_PREFLUSH.
971 bio->bi_opf &= ~REQ_PREFLUSH;
974 /* done with normal IO or empty flush */
976 bio->bi_status = io_error;
982 * Decrements the number of outstanding ios that a bio has been
983 * cloned into, completing the original io if necc.
985 static inline void __dm_io_dec_pending(struct dm_io *io)
987 if (atomic_dec_and_test(&io->io_count))
991 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
995 /* Push-back supersedes any I/O errors */
996 spin_lock_irqsave(&io->lock, flags);
997 if (!(io->status == BLK_STS_DM_REQUEUE &&
998 __noflush_suspending(io->md))) {
1001 spin_unlock_irqrestore(&io->lock, flags);
1004 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1006 if (unlikely(error))
1007 dm_io_set_error(io, error);
1009 __dm_io_dec_pending(io);
1012 void disable_discard(struct mapped_device *md)
1014 struct queue_limits *limits = dm_get_queue_limits(md);
1016 /* device doesn't really support DISCARD, disable it */
1017 limits->max_discard_sectors = 0;
1020 void disable_write_zeroes(struct mapped_device *md)
1022 struct queue_limits *limits = dm_get_queue_limits(md);
1024 /* device doesn't really support WRITE ZEROES, disable it */
1025 limits->max_write_zeroes_sectors = 0;
1028 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1030 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1033 static void clone_endio(struct bio *bio)
1035 blk_status_t error = bio->bi_status;
1036 struct dm_target_io *tio = clone_to_tio(bio);
1037 struct dm_target *ti = tio->ti;
1038 dm_endio_fn endio = ti->type->end_io;
1039 struct dm_io *io = tio->io;
1040 struct mapped_device *md = io->md;
1042 if (unlikely(error == BLK_STS_TARGET)) {
1043 if (bio_op(bio) == REQ_OP_DISCARD &&
1044 !bdev_max_discard_sectors(bio->bi_bdev))
1045 disable_discard(md);
1046 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1047 !bdev_write_zeroes_sectors(bio->bi_bdev))
1048 disable_write_zeroes(md);
1051 if (static_branch_unlikely(&zoned_enabled) &&
1052 unlikely(blk_queue_is_zoned(bdev_get_queue(bio->bi_bdev))))
1053 dm_zone_endio(io, bio);
1056 int r = endio(ti, bio, &error);
1058 case DM_ENDIO_REQUEUE:
1059 if (static_branch_unlikely(&zoned_enabled)) {
1061 * Requeuing writes to a sequential zone of a zoned
1062 * target will break the sequential write pattern:
1065 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1066 error = BLK_STS_IOERR;
1068 error = BLK_STS_DM_REQUEUE;
1070 error = BLK_STS_DM_REQUEUE;
1074 case DM_ENDIO_INCOMPLETE:
1075 /* The target will handle the io */
1078 DMWARN("unimplemented target endio return value: %d", r);
1083 if (static_branch_unlikely(&swap_bios_enabled) &&
1084 unlikely(swap_bios_limit(ti, bio)))
1085 up(&md->swap_bios_semaphore);
1088 dm_io_dec_pending(io, error);
1092 * Return maximum size of I/O possible at the supplied sector up to the current
1095 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1096 sector_t target_offset)
1098 return ti->len - target_offset;
1101 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1103 sector_t target_offset = dm_target_offset(ti, sector);
1104 sector_t len = max_io_len_target_boundary(ti, target_offset);
1107 * Does the target need to split IO even further?
1108 * - varied (per target) IO splitting is a tenet of DM; this
1109 * explains why stacked chunk_sectors based splitting via
1110 * blk_queue_split() isn't possible here.
1112 if (!ti->max_io_len)
1114 return min_t(sector_t, len,
1115 min(queue_max_sectors(ti->table->md->queue),
1116 blk_chunk_sectors_left(target_offset, ti->max_io_len)));
1119 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1121 if (len > UINT_MAX) {
1122 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1123 (unsigned long long)len, UINT_MAX);
1124 ti->error = "Maximum size of target IO is too large";
1128 ti->max_io_len = (uint32_t) len;
1132 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1134 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1135 sector_t sector, int *srcu_idx)
1136 __acquires(md->io_barrier)
1138 struct dm_table *map;
1139 struct dm_target *ti;
1141 map = dm_get_live_table(md, srcu_idx);
1145 ti = dm_table_find_target(map, sector);
1152 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1153 long nr_pages, enum dax_access_mode mode, void **kaddr,
1156 struct mapped_device *md = dax_get_private(dax_dev);
1157 sector_t sector = pgoff * PAGE_SECTORS;
1158 struct dm_target *ti;
1159 long len, ret = -EIO;
1162 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1166 if (!ti->type->direct_access)
1168 len = max_io_len(ti, sector) / PAGE_SECTORS;
1171 nr_pages = min(len, nr_pages);
1172 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1175 dm_put_live_table(md, srcu_idx);
1180 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1183 struct mapped_device *md = dax_get_private(dax_dev);
1184 sector_t sector = pgoff * PAGE_SECTORS;
1185 struct dm_target *ti;
1189 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1193 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1195 * ->zero_page_range() is mandatory dax operation. If we are
1196 * here, something is wrong.
1200 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1202 dm_put_live_table(md, srcu_idx);
1207 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1208 void *addr, size_t bytes, struct iov_iter *i)
1210 struct mapped_device *md = dax_get_private(dax_dev);
1211 sector_t sector = pgoff * PAGE_SECTORS;
1212 struct dm_target *ti;
1216 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1217 if (!ti || !ti->type->dax_recovery_write)
1220 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1222 dm_put_live_table(md, srcu_idx);
1227 * A target may call dm_accept_partial_bio only from the map routine. It is
1228 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1229 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1230 * __send_duplicate_bios().
1232 * dm_accept_partial_bio informs the dm that the target only wants to process
1233 * additional n_sectors sectors of the bio and the rest of the data should be
1234 * sent in a next bio.
1236 * A diagram that explains the arithmetics:
1237 * +--------------------+---------------+-------+
1239 * +--------------------+---------------+-------+
1241 * <-------------- *tio->len_ptr --------------->
1242 * <----- bio_sectors ----->
1245 * Region 1 was already iterated over with bio_advance or similar function.
1246 * (it may be empty if the target doesn't use bio_advance)
1247 * Region 2 is the remaining bio size that the target wants to process.
1248 * (it may be empty if region 1 is non-empty, although there is no reason
1250 * The target requires that region 3 is to be sent in the next bio.
1252 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1253 * the partially processed part (the sum of regions 1+2) must be the same for all
1254 * copies of the bio.
1256 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1258 struct dm_target_io *tio = clone_to_tio(bio);
1259 unsigned bio_sectors = bio_sectors(bio);
1261 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1262 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1263 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1264 BUG_ON(bio_sectors > *tio->len_ptr);
1265 BUG_ON(n_sectors > bio_sectors);
1267 *tio->len_ptr -= bio_sectors - n_sectors;
1268 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1271 * __split_and_process_bio() may have already saved mapped part
1272 * for accounting but it is being reduced so update accordingly.
1274 dm_io_set_flag(tio->io, DM_IO_WAS_SPLIT);
1275 tio->io->sectors = n_sectors;
1277 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1280 * @clone: clone bio that DM core passed to target's .map function
1281 * @tgt_clone: clone of @clone bio that target needs submitted
1283 * Targets should use this interface to submit bios they take
1284 * ownership of when returning DM_MAPIO_SUBMITTED.
1286 * Target should also enable ti->accounts_remapped_io
1288 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1290 struct dm_target_io *tio = clone_to_tio(clone);
1291 struct dm_io *io = tio->io;
1293 /* establish bio that will get submitted */
1298 * Account io->origin_bio to DM dev on behalf of target
1299 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1301 dm_start_io_acct(io, clone);
1303 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1305 submit_bio_noacct(tgt_clone);
1307 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1309 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1311 mutex_lock(&md->swap_bios_lock);
1312 while (latch < md->swap_bios) {
1314 down(&md->swap_bios_semaphore);
1317 while (latch > md->swap_bios) {
1319 up(&md->swap_bios_semaphore);
1322 mutex_unlock(&md->swap_bios_lock);
1325 static void __map_bio(struct bio *clone)
1327 struct dm_target_io *tio = clone_to_tio(clone);
1328 struct dm_target *ti = tio->ti;
1329 struct dm_io *io = tio->io;
1330 struct mapped_device *md = io->md;
1333 clone->bi_end_io = clone_endio;
1338 tio->old_sector = clone->bi_iter.bi_sector;
1340 if (static_branch_unlikely(&swap_bios_enabled) &&
1341 unlikely(swap_bios_limit(ti, clone))) {
1342 int latch = get_swap_bios();
1343 if (unlikely(latch != md->swap_bios))
1344 __set_swap_bios_limit(md, latch);
1345 down(&md->swap_bios_semaphore);
1348 if (static_branch_unlikely(&zoned_enabled)) {
1350 * Check if the IO needs a special mapping due to zone append
1351 * emulation on zoned target. In this case, dm_zone_map_bio()
1352 * calls the target map operation.
1354 if (unlikely(dm_emulate_zone_append(md)))
1355 r = dm_zone_map_bio(tio);
1357 r = ti->type->map(ti, clone);
1359 r = ti->type->map(ti, clone);
1362 case DM_MAPIO_SUBMITTED:
1363 /* target has assumed ownership of this io */
1364 if (!ti->accounts_remapped_io)
1365 dm_start_io_acct(io, clone);
1367 case DM_MAPIO_REMAPPED:
1368 dm_submit_bio_remap(clone, NULL);
1371 case DM_MAPIO_REQUEUE:
1372 if (static_branch_unlikely(&swap_bios_enabled) &&
1373 unlikely(swap_bios_limit(ti, clone)))
1374 up(&md->swap_bios_semaphore);
1376 if (r == DM_MAPIO_KILL)
1377 dm_io_dec_pending(io, BLK_STS_IOERR);
1379 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1382 DMWARN("unimplemented target map return value: %d", r);
1387 static void setup_split_accounting(struct clone_info *ci, unsigned len)
1389 struct dm_io *io = ci->io;
1391 if (ci->sector_count > len) {
1393 * Split needed, save the mapped part for accounting.
1394 * NOTE: dm_accept_partial_bio() will update accordingly.
1396 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1400 if (static_branch_unlikely(&stats_enabled) &&
1401 unlikely(dm_stats_used(&io->md->stats))) {
1403 * Save bi_sector in terms of its offset from end of
1404 * original bio, only needed for DM-stats' benefit.
1405 * - saved regardless of whether split needed so that
1406 * dm_accept_partial_bio() doesn't need to.
1408 io->sector_offset = bio_end_sector(ci->bio) - ci->sector;
1412 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1413 struct dm_target *ti, unsigned num_bios)
1418 for (try = 0; try < 2; try++) {
1422 mutex_lock(&ci->io->md->table_devices_lock);
1423 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1424 bio = alloc_tio(ci, ti, bio_nr, NULL,
1425 try ? GFP_NOIO : GFP_NOWAIT);
1429 bio_list_add(blist, bio);
1432 mutex_unlock(&ci->io->md->table_devices_lock);
1433 if (bio_nr == num_bios)
1436 while ((bio = bio_list_pop(blist)))
1441 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1442 unsigned num_bios, unsigned *len)
1444 struct bio_list blist = BIO_EMPTY_LIST;
1453 setup_split_accounting(ci, *len);
1454 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1459 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1460 alloc_multiple_bios(&blist, ci, ti, num_bios);
1461 while ((clone = bio_list_pop(&blist))) {
1462 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1472 static void __send_empty_flush(struct clone_info *ci)
1474 unsigned target_nr = 0;
1475 struct dm_target *ti;
1476 struct bio flush_bio;
1479 * Use an on-stack bio for this, it's safe since we don't
1480 * need to reference it after submit. It's just used as
1481 * the basis for the clone(s).
1483 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1484 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1486 ci->bio = &flush_bio;
1487 ci->sector_count = 0;
1488 ci->io->tio.clone.bi_iter.bi_size = 0;
1490 while ((ti = dm_table_get_target(ci->map, target_nr++))) {
1493 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1494 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1495 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1499 * alloc_io() takes one extra reference for submission, so the
1500 * reference won't reach 0 without the following subtraction
1502 atomic_sub(1, &ci->io->io_count);
1504 bio_uninit(ci->bio);
1507 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1513 len = min_t(sector_t, ci->sector_count,
1514 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1516 atomic_add(num_bios, &ci->io->io_count);
1517 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1519 * alloc_io() takes one extra reference for submission, so the
1520 * reference won't reach 0 without the following (+1) subtraction
1522 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1525 ci->sector_count -= len;
1528 static bool is_abnormal_io(struct bio *bio)
1530 unsigned int op = bio_op(bio);
1532 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1534 case REQ_OP_DISCARD:
1535 case REQ_OP_SECURE_ERASE:
1536 case REQ_OP_WRITE_ZEROES:
1546 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1547 struct dm_target *ti)
1549 unsigned num_bios = 0;
1551 switch (bio_op(ci->bio)) {
1552 case REQ_OP_DISCARD:
1553 num_bios = ti->num_discard_bios;
1555 case REQ_OP_SECURE_ERASE:
1556 num_bios = ti->num_secure_erase_bios;
1558 case REQ_OP_WRITE_ZEROES:
1559 num_bios = ti->num_write_zeroes_bios;
1564 * Even though the device advertised support for this type of
1565 * request, that does not mean every target supports it, and
1566 * reconfiguration might also have changed that since the
1567 * check was performed.
1569 if (unlikely(!num_bios))
1570 return BLK_STS_NOTSUPP;
1572 __send_changing_extent_only(ci, ti, num_bios);
1577 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1578 * associated with this bio, and this bio's bi_private needs to be
1579 * stored in dm_io->data before the reuse.
1581 * bio->bi_private is owned by fs or upper layer, so block layer won't
1582 * touch it after splitting. Meantime it won't be changed by anyone after
1583 * bio is submitted. So this reuse is safe.
1585 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1587 return (struct dm_io **)&bio->bi_private;
1590 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1592 struct dm_io **head = dm_poll_list_head(bio);
1594 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1595 bio->bi_opf |= REQ_DM_POLL_LIST;
1597 * Save .bi_private into dm_io, so that we can reuse
1598 * .bi_private as dm_io list head for storing dm_io list
1600 io->data = bio->bi_private;
1602 /* tell block layer to poll for completion */
1603 bio->bi_cookie = ~BLK_QC_T_NONE;
1608 * bio recursed due to split, reuse original poll list,
1609 * and save bio->bi_private too.
1611 io->data = (*head)->data;
1619 * Select the correct strategy for processing a non-flush bio.
1621 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1624 struct dm_target *ti;
1627 ti = dm_table_find_target(ci->map, ci->sector);
1629 return BLK_STS_IOERR;
1631 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1632 unlikely(!dm_target_supports_nowait(ti->type)))
1633 return BLK_STS_NOTSUPP;
1635 if (unlikely(ci->is_abnormal_io))
1636 return __process_abnormal_io(ci, ti);
1639 * Only support bio polling for normal IO, and the target io is
1640 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1642 ci->submit_as_polled = ci->bio->bi_opf & REQ_POLLED;
1644 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1645 setup_split_accounting(ci, len);
1646 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1650 ci->sector_count -= len;
1655 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1656 struct dm_table *map, struct bio *bio, bool is_abnormal)
1659 ci->io = alloc_io(md, bio);
1661 ci->is_abnormal_io = is_abnormal;
1662 ci->submit_as_polled = false;
1663 ci->sector = bio->bi_iter.bi_sector;
1664 ci->sector_count = bio_sectors(bio);
1666 /* Shouldn't happen but sector_count was being set to 0 so... */
1667 if (static_branch_unlikely(&zoned_enabled) &&
1668 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1669 ci->sector_count = 0;
1673 * Entry point to split a bio into clones and submit them to the targets.
1675 static void dm_split_and_process_bio(struct mapped_device *md,
1676 struct dm_table *map, struct bio *bio)
1678 struct clone_info ci;
1680 blk_status_t error = BLK_STS_OK;
1683 is_abnormal = is_abnormal_io(bio);
1684 if (unlikely(is_abnormal)) {
1686 * Use blk_queue_split() for abnormal IO (e.g. discard, etc)
1687 * otherwise associated queue_limits won't be imposed.
1689 blk_queue_split(&bio);
1692 init_clone_info(&ci, md, map, bio, is_abnormal);
1695 if (bio->bi_opf & REQ_PREFLUSH) {
1696 __send_empty_flush(&ci);
1697 /* dm_io_complete submits any data associated with flush */
1701 error = __split_and_process_bio(&ci);
1702 if (error || !ci.sector_count)
1705 * Remainder must be passed to submit_bio_noacct() so it gets handled
1706 * *after* bios already submitted have been completely processed.
1708 WARN_ON_ONCE(!dm_io_flagged(io, DM_IO_WAS_SPLIT));
1709 io->split_bio = bio_split(bio, io->sectors, GFP_NOIO,
1710 &md->queue->bio_split);
1711 bio_chain(io->split_bio, bio);
1712 trace_block_split(io->split_bio, bio->bi_iter.bi_sector);
1713 submit_bio_noacct(bio);
1716 * Drop the extra reference count for non-POLLED bio, and hold one
1717 * reference for POLLED bio, which will be released in dm_poll_bio
1719 * Add every dm_io instance into the dm_io list head which is stored
1720 * in bio->bi_private, so that dm_poll_bio can poll them all.
1722 if (error || !ci.submit_as_polled) {
1724 * In case of submission failure, the extra reference for
1725 * submitting io isn't consumed yet
1728 atomic_dec(&io->io_count);
1729 dm_io_dec_pending(io, error);
1731 dm_queue_poll_io(bio, io);
1734 static void dm_submit_bio(struct bio *bio)
1736 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1738 struct dm_table *map;
1739 unsigned bio_opf = bio->bi_opf;
1741 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1743 /* If suspended, or map not yet available, queue this IO for later */
1744 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1746 if (bio->bi_opf & REQ_NOWAIT)
1747 bio_wouldblock_error(bio);
1748 else if (bio->bi_opf & REQ_RAHEAD)
1755 dm_split_and_process_bio(md, map, bio);
1757 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1760 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1763 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1765 /* don't poll if the mapped io is done */
1766 if (atomic_read(&io->io_count) > 1)
1767 bio_poll(&io->tio.clone, iob, flags);
1769 /* bio_poll holds the last reference */
1770 return atomic_read(&io->io_count) == 1;
1773 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1776 struct dm_io **head = dm_poll_list_head(bio);
1777 struct dm_io *list = *head;
1778 struct dm_io *tmp = NULL;
1779 struct dm_io *curr, *next;
1781 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1782 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1785 WARN_ON_ONCE(!list);
1788 * Restore .bi_private before possibly completing dm_io.
1790 * bio_poll() is only possible once @bio has been completely
1791 * submitted via submit_bio_noacct()'s depth-first submission.
1792 * So there is no dm_queue_poll_io() race associated with
1793 * clearing REQ_DM_POLL_LIST here.
1795 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1796 bio->bi_private = list->data;
1798 for (curr = list, next = curr->next; curr; curr = next, next =
1799 curr ? curr->next : NULL) {
1800 if (dm_poll_dm_io(curr, iob, flags)) {
1802 * clone_endio() has already occurred, so no
1803 * error handling is needed here.
1805 __dm_io_dec_pending(curr);
1814 bio->bi_opf |= REQ_DM_POLL_LIST;
1815 /* Reset bio->bi_private to dm_io list head */
1822 /*-----------------------------------------------------------------
1823 * An IDR is used to keep track of allocated minor numbers.
1824 *---------------------------------------------------------------*/
1825 static void free_minor(int minor)
1827 spin_lock(&_minor_lock);
1828 idr_remove(&_minor_idr, minor);
1829 spin_unlock(&_minor_lock);
1833 * See if the device with a specific minor # is free.
1835 static int specific_minor(int minor)
1839 if (minor >= (1 << MINORBITS))
1842 idr_preload(GFP_KERNEL);
1843 spin_lock(&_minor_lock);
1845 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1847 spin_unlock(&_minor_lock);
1850 return r == -ENOSPC ? -EBUSY : r;
1854 static int next_free_minor(int *minor)
1858 idr_preload(GFP_KERNEL);
1859 spin_lock(&_minor_lock);
1861 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1863 spin_unlock(&_minor_lock);
1871 static const struct block_device_operations dm_blk_dops;
1872 static const struct block_device_operations dm_rq_blk_dops;
1873 static const struct dax_operations dm_dax_ops;
1875 static void dm_wq_work(struct work_struct *work);
1877 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1878 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1880 dm_destroy_crypto_profile(q->crypto_profile);
1883 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1885 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1888 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1890 static void cleanup_mapped_device(struct mapped_device *md)
1893 destroy_workqueue(md->wq);
1894 dm_free_md_mempools(md->mempools);
1897 dax_remove_host(md->disk);
1898 kill_dax(md->dax_dev);
1899 put_dax(md->dax_dev);
1903 dm_cleanup_zoned_dev(md);
1905 spin_lock(&_minor_lock);
1906 md->disk->private_data = NULL;
1907 spin_unlock(&_minor_lock);
1908 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1910 del_gendisk(md->disk);
1912 dm_queue_destroy_crypto_profile(md->queue);
1916 if (md->pending_io) {
1917 free_percpu(md->pending_io);
1918 md->pending_io = NULL;
1921 cleanup_srcu_struct(&md->io_barrier);
1923 mutex_destroy(&md->suspend_lock);
1924 mutex_destroy(&md->type_lock);
1925 mutex_destroy(&md->table_devices_lock);
1926 mutex_destroy(&md->swap_bios_lock);
1928 dm_mq_cleanup_mapped_device(md);
1932 * Allocate and initialise a blank device with a given minor.
1934 static struct mapped_device *alloc_dev(int minor)
1936 int r, numa_node_id = dm_get_numa_node();
1937 struct mapped_device *md;
1940 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1942 DMWARN("unable to allocate device, out of memory.");
1946 if (!try_module_get(THIS_MODULE))
1947 goto bad_module_get;
1949 /* get a minor number for the dev */
1950 if (minor == DM_ANY_MINOR)
1951 r = next_free_minor(&minor);
1953 r = specific_minor(minor);
1957 r = init_srcu_struct(&md->io_barrier);
1959 goto bad_io_barrier;
1961 md->numa_node_id = numa_node_id;
1962 md->init_tio_pdu = false;
1963 md->type = DM_TYPE_NONE;
1964 mutex_init(&md->suspend_lock);
1965 mutex_init(&md->type_lock);
1966 mutex_init(&md->table_devices_lock);
1967 spin_lock_init(&md->deferred_lock);
1968 atomic_set(&md->holders, 1);
1969 atomic_set(&md->open_count, 0);
1970 atomic_set(&md->event_nr, 0);
1971 atomic_set(&md->uevent_seq, 0);
1972 INIT_LIST_HEAD(&md->uevent_list);
1973 INIT_LIST_HEAD(&md->table_devices);
1974 spin_lock_init(&md->uevent_lock);
1977 * default to bio-based until DM table is loaded and md->type
1978 * established. If request-based table is loaded: blk-mq will
1979 * override accordingly.
1981 md->disk = blk_alloc_disk(md->numa_node_id);
1984 md->queue = md->disk->queue;
1986 init_waitqueue_head(&md->wait);
1987 INIT_WORK(&md->work, dm_wq_work);
1988 init_waitqueue_head(&md->eventq);
1989 init_completion(&md->kobj_holder.completion);
1991 md->swap_bios = get_swap_bios();
1992 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1993 mutex_init(&md->swap_bios_lock);
1995 md->disk->major = _major;
1996 md->disk->first_minor = minor;
1997 md->disk->minors = 1;
1998 md->disk->flags |= GENHD_FL_NO_PART;
1999 md->disk->fops = &dm_blk_dops;
2000 md->disk->queue = md->queue;
2001 md->disk->private_data = md;
2002 sprintf(md->disk->disk_name, "dm-%d", minor);
2004 if (IS_ENABLED(CONFIG_FS_DAX)) {
2005 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2006 if (IS_ERR(md->dax_dev)) {
2010 set_dax_nocache(md->dax_dev);
2011 set_dax_nomc(md->dax_dev);
2012 if (dax_add_host(md->dax_dev, md->disk))
2016 format_dev_t(md->name, MKDEV(_major, minor));
2018 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2022 md->pending_io = alloc_percpu(unsigned long);
2023 if (!md->pending_io)
2026 dm_stats_init(&md->stats);
2028 /* Populate the mapping, nobody knows we exist yet */
2029 spin_lock(&_minor_lock);
2030 old_md = idr_replace(&_minor_idr, md, minor);
2031 spin_unlock(&_minor_lock);
2033 BUG_ON(old_md != MINOR_ALLOCED);
2038 cleanup_mapped_device(md);
2042 module_put(THIS_MODULE);
2048 static void unlock_fs(struct mapped_device *md);
2050 static void free_dev(struct mapped_device *md)
2052 int minor = MINOR(disk_devt(md->disk));
2056 cleanup_mapped_device(md);
2058 free_table_devices(&md->table_devices);
2059 dm_stats_cleanup(&md->stats);
2062 module_put(THIS_MODULE);
2067 * Bind a table to the device.
2069 static void event_callback(void *context)
2071 unsigned long flags;
2073 struct mapped_device *md = (struct mapped_device *) context;
2075 spin_lock_irqsave(&md->uevent_lock, flags);
2076 list_splice_init(&md->uevent_list, &uevents);
2077 spin_unlock_irqrestore(&md->uevent_lock, flags);
2079 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2081 atomic_inc(&md->event_nr);
2082 wake_up(&md->eventq);
2083 dm_issue_global_event();
2087 * Returns old map, which caller must destroy.
2089 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2090 struct queue_limits *limits)
2092 struct dm_table *old_map;
2096 lockdep_assert_held(&md->suspend_lock);
2098 size = dm_table_get_size(t);
2101 * Wipe any geometry if the size of the table changed.
2103 if (size != dm_get_size(md))
2104 memset(&md->geometry, 0, sizeof(md->geometry));
2106 if (!get_capacity(md->disk))
2107 set_capacity(md->disk, size);
2109 set_capacity_and_notify(md->disk, size);
2111 dm_table_event_callback(t, event_callback, md);
2113 if (dm_table_request_based(t)) {
2115 * Leverage the fact that request-based DM targets are
2116 * immutable singletons - used to optimize dm_mq_queue_rq.
2118 md->immutable_target = dm_table_get_immutable_target(t);
2121 * There is no need to reload with request-based dm because the
2122 * size of front_pad doesn't change.
2124 * Note for future: If you are to reload bioset, prep-ed
2125 * requests in the queue may refer to bio from the old bioset,
2126 * so you must walk through the queue to unprep.
2128 if (!md->mempools) {
2129 md->mempools = t->mempools;
2134 * The md may already have mempools that need changing.
2135 * If so, reload bioset because front_pad may have changed
2136 * because a different table was loaded.
2138 dm_free_md_mempools(md->mempools);
2139 md->mempools = t->mempools;
2143 ret = dm_table_set_restrictions(t, md->queue, limits);
2145 old_map = ERR_PTR(ret);
2149 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2150 rcu_assign_pointer(md->map, (void *)t);
2151 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2160 * Returns unbound table for the caller to free.
2162 static struct dm_table *__unbind(struct mapped_device *md)
2164 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2169 dm_table_event_callback(map, NULL, NULL);
2170 RCU_INIT_POINTER(md->map, NULL);
2177 * Constructor for a new device.
2179 int dm_create(int minor, struct mapped_device **result)
2181 struct mapped_device *md;
2183 md = alloc_dev(minor);
2187 dm_ima_reset_data(md);
2194 * Functions to manage md->type.
2195 * All are required to hold md->type_lock.
2197 void dm_lock_md_type(struct mapped_device *md)
2199 mutex_lock(&md->type_lock);
2202 void dm_unlock_md_type(struct mapped_device *md)
2204 mutex_unlock(&md->type_lock);
2207 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2209 BUG_ON(!mutex_is_locked(&md->type_lock));
2213 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2218 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2220 return md->immutable_target_type;
2224 * The queue_limits are only valid as long as you have a reference
2227 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2229 BUG_ON(!atomic_read(&md->holders));
2230 return &md->queue->limits;
2232 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2235 * Setup the DM device's queue based on md's type
2237 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2239 enum dm_queue_mode type = dm_table_get_type(t);
2240 struct queue_limits limits;
2244 case DM_TYPE_REQUEST_BASED:
2245 md->disk->fops = &dm_rq_blk_dops;
2246 r = dm_mq_init_request_queue(md, t);
2248 DMERR("Cannot initialize queue for request-based dm mapped device");
2252 case DM_TYPE_BIO_BASED:
2253 case DM_TYPE_DAX_BIO_BASED:
2260 r = dm_calculate_queue_limits(t, &limits);
2262 DMERR("Cannot calculate initial queue limits");
2265 r = dm_table_set_restrictions(t, md->queue, &limits);
2269 r = add_disk(md->disk);
2273 r = dm_sysfs_init(md);
2275 del_gendisk(md->disk);
2282 struct mapped_device *dm_get_md(dev_t dev)
2284 struct mapped_device *md;
2285 unsigned minor = MINOR(dev);
2287 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2290 spin_lock(&_minor_lock);
2292 md = idr_find(&_minor_idr, minor);
2293 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2294 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2300 spin_unlock(&_minor_lock);
2304 EXPORT_SYMBOL_GPL(dm_get_md);
2306 void *dm_get_mdptr(struct mapped_device *md)
2308 return md->interface_ptr;
2311 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2313 md->interface_ptr = ptr;
2316 void dm_get(struct mapped_device *md)
2318 atomic_inc(&md->holders);
2319 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2322 int dm_hold(struct mapped_device *md)
2324 spin_lock(&_minor_lock);
2325 if (test_bit(DMF_FREEING, &md->flags)) {
2326 spin_unlock(&_minor_lock);
2330 spin_unlock(&_minor_lock);
2333 EXPORT_SYMBOL_GPL(dm_hold);
2335 const char *dm_device_name(struct mapped_device *md)
2339 EXPORT_SYMBOL_GPL(dm_device_name);
2341 static void __dm_destroy(struct mapped_device *md, bool wait)
2343 struct dm_table *map;
2348 spin_lock(&_minor_lock);
2349 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2350 set_bit(DMF_FREEING, &md->flags);
2351 spin_unlock(&_minor_lock);
2353 blk_mark_disk_dead(md->disk);
2356 * Take suspend_lock so that presuspend and postsuspend methods
2357 * do not race with internal suspend.
2359 mutex_lock(&md->suspend_lock);
2360 map = dm_get_live_table(md, &srcu_idx);
2361 if (!dm_suspended_md(md)) {
2362 dm_table_presuspend_targets(map);
2363 set_bit(DMF_SUSPENDED, &md->flags);
2364 set_bit(DMF_POST_SUSPENDING, &md->flags);
2365 dm_table_postsuspend_targets(map);
2367 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2368 dm_put_live_table(md, srcu_idx);
2369 mutex_unlock(&md->suspend_lock);
2372 * Rare, but there may be I/O requests still going to complete,
2373 * for example. Wait for all references to disappear.
2374 * No one should increment the reference count of the mapped_device,
2375 * after the mapped_device state becomes DMF_FREEING.
2378 while (atomic_read(&md->holders))
2380 else if (atomic_read(&md->holders))
2381 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2382 dm_device_name(md), atomic_read(&md->holders));
2384 dm_table_destroy(__unbind(md));
2388 void dm_destroy(struct mapped_device *md)
2390 __dm_destroy(md, true);
2393 void dm_destroy_immediate(struct mapped_device *md)
2395 __dm_destroy(md, false);
2398 void dm_put(struct mapped_device *md)
2400 atomic_dec(&md->holders);
2402 EXPORT_SYMBOL_GPL(dm_put);
2404 static bool dm_in_flight_bios(struct mapped_device *md)
2407 unsigned long sum = 0;
2409 for_each_possible_cpu(cpu)
2410 sum += *per_cpu_ptr(md->pending_io, cpu);
2415 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2421 prepare_to_wait(&md->wait, &wait, task_state);
2423 if (!dm_in_flight_bios(md))
2426 if (signal_pending_state(task_state, current)) {
2433 finish_wait(&md->wait, &wait);
2440 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2444 if (!queue_is_mq(md->queue))
2445 return dm_wait_for_bios_completion(md, task_state);
2448 if (!blk_mq_queue_inflight(md->queue))
2451 if (signal_pending_state(task_state, current)) {
2463 * Process the deferred bios
2465 static void dm_wq_work(struct work_struct *work)
2467 struct mapped_device *md = container_of(work, struct mapped_device, work);
2470 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2471 spin_lock_irq(&md->deferred_lock);
2472 bio = bio_list_pop(&md->deferred);
2473 spin_unlock_irq(&md->deferred_lock);
2478 submit_bio_noacct(bio);
2482 static void dm_queue_flush(struct mapped_device *md)
2484 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2485 smp_mb__after_atomic();
2486 queue_work(md->wq, &md->work);
2490 * Swap in a new table, returning the old one for the caller to destroy.
2492 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2494 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2495 struct queue_limits limits;
2498 mutex_lock(&md->suspend_lock);
2500 /* device must be suspended */
2501 if (!dm_suspended_md(md))
2505 * If the new table has no data devices, retain the existing limits.
2506 * This helps multipath with queue_if_no_path if all paths disappear,
2507 * then new I/O is queued based on these limits, and then some paths
2510 if (dm_table_has_no_data_devices(table)) {
2511 live_map = dm_get_live_table_fast(md);
2513 limits = md->queue->limits;
2514 dm_put_live_table_fast(md);
2518 r = dm_calculate_queue_limits(table, &limits);
2525 map = __bind(md, table, &limits);
2526 dm_issue_global_event();
2529 mutex_unlock(&md->suspend_lock);
2534 * Functions to lock and unlock any filesystem running on the
2537 static int lock_fs(struct mapped_device *md)
2541 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2543 r = freeze_bdev(md->disk->part0);
2545 set_bit(DMF_FROZEN, &md->flags);
2549 static void unlock_fs(struct mapped_device *md)
2551 if (!test_bit(DMF_FROZEN, &md->flags))
2553 thaw_bdev(md->disk->part0);
2554 clear_bit(DMF_FROZEN, &md->flags);
2558 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2559 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2560 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2562 * If __dm_suspend returns 0, the device is completely quiescent
2563 * now. There is no request-processing activity. All new requests
2564 * are being added to md->deferred list.
2566 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2567 unsigned suspend_flags, unsigned int task_state,
2568 int dmf_suspended_flag)
2570 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2571 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2574 lockdep_assert_held(&md->suspend_lock);
2577 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2578 * This flag is cleared before dm_suspend returns.
2581 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2583 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2586 * This gets reverted if there's an error later and the targets
2587 * provide the .presuspend_undo hook.
2589 dm_table_presuspend_targets(map);
2592 * Flush I/O to the device.
2593 * Any I/O submitted after lock_fs() may not be flushed.
2594 * noflush takes precedence over do_lockfs.
2595 * (lock_fs() flushes I/Os and waits for them to complete.)
2597 if (!noflush && do_lockfs) {
2600 dm_table_presuspend_undo_targets(map);
2606 * Here we must make sure that no processes are submitting requests
2607 * to target drivers i.e. no one may be executing
2608 * dm_split_and_process_bio from dm_submit_bio.
2610 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2611 * we take the write lock. To prevent any process from reentering
2612 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2613 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2614 * flush_workqueue(md->wq).
2616 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2618 synchronize_srcu(&md->io_barrier);
2621 * Stop md->queue before flushing md->wq in case request-based
2622 * dm defers requests to md->wq from md->queue.
2624 if (dm_request_based(md))
2625 dm_stop_queue(md->queue);
2627 flush_workqueue(md->wq);
2630 * At this point no more requests are entering target request routines.
2631 * We call dm_wait_for_completion to wait for all existing requests
2634 r = dm_wait_for_completion(md, task_state);
2636 set_bit(dmf_suspended_flag, &md->flags);
2639 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2641 synchronize_srcu(&md->io_barrier);
2643 /* were we interrupted ? */
2647 if (dm_request_based(md))
2648 dm_start_queue(md->queue);
2651 dm_table_presuspend_undo_targets(map);
2652 /* pushback list is already flushed, so skip flush */
2659 * We need to be able to change a mapping table under a mounted
2660 * filesystem. For example we might want to move some data in
2661 * the background. Before the table can be swapped with
2662 * dm_bind_table, dm_suspend must be called to flush any in
2663 * flight bios and ensure that any further io gets deferred.
2666 * Suspend mechanism in request-based dm.
2668 * 1. Flush all I/Os by lock_fs() if needed.
2669 * 2. Stop dispatching any I/O by stopping the request_queue.
2670 * 3. Wait for all in-flight I/Os to be completed or requeued.
2672 * To abort suspend, start the request_queue.
2674 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2676 struct dm_table *map = NULL;
2680 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2682 if (dm_suspended_md(md)) {
2687 if (dm_suspended_internally_md(md)) {
2688 /* already internally suspended, wait for internal resume */
2689 mutex_unlock(&md->suspend_lock);
2690 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2696 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2698 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2702 set_bit(DMF_POST_SUSPENDING, &md->flags);
2703 dm_table_postsuspend_targets(map);
2704 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2707 mutex_unlock(&md->suspend_lock);
2711 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2714 int r = dm_table_resume_targets(map);
2722 * Flushing deferred I/Os must be done after targets are resumed
2723 * so that mapping of targets can work correctly.
2724 * Request-based dm is queueing the deferred I/Os in its request_queue.
2726 if (dm_request_based(md))
2727 dm_start_queue(md->queue);
2734 int dm_resume(struct mapped_device *md)
2737 struct dm_table *map = NULL;
2741 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2743 if (!dm_suspended_md(md))
2746 if (dm_suspended_internally_md(md)) {
2747 /* already internally suspended, wait for internal resume */
2748 mutex_unlock(&md->suspend_lock);
2749 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2755 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2756 if (!map || !dm_table_get_size(map))
2759 r = __dm_resume(md, map);
2763 clear_bit(DMF_SUSPENDED, &md->flags);
2765 mutex_unlock(&md->suspend_lock);
2771 * Internal suspend/resume works like userspace-driven suspend. It waits
2772 * until all bios finish and prevents issuing new bios to the target drivers.
2773 * It may be used only from the kernel.
2776 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2778 struct dm_table *map = NULL;
2780 lockdep_assert_held(&md->suspend_lock);
2782 if (md->internal_suspend_count++)
2783 return; /* nested internal suspend */
2785 if (dm_suspended_md(md)) {
2786 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2787 return; /* nest suspend */
2790 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2793 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2794 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2795 * would require changing .presuspend to return an error -- avoid this
2796 * until there is a need for more elaborate variants of internal suspend.
2798 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2799 DMF_SUSPENDED_INTERNALLY);
2801 set_bit(DMF_POST_SUSPENDING, &md->flags);
2802 dm_table_postsuspend_targets(map);
2803 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2806 static void __dm_internal_resume(struct mapped_device *md)
2808 BUG_ON(!md->internal_suspend_count);
2810 if (--md->internal_suspend_count)
2811 return; /* resume from nested internal suspend */
2813 if (dm_suspended_md(md))
2814 goto done; /* resume from nested suspend */
2817 * NOTE: existing callers don't need to call dm_table_resume_targets
2818 * (which may fail -- so best to avoid it for now by passing NULL map)
2820 (void) __dm_resume(md, NULL);
2823 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2824 smp_mb__after_atomic();
2825 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2828 void dm_internal_suspend_noflush(struct mapped_device *md)
2830 mutex_lock(&md->suspend_lock);
2831 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2832 mutex_unlock(&md->suspend_lock);
2834 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2836 void dm_internal_resume(struct mapped_device *md)
2838 mutex_lock(&md->suspend_lock);
2839 __dm_internal_resume(md);
2840 mutex_unlock(&md->suspend_lock);
2842 EXPORT_SYMBOL_GPL(dm_internal_resume);
2845 * Fast variants of internal suspend/resume hold md->suspend_lock,
2846 * which prevents interaction with userspace-driven suspend.
2849 void dm_internal_suspend_fast(struct mapped_device *md)
2851 mutex_lock(&md->suspend_lock);
2852 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2855 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2856 synchronize_srcu(&md->io_barrier);
2857 flush_workqueue(md->wq);
2858 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2860 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2862 void dm_internal_resume_fast(struct mapped_device *md)
2864 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2870 mutex_unlock(&md->suspend_lock);
2872 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2874 /*-----------------------------------------------------------------
2875 * Event notification.
2876 *---------------------------------------------------------------*/
2877 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2882 char udev_cookie[DM_COOKIE_LENGTH];
2883 char *envp[] = { udev_cookie, NULL };
2885 noio_flag = memalloc_noio_save();
2888 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2890 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2891 DM_COOKIE_ENV_VAR_NAME, cookie);
2892 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2896 memalloc_noio_restore(noio_flag);
2901 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2903 return atomic_add_return(1, &md->uevent_seq);
2906 uint32_t dm_get_event_nr(struct mapped_device *md)
2908 return atomic_read(&md->event_nr);
2911 int dm_wait_event(struct mapped_device *md, int event_nr)
2913 return wait_event_interruptible(md->eventq,
2914 (event_nr != atomic_read(&md->event_nr)));
2917 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2919 unsigned long flags;
2921 spin_lock_irqsave(&md->uevent_lock, flags);
2922 list_add(elist, &md->uevent_list);
2923 spin_unlock_irqrestore(&md->uevent_lock, flags);
2927 * The gendisk is only valid as long as you have a reference
2930 struct gendisk *dm_disk(struct mapped_device *md)
2934 EXPORT_SYMBOL_GPL(dm_disk);
2936 struct kobject *dm_kobject(struct mapped_device *md)
2938 return &md->kobj_holder.kobj;
2941 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2943 struct mapped_device *md;
2945 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2947 spin_lock(&_minor_lock);
2948 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2954 spin_unlock(&_minor_lock);
2959 int dm_suspended_md(struct mapped_device *md)
2961 return test_bit(DMF_SUSPENDED, &md->flags);
2964 static int dm_post_suspending_md(struct mapped_device *md)
2966 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2969 int dm_suspended_internally_md(struct mapped_device *md)
2971 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2974 int dm_test_deferred_remove_flag(struct mapped_device *md)
2976 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2979 int dm_suspended(struct dm_target *ti)
2981 return dm_suspended_md(ti->table->md);
2983 EXPORT_SYMBOL_GPL(dm_suspended);
2985 int dm_post_suspending(struct dm_target *ti)
2987 return dm_post_suspending_md(ti->table->md);
2989 EXPORT_SYMBOL_GPL(dm_post_suspending);
2991 int dm_noflush_suspending(struct dm_target *ti)
2993 return __noflush_suspending(ti->table->md);
2995 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2997 void dm_free_md_mempools(struct dm_md_mempools *pools)
3002 bioset_exit(&pools->bs);
3003 bioset_exit(&pools->io_bs);
3015 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3018 struct mapped_device *md = bdev->bd_disk->private_data;
3019 struct dm_table *table;
3020 struct dm_target *ti;
3021 int ret = -ENOTTY, srcu_idx;
3023 table = dm_get_live_table(md, &srcu_idx);
3024 if (!table || !dm_table_get_size(table))
3027 /* We only support devices that have a single target */
3028 if (dm_table_get_num_targets(table) != 1)
3030 ti = dm_table_get_target(table, 0);
3033 if (!ti->type->iterate_devices)
3036 ret = ti->type->iterate_devices(ti, fn, data);
3038 dm_put_live_table(md, srcu_idx);
3043 * For register / unregister we need to manually call out to every path.
3045 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3046 sector_t start, sector_t len, void *data)
3048 struct dm_pr *pr = data;
3049 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3051 if (!ops || !ops->pr_register)
3053 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3056 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3067 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3068 if (ret && new_key) {
3069 /* unregister all paths if we failed to register any path */
3070 pr.old_key = new_key;
3073 pr.fail_early = false;
3074 dm_call_pr(bdev, __dm_pr_register, &pr);
3080 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3083 struct mapped_device *md = bdev->bd_disk->private_data;
3084 const struct pr_ops *ops;
3087 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3091 ops = bdev->bd_disk->fops->pr_ops;
3092 if (ops && ops->pr_reserve)
3093 r = ops->pr_reserve(bdev, key, type, flags);
3097 dm_unprepare_ioctl(md, srcu_idx);
3101 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3103 struct mapped_device *md = bdev->bd_disk->private_data;
3104 const struct pr_ops *ops;
3107 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3111 ops = bdev->bd_disk->fops->pr_ops;
3112 if (ops && ops->pr_release)
3113 r = ops->pr_release(bdev, key, type);
3117 dm_unprepare_ioctl(md, srcu_idx);
3121 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3122 enum pr_type type, bool abort)
3124 struct mapped_device *md = bdev->bd_disk->private_data;
3125 const struct pr_ops *ops;
3128 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3132 ops = bdev->bd_disk->fops->pr_ops;
3133 if (ops && ops->pr_preempt)
3134 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3138 dm_unprepare_ioctl(md, srcu_idx);
3142 static int dm_pr_clear(struct block_device *bdev, u64 key)
3144 struct mapped_device *md = bdev->bd_disk->private_data;
3145 const struct pr_ops *ops;
3148 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3152 ops = bdev->bd_disk->fops->pr_ops;
3153 if (ops && ops->pr_clear)
3154 r = ops->pr_clear(bdev, key);
3158 dm_unprepare_ioctl(md, srcu_idx);
3162 static const struct pr_ops dm_pr_ops = {
3163 .pr_register = dm_pr_register,
3164 .pr_reserve = dm_pr_reserve,
3165 .pr_release = dm_pr_release,
3166 .pr_preempt = dm_pr_preempt,
3167 .pr_clear = dm_pr_clear,
3170 static const struct block_device_operations dm_blk_dops = {
3171 .submit_bio = dm_submit_bio,
3172 .poll_bio = dm_poll_bio,
3173 .open = dm_blk_open,
3174 .release = dm_blk_close,
3175 .ioctl = dm_blk_ioctl,
3176 .getgeo = dm_blk_getgeo,
3177 .report_zones = dm_blk_report_zones,
3178 .pr_ops = &dm_pr_ops,
3179 .owner = THIS_MODULE
3182 static const struct block_device_operations dm_rq_blk_dops = {
3183 .open = dm_blk_open,
3184 .release = dm_blk_close,
3185 .ioctl = dm_blk_ioctl,
3186 .getgeo = dm_blk_getgeo,
3187 .pr_ops = &dm_pr_ops,
3188 .owner = THIS_MODULE
3191 static const struct dax_operations dm_dax_ops = {
3192 .direct_access = dm_dax_direct_access,
3193 .zero_page_range = dm_dax_zero_page_range,
3194 .recovery_write = dm_dax_recovery_write,
3200 module_init(dm_init);
3201 module_exit(dm_exit);
3203 module_param(major, uint, 0);
3204 MODULE_PARM_DESC(major, "The major number of the device mapper");
3206 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3207 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3209 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3210 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3212 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3213 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3215 MODULE_DESCRIPTION(DM_NAME " driver");
3216 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3217 MODULE_LICENSE("GPL");