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 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
92 #define DM_IO_BIO_OFFSET \
93 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
95 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
97 return container_of(clone, struct dm_target_io, clone);
100 void *dm_per_bio_data(struct bio *bio, size_t data_size)
102 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
103 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
104 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
106 EXPORT_SYMBOL_GPL(dm_per_bio_data);
108 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
110 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
111 if (io->magic == DM_IO_MAGIC)
112 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
113 BUG_ON(io->magic != DM_TIO_MAGIC);
114 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
116 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
118 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
120 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
122 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
124 #define MINOR_ALLOCED ((void *)-1)
126 #define DM_NUMA_NODE NUMA_NO_NODE
127 static int dm_numa_node = DM_NUMA_NODE;
129 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
130 static int swap_bios = DEFAULT_SWAP_BIOS;
131 static int get_swap_bios(void)
133 int latch = READ_ONCE(swap_bios);
134 if (unlikely(latch <= 0))
135 latch = DEFAULT_SWAP_BIOS;
139 struct table_device {
140 struct list_head list;
142 struct dm_dev dm_dev;
146 * Bio-based DM's mempools' reserved IOs set by the user.
148 #define RESERVED_BIO_BASED_IOS 16
149 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
151 static int __dm_get_module_param_int(int *module_param, int min, int max)
153 int param = READ_ONCE(*module_param);
154 int modified_param = 0;
155 bool modified = true;
158 modified_param = min;
159 else if (param > max)
160 modified_param = max;
165 (void)cmpxchg(module_param, param, modified_param);
166 param = modified_param;
172 unsigned __dm_get_module_param(unsigned *module_param,
173 unsigned def, unsigned max)
175 unsigned param = READ_ONCE(*module_param);
176 unsigned modified_param = 0;
179 modified_param = def;
180 else if (param > max)
181 modified_param = max;
183 if (modified_param) {
184 (void)cmpxchg(module_param, param, modified_param);
185 param = modified_param;
191 unsigned dm_get_reserved_bio_based_ios(void)
193 return __dm_get_module_param(&reserved_bio_based_ios,
194 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
196 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
198 static unsigned dm_get_numa_node(void)
200 return __dm_get_module_param_int(&dm_numa_node,
201 DM_NUMA_NODE, num_online_nodes() - 1);
204 static int __init local_init(void)
208 r = dm_uevent_init();
212 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
213 if (!deferred_remove_workqueue) {
215 goto out_uevent_exit;
219 r = register_blkdev(_major, _name);
221 goto out_free_workqueue;
229 destroy_workqueue(deferred_remove_workqueue);
236 static void local_exit(void)
238 flush_scheduled_work();
239 destroy_workqueue(deferred_remove_workqueue);
241 unregister_blkdev(_major, _name);
246 DMINFO("cleaned up");
249 static int (*_inits[])(void) __initdata = {
260 static void (*_exits[])(void) = {
271 static int __init dm_init(void)
273 const int count = ARRAY_SIZE(_inits);
276 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
277 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
278 " Duplicate IMA measurements will not be recorded in the IMA log.");
281 for (i = 0; i < count; i++) {
295 static void __exit dm_exit(void)
297 int i = ARRAY_SIZE(_exits);
303 * Should be empty by this point.
305 idr_destroy(&_minor_idr);
309 * Block device functions
311 int dm_deleting_md(struct mapped_device *md)
313 return test_bit(DMF_DELETING, &md->flags);
316 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
318 struct mapped_device *md;
320 spin_lock(&_minor_lock);
322 md = bdev->bd_disk->private_data;
326 if (test_bit(DMF_FREEING, &md->flags) ||
327 dm_deleting_md(md)) {
333 atomic_inc(&md->open_count);
335 spin_unlock(&_minor_lock);
337 return md ? 0 : -ENXIO;
340 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
342 struct mapped_device *md;
344 spin_lock(&_minor_lock);
346 md = disk->private_data;
350 if (atomic_dec_and_test(&md->open_count) &&
351 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
352 queue_work(deferred_remove_workqueue, &deferred_remove_work);
356 spin_unlock(&_minor_lock);
359 int dm_open_count(struct mapped_device *md)
361 return atomic_read(&md->open_count);
365 * Guarantees nothing is using the device before it's deleted.
367 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
371 spin_lock(&_minor_lock);
373 if (dm_open_count(md)) {
376 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
377 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
380 set_bit(DMF_DELETING, &md->flags);
382 spin_unlock(&_minor_lock);
387 int dm_cancel_deferred_remove(struct mapped_device *md)
391 spin_lock(&_minor_lock);
393 if (test_bit(DMF_DELETING, &md->flags))
396 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
398 spin_unlock(&_minor_lock);
403 static void do_deferred_remove(struct work_struct *w)
405 dm_deferred_remove();
408 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
410 struct mapped_device *md = bdev->bd_disk->private_data;
412 return dm_get_geometry(md, geo);
415 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
416 struct block_device **bdev)
418 struct dm_target *tgt;
419 struct dm_table *map;
424 map = dm_get_live_table(md, srcu_idx);
425 if (!map || !dm_table_get_size(map))
428 /* We only support devices that have a single target */
429 if (dm_table_get_num_targets(map) != 1)
432 tgt = dm_table_get_target(map, 0);
433 if (!tgt->type->prepare_ioctl)
436 if (dm_suspended_md(md))
439 r = tgt->type->prepare_ioctl(tgt, bdev);
440 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
441 dm_put_live_table(md, *srcu_idx);
449 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
451 dm_put_live_table(md, srcu_idx);
454 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
455 unsigned int cmd, unsigned long arg)
457 struct mapped_device *md = bdev->bd_disk->private_data;
460 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
466 * Target determined this ioctl is being issued against a
467 * subset of the parent bdev; require extra privileges.
469 if (!capable(CAP_SYS_RAWIO)) {
471 "%s: sending ioctl %x to DM device without required privilege.",
478 if (!bdev->bd_disk->fops->ioctl)
481 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
483 dm_unprepare_ioctl(md, srcu_idx);
487 u64 dm_start_time_ns_from_clone(struct bio *bio)
489 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
491 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
493 static bool bio_is_flush_with_data(struct bio *bio)
495 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
498 static void dm_io_acct(struct dm_io *io, bool end)
500 struct dm_stats_aux *stats_aux = &io->stats_aux;
501 unsigned long start_time = io->start_time;
502 struct mapped_device *md = io->md;
503 struct bio *bio = io->orig_bio;
504 unsigned int sectors;
507 * If REQ_PREFLUSH set, don't account payload, it will be
508 * submitted (and accounted) after this flush completes.
510 if (bio_is_flush_with_data(bio))
512 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
513 sectors = bio_sectors(bio);
515 sectors = io->sectors;
518 bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
521 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
523 if (static_branch_unlikely(&stats_enabled) &&
524 unlikely(dm_stats_used(&md->stats))) {
527 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
528 sector = bio->bi_iter.bi_sector;
530 sector = bio_end_sector(bio) - io->sector_offset;
532 dm_stats_account_io(&md->stats, bio_data_dir(bio),
534 end, start_time, stats_aux);
538 static void __dm_start_io_acct(struct dm_io *io)
540 dm_io_acct(io, false);
543 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
546 * Ensure IO accounting is only ever started once.
548 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
551 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
552 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
553 dm_io_set_flag(io, DM_IO_ACCOUNTED);
556 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
557 spin_lock_irqsave(&io->lock, flags);
558 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
559 spin_unlock_irqrestore(&io->lock, flags);
562 dm_io_set_flag(io, DM_IO_ACCOUNTED);
563 spin_unlock_irqrestore(&io->lock, flags);
566 __dm_start_io_acct(io);
569 static void dm_end_io_acct(struct dm_io *io)
571 dm_io_acct(io, true);
574 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
577 struct dm_target_io *tio;
580 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
581 /* Set default bdev, but target must bio_set_dev() before issuing IO */
582 clone->bi_bdev = md->disk->part0;
584 tio = clone_to_tio(clone);
586 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
589 io = container_of(tio, struct dm_io, tio);
590 io->magic = DM_IO_MAGIC;
591 io->status = BLK_STS_OK;
593 /* one ref is for submission, the other is for completion */
594 atomic_set(&io->io_count, 2);
595 this_cpu_inc(*md->pending_io);
598 spin_lock_init(&io->lock);
599 io->start_time = jiffies;
602 if (static_branch_unlikely(&stats_enabled))
603 dm_stats_record_start(&md->stats, &io->stats_aux);
608 static void free_io(struct dm_io *io)
610 bio_put(&io->tio.clone);
613 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
614 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
616 struct dm_target_io *tio;
619 if (!ci->io->tio.io) {
620 /* the dm_target_io embedded in ci->io is available */
622 /* alloc_io() already initialized embedded clone */
625 struct mapped_device *md = ci->io->md;
627 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
631 /* Set default bdev, but target must bio_set_dev() before issuing IO */
632 clone->bi_bdev = md->disk->part0;
634 /* REQ_DM_POLL_LIST shouldn't be inherited */
635 clone->bi_opf &= ~REQ_DM_POLL_LIST;
637 tio = clone_to_tio(clone);
638 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
641 tio->magic = DM_TIO_MAGIC;
644 tio->target_bio_nr = target_bio_nr;
649 clone->bi_iter.bi_size = to_bytes(*len);
650 if (bio_integrity(clone))
651 bio_integrity_trim(clone);
657 static void free_tio(struct bio *clone)
659 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
665 * Add the bio to the list of deferred io.
667 static void queue_io(struct mapped_device *md, struct bio *bio)
671 spin_lock_irqsave(&md->deferred_lock, flags);
672 bio_list_add(&md->deferred, bio);
673 spin_unlock_irqrestore(&md->deferred_lock, flags);
674 queue_work(md->wq, &md->work);
678 * Everyone (including functions in this file), should use this
679 * function to access the md->map field, and make sure they call
680 * dm_put_live_table() when finished.
682 struct dm_table *dm_get_live_table(struct mapped_device *md,
683 int *srcu_idx) __acquires(md->io_barrier)
685 *srcu_idx = srcu_read_lock(&md->io_barrier);
687 return srcu_dereference(md->map, &md->io_barrier);
690 void dm_put_live_table(struct mapped_device *md,
691 int srcu_idx) __releases(md->io_barrier)
693 srcu_read_unlock(&md->io_barrier, srcu_idx);
696 void dm_sync_table(struct mapped_device *md)
698 synchronize_srcu(&md->io_barrier);
699 synchronize_rcu_expedited();
703 * A fast alternative to dm_get_live_table/dm_put_live_table.
704 * The caller must not block between these two functions.
706 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
709 return rcu_dereference(md->map);
712 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
717 static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
718 int *srcu_idx, unsigned bio_opf)
720 if (bio_opf & REQ_NOWAIT)
721 return dm_get_live_table_fast(md);
723 return dm_get_live_table(md, srcu_idx);
726 static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
729 if (bio_opf & REQ_NOWAIT)
730 dm_put_live_table_fast(md);
732 dm_put_live_table(md, srcu_idx);
735 static char *_dm_claim_ptr = "I belong to device-mapper";
738 * Open a table device so we can use it as a map destination.
740 static int open_table_device(struct table_device *td, dev_t dev,
741 struct mapped_device *md)
743 struct block_device *bdev;
747 BUG_ON(td->dm_dev.bdev);
749 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
751 return PTR_ERR(bdev);
753 r = bd_link_disk_holder(bdev, dm_disk(md));
755 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
759 td->dm_dev.bdev = bdev;
760 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off);
765 * Close a table device that we've been using.
767 static void close_table_device(struct table_device *td, struct mapped_device *md)
769 if (!td->dm_dev.bdev)
772 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
773 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
774 put_dax(td->dm_dev.dax_dev);
775 td->dm_dev.bdev = NULL;
776 td->dm_dev.dax_dev = NULL;
779 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
782 struct table_device *td;
784 list_for_each_entry(td, l, list)
785 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
791 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
792 struct dm_dev **result)
795 struct table_device *td;
797 mutex_lock(&md->table_devices_lock);
798 td = find_table_device(&md->table_devices, dev, mode);
800 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
802 mutex_unlock(&md->table_devices_lock);
806 td->dm_dev.mode = mode;
807 td->dm_dev.bdev = NULL;
809 if ((r = open_table_device(td, dev, md))) {
810 mutex_unlock(&md->table_devices_lock);
815 format_dev_t(td->dm_dev.name, dev);
817 refcount_set(&td->count, 1);
818 list_add(&td->list, &md->table_devices);
820 refcount_inc(&td->count);
822 mutex_unlock(&md->table_devices_lock);
824 *result = &td->dm_dev;
828 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
830 struct table_device *td = container_of(d, struct table_device, dm_dev);
832 mutex_lock(&md->table_devices_lock);
833 if (refcount_dec_and_test(&td->count)) {
834 close_table_device(td, md);
838 mutex_unlock(&md->table_devices_lock);
841 static void free_table_devices(struct list_head *devices)
843 struct list_head *tmp, *next;
845 list_for_each_safe(tmp, next, devices) {
846 struct table_device *td = list_entry(tmp, struct table_device, list);
848 DMWARN("dm_destroy: %s still exists with %d references",
849 td->dm_dev.name, refcount_read(&td->count));
855 * Get the geometry associated with a dm device
857 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
865 * Set the geometry of a device.
867 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
869 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
871 if (geo->start > sz) {
872 DMWARN("Start sector is beyond the geometry limits.");
881 static int __noflush_suspending(struct mapped_device *md)
883 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
886 static void dm_io_complete(struct dm_io *io)
888 blk_status_t io_error;
889 struct mapped_device *md = io->md;
890 struct bio *bio = io->orig_bio;
892 if (io->status == BLK_STS_DM_REQUEUE) {
895 * Target requested pushing back the I/O.
897 spin_lock_irqsave(&md->deferred_lock, flags);
898 if (__noflush_suspending(md) &&
899 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
900 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
901 bio_list_add_head(&md->deferred, bio);
904 * noflush suspend was interrupted or this is
905 * a write to a zoned target.
907 io->status = BLK_STS_IOERR;
909 spin_unlock_irqrestore(&md->deferred_lock, flags);
912 io_error = io->status;
913 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
915 else if (!io_error) {
917 * Must handle target that DM_MAPIO_SUBMITTED only to
918 * then bio_endio() rather than dm_submit_bio_remap()
920 __dm_start_io_acct(io);
925 this_cpu_dec(*md->pending_io);
927 /* nudge anyone waiting on suspend queue */
928 if (unlikely(wq_has_sleeper(&md->wait)))
931 if (io_error == BLK_STS_DM_REQUEUE || io_error == BLK_STS_AGAIN) {
932 if (bio->bi_opf & REQ_POLLED) {
934 * Upper layer won't help us poll split bio (io->orig_bio
935 * may only reflect a subset of the pre-split original)
936 * so clear REQ_POLLED in case of requeue.
938 bio_clear_polled(bio);
939 if (io_error == BLK_STS_AGAIN) {
940 /* io_uring doesn't handle BLK_STS_AGAIN (yet) */
945 if (io_error == BLK_STS_DM_REQUEUE)
949 if (bio_is_flush_with_data(bio)) {
951 * Preflush done for flush with data, reissue
952 * without REQ_PREFLUSH.
954 bio->bi_opf &= ~REQ_PREFLUSH;
957 /* done with normal IO or empty flush */
959 bio->bi_status = io_error;
965 * Decrements the number of outstanding ios that a bio has been
966 * cloned into, completing the original io if necc.
968 static inline void __dm_io_dec_pending(struct dm_io *io)
970 if (atomic_dec_and_test(&io->io_count))
974 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
978 /* Push-back supersedes any I/O errors */
979 spin_lock_irqsave(&io->lock, flags);
980 if (!(io->status == BLK_STS_DM_REQUEUE &&
981 __noflush_suspending(io->md))) {
984 spin_unlock_irqrestore(&io->lock, flags);
987 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
990 dm_io_set_error(io, error);
992 __dm_io_dec_pending(io);
995 void disable_discard(struct mapped_device *md)
997 struct queue_limits *limits = dm_get_queue_limits(md);
999 /* device doesn't really support DISCARD, disable it */
1000 limits->max_discard_sectors = 0;
1003 void disable_write_zeroes(struct mapped_device *md)
1005 struct queue_limits *limits = dm_get_queue_limits(md);
1007 /* device doesn't really support WRITE ZEROES, disable it */
1008 limits->max_write_zeroes_sectors = 0;
1011 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1013 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1016 static void clone_endio(struct bio *bio)
1018 blk_status_t error = bio->bi_status;
1019 struct dm_target_io *tio = clone_to_tio(bio);
1020 struct dm_target *ti = tio->ti;
1021 dm_endio_fn endio = ti->type->end_io;
1022 struct dm_io *io = tio->io;
1023 struct mapped_device *md = io->md;
1025 if (unlikely(error == BLK_STS_TARGET)) {
1026 if (bio_op(bio) == REQ_OP_DISCARD &&
1027 !bdev_max_discard_sectors(bio->bi_bdev))
1028 disable_discard(md);
1029 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1030 !bdev_write_zeroes_sectors(bio->bi_bdev))
1031 disable_write_zeroes(md);
1034 if (static_branch_unlikely(&zoned_enabled) &&
1035 unlikely(blk_queue_is_zoned(bdev_get_queue(bio->bi_bdev))))
1036 dm_zone_endio(io, bio);
1039 int r = endio(ti, bio, &error);
1041 case DM_ENDIO_REQUEUE:
1042 if (static_branch_unlikely(&zoned_enabled)) {
1044 * Requeuing writes to a sequential zone of a zoned
1045 * target will break the sequential write pattern:
1048 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1049 error = BLK_STS_IOERR;
1051 error = BLK_STS_DM_REQUEUE;
1053 error = BLK_STS_DM_REQUEUE;
1057 case DM_ENDIO_INCOMPLETE:
1058 /* The target will handle the io */
1061 DMWARN("unimplemented target endio return value: %d", r);
1066 if (static_branch_unlikely(&swap_bios_enabled) &&
1067 unlikely(swap_bios_limit(ti, bio)))
1068 up(&md->swap_bios_semaphore);
1071 dm_io_dec_pending(io, error);
1075 * Return maximum size of I/O possible at the supplied sector up to the current
1078 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1079 sector_t target_offset)
1081 return ti->len - target_offset;
1084 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1086 sector_t target_offset = dm_target_offset(ti, sector);
1087 sector_t len = max_io_len_target_boundary(ti, target_offset);
1091 * Does the target need to split IO even further?
1092 * - varied (per target) IO splitting is a tenet of DM; this
1093 * explains why stacked chunk_sectors based splitting via
1094 * blk_max_size_offset() isn't possible here. So pass in
1095 * ti->max_io_len to override stacked chunk_sectors.
1097 if (ti->max_io_len) {
1098 max_len = blk_max_size_offset(ti->table->md->queue,
1099 target_offset, ti->max_io_len);
1107 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1109 if (len > UINT_MAX) {
1110 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1111 (unsigned long long)len, UINT_MAX);
1112 ti->error = "Maximum size of target IO is too large";
1116 ti->max_io_len = (uint32_t) len;
1120 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1122 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1123 sector_t sector, int *srcu_idx)
1124 __acquires(md->io_barrier)
1126 struct dm_table *map;
1127 struct dm_target *ti;
1129 map = dm_get_live_table(md, srcu_idx);
1133 ti = dm_table_find_target(map, sector);
1140 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1141 long nr_pages, enum dax_access_mode mode, void **kaddr,
1144 struct mapped_device *md = dax_get_private(dax_dev);
1145 sector_t sector = pgoff * PAGE_SECTORS;
1146 struct dm_target *ti;
1147 long len, ret = -EIO;
1150 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1154 if (!ti->type->direct_access)
1156 len = max_io_len(ti, sector) / PAGE_SECTORS;
1159 nr_pages = min(len, nr_pages);
1160 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1163 dm_put_live_table(md, srcu_idx);
1168 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1171 struct mapped_device *md = dax_get_private(dax_dev);
1172 sector_t sector = pgoff * PAGE_SECTORS;
1173 struct dm_target *ti;
1177 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1181 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1183 * ->zero_page_range() is mandatory dax operation. If we are
1184 * here, something is wrong.
1188 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1190 dm_put_live_table(md, srcu_idx);
1195 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1196 void *addr, size_t bytes, struct iov_iter *i)
1198 struct mapped_device *md = dax_get_private(dax_dev);
1199 sector_t sector = pgoff * PAGE_SECTORS;
1200 struct dm_target *ti;
1204 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1205 if (!ti || !ti->type->dax_recovery_write)
1208 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1210 dm_put_live_table(md, srcu_idx);
1215 * A target may call dm_accept_partial_bio only from the map routine. It is
1216 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1217 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1218 * __send_duplicate_bios().
1220 * dm_accept_partial_bio informs the dm that the target only wants to process
1221 * additional n_sectors sectors of the bio and the rest of the data should be
1222 * sent in a next bio.
1224 * A diagram that explains the arithmetics:
1225 * +--------------------+---------------+-------+
1227 * +--------------------+---------------+-------+
1229 * <-------------- *tio->len_ptr --------------->
1230 * <----- bio_sectors ----->
1233 * Region 1 was already iterated over with bio_advance or similar function.
1234 * (it may be empty if the target doesn't use bio_advance)
1235 * Region 2 is the remaining bio size that the target wants to process.
1236 * (it may be empty if region 1 is non-empty, although there is no reason
1238 * The target requires that region 3 is to be sent in the next bio.
1240 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1241 * the partially processed part (the sum of regions 1+2) must be the same for all
1242 * copies of the bio.
1244 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1246 struct dm_target_io *tio = clone_to_tio(bio);
1247 unsigned bio_sectors = bio_sectors(bio);
1249 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1250 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1251 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1252 BUG_ON(bio_sectors > *tio->len_ptr);
1253 BUG_ON(n_sectors > bio_sectors);
1255 *tio->len_ptr -= bio_sectors - n_sectors;
1256 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1259 * __split_and_process_bio() may have already saved mapped part
1260 * for accounting but it is being reduced so update accordingly.
1262 dm_io_set_flag(tio->io, DM_IO_WAS_SPLIT);
1263 tio->io->sectors = n_sectors;
1265 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1268 * @clone: clone bio that DM core passed to target's .map function
1269 * @tgt_clone: clone of @clone bio that target needs submitted
1271 * Targets should use this interface to submit bios they take
1272 * ownership of when returning DM_MAPIO_SUBMITTED.
1274 * Target should also enable ti->accounts_remapped_io
1276 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1278 struct dm_target_io *tio = clone_to_tio(clone);
1279 struct dm_io *io = tio->io;
1281 /* establish bio that will get submitted */
1286 * Account io->origin_bio to DM dev on behalf of target
1287 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1289 dm_start_io_acct(io, clone);
1291 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1293 submit_bio_noacct(tgt_clone);
1295 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1297 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1299 mutex_lock(&md->swap_bios_lock);
1300 while (latch < md->swap_bios) {
1302 down(&md->swap_bios_semaphore);
1305 while (latch > md->swap_bios) {
1307 up(&md->swap_bios_semaphore);
1310 mutex_unlock(&md->swap_bios_lock);
1313 static void __map_bio(struct bio *clone)
1315 struct dm_target_io *tio = clone_to_tio(clone);
1316 struct dm_target *ti = tio->ti;
1317 struct dm_io *io = tio->io;
1318 struct mapped_device *md = io->md;
1321 clone->bi_end_io = clone_endio;
1326 tio->old_sector = clone->bi_iter.bi_sector;
1328 if (static_branch_unlikely(&swap_bios_enabled) &&
1329 unlikely(swap_bios_limit(ti, clone))) {
1330 int latch = get_swap_bios();
1331 if (unlikely(latch != md->swap_bios))
1332 __set_swap_bios_limit(md, latch);
1333 down(&md->swap_bios_semaphore);
1336 if (static_branch_unlikely(&zoned_enabled)) {
1338 * Check if the IO needs a special mapping due to zone append
1339 * emulation on zoned target. In this case, dm_zone_map_bio()
1340 * calls the target map operation.
1342 if (unlikely(dm_emulate_zone_append(md)))
1343 r = dm_zone_map_bio(tio);
1345 r = ti->type->map(ti, clone);
1347 r = ti->type->map(ti, clone);
1350 case DM_MAPIO_SUBMITTED:
1351 /* target has assumed ownership of this io */
1352 if (!ti->accounts_remapped_io)
1353 dm_start_io_acct(io, clone);
1355 case DM_MAPIO_REMAPPED:
1356 dm_submit_bio_remap(clone, NULL);
1359 case DM_MAPIO_REQUEUE:
1360 if (static_branch_unlikely(&swap_bios_enabled) &&
1361 unlikely(swap_bios_limit(ti, clone)))
1362 up(&md->swap_bios_semaphore);
1364 if (r == DM_MAPIO_KILL)
1365 dm_io_dec_pending(io, BLK_STS_IOERR);
1367 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1370 DMWARN("unimplemented target map return value: %d", r);
1375 static void setup_split_accounting(struct clone_info *ci, unsigned len)
1377 struct dm_io *io = ci->io;
1379 if (ci->sector_count > len) {
1381 * Split needed, save the mapped part for accounting.
1382 * NOTE: dm_accept_partial_bio() will update accordingly.
1384 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1388 if (static_branch_unlikely(&stats_enabled) &&
1389 unlikely(dm_stats_used(&io->md->stats))) {
1391 * Save bi_sector in terms of its offset from end of
1392 * original bio, only needed for DM-stats' benefit.
1393 * - saved regardless of whether split needed so that
1394 * dm_accept_partial_bio() doesn't need to.
1396 io->sector_offset = bio_end_sector(ci->bio) - ci->sector;
1400 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1401 struct dm_target *ti, unsigned num_bios)
1406 for (try = 0; try < 2; try++) {
1410 mutex_lock(&ci->io->md->table_devices_lock);
1411 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1412 bio = alloc_tio(ci, ti, bio_nr, NULL,
1413 try ? GFP_NOIO : GFP_NOWAIT);
1417 bio_list_add(blist, bio);
1420 mutex_unlock(&ci->io->md->table_devices_lock);
1421 if (bio_nr == num_bios)
1424 while ((bio = bio_list_pop(blist)))
1429 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1430 unsigned num_bios, unsigned *len)
1432 struct bio_list blist = BIO_EMPTY_LIST;
1441 setup_split_accounting(ci, *len);
1442 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1447 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1448 alloc_multiple_bios(&blist, ci, ti, num_bios);
1449 while ((clone = bio_list_pop(&blist))) {
1450 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1460 static void __send_empty_flush(struct clone_info *ci)
1462 unsigned target_nr = 0;
1463 struct dm_target *ti;
1464 struct bio flush_bio;
1467 * Use an on-stack bio for this, it's safe since we don't
1468 * need to reference it after submit. It's just used as
1469 * the basis for the clone(s).
1471 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1472 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1474 ci->bio = &flush_bio;
1475 ci->sector_count = 0;
1476 ci->io->tio.clone.bi_iter.bi_size = 0;
1478 while ((ti = dm_table_get_target(ci->map, target_nr++))) {
1481 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1482 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1483 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1487 * alloc_io() takes one extra reference for submission, so the
1488 * reference won't reach 0 without the following subtraction
1490 atomic_sub(1, &ci->io->io_count);
1492 bio_uninit(ci->bio);
1495 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1501 len = min_t(sector_t, ci->sector_count,
1502 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1504 atomic_add(num_bios, &ci->io->io_count);
1505 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1507 * alloc_io() takes one extra reference for submission, so the
1508 * reference won't reach 0 without the following (+1) subtraction
1510 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1513 ci->sector_count -= len;
1516 static bool is_abnormal_io(struct bio *bio)
1518 unsigned int op = bio_op(bio);
1520 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1522 case REQ_OP_DISCARD:
1523 case REQ_OP_SECURE_ERASE:
1524 case REQ_OP_WRITE_ZEROES:
1534 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1535 struct dm_target *ti)
1537 unsigned num_bios = 0;
1539 switch (bio_op(ci->bio)) {
1540 case REQ_OP_DISCARD:
1541 num_bios = ti->num_discard_bios;
1543 case REQ_OP_SECURE_ERASE:
1544 num_bios = ti->num_secure_erase_bios;
1546 case REQ_OP_WRITE_ZEROES:
1547 num_bios = ti->num_write_zeroes_bios;
1552 * Even though the device advertised support for this type of
1553 * request, that does not mean every target supports it, and
1554 * reconfiguration might also have changed that since the
1555 * check was performed.
1557 if (unlikely(!num_bios))
1558 return BLK_STS_NOTSUPP;
1560 __send_changing_extent_only(ci, ti, num_bios);
1565 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1566 * associated with this bio, and this bio's bi_private needs to be
1567 * stored in dm_io->data before the reuse.
1569 * bio->bi_private is owned by fs or upper layer, so block layer won't
1570 * touch it after splitting. Meantime it won't be changed by anyone after
1571 * bio is submitted. So this reuse is safe.
1573 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1575 return (struct dm_io **)&bio->bi_private;
1578 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1580 struct dm_io **head = dm_poll_list_head(bio);
1582 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1583 bio->bi_opf |= REQ_DM_POLL_LIST;
1585 * Save .bi_private into dm_io, so that we can reuse
1586 * .bi_private as dm_io list head for storing dm_io list
1588 io->data = bio->bi_private;
1590 /* tell block layer to poll for completion */
1591 bio->bi_cookie = ~BLK_QC_T_NONE;
1596 * bio recursed due to split, reuse original poll list,
1597 * and save bio->bi_private too.
1599 io->data = (*head)->data;
1607 * Select the correct strategy for processing a non-flush bio.
1609 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1612 struct dm_target *ti;
1615 ti = dm_table_find_target(ci->map, ci->sector);
1617 return BLK_STS_IOERR;
1619 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1620 unlikely(!dm_target_supports_nowait(ti->type)))
1621 return BLK_STS_NOTSUPP;
1623 if (unlikely(ci->is_abnormal_io))
1624 return __process_abnormal_io(ci, ti);
1627 * Only support bio polling for normal IO, and the target io is
1628 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1630 ci->submit_as_polled = ci->bio->bi_opf & REQ_POLLED;
1632 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1633 setup_split_accounting(ci, len);
1634 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1638 ci->sector_count -= len;
1643 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1644 struct dm_table *map, struct bio *bio, bool is_abnormal)
1647 ci->io = alloc_io(md, bio);
1649 ci->is_abnormal_io = is_abnormal;
1650 ci->submit_as_polled = false;
1651 ci->sector = bio->bi_iter.bi_sector;
1652 ci->sector_count = bio_sectors(bio);
1654 /* Shouldn't happen but sector_count was being set to 0 so... */
1655 if (static_branch_unlikely(&zoned_enabled) &&
1656 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1657 ci->sector_count = 0;
1661 * Entry point to split a bio into clones and submit them to the targets.
1663 static void dm_split_and_process_bio(struct mapped_device *md,
1664 struct dm_table *map, struct bio *bio)
1666 struct clone_info ci;
1668 blk_status_t error = BLK_STS_OK;
1671 is_abnormal = is_abnormal_io(bio);
1672 if (unlikely(is_abnormal)) {
1674 * Use blk_queue_split() for abnormal IO (e.g. discard, etc)
1675 * otherwise associated queue_limits won't be imposed.
1677 blk_queue_split(&bio);
1680 init_clone_info(&ci, md, map, bio, is_abnormal);
1683 if (bio->bi_opf & REQ_PREFLUSH) {
1684 __send_empty_flush(&ci);
1685 /* dm_io_complete submits any data associated with flush */
1689 error = __split_and_process_bio(&ci);
1690 if (error || !ci.sector_count)
1693 * Remainder must be passed to submit_bio_noacct() so it gets handled
1694 * *after* bios already submitted have been completely processed.
1696 bio_trim(bio, io->sectors, ci.sector_count);
1697 trace_block_split(bio, bio->bi_iter.bi_sector);
1698 bio_inc_remaining(bio);
1699 submit_bio_noacct(bio);
1702 * Drop the extra reference count for non-POLLED bio, and hold one
1703 * reference for POLLED bio, which will be released in dm_poll_bio
1705 * Add every dm_io instance into the dm_io list head which is stored
1706 * in bio->bi_private, so that dm_poll_bio can poll them all.
1708 if (error || !ci.submit_as_polled) {
1710 * In case of submission failure, the extra reference for
1711 * submitting io isn't consumed yet
1714 atomic_dec(&io->io_count);
1715 dm_io_dec_pending(io, error);
1717 dm_queue_poll_io(bio, io);
1720 static void dm_submit_bio(struct bio *bio)
1722 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1724 struct dm_table *map;
1725 unsigned bio_opf = bio->bi_opf;
1727 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1729 /* If suspended, or map not yet available, queue this IO for later */
1730 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1732 if (bio->bi_opf & REQ_NOWAIT)
1733 bio_wouldblock_error(bio);
1734 else if (bio->bi_opf & REQ_RAHEAD)
1741 dm_split_and_process_bio(md, map, bio);
1743 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1746 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1749 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1751 /* don't poll if the mapped io is done */
1752 if (atomic_read(&io->io_count) > 1)
1753 bio_poll(&io->tio.clone, iob, flags);
1755 /* bio_poll holds the last reference */
1756 return atomic_read(&io->io_count) == 1;
1759 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1762 struct dm_io **head = dm_poll_list_head(bio);
1763 struct dm_io *list = *head;
1764 struct dm_io *tmp = NULL;
1765 struct dm_io *curr, *next;
1767 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1768 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1771 WARN_ON_ONCE(!list);
1774 * Restore .bi_private before possibly completing dm_io.
1776 * bio_poll() is only possible once @bio has been completely
1777 * submitted via submit_bio_noacct()'s depth-first submission.
1778 * So there is no dm_queue_poll_io() race associated with
1779 * clearing REQ_DM_POLL_LIST here.
1781 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1782 bio->bi_private = list->data;
1784 for (curr = list, next = curr->next; curr; curr = next, next =
1785 curr ? curr->next : NULL) {
1786 if (dm_poll_dm_io(curr, iob, flags)) {
1788 * clone_endio() has already occurred, so no
1789 * error handling is needed here.
1791 __dm_io_dec_pending(curr);
1800 bio->bi_opf |= REQ_DM_POLL_LIST;
1801 /* Reset bio->bi_private to dm_io list head */
1808 /*-----------------------------------------------------------------
1809 * An IDR is used to keep track of allocated minor numbers.
1810 *---------------------------------------------------------------*/
1811 static void free_minor(int minor)
1813 spin_lock(&_minor_lock);
1814 idr_remove(&_minor_idr, minor);
1815 spin_unlock(&_minor_lock);
1819 * See if the device with a specific minor # is free.
1821 static int specific_minor(int minor)
1825 if (minor >= (1 << MINORBITS))
1828 idr_preload(GFP_KERNEL);
1829 spin_lock(&_minor_lock);
1831 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1833 spin_unlock(&_minor_lock);
1836 return r == -ENOSPC ? -EBUSY : r;
1840 static int next_free_minor(int *minor)
1844 idr_preload(GFP_KERNEL);
1845 spin_lock(&_minor_lock);
1847 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1849 spin_unlock(&_minor_lock);
1857 static const struct block_device_operations dm_blk_dops;
1858 static const struct block_device_operations dm_rq_blk_dops;
1859 static const struct dax_operations dm_dax_ops;
1861 static void dm_wq_work(struct work_struct *work);
1863 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1864 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1866 dm_destroy_crypto_profile(q->crypto_profile);
1869 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1871 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1874 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1876 static void cleanup_mapped_device(struct mapped_device *md)
1879 destroy_workqueue(md->wq);
1880 dm_free_md_mempools(md->mempools);
1883 dax_remove_host(md->disk);
1884 kill_dax(md->dax_dev);
1885 put_dax(md->dax_dev);
1889 dm_cleanup_zoned_dev(md);
1891 spin_lock(&_minor_lock);
1892 md->disk->private_data = NULL;
1893 spin_unlock(&_minor_lock);
1894 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1896 del_gendisk(md->disk);
1898 dm_queue_destroy_crypto_profile(md->queue);
1899 blk_cleanup_disk(md->disk);
1902 if (md->pending_io) {
1903 free_percpu(md->pending_io);
1904 md->pending_io = NULL;
1907 cleanup_srcu_struct(&md->io_barrier);
1909 mutex_destroy(&md->suspend_lock);
1910 mutex_destroy(&md->type_lock);
1911 mutex_destroy(&md->table_devices_lock);
1912 mutex_destroy(&md->swap_bios_lock);
1914 dm_mq_cleanup_mapped_device(md);
1918 * Allocate and initialise a blank device with a given minor.
1920 static struct mapped_device *alloc_dev(int minor)
1922 int r, numa_node_id = dm_get_numa_node();
1923 struct mapped_device *md;
1926 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1928 DMWARN("unable to allocate device, out of memory.");
1932 if (!try_module_get(THIS_MODULE))
1933 goto bad_module_get;
1935 /* get a minor number for the dev */
1936 if (minor == DM_ANY_MINOR)
1937 r = next_free_minor(&minor);
1939 r = specific_minor(minor);
1943 r = init_srcu_struct(&md->io_barrier);
1945 goto bad_io_barrier;
1947 md->numa_node_id = numa_node_id;
1948 md->init_tio_pdu = false;
1949 md->type = DM_TYPE_NONE;
1950 mutex_init(&md->suspend_lock);
1951 mutex_init(&md->type_lock);
1952 mutex_init(&md->table_devices_lock);
1953 spin_lock_init(&md->deferred_lock);
1954 atomic_set(&md->holders, 1);
1955 atomic_set(&md->open_count, 0);
1956 atomic_set(&md->event_nr, 0);
1957 atomic_set(&md->uevent_seq, 0);
1958 INIT_LIST_HEAD(&md->uevent_list);
1959 INIT_LIST_HEAD(&md->table_devices);
1960 spin_lock_init(&md->uevent_lock);
1963 * default to bio-based until DM table is loaded and md->type
1964 * established. If request-based table is loaded: blk-mq will
1965 * override accordingly.
1967 md->disk = blk_alloc_disk(md->numa_node_id);
1970 md->queue = md->disk->queue;
1972 init_waitqueue_head(&md->wait);
1973 INIT_WORK(&md->work, dm_wq_work);
1974 init_waitqueue_head(&md->eventq);
1975 init_completion(&md->kobj_holder.completion);
1977 md->swap_bios = get_swap_bios();
1978 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1979 mutex_init(&md->swap_bios_lock);
1981 md->disk->major = _major;
1982 md->disk->first_minor = minor;
1983 md->disk->minors = 1;
1984 md->disk->flags |= GENHD_FL_NO_PART;
1985 md->disk->fops = &dm_blk_dops;
1986 md->disk->queue = md->queue;
1987 md->disk->private_data = md;
1988 sprintf(md->disk->disk_name, "dm-%d", minor);
1990 if (IS_ENABLED(CONFIG_FS_DAX)) {
1991 md->dax_dev = alloc_dax(md, &dm_dax_ops);
1992 if (IS_ERR(md->dax_dev)) {
1996 set_dax_nocache(md->dax_dev);
1997 set_dax_nomc(md->dax_dev);
1998 if (dax_add_host(md->dax_dev, md->disk))
2002 format_dev_t(md->name, MKDEV(_major, minor));
2004 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2008 md->pending_io = alloc_percpu(unsigned long);
2009 if (!md->pending_io)
2012 dm_stats_init(&md->stats);
2014 /* Populate the mapping, nobody knows we exist yet */
2015 spin_lock(&_minor_lock);
2016 old_md = idr_replace(&_minor_idr, md, minor);
2017 spin_unlock(&_minor_lock);
2019 BUG_ON(old_md != MINOR_ALLOCED);
2024 cleanup_mapped_device(md);
2028 module_put(THIS_MODULE);
2034 static void unlock_fs(struct mapped_device *md);
2036 static void free_dev(struct mapped_device *md)
2038 int minor = MINOR(disk_devt(md->disk));
2042 cleanup_mapped_device(md);
2044 free_table_devices(&md->table_devices);
2045 dm_stats_cleanup(&md->stats);
2048 module_put(THIS_MODULE);
2053 * Bind a table to the device.
2055 static void event_callback(void *context)
2057 unsigned long flags;
2059 struct mapped_device *md = (struct mapped_device *) context;
2061 spin_lock_irqsave(&md->uevent_lock, flags);
2062 list_splice_init(&md->uevent_list, &uevents);
2063 spin_unlock_irqrestore(&md->uevent_lock, flags);
2065 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2067 atomic_inc(&md->event_nr);
2068 wake_up(&md->eventq);
2069 dm_issue_global_event();
2073 * Returns old map, which caller must destroy.
2075 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2076 struct queue_limits *limits)
2078 struct dm_table *old_map;
2082 lockdep_assert_held(&md->suspend_lock);
2084 size = dm_table_get_size(t);
2087 * Wipe any geometry if the size of the table changed.
2089 if (size != dm_get_size(md))
2090 memset(&md->geometry, 0, sizeof(md->geometry));
2092 if (!get_capacity(md->disk))
2093 set_capacity(md->disk, size);
2095 set_capacity_and_notify(md->disk, size);
2097 dm_table_event_callback(t, event_callback, md);
2099 if (dm_table_request_based(t)) {
2101 * Leverage the fact that request-based DM targets are
2102 * immutable singletons - used to optimize dm_mq_queue_rq.
2104 md->immutable_target = dm_table_get_immutable_target(t);
2107 * There is no need to reload with request-based dm because the
2108 * size of front_pad doesn't change.
2110 * Note for future: If you are to reload bioset, prep-ed
2111 * requests in the queue may refer to bio from the old bioset,
2112 * so you must walk through the queue to unprep.
2114 if (!md->mempools) {
2115 md->mempools = t->mempools;
2120 * The md may already have mempools that need changing.
2121 * If so, reload bioset because front_pad may have changed
2122 * because a different table was loaded.
2124 dm_free_md_mempools(md->mempools);
2125 md->mempools = t->mempools;
2129 ret = dm_table_set_restrictions(t, md->queue, limits);
2131 old_map = ERR_PTR(ret);
2135 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2136 rcu_assign_pointer(md->map, (void *)t);
2137 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2146 * Returns unbound table for the caller to free.
2148 static struct dm_table *__unbind(struct mapped_device *md)
2150 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2155 dm_table_event_callback(map, NULL, NULL);
2156 RCU_INIT_POINTER(md->map, NULL);
2163 * Constructor for a new device.
2165 int dm_create(int minor, struct mapped_device **result)
2167 struct mapped_device *md;
2169 md = alloc_dev(minor);
2173 dm_ima_reset_data(md);
2180 * Functions to manage md->type.
2181 * All are required to hold md->type_lock.
2183 void dm_lock_md_type(struct mapped_device *md)
2185 mutex_lock(&md->type_lock);
2188 void dm_unlock_md_type(struct mapped_device *md)
2190 mutex_unlock(&md->type_lock);
2193 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2195 BUG_ON(!mutex_is_locked(&md->type_lock));
2199 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2204 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2206 return md->immutable_target_type;
2210 * The queue_limits are only valid as long as you have a reference
2213 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2215 BUG_ON(!atomic_read(&md->holders));
2216 return &md->queue->limits;
2218 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2221 * Setup the DM device's queue based on md's type
2223 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2225 enum dm_queue_mode type = dm_table_get_type(t);
2226 struct queue_limits limits;
2230 case DM_TYPE_REQUEST_BASED:
2231 md->disk->fops = &dm_rq_blk_dops;
2232 r = dm_mq_init_request_queue(md, t);
2234 DMERR("Cannot initialize queue for request-based dm mapped device");
2238 case DM_TYPE_BIO_BASED:
2239 case DM_TYPE_DAX_BIO_BASED:
2246 r = dm_calculate_queue_limits(t, &limits);
2248 DMERR("Cannot calculate initial queue limits");
2251 r = dm_table_set_restrictions(t, md->queue, &limits);
2255 r = add_disk(md->disk);
2259 r = dm_sysfs_init(md);
2261 del_gendisk(md->disk);
2268 struct mapped_device *dm_get_md(dev_t dev)
2270 struct mapped_device *md;
2271 unsigned minor = MINOR(dev);
2273 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2276 spin_lock(&_minor_lock);
2278 md = idr_find(&_minor_idr, minor);
2279 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2280 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2286 spin_unlock(&_minor_lock);
2290 EXPORT_SYMBOL_GPL(dm_get_md);
2292 void *dm_get_mdptr(struct mapped_device *md)
2294 return md->interface_ptr;
2297 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2299 md->interface_ptr = ptr;
2302 void dm_get(struct mapped_device *md)
2304 atomic_inc(&md->holders);
2305 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2308 int dm_hold(struct mapped_device *md)
2310 spin_lock(&_minor_lock);
2311 if (test_bit(DMF_FREEING, &md->flags)) {
2312 spin_unlock(&_minor_lock);
2316 spin_unlock(&_minor_lock);
2319 EXPORT_SYMBOL_GPL(dm_hold);
2321 const char *dm_device_name(struct mapped_device *md)
2325 EXPORT_SYMBOL_GPL(dm_device_name);
2327 static void __dm_destroy(struct mapped_device *md, bool wait)
2329 struct dm_table *map;
2334 spin_lock(&_minor_lock);
2335 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2336 set_bit(DMF_FREEING, &md->flags);
2337 spin_unlock(&_minor_lock);
2339 blk_mark_disk_dead(md->disk);
2342 * Take suspend_lock so that presuspend and postsuspend methods
2343 * do not race with internal suspend.
2345 mutex_lock(&md->suspend_lock);
2346 map = dm_get_live_table(md, &srcu_idx);
2347 if (!dm_suspended_md(md)) {
2348 dm_table_presuspend_targets(map);
2349 set_bit(DMF_SUSPENDED, &md->flags);
2350 set_bit(DMF_POST_SUSPENDING, &md->flags);
2351 dm_table_postsuspend_targets(map);
2353 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2354 dm_put_live_table(md, srcu_idx);
2355 mutex_unlock(&md->suspend_lock);
2358 * Rare, but there may be I/O requests still going to complete,
2359 * for example. Wait for all references to disappear.
2360 * No one should increment the reference count of the mapped_device,
2361 * after the mapped_device state becomes DMF_FREEING.
2364 while (atomic_read(&md->holders))
2366 else if (atomic_read(&md->holders))
2367 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2368 dm_device_name(md), atomic_read(&md->holders));
2370 dm_table_destroy(__unbind(md));
2374 void dm_destroy(struct mapped_device *md)
2376 __dm_destroy(md, true);
2379 void dm_destroy_immediate(struct mapped_device *md)
2381 __dm_destroy(md, false);
2384 void dm_put(struct mapped_device *md)
2386 atomic_dec(&md->holders);
2388 EXPORT_SYMBOL_GPL(dm_put);
2390 static bool dm_in_flight_bios(struct mapped_device *md)
2393 unsigned long sum = 0;
2395 for_each_possible_cpu(cpu)
2396 sum += *per_cpu_ptr(md->pending_io, cpu);
2401 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2407 prepare_to_wait(&md->wait, &wait, task_state);
2409 if (!dm_in_flight_bios(md))
2412 if (signal_pending_state(task_state, current)) {
2419 finish_wait(&md->wait, &wait);
2426 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2430 if (!queue_is_mq(md->queue))
2431 return dm_wait_for_bios_completion(md, task_state);
2434 if (!blk_mq_queue_inflight(md->queue))
2437 if (signal_pending_state(task_state, current)) {
2449 * Process the deferred bios
2451 static void dm_wq_work(struct work_struct *work)
2453 struct mapped_device *md = container_of(work, struct mapped_device, work);
2456 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2457 spin_lock_irq(&md->deferred_lock);
2458 bio = bio_list_pop(&md->deferred);
2459 spin_unlock_irq(&md->deferred_lock);
2464 submit_bio_noacct(bio);
2468 static void dm_queue_flush(struct mapped_device *md)
2470 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2471 smp_mb__after_atomic();
2472 queue_work(md->wq, &md->work);
2476 * Swap in a new table, returning the old one for the caller to destroy.
2478 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2480 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2481 struct queue_limits limits;
2484 mutex_lock(&md->suspend_lock);
2486 /* device must be suspended */
2487 if (!dm_suspended_md(md))
2491 * If the new table has no data devices, retain the existing limits.
2492 * This helps multipath with queue_if_no_path if all paths disappear,
2493 * then new I/O is queued based on these limits, and then some paths
2496 if (dm_table_has_no_data_devices(table)) {
2497 live_map = dm_get_live_table_fast(md);
2499 limits = md->queue->limits;
2500 dm_put_live_table_fast(md);
2504 r = dm_calculate_queue_limits(table, &limits);
2511 map = __bind(md, table, &limits);
2512 dm_issue_global_event();
2515 mutex_unlock(&md->suspend_lock);
2520 * Functions to lock and unlock any filesystem running on the
2523 static int lock_fs(struct mapped_device *md)
2527 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2529 r = freeze_bdev(md->disk->part0);
2531 set_bit(DMF_FROZEN, &md->flags);
2535 static void unlock_fs(struct mapped_device *md)
2537 if (!test_bit(DMF_FROZEN, &md->flags))
2539 thaw_bdev(md->disk->part0);
2540 clear_bit(DMF_FROZEN, &md->flags);
2544 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2545 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2546 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2548 * If __dm_suspend returns 0, the device is completely quiescent
2549 * now. There is no request-processing activity. All new requests
2550 * are being added to md->deferred list.
2552 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2553 unsigned suspend_flags, unsigned int task_state,
2554 int dmf_suspended_flag)
2556 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2557 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2560 lockdep_assert_held(&md->suspend_lock);
2563 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2564 * This flag is cleared before dm_suspend returns.
2567 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2569 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2572 * This gets reverted if there's an error later and the targets
2573 * provide the .presuspend_undo hook.
2575 dm_table_presuspend_targets(map);
2578 * Flush I/O to the device.
2579 * Any I/O submitted after lock_fs() may not be flushed.
2580 * noflush takes precedence over do_lockfs.
2581 * (lock_fs() flushes I/Os and waits for them to complete.)
2583 if (!noflush && do_lockfs) {
2586 dm_table_presuspend_undo_targets(map);
2592 * Here we must make sure that no processes are submitting requests
2593 * to target drivers i.e. no one may be executing
2594 * dm_split_and_process_bio from dm_submit_bio.
2596 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2597 * we take the write lock. To prevent any process from reentering
2598 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2599 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2600 * flush_workqueue(md->wq).
2602 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2604 synchronize_srcu(&md->io_barrier);
2607 * Stop md->queue before flushing md->wq in case request-based
2608 * dm defers requests to md->wq from md->queue.
2610 if (dm_request_based(md))
2611 dm_stop_queue(md->queue);
2613 flush_workqueue(md->wq);
2616 * At this point no more requests are entering target request routines.
2617 * We call dm_wait_for_completion to wait for all existing requests
2620 r = dm_wait_for_completion(md, task_state);
2622 set_bit(dmf_suspended_flag, &md->flags);
2625 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2627 synchronize_srcu(&md->io_barrier);
2629 /* were we interrupted ? */
2633 if (dm_request_based(md))
2634 dm_start_queue(md->queue);
2637 dm_table_presuspend_undo_targets(map);
2638 /* pushback list is already flushed, so skip flush */
2645 * We need to be able to change a mapping table under a mounted
2646 * filesystem. For example we might want to move some data in
2647 * the background. Before the table can be swapped with
2648 * dm_bind_table, dm_suspend must be called to flush any in
2649 * flight bios and ensure that any further io gets deferred.
2652 * Suspend mechanism in request-based dm.
2654 * 1. Flush all I/Os by lock_fs() if needed.
2655 * 2. Stop dispatching any I/O by stopping the request_queue.
2656 * 3. Wait for all in-flight I/Os to be completed or requeued.
2658 * To abort suspend, start the request_queue.
2660 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2662 struct dm_table *map = NULL;
2666 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2668 if (dm_suspended_md(md)) {
2673 if (dm_suspended_internally_md(md)) {
2674 /* already internally suspended, wait for internal resume */
2675 mutex_unlock(&md->suspend_lock);
2676 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2682 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2684 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2688 set_bit(DMF_POST_SUSPENDING, &md->flags);
2689 dm_table_postsuspend_targets(map);
2690 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2693 mutex_unlock(&md->suspend_lock);
2697 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2700 int r = dm_table_resume_targets(map);
2708 * Flushing deferred I/Os must be done after targets are resumed
2709 * so that mapping of targets can work correctly.
2710 * Request-based dm is queueing the deferred I/Os in its request_queue.
2712 if (dm_request_based(md))
2713 dm_start_queue(md->queue);
2720 int dm_resume(struct mapped_device *md)
2723 struct dm_table *map = NULL;
2727 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2729 if (!dm_suspended_md(md))
2732 if (dm_suspended_internally_md(md)) {
2733 /* already internally suspended, wait for internal resume */
2734 mutex_unlock(&md->suspend_lock);
2735 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2741 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2742 if (!map || !dm_table_get_size(map))
2745 r = __dm_resume(md, map);
2749 clear_bit(DMF_SUSPENDED, &md->flags);
2751 mutex_unlock(&md->suspend_lock);
2757 * Internal suspend/resume works like userspace-driven suspend. It waits
2758 * until all bios finish and prevents issuing new bios to the target drivers.
2759 * It may be used only from the kernel.
2762 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2764 struct dm_table *map = NULL;
2766 lockdep_assert_held(&md->suspend_lock);
2768 if (md->internal_suspend_count++)
2769 return; /* nested internal suspend */
2771 if (dm_suspended_md(md)) {
2772 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2773 return; /* nest suspend */
2776 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2779 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2780 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2781 * would require changing .presuspend to return an error -- avoid this
2782 * until there is a need for more elaborate variants of internal suspend.
2784 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2785 DMF_SUSPENDED_INTERNALLY);
2787 set_bit(DMF_POST_SUSPENDING, &md->flags);
2788 dm_table_postsuspend_targets(map);
2789 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2792 static void __dm_internal_resume(struct mapped_device *md)
2794 BUG_ON(!md->internal_suspend_count);
2796 if (--md->internal_suspend_count)
2797 return; /* resume from nested internal suspend */
2799 if (dm_suspended_md(md))
2800 goto done; /* resume from nested suspend */
2803 * NOTE: existing callers don't need to call dm_table_resume_targets
2804 * (which may fail -- so best to avoid it for now by passing NULL map)
2806 (void) __dm_resume(md, NULL);
2809 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2810 smp_mb__after_atomic();
2811 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2814 void dm_internal_suspend_noflush(struct mapped_device *md)
2816 mutex_lock(&md->suspend_lock);
2817 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2818 mutex_unlock(&md->suspend_lock);
2820 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2822 void dm_internal_resume(struct mapped_device *md)
2824 mutex_lock(&md->suspend_lock);
2825 __dm_internal_resume(md);
2826 mutex_unlock(&md->suspend_lock);
2828 EXPORT_SYMBOL_GPL(dm_internal_resume);
2831 * Fast variants of internal suspend/resume hold md->suspend_lock,
2832 * which prevents interaction with userspace-driven suspend.
2835 void dm_internal_suspend_fast(struct mapped_device *md)
2837 mutex_lock(&md->suspend_lock);
2838 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2841 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2842 synchronize_srcu(&md->io_barrier);
2843 flush_workqueue(md->wq);
2844 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2846 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2848 void dm_internal_resume_fast(struct mapped_device *md)
2850 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2856 mutex_unlock(&md->suspend_lock);
2858 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2860 /*-----------------------------------------------------------------
2861 * Event notification.
2862 *---------------------------------------------------------------*/
2863 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2868 char udev_cookie[DM_COOKIE_LENGTH];
2869 char *envp[] = { udev_cookie, NULL };
2871 noio_flag = memalloc_noio_save();
2874 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2876 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2877 DM_COOKIE_ENV_VAR_NAME, cookie);
2878 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2882 memalloc_noio_restore(noio_flag);
2887 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2889 return atomic_add_return(1, &md->uevent_seq);
2892 uint32_t dm_get_event_nr(struct mapped_device *md)
2894 return atomic_read(&md->event_nr);
2897 int dm_wait_event(struct mapped_device *md, int event_nr)
2899 return wait_event_interruptible(md->eventq,
2900 (event_nr != atomic_read(&md->event_nr)));
2903 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2905 unsigned long flags;
2907 spin_lock_irqsave(&md->uevent_lock, flags);
2908 list_add(elist, &md->uevent_list);
2909 spin_unlock_irqrestore(&md->uevent_lock, flags);
2913 * The gendisk is only valid as long as you have a reference
2916 struct gendisk *dm_disk(struct mapped_device *md)
2920 EXPORT_SYMBOL_GPL(dm_disk);
2922 struct kobject *dm_kobject(struct mapped_device *md)
2924 return &md->kobj_holder.kobj;
2927 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2929 struct mapped_device *md;
2931 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2933 spin_lock(&_minor_lock);
2934 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2940 spin_unlock(&_minor_lock);
2945 int dm_suspended_md(struct mapped_device *md)
2947 return test_bit(DMF_SUSPENDED, &md->flags);
2950 static int dm_post_suspending_md(struct mapped_device *md)
2952 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2955 int dm_suspended_internally_md(struct mapped_device *md)
2957 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2960 int dm_test_deferred_remove_flag(struct mapped_device *md)
2962 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2965 int dm_suspended(struct dm_target *ti)
2967 return dm_suspended_md(ti->table->md);
2969 EXPORT_SYMBOL_GPL(dm_suspended);
2971 int dm_post_suspending(struct dm_target *ti)
2973 return dm_post_suspending_md(ti->table->md);
2975 EXPORT_SYMBOL_GPL(dm_post_suspending);
2977 int dm_noflush_suspending(struct dm_target *ti)
2979 return __noflush_suspending(ti->table->md);
2981 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2983 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2984 unsigned per_io_data_size, unsigned min_pool_size,
2985 bool integrity, bool poll)
2987 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2988 unsigned int pool_size = 0;
2989 unsigned int front_pad, io_front_pad;
2996 case DM_TYPE_BIO_BASED:
2997 case DM_TYPE_DAX_BIO_BASED:
2998 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2999 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
3000 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
3001 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, poll ? BIOSET_PERCPU_CACHE : 0);
3004 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3007 case DM_TYPE_REQUEST_BASED:
3008 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3009 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3010 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3016 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3020 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3026 dm_free_md_mempools(pools);
3031 void dm_free_md_mempools(struct dm_md_mempools *pools)
3036 bioset_exit(&pools->bs);
3037 bioset_exit(&pools->io_bs);
3049 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3052 struct mapped_device *md = bdev->bd_disk->private_data;
3053 struct dm_table *table;
3054 struct dm_target *ti;
3055 int ret = -ENOTTY, srcu_idx;
3057 table = dm_get_live_table(md, &srcu_idx);
3058 if (!table || !dm_table_get_size(table))
3061 /* We only support devices that have a single target */
3062 if (dm_table_get_num_targets(table) != 1)
3064 ti = dm_table_get_target(table, 0);
3067 if (!ti->type->iterate_devices)
3070 ret = ti->type->iterate_devices(ti, fn, data);
3072 dm_put_live_table(md, srcu_idx);
3077 * For register / unregister we need to manually call out to every path.
3079 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3080 sector_t start, sector_t len, void *data)
3082 struct dm_pr *pr = data;
3083 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3085 if (!ops || !ops->pr_register)
3087 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3090 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3101 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3102 if (ret && new_key) {
3103 /* unregister all paths if we failed to register any path */
3104 pr.old_key = new_key;
3107 pr.fail_early = false;
3108 dm_call_pr(bdev, __dm_pr_register, &pr);
3114 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3117 struct mapped_device *md = bdev->bd_disk->private_data;
3118 const struct pr_ops *ops;
3121 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3125 ops = bdev->bd_disk->fops->pr_ops;
3126 if (ops && ops->pr_reserve)
3127 r = ops->pr_reserve(bdev, key, type, flags);
3131 dm_unprepare_ioctl(md, srcu_idx);
3135 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3137 struct mapped_device *md = bdev->bd_disk->private_data;
3138 const struct pr_ops *ops;
3141 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3145 ops = bdev->bd_disk->fops->pr_ops;
3146 if (ops && ops->pr_release)
3147 r = ops->pr_release(bdev, key, type);
3151 dm_unprepare_ioctl(md, srcu_idx);
3155 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3156 enum pr_type type, bool abort)
3158 struct mapped_device *md = bdev->bd_disk->private_data;
3159 const struct pr_ops *ops;
3162 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3166 ops = bdev->bd_disk->fops->pr_ops;
3167 if (ops && ops->pr_preempt)
3168 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3172 dm_unprepare_ioctl(md, srcu_idx);
3176 static int dm_pr_clear(struct block_device *bdev, u64 key)
3178 struct mapped_device *md = bdev->bd_disk->private_data;
3179 const struct pr_ops *ops;
3182 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3186 ops = bdev->bd_disk->fops->pr_ops;
3187 if (ops && ops->pr_clear)
3188 r = ops->pr_clear(bdev, key);
3192 dm_unprepare_ioctl(md, srcu_idx);
3196 static const struct pr_ops dm_pr_ops = {
3197 .pr_register = dm_pr_register,
3198 .pr_reserve = dm_pr_reserve,
3199 .pr_release = dm_pr_release,
3200 .pr_preempt = dm_pr_preempt,
3201 .pr_clear = dm_pr_clear,
3204 static const struct block_device_operations dm_blk_dops = {
3205 .submit_bio = dm_submit_bio,
3206 .poll_bio = dm_poll_bio,
3207 .open = dm_blk_open,
3208 .release = dm_blk_close,
3209 .ioctl = dm_blk_ioctl,
3210 .getgeo = dm_blk_getgeo,
3211 .report_zones = dm_blk_report_zones,
3212 .pr_ops = &dm_pr_ops,
3213 .owner = THIS_MODULE
3216 static const struct block_device_operations dm_rq_blk_dops = {
3217 .open = dm_blk_open,
3218 .release = dm_blk_close,
3219 .ioctl = dm_blk_ioctl,
3220 .getgeo = dm_blk_getgeo,
3221 .pr_ops = &dm_pr_ops,
3222 .owner = THIS_MODULE
3225 static const struct dax_operations dm_dax_ops = {
3226 .direct_access = dm_dax_direct_access,
3227 .zero_page_range = dm_dax_zero_page_range,
3228 .recovery_write = dm_dax_recovery_write,
3234 module_init(dm_init);
3235 module_exit(dm_exit);
3237 module_param(major, uint, 0);
3238 MODULE_PARM_DESC(major, "The major number of the device mapper");
3240 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3241 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3243 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3244 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3246 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3247 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3249 MODULE_DESCRIPTION(DM_NAME " driver");
3250 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3251 MODULE_LICENSE("GPL");