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);
75 * One of these is allocated (on-stack) per original bio.
82 unsigned sector_count;
83 bool submit_as_polled;
86 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
87 #define DM_IO_BIO_OFFSET \
88 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
90 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
92 return container_of(clone, struct dm_target_io, clone);
95 void *dm_per_bio_data(struct bio *bio, size_t data_size)
97 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
98 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
99 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
101 EXPORT_SYMBOL_GPL(dm_per_bio_data);
103 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
105 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
106 if (io->magic == DM_IO_MAGIC)
107 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
108 BUG_ON(io->magic != DM_TIO_MAGIC);
109 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
111 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
113 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
115 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
117 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
119 #define MINOR_ALLOCED ((void *)-1)
121 #define DM_NUMA_NODE NUMA_NO_NODE
122 static int dm_numa_node = DM_NUMA_NODE;
124 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
125 static int swap_bios = DEFAULT_SWAP_BIOS;
126 static int get_swap_bios(void)
128 int latch = READ_ONCE(swap_bios);
129 if (unlikely(latch <= 0))
130 latch = DEFAULT_SWAP_BIOS;
135 * For mempools pre-allocation at the table loading time.
137 struct dm_md_mempools {
139 struct bio_set io_bs;
142 struct table_device {
143 struct list_head list;
145 struct dm_dev dm_dev;
149 * Bio-based DM's mempools' reserved IOs set by the user.
151 #define RESERVED_BIO_BASED_IOS 16
152 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
154 static int __dm_get_module_param_int(int *module_param, int min, int max)
156 int param = READ_ONCE(*module_param);
157 int modified_param = 0;
158 bool modified = true;
161 modified_param = min;
162 else if (param > max)
163 modified_param = max;
168 (void)cmpxchg(module_param, param, modified_param);
169 param = modified_param;
175 unsigned __dm_get_module_param(unsigned *module_param,
176 unsigned def, unsigned max)
178 unsigned param = READ_ONCE(*module_param);
179 unsigned modified_param = 0;
182 modified_param = def;
183 else if (param > max)
184 modified_param = max;
186 if (modified_param) {
187 (void)cmpxchg(module_param, param, modified_param);
188 param = modified_param;
194 unsigned dm_get_reserved_bio_based_ios(void)
196 return __dm_get_module_param(&reserved_bio_based_ios,
197 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
199 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
201 static unsigned dm_get_numa_node(void)
203 return __dm_get_module_param_int(&dm_numa_node,
204 DM_NUMA_NODE, num_online_nodes() - 1);
207 static int __init local_init(void)
211 r = dm_uevent_init();
215 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
216 if (!deferred_remove_workqueue) {
218 goto out_uevent_exit;
222 r = register_blkdev(_major, _name);
224 goto out_free_workqueue;
232 destroy_workqueue(deferred_remove_workqueue);
239 static void local_exit(void)
241 flush_scheduled_work();
242 destroy_workqueue(deferred_remove_workqueue);
244 unregister_blkdev(_major, _name);
249 DMINFO("cleaned up");
252 static int (*_inits[])(void) __initdata = {
263 static void (*_exits[])(void) = {
274 static int __init dm_init(void)
276 const int count = ARRAY_SIZE(_inits);
279 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
280 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
281 " Duplicate IMA measurements will not be recorded in the IMA log.");
284 for (i = 0; i < count; i++) {
298 static void __exit dm_exit(void)
300 int i = ARRAY_SIZE(_exits);
306 * Should be empty by this point.
308 idr_destroy(&_minor_idr);
312 * Block device functions
314 int dm_deleting_md(struct mapped_device *md)
316 return test_bit(DMF_DELETING, &md->flags);
319 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
321 struct mapped_device *md;
323 spin_lock(&_minor_lock);
325 md = bdev->bd_disk->private_data;
329 if (test_bit(DMF_FREEING, &md->flags) ||
330 dm_deleting_md(md)) {
336 atomic_inc(&md->open_count);
338 spin_unlock(&_minor_lock);
340 return md ? 0 : -ENXIO;
343 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
345 struct mapped_device *md;
347 spin_lock(&_minor_lock);
349 md = disk->private_data;
353 if (atomic_dec_and_test(&md->open_count) &&
354 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
355 queue_work(deferred_remove_workqueue, &deferred_remove_work);
359 spin_unlock(&_minor_lock);
362 int dm_open_count(struct mapped_device *md)
364 return atomic_read(&md->open_count);
368 * Guarantees nothing is using the device before it's deleted.
370 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
374 spin_lock(&_minor_lock);
376 if (dm_open_count(md)) {
379 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
380 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
383 set_bit(DMF_DELETING, &md->flags);
385 spin_unlock(&_minor_lock);
390 int dm_cancel_deferred_remove(struct mapped_device *md)
394 spin_lock(&_minor_lock);
396 if (test_bit(DMF_DELETING, &md->flags))
399 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
401 spin_unlock(&_minor_lock);
406 static void do_deferred_remove(struct work_struct *w)
408 dm_deferred_remove();
411 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
413 struct mapped_device *md = bdev->bd_disk->private_data;
415 return dm_get_geometry(md, geo);
418 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
419 struct block_device **bdev)
421 struct dm_target *tgt;
422 struct dm_table *map;
427 map = dm_get_live_table(md, srcu_idx);
428 if (!map || !dm_table_get_size(map))
431 /* We only support devices that have a single target */
432 if (dm_table_get_num_targets(map) != 1)
435 tgt = dm_table_get_target(map, 0);
436 if (!tgt->type->prepare_ioctl)
439 if (dm_suspended_md(md))
442 r = tgt->type->prepare_ioctl(tgt, bdev);
443 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
444 dm_put_live_table(md, *srcu_idx);
452 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
454 dm_put_live_table(md, srcu_idx);
457 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
458 unsigned int cmd, unsigned long arg)
460 struct mapped_device *md = bdev->bd_disk->private_data;
463 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
469 * Target determined this ioctl is being issued against a
470 * subset of the parent bdev; require extra privileges.
472 if (!capable(CAP_SYS_RAWIO)) {
474 "%s: sending ioctl %x to DM device without required privilege.",
481 if (!bdev->bd_disk->fops->ioctl)
484 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
486 dm_unprepare_ioctl(md, srcu_idx);
490 u64 dm_start_time_ns_from_clone(struct bio *bio)
492 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
494 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
496 static bool bio_is_flush_with_data(struct bio *bio)
498 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
501 static void dm_io_acct(bool end, struct mapped_device *md, struct bio *bio,
502 unsigned long start_time, struct dm_stats_aux *stats_aux)
504 bool is_flush_with_data;
505 unsigned int bi_size;
507 /* If REQ_PREFLUSH set save any payload but do not account it */
508 is_flush_with_data = bio_is_flush_with_data(bio);
509 if (is_flush_with_data) {
510 bi_size = bio->bi_iter.bi_size;
511 bio->bi_iter.bi_size = 0;
515 bio_start_io_acct_time(bio, start_time);
517 bio_end_io_acct(bio, start_time);
519 if (unlikely(dm_stats_used(&md->stats)))
520 dm_stats_account_io(&md->stats, bio_data_dir(bio),
521 bio->bi_iter.bi_sector, bio_sectors(bio),
522 end, start_time, stats_aux);
524 /* Restore bio's payload so it does get accounted upon requeue */
525 if (is_flush_with_data)
526 bio->bi_iter.bi_size = bi_size;
529 static void __dm_start_io_acct(struct dm_io *io, struct bio *bio)
531 dm_io_acct(false, io->md, bio, io->start_time, &io->stats_aux);
534 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
536 /* Must account IO to DM device in terms of orig_bio */
537 struct bio *bio = io->orig_bio;
540 * Ensure IO accounting is only ever started once.
541 * Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO.
544 likely(!dm_tio_flagged(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO))) {
545 if (WARN_ON_ONCE(dm_io_flagged(io, DM_IO_ACCOUNTED)))
547 dm_io_set_flag(io, DM_IO_ACCOUNTED);
550 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
552 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
553 spin_lock_irqsave(&io->lock, flags);
554 dm_io_set_flag(io, DM_IO_ACCOUNTED);
555 spin_unlock_irqrestore(&io->lock, flags);
558 __dm_start_io_acct(io, bio);
561 static void dm_end_io_acct(struct dm_io *io, struct bio *bio)
563 dm_io_acct(true, io->md, bio, io->start_time, &io->stats_aux);
566 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
569 struct dm_target_io *tio;
572 clone = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, &md->io_bs);
574 tio = clone_to_tio(clone);
576 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
579 io = container_of(tio, struct dm_io, tio);
580 io->magic = DM_IO_MAGIC;
582 atomic_set(&io->io_count, 1);
583 this_cpu_inc(*md->pending_io);
586 io->map_task = current;
587 spin_lock_init(&io->lock);
588 io->start_time = jiffies;
591 dm_stats_record_start(&md->stats, &io->stats_aux);
596 static void free_io(struct dm_io *io)
598 bio_put(&io->tio.clone);
601 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
602 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
604 struct dm_target_io *tio;
607 if (!ci->io->tio.io) {
608 /* the dm_target_io embedded in ci->io is available */
610 /* alloc_io() already initialized embedded clone */
613 clone = bio_alloc_clone(ci->bio->bi_bdev, ci->bio,
614 gfp_mask, &ci->io->md->bs);
618 /* REQ_DM_POLL_LIST shouldn't be inherited */
619 clone->bi_opf &= ~REQ_DM_POLL_LIST;
621 tio = clone_to_tio(clone);
622 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
625 tio->magic = DM_TIO_MAGIC;
628 tio->target_bio_nr = target_bio_nr;
633 clone->bi_iter.bi_size = to_bytes(*len);
634 if (bio_integrity(clone))
635 bio_integrity_trim(clone);
641 static void free_tio(struct bio *clone)
643 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
649 * Add the bio to the list of deferred io.
651 static void queue_io(struct mapped_device *md, struct bio *bio)
655 spin_lock_irqsave(&md->deferred_lock, flags);
656 bio_list_add(&md->deferred, bio);
657 spin_unlock_irqrestore(&md->deferred_lock, flags);
658 queue_work(md->wq, &md->work);
662 * Everyone (including functions in this file), should use this
663 * function to access the md->map field, and make sure they call
664 * dm_put_live_table() when finished.
666 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
668 *srcu_idx = srcu_read_lock(&md->io_barrier);
670 return srcu_dereference(md->map, &md->io_barrier);
673 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
675 srcu_read_unlock(&md->io_barrier, srcu_idx);
678 void dm_sync_table(struct mapped_device *md)
680 synchronize_srcu(&md->io_barrier);
681 synchronize_rcu_expedited();
685 * A fast alternative to dm_get_live_table/dm_put_live_table.
686 * The caller must not block between these two functions.
688 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
691 return rcu_dereference(md->map);
694 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
699 static char *_dm_claim_ptr = "I belong to device-mapper";
702 * Open a table device so we can use it as a map destination.
704 static int open_table_device(struct table_device *td, dev_t dev,
705 struct mapped_device *md)
707 struct block_device *bdev;
711 BUG_ON(td->dm_dev.bdev);
713 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
715 return PTR_ERR(bdev);
717 r = bd_link_disk_holder(bdev, dm_disk(md));
719 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
723 td->dm_dev.bdev = bdev;
724 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off);
729 * Close a table device that we've been using.
731 static void close_table_device(struct table_device *td, struct mapped_device *md)
733 if (!td->dm_dev.bdev)
736 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
737 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
738 put_dax(td->dm_dev.dax_dev);
739 td->dm_dev.bdev = NULL;
740 td->dm_dev.dax_dev = NULL;
743 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
746 struct table_device *td;
748 list_for_each_entry(td, l, list)
749 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
755 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
756 struct dm_dev **result)
759 struct table_device *td;
761 mutex_lock(&md->table_devices_lock);
762 td = find_table_device(&md->table_devices, dev, mode);
764 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
766 mutex_unlock(&md->table_devices_lock);
770 td->dm_dev.mode = mode;
771 td->dm_dev.bdev = NULL;
773 if ((r = open_table_device(td, dev, md))) {
774 mutex_unlock(&md->table_devices_lock);
779 format_dev_t(td->dm_dev.name, dev);
781 refcount_set(&td->count, 1);
782 list_add(&td->list, &md->table_devices);
784 refcount_inc(&td->count);
786 mutex_unlock(&md->table_devices_lock);
788 *result = &td->dm_dev;
792 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
794 struct table_device *td = container_of(d, struct table_device, dm_dev);
796 mutex_lock(&md->table_devices_lock);
797 if (refcount_dec_and_test(&td->count)) {
798 close_table_device(td, md);
802 mutex_unlock(&md->table_devices_lock);
805 static void free_table_devices(struct list_head *devices)
807 struct list_head *tmp, *next;
809 list_for_each_safe(tmp, next, devices) {
810 struct table_device *td = list_entry(tmp, struct table_device, list);
812 DMWARN("dm_destroy: %s still exists with %d references",
813 td->dm_dev.name, refcount_read(&td->count));
819 * Get the geometry associated with a dm device
821 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
829 * Set the geometry of a device.
831 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
833 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
835 if (geo->start > sz) {
836 DMWARN("Start sector is beyond the geometry limits.");
845 static int __noflush_suspending(struct mapped_device *md)
847 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
850 static void dm_io_complete(struct dm_io *io)
852 blk_status_t io_error;
853 struct mapped_device *md = io->md;
854 struct bio *bio = io->orig_bio;
856 if (io->status == BLK_STS_DM_REQUEUE) {
859 * Target requested pushing back the I/O.
861 spin_lock_irqsave(&md->deferred_lock, flags);
862 if (__noflush_suspending(md) &&
863 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
864 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
865 bio_list_add_head(&md->deferred, bio);
868 * noflush suspend was interrupted or this is
869 * a write to a zoned target.
871 io->status = BLK_STS_IOERR;
873 spin_unlock_irqrestore(&md->deferred_lock, flags);
876 io_error = io->status;
877 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
878 dm_end_io_acct(io, bio);
879 else if (!io_error) {
881 * Must handle target that DM_MAPIO_SUBMITTED only to
882 * then bio_endio() rather than dm_submit_bio_remap()
884 __dm_start_io_acct(io, bio);
885 dm_end_io_acct(io, bio);
889 this_cpu_dec(*md->pending_io);
891 /* nudge anyone waiting on suspend queue */
892 if (unlikely(wq_has_sleeper(&md->wait)))
895 if (io_error == BLK_STS_DM_REQUEUE || io_error == BLK_STS_AGAIN) {
896 if (bio->bi_opf & REQ_POLLED) {
898 * Upper layer won't help us poll split bio (io->orig_bio
899 * may only reflect a subset of the pre-split original)
900 * so clear REQ_POLLED in case of requeue.
902 bio->bi_opf &= ~REQ_POLLED;
903 if (io_error == BLK_STS_AGAIN) {
904 /* io_uring doesn't handle BLK_STS_AGAIN (yet) */
911 if (bio_is_flush_with_data(bio)) {
913 * Preflush done for flush with data, reissue
914 * without REQ_PREFLUSH.
916 bio->bi_opf &= ~REQ_PREFLUSH;
919 /* done with normal IO or empty flush */
921 bio->bi_status = io_error;
926 static inline bool dm_tio_is_normal(struct dm_target_io *tio)
928 return (dm_tio_flagged(tio, DM_TIO_INSIDE_DM_IO) &&
929 !dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
933 * Decrements the number of outstanding ios that a bio has been
934 * cloned into, completing the original io if necc.
936 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
938 /* Push-back supersedes any I/O errors */
939 if (unlikely(error)) {
941 spin_lock_irqsave(&io->lock, flags);
942 if (!(io->status == BLK_STS_DM_REQUEUE &&
943 __noflush_suspending(io->md)))
945 spin_unlock_irqrestore(&io->lock, flags);
948 if (atomic_dec_and_test(&io->io_count))
952 void disable_discard(struct mapped_device *md)
954 struct queue_limits *limits = dm_get_queue_limits(md);
956 /* device doesn't really support DISCARD, disable it */
957 limits->max_discard_sectors = 0;
960 void disable_write_zeroes(struct mapped_device *md)
962 struct queue_limits *limits = dm_get_queue_limits(md);
964 /* device doesn't really support WRITE ZEROES, disable it */
965 limits->max_write_zeroes_sectors = 0;
968 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
970 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
973 static void clone_endio(struct bio *bio)
975 blk_status_t error = bio->bi_status;
976 struct dm_target_io *tio = clone_to_tio(bio);
977 struct dm_io *io = tio->io;
978 struct mapped_device *md = tio->io->md;
979 dm_endio_fn endio = tio->ti->type->end_io;
980 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
982 if (unlikely(error == BLK_STS_TARGET)) {
983 if (bio_op(bio) == REQ_OP_DISCARD &&
984 !bdev_max_discard_sectors(bio->bi_bdev))
986 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
987 !q->limits.max_write_zeroes_sectors)
988 disable_write_zeroes(md);
991 if (blk_queue_is_zoned(q))
992 dm_zone_endio(io, bio);
995 int r = endio(tio->ti, bio, &error);
997 case DM_ENDIO_REQUEUE:
999 * Requeuing writes to a sequential zone of a zoned
1000 * target will break the sequential write pattern:
1003 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1004 error = BLK_STS_IOERR;
1006 error = BLK_STS_DM_REQUEUE;
1010 case DM_ENDIO_INCOMPLETE:
1011 /* The target will handle the io */
1014 DMWARN("unimplemented target endio return value: %d", r);
1019 if (unlikely(swap_bios_limit(tio->ti, bio))) {
1020 struct mapped_device *md = io->md;
1021 up(&md->swap_bios_semaphore);
1025 dm_io_dec_pending(io, error);
1029 * Return maximum size of I/O possible at the supplied sector up to the current
1032 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1033 sector_t target_offset)
1035 return ti->len - target_offset;
1038 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1040 sector_t target_offset = dm_target_offset(ti, sector);
1041 sector_t len = max_io_len_target_boundary(ti, target_offset);
1045 * Does the target need to split IO even further?
1046 * - varied (per target) IO splitting is a tenet of DM; this
1047 * explains why stacked chunk_sectors based splitting via
1048 * blk_max_size_offset() isn't possible here. So pass in
1049 * ti->max_io_len to override stacked chunk_sectors.
1051 if (ti->max_io_len) {
1052 max_len = blk_max_size_offset(ti->table->md->queue,
1053 target_offset, ti->max_io_len);
1061 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1063 if (len > UINT_MAX) {
1064 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1065 (unsigned long long)len, UINT_MAX);
1066 ti->error = "Maximum size of target IO is too large";
1070 ti->max_io_len = (uint32_t) len;
1074 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1076 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1077 sector_t sector, int *srcu_idx)
1078 __acquires(md->io_barrier)
1080 struct dm_table *map;
1081 struct dm_target *ti;
1083 map = dm_get_live_table(md, srcu_idx);
1087 ti = dm_table_find_target(map, sector);
1094 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1095 long nr_pages, void **kaddr, pfn_t *pfn)
1097 struct mapped_device *md = dax_get_private(dax_dev);
1098 sector_t sector = pgoff * PAGE_SECTORS;
1099 struct dm_target *ti;
1100 long len, ret = -EIO;
1103 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1107 if (!ti->type->direct_access)
1109 len = max_io_len(ti, sector) / PAGE_SECTORS;
1112 nr_pages = min(len, nr_pages);
1113 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1116 dm_put_live_table(md, srcu_idx);
1121 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1124 struct mapped_device *md = dax_get_private(dax_dev);
1125 sector_t sector = pgoff * PAGE_SECTORS;
1126 struct dm_target *ti;
1130 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1134 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1136 * ->zero_page_range() is mandatory dax operation. If we are
1137 * here, something is wrong.
1141 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1143 dm_put_live_table(md, srcu_idx);
1149 * A target may call dm_accept_partial_bio only from the map routine. It is
1150 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1151 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1152 * __send_duplicate_bios().
1154 * dm_accept_partial_bio informs the dm that the target only wants to process
1155 * additional n_sectors sectors of the bio and the rest of the data should be
1156 * sent in a next bio.
1158 * A diagram that explains the arithmetics:
1159 * +--------------------+---------------+-------+
1161 * +--------------------+---------------+-------+
1163 * <-------------- *tio->len_ptr --------------->
1164 * <------- bi_size ------->
1167 * Region 1 was already iterated over with bio_advance or similar function.
1168 * (it may be empty if the target doesn't use bio_advance)
1169 * Region 2 is the remaining bio size that the target wants to process.
1170 * (it may be empty if region 1 is non-empty, although there is no reason
1172 * The target requires that region 3 is to be sent in the next bio.
1174 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1175 * the partially processed part (the sum of regions 1+2) must be the same for all
1176 * copies of the bio.
1178 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1180 struct dm_target_io *tio = clone_to_tio(bio);
1181 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1183 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1184 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1185 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1186 BUG_ON(bi_size > *tio->len_ptr);
1187 BUG_ON(n_sectors > bi_size);
1189 *tio->len_ptr -= bi_size - n_sectors;
1190 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1192 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1194 static inline void __dm_submit_bio_remap(struct bio *clone,
1195 dev_t dev, sector_t old_sector)
1197 trace_block_bio_remap(clone, dev, old_sector);
1198 submit_bio_noacct(clone);
1202 * @clone: clone bio that DM core passed to target's .map function
1203 * @tgt_clone: clone of @clone bio that target needs submitted
1205 * Targets should use this interface to submit bios they take
1206 * ownership of when returning DM_MAPIO_SUBMITTED.
1208 * Target should also enable ti->accounts_remapped_io
1210 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1212 struct dm_target_io *tio = clone_to_tio(clone);
1213 struct dm_io *io = tio->io;
1215 WARN_ON_ONCE(!tio->ti->accounts_remapped_io);
1217 /* establish bio that will get submitted */
1222 * Account io->origin_bio to DM dev on behalf of target
1223 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1225 if (io->map_task == current) {
1226 /* Still in target's map function */
1227 dm_io_set_flag(io, DM_IO_START_ACCT);
1230 * Called by another thread, managed by DM target,
1231 * wait for dm_split_and_process_bio() to store
1234 while (unlikely(!smp_load_acquire(&io->orig_bio)))
1236 dm_start_io_acct(io, clone);
1239 __dm_submit_bio_remap(tgt_clone, disk_devt(io->md->disk),
1242 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1244 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1246 mutex_lock(&md->swap_bios_lock);
1247 while (latch < md->swap_bios) {
1249 down(&md->swap_bios_semaphore);
1252 while (latch > md->swap_bios) {
1254 up(&md->swap_bios_semaphore);
1257 mutex_unlock(&md->swap_bios_lock);
1260 static void __map_bio(struct bio *clone)
1262 struct dm_target_io *tio = clone_to_tio(clone);
1264 struct dm_io *io = tio->io;
1265 struct dm_target *ti = tio->ti;
1267 clone->bi_end_io = clone_endio;
1272 dm_io_inc_pending(io);
1273 tio->old_sector = clone->bi_iter.bi_sector;
1275 if (unlikely(swap_bios_limit(ti, clone))) {
1276 struct mapped_device *md = io->md;
1277 int latch = get_swap_bios();
1278 if (unlikely(latch != md->swap_bios))
1279 __set_swap_bios_limit(md, latch);
1280 down(&md->swap_bios_semaphore);
1284 * Check if the IO needs a special mapping due to zone append emulation
1285 * on zoned target. In this case, dm_zone_map_bio() calls the target
1288 if (dm_emulate_zone_append(io->md))
1289 r = dm_zone_map_bio(tio);
1291 r = ti->type->map(ti, clone);
1294 case DM_MAPIO_SUBMITTED:
1295 /* target has assumed ownership of this io */
1296 if (!ti->accounts_remapped_io)
1297 dm_io_set_flag(io, DM_IO_START_ACCT);
1299 case DM_MAPIO_REMAPPED:
1301 * the bio has been remapped so dispatch it, but defer
1302 * dm_start_io_acct() until after possible bio_split().
1304 __dm_submit_bio_remap(clone, disk_devt(io->md->disk),
1306 dm_io_set_flag(io, DM_IO_START_ACCT);
1309 case DM_MAPIO_REQUEUE:
1310 if (unlikely(swap_bios_limit(ti, clone)))
1311 up(&io->md->swap_bios_semaphore);
1313 if (r == DM_MAPIO_KILL)
1314 dm_io_dec_pending(io, BLK_STS_IOERR);
1316 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1319 DMWARN("unimplemented target map return value: %d", r);
1324 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1325 struct dm_target *ti, unsigned num_bios)
1330 for (try = 0; try < 2; try++) {
1334 mutex_lock(&ci->io->md->table_devices_lock);
1335 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1336 bio = alloc_tio(ci, ti, bio_nr, NULL,
1337 try ? GFP_NOIO : GFP_NOWAIT);
1341 bio_list_add(blist, bio);
1344 mutex_unlock(&ci->io->md->table_devices_lock);
1345 if (bio_nr == num_bios)
1348 while ((bio = bio_list_pop(blist)))
1353 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1354 unsigned num_bios, unsigned *len)
1356 struct bio_list blist = BIO_EMPTY_LIST;
1363 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1367 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1368 alloc_multiple_bios(&blist, ci, ti, num_bios);
1369 while ((clone = bio_list_pop(&blist))) {
1370 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1377 static void __send_empty_flush(struct clone_info *ci)
1379 unsigned target_nr = 0;
1380 struct dm_target *ti;
1381 struct bio flush_bio;
1384 * Use an on-stack bio for this, it's safe since we don't
1385 * need to reference it after submit. It's just used as
1386 * the basis for the clone(s).
1388 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1389 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1391 ci->bio = &flush_bio;
1392 ci->sector_count = 0;
1393 ci->io->tio.clone.bi_iter.bi_size = 0;
1395 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1396 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1398 bio_uninit(ci->bio);
1401 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1406 len = min_t(sector_t, ci->sector_count,
1407 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1409 __send_duplicate_bios(ci, ti, num_bios, &len);
1412 ci->sector_count -= len;
1415 static bool is_abnormal_io(struct bio *bio)
1419 switch (bio_op(bio)) {
1420 case REQ_OP_DISCARD:
1421 case REQ_OP_SECURE_ERASE:
1422 case REQ_OP_WRITE_ZEROES:
1430 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1433 unsigned num_bios = 0;
1435 switch (bio_op(ci->bio)) {
1436 case REQ_OP_DISCARD:
1437 num_bios = ti->num_discard_bios;
1439 case REQ_OP_SECURE_ERASE:
1440 num_bios = ti->num_secure_erase_bios;
1442 case REQ_OP_WRITE_ZEROES:
1443 num_bios = ti->num_write_zeroes_bios;
1450 * Even though the device advertised support for this type of
1451 * request, that does not mean every target supports it, and
1452 * reconfiguration might also have changed that since the
1453 * check was performed.
1456 *result = -EOPNOTSUPP;
1458 __send_changing_extent_only(ci, ti, num_bios);
1465 * Reuse ->bi_private as hlist head for storing all dm_io instances
1466 * associated with this bio, and this bio's bi_private needs to be
1467 * stored in dm_io->data before the reuse.
1469 * bio->bi_private is owned by fs or upper layer, so block layer won't
1470 * touch it after splitting. Meantime it won't be changed by anyone after
1471 * bio is submitted. So this reuse is safe.
1473 static inline struct hlist_head *dm_get_bio_hlist_head(struct bio *bio)
1475 return (struct hlist_head *)&bio->bi_private;
1478 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1480 struct hlist_head *head = dm_get_bio_hlist_head(bio);
1482 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1483 bio->bi_opf |= REQ_DM_POLL_LIST;
1485 * Save .bi_private into dm_io, so that we can reuse
1486 * .bi_private as hlist head for storing dm_io list
1488 io->data = bio->bi_private;
1490 INIT_HLIST_HEAD(head);
1492 /* tell block layer to poll for completion */
1493 bio->bi_cookie = ~BLK_QC_T_NONE;
1496 * bio recursed due to split, reuse original poll list,
1497 * and save bio->bi_private too.
1499 io->data = hlist_entry(head->first, struct dm_io, node)->data;
1502 hlist_add_head(&io->node, head);
1506 * Select the correct strategy for processing a non-flush bio.
1508 static int __split_and_process_bio(struct clone_info *ci)
1511 struct dm_target *ti;
1515 ti = dm_table_find_target(ci->map, ci->sector);
1519 if (__process_abnormal_io(ci, ti, &r))
1523 * Only support bio polling for normal IO, and the target io is
1524 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1526 ci->submit_as_polled = ci->bio->bi_opf & REQ_POLLED;
1528 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1529 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1533 ci->sector_count -= len;
1538 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1539 struct dm_table *map, struct bio *bio)
1542 ci->io = alloc_io(md, bio);
1544 ci->submit_as_polled = false;
1545 ci->sector = bio->bi_iter.bi_sector;
1546 ci->sector_count = bio_sectors(bio);
1548 /* Shouldn't happen but sector_count was being set to 0 so... */
1549 if (WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1550 ci->sector_count = 0;
1554 * Entry point to split a bio into clones and submit them to the targets.
1556 static void dm_split_and_process_bio(struct mapped_device *md,
1557 struct dm_table *map, struct bio *bio)
1559 struct clone_info ci;
1560 struct bio *orig_bio = NULL;
1563 init_clone_info(&ci, md, map, bio);
1565 if (bio->bi_opf & REQ_PREFLUSH) {
1566 __send_empty_flush(&ci);
1567 /* dm_io_complete submits any data associated with flush */
1571 error = __split_and_process_bio(&ci);
1572 ci.io->map_task = NULL;
1573 if (error || !ci.sector_count)
1577 * Remainder must be passed to submit_bio_noacct() so it gets handled
1578 * *after* bios already submitted have been completely processed.
1579 * We take a clone of the original to store in ci.io->orig_bio to be
1580 * used by dm_end_io_acct() and for dm_io_complete() to use for
1581 * completion handling.
1583 orig_bio = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1584 GFP_NOIO, &md->queue->bio_split);
1585 bio_chain(orig_bio, bio);
1586 trace_block_split(orig_bio, bio->bi_iter.bi_sector);
1587 submit_bio_noacct(bio);
1591 smp_store_release(&ci.io->orig_bio, orig_bio);
1592 if (dm_io_flagged(ci.io, DM_IO_START_ACCT))
1593 dm_start_io_acct(ci.io, NULL);
1596 * Drop the extra reference count for non-POLLED bio, and hold one
1597 * reference for POLLED bio, which will be released in dm_poll_bio
1599 * Add every dm_io instance into the hlist_head which is stored in
1600 * bio->bi_private, so that dm_poll_bio can poll them all.
1602 if (error || !ci.submit_as_polled)
1603 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1605 dm_queue_poll_io(bio, ci.io);
1608 static void dm_submit_bio(struct bio *bio)
1610 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1612 struct dm_table *map;
1614 map = dm_get_live_table(md, &srcu_idx);
1616 /* If suspended, or map not yet available, queue this IO for later */
1617 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1619 if (bio->bi_opf & REQ_NOWAIT)
1620 bio_wouldblock_error(bio);
1621 else if (bio->bi_opf & REQ_RAHEAD)
1629 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1630 * otherwise associated queue_limits won't be imposed.
1632 if (is_abnormal_io(bio))
1633 blk_queue_split(&bio);
1635 dm_split_and_process_bio(md, map, bio);
1637 dm_put_live_table(md, srcu_idx);
1640 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1643 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1645 /* don't poll if the mapped io is done */
1646 if (atomic_read(&io->io_count) > 1)
1647 bio_poll(&io->tio.clone, iob, flags);
1649 /* bio_poll holds the last reference */
1650 return atomic_read(&io->io_count) == 1;
1653 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1656 struct hlist_head *head = dm_get_bio_hlist_head(bio);
1657 struct hlist_head tmp = HLIST_HEAD_INIT;
1658 struct hlist_node *next;
1661 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1662 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1665 WARN_ON_ONCE(hlist_empty(head));
1667 hlist_move_list(head, &tmp);
1670 * Restore .bi_private before possibly completing dm_io.
1672 * bio_poll() is only possible once @bio has been completely
1673 * submitted via submit_bio_noacct()'s depth-first submission.
1674 * So there is no dm_queue_poll_io() race associated with
1675 * clearing REQ_DM_POLL_LIST here.
1677 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1678 bio->bi_private = hlist_entry(tmp.first, struct dm_io, node)->data;
1680 hlist_for_each_entry_safe(io, next, &tmp, node) {
1681 if (dm_poll_dm_io(io, iob, flags)) {
1682 hlist_del_init(&io->node);
1684 * clone_endio() has already occurred, so passing
1685 * error as 0 here doesn't override io->status
1687 dm_io_dec_pending(io, 0);
1692 if (!hlist_empty(&tmp)) {
1693 bio->bi_opf |= REQ_DM_POLL_LIST;
1694 /* Reset bio->bi_private to dm_io list head */
1695 hlist_move_list(&tmp, head);
1701 /*-----------------------------------------------------------------
1702 * An IDR is used to keep track of allocated minor numbers.
1703 *---------------------------------------------------------------*/
1704 static void free_minor(int minor)
1706 spin_lock(&_minor_lock);
1707 idr_remove(&_minor_idr, minor);
1708 spin_unlock(&_minor_lock);
1712 * See if the device with a specific minor # is free.
1714 static int specific_minor(int minor)
1718 if (minor >= (1 << MINORBITS))
1721 idr_preload(GFP_KERNEL);
1722 spin_lock(&_minor_lock);
1724 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1726 spin_unlock(&_minor_lock);
1729 return r == -ENOSPC ? -EBUSY : r;
1733 static int next_free_minor(int *minor)
1737 idr_preload(GFP_KERNEL);
1738 spin_lock(&_minor_lock);
1740 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1742 spin_unlock(&_minor_lock);
1750 static const struct block_device_operations dm_blk_dops;
1751 static const struct block_device_operations dm_rq_blk_dops;
1752 static const struct dax_operations dm_dax_ops;
1754 static void dm_wq_work(struct work_struct *work);
1756 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1757 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1759 dm_destroy_crypto_profile(q->crypto_profile);
1762 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1764 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1767 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1769 static void cleanup_mapped_device(struct mapped_device *md)
1772 destroy_workqueue(md->wq);
1773 bioset_exit(&md->bs);
1774 bioset_exit(&md->io_bs);
1777 dax_remove_host(md->disk);
1778 kill_dax(md->dax_dev);
1779 put_dax(md->dax_dev);
1783 dm_cleanup_zoned_dev(md);
1785 spin_lock(&_minor_lock);
1786 md->disk->private_data = NULL;
1787 spin_unlock(&_minor_lock);
1788 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1790 del_gendisk(md->disk);
1792 dm_queue_destroy_crypto_profile(md->queue);
1793 blk_cleanup_disk(md->disk);
1796 if (md->pending_io) {
1797 free_percpu(md->pending_io);
1798 md->pending_io = NULL;
1801 cleanup_srcu_struct(&md->io_barrier);
1803 mutex_destroy(&md->suspend_lock);
1804 mutex_destroy(&md->type_lock);
1805 mutex_destroy(&md->table_devices_lock);
1806 mutex_destroy(&md->swap_bios_lock);
1808 dm_mq_cleanup_mapped_device(md);
1812 * Allocate and initialise a blank device with a given minor.
1814 static struct mapped_device *alloc_dev(int minor)
1816 int r, numa_node_id = dm_get_numa_node();
1817 struct mapped_device *md;
1820 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1822 DMWARN("unable to allocate device, out of memory.");
1826 if (!try_module_get(THIS_MODULE))
1827 goto bad_module_get;
1829 /* get a minor number for the dev */
1830 if (minor == DM_ANY_MINOR)
1831 r = next_free_minor(&minor);
1833 r = specific_minor(minor);
1837 r = init_srcu_struct(&md->io_barrier);
1839 goto bad_io_barrier;
1841 md->numa_node_id = numa_node_id;
1842 md->init_tio_pdu = false;
1843 md->type = DM_TYPE_NONE;
1844 mutex_init(&md->suspend_lock);
1845 mutex_init(&md->type_lock);
1846 mutex_init(&md->table_devices_lock);
1847 spin_lock_init(&md->deferred_lock);
1848 atomic_set(&md->holders, 1);
1849 atomic_set(&md->open_count, 0);
1850 atomic_set(&md->event_nr, 0);
1851 atomic_set(&md->uevent_seq, 0);
1852 INIT_LIST_HEAD(&md->uevent_list);
1853 INIT_LIST_HEAD(&md->table_devices);
1854 spin_lock_init(&md->uevent_lock);
1857 * default to bio-based until DM table is loaded and md->type
1858 * established. If request-based table is loaded: blk-mq will
1859 * override accordingly.
1861 md->disk = blk_alloc_disk(md->numa_node_id);
1864 md->queue = md->disk->queue;
1866 init_waitqueue_head(&md->wait);
1867 INIT_WORK(&md->work, dm_wq_work);
1868 init_waitqueue_head(&md->eventq);
1869 init_completion(&md->kobj_holder.completion);
1871 md->swap_bios = get_swap_bios();
1872 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1873 mutex_init(&md->swap_bios_lock);
1875 md->disk->major = _major;
1876 md->disk->first_minor = minor;
1877 md->disk->minors = 1;
1878 md->disk->flags |= GENHD_FL_NO_PART;
1879 md->disk->fops = &dm_blk_dops;
1880 md->disk->queue = md->queue;
1881 md->disk->private_data = md;
1882 sprintf(md->disk->disk_name, "dm-%d", minor);
1884 if (IS_ENABLED(CONFIG_FS_DAX)) {
1885 md->dax_dev = alloc_dax(md, &dm_dax_ops);
1886 if (IS_ERR(md->dax_dev)) {
1890 set_dax_nocache(md->dax_dev);
1891 set_dax_nomc(md->dax_dev);
1892 if (dax_add_host(md->dax_dev, md->disk))
1896 format_dev_t(md->name, MKDEV(_major, minor));
1898 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
1902 md->pending_io = alloc_percpu(unsigned long);
1903 if (!md->pending_io)
1906 dm_stats_init(&md->stats);
1908 /* Populate the mapping, nobody knows we exist yet */
1909 spin_lock(&_minor_lock);
1910 old_md = idr_replace(&_minor_idr, md, minor);
1911 spin_unlock(&_minor_lock);
1913 BUG_ON(old_md != MINOR_ALLOCED);
1918 cleanup_mapped_device(md);
1922 module_put(THIS_MODULE);
1928 static void unlock_fs(struct mapped_device *md);
1930 static void free_dev(struct mapped_device *md)
1932 int minor = MINOR(disk_devt(md->disk));
1936 cleanup_mapped_device(md);
1938 free_table_devices(&md->table_devices);
1939 dm_stats_cleanup(&md->stats);
1942 module_put(THIS_MODULE);
1946 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1948 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1951 if (dm_table_bio_based(t)) {
1953 * The md may already have mempools that need changing.
1954 * If so, reload bioset because front_pad may have changed
1955 * because a different table was loaded.
1957 bioset_exit(&md->bs);
1958 bioset_exit(&md->io_bs);
1960 } else if (bioset_initialized(&md->bs)) {
1962 * There's no need to reload with request-based dm
1963 * because the size of front_pad doesn't change.
1964 * Note for future: If you are to reload bioset,
1965 * prep-ed requests in the queue may refer
1966 * to bio from the old bioset, so you must walk
1967 * through the queue to unprep.
1973 bioset_initialized(&md->bs) ||
1974 bioset_initialized(&md->io_bs));
1976 ret = bioset_init_from_src(&md->bs, &p->bs);
1979 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1981 bioset_exit(&md->bs);
1983 /* mempool bind completed, no longer need any mempools in the table */
1984 dm_table_free_md_mempools(t);
1989 * Bind a table to the device.
1991 static void event_callback(void *context)
1993 unsigned long flags;
1995 struct mapped_device *md = (struct mapped_device *) context;
1997 spin_lock_irqsave(&md->uevent_lock, flags);
1998 list_splice_init(&md->uevent_list, &uevents);
1999 spin_unlock_irqrestore(&md->uevent_lock, flags);
2001 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2003 atomic_inc(&md->event_nr);
2004 wake_up(&md->eventq);
2005 dm_issue_global_event();
2009 * Returns old map, which caller must destroy.
2011 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2012 struct queue_limits *limits)
2014 struct dm_table *old_map;
2018 lockdep_assert_held(&md->suspend_lock);
2020 size = dm_table_get_size(t);
2023 * Wipe any geometry if the size of the table changed.
2025 if (size != dm_get_size(md))
2026 memset(&md->geometry, 0, sizeof(md->geometry));
2028 if (!get_capacity(md->disk))
2029 set_capacity(md->disk, size);
2031 set_capacity_and_notify(md->disk, size);
2033 dm_table_event_callback(t, event_callback, md);
2035 if (dm_table_request_based(t)) {
2037 * Leverage the fact that request-based DM targets are
2038 * immutable singletons - used to optimize dm_mq_queue_rq.
2040 md->immutable_target = dm_table_get_immutable_target(t);
2043 ret = __bind_mempools(md, t);
2045 old_map = ERR_PTR(ret);
2049 ret = dm_table_set_restrictions(t, md->queue, limits);
2051 old_map = ERR_PTR(ret);
2055 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2056 rcu_assign_pointer(md->map, (void *)t);
2057 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2066 * Returns unbound table for the caller to free.
2068 static struct dm_table *__unbind(struct mapped_device *md)
2070 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2075 dm_table_event_callback(map, NULL, NULL);
2076 RCU_INIT_POINTER(md->map, NULL);
2083 * Constructor for a new device.
2085 int dm_create(int minor, struct mapped_device **result)
2087 struct mapped_device *md;
2089 md = alloc_dev(minor);
2093 dm_ima_reset_data(md);
2100 * Functions to manage md->type.
2101 * All are required to hold md->type_lock.
2103 void dm_lock_md_type(struct mapped_device *md)
2105 mutex_lock(&md->type_lock);
2108 void dm_unlock_md_type(struct mapped_device *md)
2110 mutex_unlock(&md->type_lock);
2113 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2115 BUG_ON(!mutex_is_locked(&md->type_lock));
2119 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2124 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2126 return md->immutable_target_type;
2130 * The queue_limits are only valid as long as you have a reference
2133 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2135 BUG_ON(!atomic_read(&md->holders));
2136 return &md->queue->limits;
2138 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2141 * Setup the DM device's queue based on md's type
2143 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2145 enum dm_queue_mode type = dm_table_get_type(t);
2146 struct queue_limits limits;
2150 case DM_TYPE_REQUEST_BASED:
2151 md->disk->fops = &dm_rq_blk_dops;
2152 r = dm_mq_init_request_queue(md, t);
2154 DMERR("Cannot initialize queue for request-based dm mapped device");
2158 case DM_TYPE_BIO_BASED:
2159 case DM_TYPE_DAX_BIO_BASED:
2166 r = dm_calculate_queue_limits(t, &limits);
2168 DMERR("Cannot calculate initial queue limits");
2171 r = dm_table_set_restrictions(t, md->queue, &limits);
2175 r = add_disk(md->disk);
2179 r = dm_sysfs_init(md);
2181 del_gendisk(md->disk);
2188 struct mapped_device *dm_get_md(dev_t dev)
2190 struct mapped_device *md;
2191 unsigned minor = MINOR(dev);
2193 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2196 spin_lock(&_minor_lock);
2198 md = idr_find(&_minor_idr, minor);
2199 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2200 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2206 spin_unlock(&_minor_lock);
2210 EXPORT_SYMBOL_GPL(dm_get_md);
2212 void *dm_get_mdptr(struct mapped_device *md)
2214 return md->interface_ptr;
2217 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2219 md->interface_ptr = ptr;
2222 void dm_get(struct mapped_device *md)
2224 atomic_inc(&md->holders);
2225 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2228 int dm_hold(struct mapped_device *md)
2230 spin_lock(&_minor_lock);
2231 if (test_bit(DMF_FREEING, &md->flags)) {
2232 spin_unlock(&_minor_lock);
2236 spin_unlock(&_minor_lock);
2239 EXPORT_SYMBOL_GPL(dm_hold);
2241 const char *dm_device_name(struct mapped_device *md)
2245 EXPORT_SYMBOL_GPL(dm_device_name);
2247 static void __dm_destroy(struct mapped_device *md, bool wait)
2249 struct dm_table *map;
2254 spin_lock(&_minor_lock);
2255 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2256 set_bit(DMF_FREEING, &md->flags);
2257 spin_unlock(&_minor_lock);
2259 blk_mark_disk_dead(md->disk);
2262 * Take suspend_lock so that presuspend and postsuspend methods
2263 * do not race with internal suspend.
2265 mutex_lock(&md->suspend_lock);
2266 map = dm_get_live_table(md, &srcu_idx);
2267 if (!dm_suspended_md(md)) {
2268 dm_table_presuspend_targets(map);
2269 set_bit(DMF_SUSPENDED, &md->flags);
2270 set_bit(DMF_POST_SUSPENDING, &md->flags);
2271 dm_table_postsuspend_targets(map);
2273 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2274 dm_put_live_table(md, srcu_idx);
2275 mutex_unlock(&md->suspend_lock);
2278 * Rare, but there may be I/O requests still going to complete,
2279 * for example. Wait for all references to disappear.
2280 * No one should increment the reference count of the mapped_device,
2281 * after the mapped_device state becomes DMF_FREEING.
2284 while (atomic_read(&md->holders))
2286 else if (atomic_read(&md->holders))
2287 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2288 dm_device_name(md), atomic_read(&md->holders));
2290 dm_table_destroy(__unbind(md));
2294 void dm_destroy(struct mapped_device *md)
2296 __dm_destroy(md, true);
2299 void dm_destroy_immediate(struct mapped_device *md)
2301 __dm_destroy(md, false);
2304 void dm_put(struct mapped_device *md)
2306 atomic_dec(&md->holders);
2308 EXPORT_SYMBOL_GPL(dm_put);
2310 static bool dm_in_flight_bios(struct mapped_device *md)
2313 unsigned long sum = 0;
2315 for_each_possible_cpu(cpu)
2316 sum += *per_cpu_ptr(md->pending_io, cpu);
2321 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2327 prepare_to_wait(&md->wait, &wait, task_state);
2329 if (!dm_in_flight_bios(md))
2332 if (signal_pending_state(task_state, current)) {
2339 finish_wait(&md->wait, &wait);
2346 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2350 if (!queue_is_mq(md->queue))
2351 return dm_wait_for_bios_completion(md, task_state);
2354 if (!blk_mq_queue_inflight(md->queue))
2357 if (signal_pending_state(task_state, current)) {
2369 * Process the deferred bios
2371 static void dm_wq_work(struct work_struct *work)
2373 struct mapped_device *md = container_of(work, struct mapped_device, work);
2376 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2377 spin_lock_irq(&md->deferred_lock);
2378 bio = bio_list_pop(&md->deferred);
2379 spin_unlock_irq(&md->deferred_lock);
2384 submit_bio_noacct(bio);
2388 static void dm_queue_flush(struct mapped_device *md)
2390 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2391 smp_mb__after_atomic();
2392 queue_work(md->wq, &md->work);
2396 * Swap in a new table, returning the old one for the caller to destroy.
2398 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2400 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2401 struct queue_limits limits;
2404 mutex_lock(&md->suspend_lock);
2406 /* device must be suspended */
2407 if (!dm_suspended_md(md))
2411 * If the new table has no data devices, retain the existing limits.
2412 * This helps multipath with queue_if_no_path if all paths disappear,
2413 * then new I/O is queued based on these limits, and then some paths
2416 if (dm_table_has_no_data_devices(table)) {
2417 live_map = dm_get_live_table_fast(md);
2419 limits = md->queue->limits;
2420 dm_put_live_table_fast(md);
2424 r = dm_calculate_queue_limits(table, &limits);
2431 map = __bind(md, table, &limits);
2432 dm_issue_global_event();
2435 mutex_unlock(&md->suspend_lock);
2440 * Functions to lock and unlock any filesystem running on the
2443 static int lock_fs(struct mapped_device *md)
2447 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2449 r = freeze_bdev(md->disk->part0);
2451 set_bit(DMF_FROZEN, &md->flags);
2455 static void unlock_fs(struct mapped_device *md)
2457 if (!test_bit(DMF_FROZEN, &md->flags))
2459 thaw_bdev(md->disk->part0);
2460 clear_bit(DMF_FROZEN, &md->flags);
2464 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2465 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2466 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2468 * If __dm_suspend returns 0, the device is completely quiescent
2469 * now. There is no request-processing activity. All new requests
2470 * are being added to md->deferred list.
2472 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2473 unsigned suspend_flags, unsigned int task_state,
2474 int dmf_suspended_flag)
2476 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2477 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2480 lockdep_assert_held(&md->suspend_lock);
2483 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2484 * This flag is cleared before dm_suspend returns.
2487 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2489 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2492 * This gets reverted if there's an error later and the targets
2493 * provide the .presuspend_undo hook.
2495 dm_table_presuspend_targets(map);
2498 * Flush I/O to the device.
2499 * Any I/O submitted after lock_fs() may not be flushed.
2500 * noflush takes precedence over do_lockfs.
2501 * (lock_fs() flushes I/Os and waits for them to complete.)
2503 if (!noflush && do_lockfs) {
2506 dm_table_presuspend_undo_targets(map);
2512 * Here we must make sure that no processes are submitting requests
2513 * to target drivers i.e. no one may be executing
2514 * dm_split_and_process_bio from dm_submit_bio.
2516 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2517 * we take the write lock. To prevent any process from reentering
2518 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2519 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2520 * flush_workqueue(md->wq).
2522 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2524 synchronize_srcu(&md->io_barrier);
2527 * Stop md->queue before flushing md->wq in case request-based
2528 * dm defers requests to md->wq from md->queue.
2530 if (dm_request_based(md))
2531 dm_stop_queue(md->queue);
2533 flush_workqueue(md->wq);
2536 * At this point no more requests are entering target request routines.
2537 * We call dm_wait_for_completion to wait for all existing requests
2540 r = dm_wait_for_completion(md, task_state);
2542 set_bit(dmf_suspended_flag, &md->flags);
2545 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2547 synchronize_srcu(&md->io_barrier);
2549 /* were we interrupted ? */
2553 if (dm_request_based(md))
2554 dm_start_queue(md->queue);
2557 dm_table_presuspend_undo_targets(map);
2558 /* pushback list is already flushed, so skip flush */
2565 * We need to be able to change a mapping table under a mounted
2566 * filesystem. For example we might want to move some data in
2567 * the background. Before the table can be swapped with
2568 * dm_bind_table, dm_suspend must be called to flush any in
2569 * flight bios and ensure that any further io gets deferred.
2572 * Suspend mechanism in request-based dm.
2574 * 1. Flush all I/Os by lock_fs() if needed.
2575 * 2. Stop dispatching any I/O by stopping the request_queue.
2576 * 3. Wait for all in-flight I/Os to be completed or requeued.
2578 * To abort suspend, start the request_queue.
2580 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2582 struct dm_table *map = NULL;
2586 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2588 if (dm_suspended_md(md)) {
2593 if (dm_suspended_internally_md(md)) {
2594 /* already internally suspended, wait for internal resume */
2595 mutex_unlock(&md->suspend_lock);
2596 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2602 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2604 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2608 set_bit(DMF_POST_SUSPENDING, &md->flags);
2609 dm_table_postsuspend_targets(map);
2610 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2613 mutex_unlock(&md->suspend_lock);
2617 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2620 int r = dm_table_resume_targets(map);
2628 * Flushing deferred I/Os must be done after targets are resumed
2629 * so that mapping of targets can work correctly.
2630 * Request-based dm is queueing the deferred I/Os in its request_queue.
2632 if (dm_request_based(md))
2633 dm_start_queue(md->queue);
2640 int dm_resume(struct mapped_device *md)
2643 struct dm_table *map = NULL;
2647 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2649 if (!dm_suspended_md(md))
2652 if (dm_suspended_internally_md(md)) {
2653 /* already internally suspended, wait for internal resume */
2654 mutex_unlock(&md->suspend_lock);
2655 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2661 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2662 if (!map || !dm_table_get_size(map))
2665 r = __dm_resume(md, map);
2669 clear_bit(DMF_SUSPENDED, &md->flags);
2671 mutex_unlock(&md->suspend_lock);
2677 * Internal suspend/resume works like userspace-driven suspend. It waits
2678 * until all bios finish and prevents issuing new bios to the target drivers.
2679 * It may be used only from the kernel.
2682 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2684 struct dm_table *map = NULL;
2686 lockdep_assert_held(&md->suspend_lock);
2688 if (md->internal_suspend_count++)
2689 return; /* nested internal suspend */
2691 if (dm_suspended_md(md)) {
2692 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2693 return; /* nest suspend */
2696 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2699 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2700 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2701 * would require changing .presuspend to return an error -- avoid this
2702 * until there is a need for more elaborate variants of internal suspend.
2704 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2705 DMF_SUSPENDED_INTERNALLY);
2707 set_bit(DMF_POST_SUSPENDING, &md->flags);
2708 dm_table_postsuspend_targets(map);
2709 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2712 static void __dm_internal_resume(struct mapped_device *md)
2714 BUG_ON(!md->internal_suspend_count);
2716 if (--md->internal_suspend_count)
2717 return; /* resume from nested internal suspend */
2719 if (dm_suspended_md(md))
2720 goto done; /* resume from nested suspend */
2723 * NOTE: existing callers don't need to call dm_table_resume_targets
2724 * (which may fail -- so best to avoid it for now by passing NULL map)
2726 (void) __dm_resume(md, NULL);
2729 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2730 smp_mb__after_atomic();
2731 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2734 void dm_internal_suspend_noflush(struct mapped_device *md)
2736 mutex_lock(&md->suspend_lock);
2737 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2738 mutex_unlock(&md->suspend_lock);
2740 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2742 void dm_internal_resume(struct mapped_device *md)
2744 mutex_lock(&md->suspend_lock);
2745 __dm_internal_resume(md);
2746 mutex_unlock(&md->suspend_lock);
2748 EXPORT_SYMBOL_GPL(dm_internal_resume);
2751 * Fast variants of internal suspend/resume hold md->suspend_lock,
2752 * which prevents interaction with userspace-driven suspend.
2755 void dm_internal_suspend_fast(struct mapped_device *md)
2757 mutex_lock(&md->suspend_lock);
2758 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2761 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2762 synchronize_srcu(&md->io_barrier);
2763 flush_workqueue(md->wq);
2764 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2766 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2768 void dm_internal_resume_fast(struct mapped_device *md)
2770 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2776 mutex_unlock(&md->suspend_lock);
2778 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2780 /*-----------------------------------------------------------------
2781 * Event notification.
2782 *---------------------------------------------------------------*/
2783 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2788 char udev_cookie[DM_COOKIE_LENGTH];
2789 char *envp[] = { udev_cookie, NULL };
2791 noio_flag = memalloc_noio_save();
2794 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2796 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2797 DM_COOKIE_ENV_VAR_NAME, cookie);
2798 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2802 memalloc_noio_restore(noio_flag);
2807 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2809 return atomic_add_return(1, &md->uevent_seq);
2812 uint32_t dm_get_event_nr(struct mapped_device *md)
2814 return atomic_read(&md->event_nr);
2817 int dm_wait_event(struct mapped_device *md, int event_nr)
2819 return wait_event_interruptible(md->eventq,
2820 (event_nr != atomic_read(&md->event_nr)));
2823 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2825 unsigned long flags;
2827 spin_lock_irqsave(&md->uevent_lock, flags);
2828 list_add(elist, &md->uevent_list);
2829 spin_unlock_irqrestore(&md->uevent_lock, flags);
2833 * The gendisk is only valid as long as you have a reference
2836 struct gendisk *dm_disk(struct mapped_device *md)
2840 EXPORT_SYMBOL_GPL(dm_disk);
2842 struct kobject *dm_kobject(struct mapped_device *md)
2844 return &md->kobj_holder.kobj;
2847 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2849 struct mapped_device *md;
2851 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2853 spin_lock(&_minor_lock);
2854 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2860 spin_unlock(&_minor_lock);
2865 int dm_suspended_md(struct mapped_device *md)
2867 return test_bit(DMF_SUSPENDED, &md->flags);
2870 static int dm_post_suspending_md(struct mapped_device *md)
2872 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2875 int dm_suspended_internally_md(struct mapped_device *md)
2877 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2880 int dm_test_deferred_remove_flag(struct mapped_device *md)
2882 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2885 int dm_suspended(struct dm_target *ti)
2887 return dm_suspended_md(ti->table->md);
2889 EXPORT_SYMBOL_GPL(dm_suspended);
2891 int dm_post_suspending(struct dm_target *ti)
2893 return dm_post_suspending_md(ti->table->md);
2895 EXPORT_SYMBOL_GPL(dm_post_suspending);
2897 int dm_noflush_suspending(struct dm_target *ti)
2899 return __noflush_suspending(ti->table->md);
2901 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2903 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2904 unsigned integrity, unsigned per_io_data_size,
2905 unsigned min_pool_size)
2907 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2908 unsigned int pool_size = 0;
2909 unsigned int front_pad, io_front_pad;
2916 case DM_TYPE_BIO_BASED:
2917 case DM_TYPE_DAX_BIO_BASED:
2918 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2919 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2920 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2921 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2924 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2927 case DM_TYPE_REQUEST_BASED:
2928 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2929 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2930 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2936 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2940 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2946 dm_free_md_mempools(pools);
2951 void dm_free_md_mempools(struct dm_md_mempools *pools)
2956 bioset_exit(&pools->bs);
2957 bioset_exit(&pools->io_bs);
2969 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2972 struct mapped_device *md = bdev->bd_disk->private_data;
2973 struct dm_table *table;
2974 struct dm_target *ti;
2975 int ret = -ENOTTY, srcu_idx;
2977 table = dm_get_live_table(md, &srcu_idx);
2978 if (!table || !dm_table_get_size(table))
2981 /* We only support devices that have a single target */
2982 if (dm_table_get_num_targets(table) != 1)
2984 ti = dm_table_get_target(table, 0);
2987 if (!ti->type->iterate_devices)
2990 ret = ti->type->iterate_devices(ti, fn, data);
2992 dm_put_live_table(md, srcu_idx);
2997 * For register / unregister we need to manually call out to every path.
2999 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3000 sector_t start, sector_t len, void *data)
3002 struct dm_pr *pr = data;
3003 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3005 if (!ops || !ops->pr_register)
3007 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3010 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3021 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3022 if (ret && new_key) {
3023 /* unregister all paths if we failed to register any path */
3024 pr.old_key = new_key;
3027 pr.fail_early = false;
3028 dm_call_pr(bdev, __dm_pr_register, &pr);
3034 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3037 struct mapped_device *md = bdev->bd_disk->private_data;
3038 const struct pr_ops *ops;
3041 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3045 ops = bdev->bd_disk->fops->pr_ops;
3046 if (ops && ops->pr_reserve)
3047 r = ops->pr_reserve(bdev, key, type, flags);
3051 dm_unprepare_ioctl(md, srcu_idx);
3055 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3057 struct mapped_device *md = bdev->bd_disk->private_data;
3058 const struct pr_ops *ops;
3061 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3065 ops = bdev->bd_disk->fops->pr_ops;
3066 if (ops && ops->pr_release)
3067 r = ops->pr_release(bdev, key, type);
3071 dm_unprepare_ioctl(md, srcu_idx);
3075 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3076 enum pr_type type, bool abort)
3078 struct mapped_device *md = bdev->bd_disk->private_data;
3079 const struct pr_ops *ops;
3082 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3086 ops = bdev->bd_disk->fops->pr_ops;
3087 if (ops && ops->pr_preempt)
3088 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3092 dm_unprepare_ioctl(md, srcu_idx);
3096 static int dm_pr_clear(struct block_device *bdev, u64 key)
3098 struct mapped_device *md = bdev->bd_disk->private_data;
3099 const struct pr_ops *ops;
3102 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3106 ops = bdev->bd_disk->fops->pr_ops;
3107 if (ops && ops->pr_clear)
3108 r = ops->pr_clear(bdev, key);
3112 dm_unprepare_ioctl(md, srcu_idx);
3116 static const struct pr_ops dm_pr_ops = {
3117 .pr_register = dm_pr_register,
3118 .pr_reserve = dm_pr_reserve,
3119 .pr_release = dm_pr_release,
3120 .pr_preempt = dm_pr_preempt,
3121 .pr_clear = dm_pr_clear,
3124 static const struct block_device_operations dm_blk_dops = {
3125 .submit_bio = dm_submit_bio,
3126 .poll_bio = dm_poll_bio,
3127 .open = dm_blk_open,
3128 .release = dm_blk_close,
3129 .ioctl = dm_blk_ioctl,
3130 .getgeo = dm_blk_getgeo,
3131 .report_zones = dm_blk_report_zones,
3132 .pr_ops = &dm_pr_ops,
3133 .owner = THIS_MODULE
3136 static const struct block_device_operations dm_rq_blk_dops = {
3137 .open = dm_blk_open,
3138 .release = dm_blk_close,
3139 .ioctl = dm_blk_ioctl,
3140 .getgeo = dm_blk_getgeo,
3141 .pr_ops = &dm_pr_ops,
3142 .owner = THIS_MODULE
3145 static const struct dax_operations dm_dax_ops = {
3146 .direct_access = dm_dax_direct_access,
3147 .zero_page_range = dm_dax_zero_page_range,
3153 module_init(dm_init);
3154 module_exit(dm_exit);
3156 module_param(major, uint, 0);
3157 MODULE_PARM_DESC(major, "The major number of the device mapper");
3159 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3160 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3162 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3163 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3165 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3166 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3168 MODULE_DESCRIPTION(DM_NAME " driver");
3169 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3170 MODULE_LICENSE("GPL");