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 int sector_count;
87 bool is_abnormal_io:1;
88 bool submit_as_polled:1;
91 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
93 return container_of(clone, struct dm_target_io, clone);
96 void *dm_per_bio_data(struct bio *bio, size_t data_size)
98 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
99 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
100 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
102 EXPORT_SYMBOL_GPL(dm_per_bio_data);
104 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
106 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
107 if (io->magic == DM_IO_MAGIC)
108 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
109 BUG_ON(io->magic != DM_TIO_MAGIC);
110 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
112 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
114 unsigned int dm_bio_get_target_bio_nr(const struct bio *bio)
116 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
118 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
120 #define MINOR_ALLOCED ((void *)-1)
122 #define DM_NUMA_NODE NUMA_NO_NODE
123 static int dm_numa_node = DM_NUMA_NODE;
125 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
126 static int swap_bios = DEFAULT_SWAP_BIOS;
127 static int get_swap_bios(void)
129 int latch = READ_ONCE(swap_bios);
130 if (unlikely(latch <= 0))
131 latch = DEFAULT_SWAP_BIOS;
135 struct table_device {
136 struct list_head list;
138 struct dm_dev dm_dev;
142 * Bio-based DM's mempools' reserved IOs set by the user.
144 #define RESERVED_BIO_BASED_IOS 16
145 static unsigned int reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
147 static int __dm_get_module_param_int(int *module_param, int min, int max)
149 int param = READ_ONCE(*module_param);
150 int modified_param = 0;
151 bool modified = true;
154 modified_param = min;
155 else if (param > max)
156 modified_param = max;
161 (void)cmpxchg(module_param, param, modified_param);
162 param = modified_param;
168 unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max)
170 unsigned int param = READ_ONCE(*module_param);
171 unsigned int modified_param = 0;
174 modified_param = def;
175 else if (param > max)
176 modified_param = max;
178 if (modified_param) {
179 (void)cmpxchg(module_param, param, modified_param);
180 param = modified_param;
186 unsigned int dm_get_reserved_bio_based_ios(void)
188 return __dm_get_module_param(&reserved_bio_based_ios,
189 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
191 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
193 static unsigned int dm_get_numa_node(void)
195 return __dm_get_module_param_int(&dm_numa_node,
196 DM_NUMA_NODE, num_online_nodes() - 1);
199 static int __init local_init(void)
203 r = dm_uevent_init();
207 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
208 if (!deferred_remove_workqueue) {
210 goto out_uevent_exit;
214 r = register_blkdev(_major, _name);
216 goto out_free_workqueue;
224 destroy_workqueue(deferred_remove_workqueue);
231 static void local_exit(void)
233 destroy_workqueue(deferred_remove_workqueue);
235 unregister_blkdev(_major, _name);
240 DMINFO("cleaned up");
243 static int (*_inits[])(void) __initdata = {
254 static void (*_exits[])(void) = {
265 static int __init dm_init(void)
267 const int count = ARRAY_SIZE(_inits);
270 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
271 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
272 " Duplicate IMA measurements will not be recorded in the IMA log.");
275 for (i = 0; i < count; i++) {
289 static void __exit dm_exit(void)
291 int i = ARRAY_SIZE(_exits);
297 * Should be empty by this point.
299 idr_destroy(&_minor_idr);
303 * Block device functions
305 int dm_deleting_md(struct mapped_device *md)
307 return test_bit(DMF_DELETING, &md->flags);
310 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
312 struct mapped_device *md;
314 spin_lock(&_minor_lock);
316 md = bdev->bd_disk->private_data;
320 if (test_bit(DMF_FREEING, &md->flags) ||
321 dm_deleting_md(md)) {
327 atomic_inc(&md->open_count);
329 spin_unlock(&_minor_lock);
331 return md ? 0 : -ENXIO;
334 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
336 struct mapped_device *md;
338 spin_lock(&_minor_lock);
340 md = disk->private_data;
344 if (atomic_dec_and_test(&md->open_count) &&
345 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
346 queue_work(deferred_remove_workqueue, &deferred_remove_work);
350 spin_unlock(&_minor_lock);
353 int dm_open_count(struct mapped_device *md)
355 return atomic_read(&md->open_count);
359 * Guarantees nothing is using the device before it's deleted.
361 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
365 spin_lock(&_minor_lock);
367 if (dm_open_count(md)) {
370 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
371 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
374 set_bit(DMF_DELETING, &md->flags);
376 spin_unlock(&_minor_lock);
381 int dm_cancel_deferred_remove(struct mapped_device *md)
385 spin_lock(&_minor_lock);
387 if (test_bit(DMF_DELETING, &md->flags))
390 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
392 spin_unlock(&_minor_lock);
397 static void do_deferred_remove(struct work_struct *w)
399 dm_deferred_remove();
402 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
404 struct mapped_device *md = bdev->bd_disk->private_data;
406 return dm_get_geometry(md, geo);
409 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
410 struct block_device **bdev)
412 struct dm_target *ti;
413 struct dm_table *map;
418 map = dm_get_live_table(md, srcu_idx);
419 if (!map || !dm_table_get_size(map))
422 /* We only support devices that have a single target */
423 if (map->num_targets != 1)
426 ti = dm_table_get_target(map, 0);
427 if (!ti->type->prepare_ioctl)
430 if (dm_suspended_md(md))
433 r = ti->type->prepare_ioctl(ti, bdev);
434 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
435 dm_put_live_table(md, *srcu_idx);
443 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
445 dm_put_live_table(md, srcu_idx);
448 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
449 unsigned int cmd, unsigned long arg)
451 struct mapped_device *md = bdev->bd_disk->private_data;
454 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
460 * Target determined this ioctl is being issued against a
461 * subset of the parent bdev; require extra privileges.
463 if (!capable(CAP_SYS_RAWIO)) {
465 "%s: sending ioctl %x to DM device without required privilege.",
472 if (!bdev->bd_disk->fops->ioctl)
475 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
477 dm_unprepare_ioctl(md, srcu_idx);
481 u64 dm_start_time_ns_from_clone(struct bio *bio)
483 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
485 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
487 static bool bio_is_flush_with_data(struct bio *bio)
489 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
492 static void dm_io_acct(struct dm_io *io, bool end)
494 struct dm_stats_aux *stats_aux = &io->stats_aux;
495 unsigned long start_time = io->start_time;
496 struct mapped_device *md = io->md;
497 struct bio *bio = io->orig_bio;
498 unsigned int sectors;
501 * If REQ_PREFLUSH set, don't account payload, it will be
502 * submitted (and accounted) after this flush completes.
504 if (bio_is_flush_with_data(bio))
506 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
507 sectors = bio_sectors(bio);
509 sectors = io->sectors;
512 bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
515 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
517 if (static_branch_unlikely(&stats_enabled) &&
518 unlikely(dm_stats_used(&md->stats))) {
521 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
522 sector = bio->bi_iter.bi_sector;
524 sector = bio_end_sector(bio) - io->sector_offset;
526 dm_stats_account_io(&md->stats, bio_data_dir(bio),
528 end, start_time, stats_aux);
532 static void __dm_start_io_acct(struct dm_io *io)
534 dm_io_acct(io, false);
537 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
540 * Ensure IO accounting is only ever started once.
542 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
545 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
546 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
547 dm_io_set_flag(io, DM_IO_ACCOUNTED);
550 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
551 spin_lock_irqsave(&io->lock, flags);
552 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
553 spin_unlock_irqrestore(&io->lock, flags);
556 dm_io_set_flag(io, DM_IO_ACCOUNTED);
557 spin_unlock_irqrestore(&io->lock, flags);
560 __dm_start_io_acct(io);
563 static void dm_end_io_acct(struct dm_io *io)
565 dm_io_acct(io, true);
568 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
571 struct dm_target_io *tio;
574 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
575 tio = clone_to_tio(clone);
577 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
580 io = container_of(tio, struct dm_io, tio);
581 io->magic = DM_IO_MAGIC;
582 io->status = BLK_STS_OK;
584 /* one ref is for submission, the other is for completion */
585 atomic_set(&io->io_count, 2);
586 this_cpu_inc(*md->pending_io);
589 spin_lock_init(&io->lock);
590 io->start_time = jiffies;
593 if (static_branch_unlikely(&stats_enabled))
594 dm_stats_record_start(&md->stats, &io->stats_aux);
599 static void free_io(struct dm_io *io)
601 bio_put(&io->tio.clone);
604 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
605 unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask)
607 struct mapped_device *md = ci->io->md;
608 struct dm_target_io *tio;
611 if (!ci->io->tio.io) {
612 /* the dm_target_io embedded in ci->io is available */
614 /* alloc_io() already initialized embedded clone */
617 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
622 /* REQ_DM_POLL_LIST shouldn't be inherited */
623 clone->bi_opf &= ~REQ_DM_POLL_LIST;
625 tio = clone_to_tio(clone);
626 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
629 tio->magic = DM_TIO_MAGIC;
632 tio->target_bio_nr = target_bio_nr;
636 /* Set default bdev, but target must bio_set_dev() before issuing IO */
637 clone->bi_bdev = md->disk->part0;
638 if (unlikely(ti->needs_bio_set_dev))
639 bio_set_dev(clone, md->disk->part0);
642 clone->bi_iter.bi_size = to_bytes(*len);
643 if (bio_integrity(clone))
644 bio_integrity_trim(clone);
650 static void free_tio(struct bio *clone)
652 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
658 * Add the bio to the list of deferred io.
660 static void queue_io(struct mapped_device *md, struct bio *bio)
664 spin_lock_irqsave(&md->deferred_lock, flags);
665 bio_list_add(&md->deferred, bio);
666 spin_unlock_irqrestore(&md->deferred_lock, flags);
667 queue_work(md->wq, &md->work);
671 * Everyone (including functions in this file), should use this
672 * function to access the md->map field, and make sure they call
673 * dm_put_live_table() when finished.
675 struct dm_table *dm_get_live_table(struct mapped_device *md,
676 int *srcu_idx) __acquires(md->io_barrier)
678 *srcu_idx = srcu_read_lock(&md->io_barrier);
680 return srcu_dereference(md->map, &md->io_barrier);
683 void dm_put_live_table(struct mapped_device *md,
684 int srcu_idx) __releases(md->io_barrier)
686 srcu_read_unlock(&md->io_barrier, srcu_idx);
689 void dm_sync_table(struct mapped_device *md)
691 synchronize_srcu(&md->io_barrier);
692 synchronize_rcu_expedited();
696 * A fast alternative to dm_get_live_table/dm_put_live_table.
697 * The caller must not block between these two functions.
699 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
702 return rcu_dereference(md->map);
705 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
710 static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
711 int *srcu_idx, blk_opf_t bio_opf)
713 if (bio_opf & REQ_NOWAIT)
714 return dm_get_live_table_fast(md);
716 return dm_get_live_table(md, srcu_idx);
719 static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
722 if (bio_opf & REQ_NOWAIT)
723 dm_put_live_table_fast(md);
725 dm_put_live_table(md, srcu_idx);
728 static char *_dm_claim_ptr = "I belong to device-mapper";
731 * Open a table device so we can use it as a map destination.
733 static struct table_device *open_table_device(struct mapped_device *md,
734 dev_t dev, fmode_t mode)
736 struct table_device *td;
737 struct block_device *bdev;
741 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
743 return ERR_PTR(-ENOMEM);
744 refcount_set(&td->count, 1);
746 bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr);
753 * We can be called before the dm disk is added. In that case we can't
754 * register the holder relation here. It will be done once add_disk was
757 if (md->disk->slave_dir) {
758 r = bd_link_disk_holder(bdev, md->disk);
763 td->dm_dev.mode = mode;
764 td->dm_dev.bdev = bdev;
765 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
766 format_dev_t(td->dm_dev.name, dev);
767 list_add(&td->list, &md->table_devices);
771 blkdev_put(bdev, mode | FMODE_EXCL);
778 * Close a table device that we've been using.
780 static void close_table_device(struct table_device *td, struct mapped_device *md)
782 if (md->disk->slave_dir)
783 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
784 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
785 put_dax(td->dm_dev.dax_dev);
790 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
793 struct table_device *td;
795 list_for_each_entry(td, l, list)
796 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
802 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
803 struct dm_dev **result)
805 struct table_device *td;
807 mutex_lock(&md->table_devices_lock);
808 td = find_table_device(&md->table_devices, dev, mode);
810 td = open_table_device(md, dev, mode);
812 mutex_unlock(&md->table_devices_lock);
816 refcount_inc(&td->count);
818 mutex_unlock(&md->table_devices_lock);
820 *result = &td->dm_dev;
824 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
826 struct table_device *td = container_of(d, struct table_device, dm_dev);
828 mutex_lock(&md->table_devices_lock);
829 if (refcount_dec_and_test(&td->count))
830 close_table_device(td, md);
831 mutex_unlock(&md->table_devices_lock);
834 static void free_table_devices(struct list_head *devices)
836 struct list_head *tmp, *next;
838 list_for_each_safe(tmp, next, devices) {
839 struct table_device *td = list_entry(tmp, struct table_device, list);
841 DMWARN("dm_destroy: %s still exists with %d references",
842 td->dm_dev.name, refcount_read(&td->count));
848 * Get the geometry associated with a dm device
850 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
858 * Set the geometry of a device.
860 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
862 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
864 if (geo->start > sz) {
865 DMERR("Start sector is beyond the geometry limits.");
874 static int __noflush_suspending(struct mapped_device *md)
876 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
879 static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
881 struct mapped_device *md = io->md;
884 struct dm_io *next = md->requeue_list;
886 md->requeue_list = io;
889 bio_list_add_head(&md->deferred, io->orig_bio);
893 static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
896 queue_work(md->wq, &md->requeue_work);
898 queue_work(md->wq, &md->work);
902 * Return true if the dm_io's original bio is requeued.
903 * io->status is updated with error if requeue disallowed.
905 static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
907 struct bio *bio = io->orig_bio;
908 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
909 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
910 (bio->bi_opf & REQ_POLLED));
911 struct mapped_device *md = io->md;
912 bool requeued = false;
914 if (handle_requeue || handle_polled_eagain) {
917 if (bio->bi_opf & REQ_POLLED) {
919 * Upper layer won't help us poll split bio
920 * (io->orig_bio may only reflect a subset of the
921 * pre-split original) so clear REQ_POLLED.
923 bio_clear_polled(bio);
927 * Target requested pushing back the I/O or
928 * polled IO hit BLK_STS_AGAIN.
930 spin_lock_irqsave(&md->deferred_lock, flags);
931 if ((__noflush_suspending(md) &&
932 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
933 handle_polled_eagain || first_stage) {
934 dm_requeue_add_io(io, first_stage);
938 * noflush suspend was interrupted or this is
939 * a write to a zoned target.
941 io->status = BLK_STS_IOERR;
943 spin_unlock_irqrestore(&md->deferred_lock, flags);
947 dm_kick_requeue(md, first_stage);
952 static void __dm_io_complete(struct dm_io *io, bool first_stage)
954 struct bio *bio = io->orig_bio;
955 struct mapped_device *md = io->md;
956 blk_status_t io_error;
959 requeued = dm_handle_requeue(io, first_stage);
960 if (requeued && first_stage)
963 io_error = io->status;
964 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
966 else if (!io_error) {
968 * Must handle target that DM_MAPIO_SUBMITTED only to
969 * then bio_endio() rather than dm_submit_bio_remap()
971 __dm_start_io_acct(io);
976 this_cpu_dec(*md->pending_io);
978 /* nudge anyone waiting on suspend queue */
979 if (unlikely(wq_has_sleeper(&md->wait)))
982 /* Return early if the original bio was requeued */
986 if (bio_is_flush_with_data(bio)) {
988 * Preflush done for flush with data, reissue
989 * without REQ_PREFLUSH.
991 bio->bi_opf &= ~REQ_PREFLUSH;
994 /* done with normal IO or empty flush */
996 bio->bi_status = io_error;
1001 static void dm_wq_requeue_work(struct work_struct *work)
1003 struct mapped_device *md = container_of(work, struct mapped_device,
1005 unsigned long flags;
1008 /* reuse deferred lock to simplify dm_handle_requeue */
1009 spin_lock_irqsave(&md->deferred_lock, flags);
1010 io = md->requeue_list;
1011 md->requeue_list = NULL;
1012 spin_unlock_irqrestore(&md->deferred_lock, flags);
1015 struct dm_io *next = io->next;
1017 dm_io_rewind(io, &md->disk->bio_split);
1020 __dm_io_complete(io, false);
1027 * Two staged requeue:
1029 * 1) io->orig_bio points to the real original bio, and the part mapped to
1030 * this io must be requeued, instead of other parts of the original bio.
1032 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1034 static void dm_io_complete(struct dm_io *io)
1039 * Only dm_io that has been split needs two stage requeue, otherwise
1040 * we may run into long bio clone chain during suspend and OOM could
1043 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1044 * also aren't handled via the first stage requeue.
1046 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1047 first_requeue = true;
1049 first_requeue = false;
1051 __dm_io_complete(io, first_requeue);
1055 * Decrements the number of outstanding ios that a bio has been
1056 * cloned into, completing the original io if necc.
1058 static inline void __dm_io_dec_pending(struct dm_io *io)
1060 if (atomic_dec_and_test(&io->io_count))
1064 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1066 unsigned long flags;
1068 /* Push-back supersedes any I/O errors */
1069 spin_lock_irqsave(&io->lock, flags);
1070 if (!(io->status == BLK_STS_DM_REQUEUE &&
1071 __noflush_suspending(io->md))) {
1074 spin_unlock_irqrestore(&io->lock, flags);
1077 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1079 if (unlikely(error))
1080 dm_io_set_error(io, error);
1082 __dm_io_dec_pending(io);
1085 void disable_discard(struct mapped_device *md)
1087 struct queue_limits *limits = dm_get_queue_limits(md);
1089 /* device doesn't really support DISCARD, disable it */
1090 limits->max_discard_sectors = 0;
1093 void disable_write_zeroes(struct mapped_device *md)
1095 struct queue_limits *limits = dm_get_queue_limits(md);
1097 /* device doesn't really support WRITE ZEROES, disable it */
1098 limits->max_write_zeroes_sectors = 0;
1101 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1103 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1106 static void clone_endio(struct bio *bio)
1108 blk_status_t error = bio->bi_status;
1109 struct dm_target_io *tio = clone_to_tio(bio);
1110 struct dm_target *ti = tio->ti;
1111 dm_endio_fn endio = ti->type->end_io;
1112 struct dm_io *io = tio->io;
1113 struct mapped_device *md = io->md;
1115 if (unlikely(error == BLK_STS_TARGET)) {
1116 if (bio_op(bio) == REQ_OP_DISCARD &&
1117 !bdev_max_discard_sectors(bio->bi_bdev))
1118 disable_discard(md);
1119 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1120 !bdev_write_zeroes_sectors(bio->bi_bdev))
1121 disable_write_zeroes(md);
1124 if (static_branch_unlikely(&zoned_enabled) &&
1125 unlikely(bdev_is_zoned(bio->bi_bdev)))
1126 dm_zone_endio(io, bio);
1129 int r = endio(ti, bio, &error);
1131 case DM_ENDIO_REQUEUE:
1132 if (static_branch_unlikely(&zoned_enabled)) {
1134 * Requeuing writes to a sequential zone of a zoned
1135 * target will break the sequential write pattern:
1138 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1139 error = BLK_STS_IOERR;
1141 error = BLK_STS_DM_REQUEUE;
1143 error = BLK_STS_DM_REQUEUE;
1147 case DM_ENDIO_INCOMPLETE:
1148 /* The target will handle the io */
1151 DMCRIT("unimplemented target endio return value: %d", r);
1156 if (static_branch_unlikely(&swap_bios_enabled) &&
1157 unlikely(swap_bios_limit(ti, bio)))
1158 up(&md->swap_bios_semaphore);
1161 dm_io_dec_pending(io, error);
1165 * Return maximum size of I/O possible at the supplied sector up to the current
1168 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1169 sector_t target_offset)
1171 return ti->len - target_offset;
1174 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1176 sector_t target_offset = dm_target_offset(ti, sector);
1177 sector_t len = max_io_len_target_boundary(ti, target_offset);
1180 * Does the target need to split IO even further?
1181 * - varied (per target) IO splitting is a tenet of DM; this
1182 * explains why stacked chunk_sectors based splitting via
1183 * bio_split_to_limits() isn't possible here.
1185 if (!ti->max_io_len)
1187 return min_t(sector_t, len,
1188 min(queue_max_sectors(ti->table->md->queue),
1189 blk_chunk_sectors_left(target_offset, ti->max_io_len)));
1192 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1194 if (len > UINT_MAX) {
1195 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1196 (unsigned long long)len, UINT_MAX);
1197 ti->error = "Maximum size of target IO is too large";
1201 ti->max_io_len = (uint32_t) len;
1205 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1207 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1208 sector_t sector, int *srcu_idx)
1209 __acquires(md->io_barrier)
1211 struct dm_table *map;
1212 struct dm_target *ti;
1214 map = dm_get_live_table(md, srcu_idx);
1218 ti = dm_table_find_target(map, sector);
1225 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1226 long nr_pages, enum dax_access_mode mode, void **kaddr,
1229 struct mapped_device *md = dax_get_private(dax_dev);
1230 sector_t sector = pgoff * PAGE_SECTORS;
1231 struct dm_target *ti;
1232 long len, ret = -EIO;
1235 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1239 if (!ti->type->direct_access)
1241 len = max_io_len(ti, sector) / PAGE_SECTORS;
1244 nr_pages = min(len, nr_pages);
1245 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1248 dm_put_live_table(md, srcu_idx);
1253 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1256 struct mapped_device *md = dax_get_private(dax_dev);
1257 sector_t sector = pgoff * PAGE_SECTORS;
1258 struct dm_target *ti;
1262 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1266 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1268 * ->zero_page_range() is mandatory dax operation. If we are
1269 * here, something is wrong.
1273 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1275 dm_put_live_table(md, srcu_idx);
1280 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1281 void *addr, size_t bytes, struct iov_iter *i)
1283 struct mapped_device *md = dax_get_private(dax_dev);
1284 sector_t sector = pgoff * PAGE_SECTORS;
1285 struct dm_target *ti;
1289 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1290 if (!ti || !ti->type->dax_recovery_write)
1293 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1295 dm_put_live_table(md, srcu_idx);
1300 * A target may call dm_accept_partial_bio only from the map routine. It is
1301 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1302 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1303 * __send_duplicate_bios().
1305 * dm_accept_partial_bio informs the dm that the target only wants to process
1306 * additional n_sectors sectors of the bio and the rest of the data should be
1307 * sent in a next bio.
1309 * A diagram that explains the arithmetics:
1310 * +--------------------+---------------+-------+
1312 * +--------------------+---------------+-------+
1314 * <-------------- *tio->len_ptr --------------->
1315 * <----- bio_sectors ----->
1318 * Region 1 was already iterated over with bio_advance or similar function.
1319 * (it may be empty if the target doesn't use bio_advance)
1320 * Region 2 is the remaining bio size that the target wants to process.
1321 * (it may be empty if region 1 is non-empty, although there is no reason
1323 * The target requires that region 3 is to be sent in the next bio.
1325 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1326 * the partially processed part (the sum of regions 1+2) must be the same for all
1327 * copies of the bio.
1329 void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors)
1331 struct dm_target_io *tio = clone_to_tio(bio);
1332 struct dm_io *io = tio->io;
1333 unsigned int bio_sectors = bio_sectors(bio);
1335 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1336 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1337 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1338 BUG_ON(bio_sectors > *tio->len_ptr);
1339 BUG_ON(n_sectors > bio_sectors);
1341 *tio->len_ptr -= bio_sectors - n_sectors;
1342 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1345 * __split_and_process_bio() may have already saved mapped part
1346 * for accounting but it is being reduced so update accordingly.
1348 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1349 io->sectors = n_sectors;
1350 io->sector_offset = bio_sectors(io->orig_bio);
1352 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1355 * @clone: clone bio that DM core passed to target's .map function
1356 * @tgt_clone: clone of @clone bio that target needs submitted
1358 * Targets should use this interface to submit bios they take
1359 * ownership of when returning DM_MAPIO_SUBMITTED.
1361 * Target should also enable ti->accounts_remapped_io
1363 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1365 struct dm_target_io *tio = clone_to_tio(clone);
1366 struct dm_io *io = tio->io;
1368 /* establish bio that will get submitted */
1373 * Account io->origin_bio to DM dev on behalf of target
1374 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1376 dm_start_io_acct(io, clone);
1378 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1380 submit_bio_noacct(tgt_clone);
1382 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1384 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1386 mutex_lock(&md->swap_bios_lock);
1387 while (latch < md->swap_bios) {
1389 down(&md->swap_bios_semaphore);
1392 while (latch > md->swap_bios) {
1394 up(&md->swap_bios_semaphore);
1397 mutex_unlock(&md->swap_bios_lock);
1400 static void __map_bio(struct bio *clone)
1402 struct dm_target_io *tio = clone_to_tio(clone);
1403 struct dm_target *ti = tio->ti;
1404 struct dm_io *io = tio->io;
1405 struct mapped_device *md = io->md;
1408 clone->bi_end_io = clone_endio;
1413 tio->old_sector = clone->bi_iter.bi_sector;
1415 if (static_branch_unlikely(&swap_bios_enabled) &&
1416 unlikely(swap_bios_limit(ti, clone))) {
1417 int latch = get_swap_bios();
1418 if (unlikely(latch != md->swap_bios))
1419 __set_swap_bios_limit(md, latch);
1420 down(&md->swap_bios_semaphore);
1423 if (static_branch_unlikely(&zoned_enabled)) {
1425 * Check if the IO needs a special mapping due to zone append
1426 * emulation on zoned target. In this case, dm_zone_map_bio()
1427 * calls the target map operation.
1429 if (unlikely(dm_emulate_zone_append(md)))
1430 r = dm_zone_map_bio(tio);
1432 r = ti->type->map(ti, clone);
1434 r = ti->type->map(ti, clone);
1437 case DM_MAPIO_SUBMITTED:
1438 /* target has assumed ownership of this io */
1439 if (!ti->accounts_remapped_io)
1440 dm_start_io_acct(io, clone);
1442 case DM_MAPIO_REMAPPED:
1443 dm_submit_bio_remap(clone, NULL);
1446 case DM_MAPIO_REQUEUE:
1447 if (static_branch_unlikely(&swap_bios_enabled) &&
1448 unlikely(swap_bios_limit(ti, clone)))
1449 up(&md->swap_bios_semaphore);
1451 if (r == DM_MAPIO_KILL)
1452 dm_io_dec_pending(io, BLK_STS_IOERR);
1454 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1457 DMCRIT("unimplemented target map return value: %d", r);
1462 static void setup_split_accounting(struct clone_info *ci, unsigned int len)
1464 struct dm_io *io = ci->io;
1466 if (ci->sector_count > len) {
1468 * Split needed, save the mapped part for accounting.
1469 * NOTE: dm_accept_partial_bio() will update accordingly.
1471 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1473 io->sector_offset = bio_sectors(ci->bio);
1477 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1478 struct dm_target *ti, unsigned int num_bios,
1484 for (try = 0; try < 2; try++) {
1488 mutex_lock(&ci->io->md->table_devices_lock);
1489 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1490 bio = alloc_tio(ci, ti, bio_nr, len,
1491 try ? GFP_NOIO : GFP_NOWAIT);
1495 bio_list_add(blist, bio);
1498 mutex_unlock(&ci->io->md->table_devices_lock);
1499 if (bio_nr == num_bios)
1502 while ((bio = bio_list_pop(blist)))
1507 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1508 unsigned int num_bios, unsigned int *len)
1510 struct bio_list blist = BIO_EMPTY_LIST;
1512 unsigned int ret = 0;
1519 setup_split_accounting(ci, *len);
1520 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1526 setup_split_accounting(ci, *len);
1527 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1528 alloc_multiple_bios(&blist, ci, ti, num_bios, len);
1529 while ((clone = bio_list_pop(&blist))) {
1530 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1540 static void __send_empty_flush(struct clone_info *ci)
1542 struct dm_table *t = ci->map;
1543 struct bio flush_bio;
1546 * Use an on-stack bio for this, it's safe since we don't
1547 * need to reference it after submit. It's just used as
1548 * the basis for the clone(s).
1550 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1551 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1553 ci->bio = &flush_bio;
1554 ci->sector_count = 0;
1555 ci->io->tio.clone.bi_iter.bi_size = 0;
1557 for (unsigned int i = 0; i < t->num_targets; i++) {
1559 struct dm_target *ti = dm_table_get_target(t, i);
1561 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1562 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1563 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1567 * alloc_io() takes one extra reference for submission, so the
1568 * reference won't reach 0 without the following subtraction
1570 atomic_sub(1, &ci->io->io_count);
1572 bio_uninit(ci->bio);
1575 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1576 unsigned int num_bios)
1578 unsigned int len, bios;
1580 len = min_t(sector_t, ci->sector_count,
1581 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1583 atomic_add(num_bios, &ci->io->io_count);
1584 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1586 * alloc_io() takes one extra reference for submission, so the
1587 * reference won't reach 0 without the following (+1) subtraction
1589 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1592 ci->sector_count -= len;
1595 static bool is_abnormal_io(struct bio *bio)
1597 enum req_op op = bio_op(bio);
1599 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1601 case REQ_OP_DISCARD:
1602 case REQ_OP_SECURE_ERASE:
1603 case REQ_OP_WRITE_ZEROES:
1613 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1614 struct dm_target *ti)
1616 unsigned int num_bios = 0;
1618 switch (bio_op(ci->bio)) {
1619 case REQ_OP_DISCARD:
1620 num_bios = ti->num_discard_bios;
1622 case REQ_OP_SECURE_ERASE:
1623 num_bios = ti->num_secure_erase_bios;
1625 case REQ_OP_WRITE_ZEROES:
1626 num_bios = ti->num_write_zeroes_bios;
1633 * Even though the device advertised support for this type of
1634 * request, that does not mean every target supports it, and
1635 * reconfiguration might also have changed that since the
1636 * check was performed.
1638 if (unlikely(!num_bios))
1639 return BLK_STS_NOTSUPP;
1641 __send_changing_extent_only(ci, ti, num_bios);
1646 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1647 * associated with this bio, and this bio's bi_private needs to be
1648 * stored in dm_io->data before the reuse.
1650 * bio->bi_private is owned by fs or upper layer, so block layer won't
1651 * touch it after splitting. Meantime it won't be changed by anyone after
1652 * bio is submitted. So this reuse is safe.
1654 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1656 return (struct dm_io **)&bio->bi_private;
1659 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1661 struct dm_io **head = dm_poll_list_head(bio);
1663 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1664 bio->bi_opf |= REQ_DM_POLL_LIST;
1666 * Save .bi_private into dm_io, so that we can reuse
1667 * .bi_private as dm_io list head for storing dm_io list
1669 io->data = bio->bi_private;
1671 /* tell block layer to poll for completion */
1672 bio->bi_cookie = ~BLK_QC_T_NONE;
1677 * bio recursed due to split, reuse original poll list,
1678 * and save bio->bi_private too.
1680 io->data = (*head)->data;
1688 * Select the correct strategy for processing a non-flush bio.
1690 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1693 struct dm_target *ti;
1696 ti = dm_table_find_target(ci->map, ci->sector);
1698 return BLK_STS_IOERR;
1700 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1701 unlikely(!dm_target_supports_nowait(ti->type)))
1702 return BLK_STS_NOTSUPP;
1704 if (unlikely(ci->is_abnormal_io))
1705 return __process_abnormal_io(ci, ti);
1708 * Only support bio polling for normal IO, and the target io is
1709 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1711 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1713 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1714 setup_split_accounting(ci, len);
1715 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1719 ci->sector_count -= len;
1724 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1725 struct dm_table *map, struct bio *bio, bool is_abnormal)
1728 ci->io = alloc_io(md, bio);
1730 ci->is_abnormal_io = is_abnormal;
1731 ci->submit_as_polled = false;
1732 ci->sector = bio->bi_iter.bi_sector;
1733 ci->sector_count = bio_sectors(bio);
1735 /* Shouldn't happen but sector_count was being set to 0 so... */
1736 if (static_branch_unlikely(&zoned_enabled) &&
1737 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1738 ci->sector_count = 0;
1742 * Entry point to split a bio into clones and submit them to the targets.
1744 static void dm_split_and_process_bio(struct mapped_device *md,
1745 struct dm_table *map, struct bio *bio)
1747 struct clone_info ci;
1749 blk_status_t error = BLK_STS_OK;
1752 is_abnormal = is_abnormal_io(bio);
1753 if (unlikely(is_abnormal)) {
1755 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1756 * otherwise associated queue_limits won't be imposed.
1758 bio = bio_split_to_limits(bio);
1763 init_clone_info(&ci, md, map, bio, is_abnormal);
1766 if (bio->bi_opf & REQ_PREFLUSH) {
1767 __send_empty_flush(&ci);
1768 /* dm_io_complete submits any data associated with flush */
1772 error = __split_and_process_bio(&ci);
1773 if (error || !ci.sector_count)
1776 * Remainder must be passed to submit_bio_noacct() so it gets handled
1777 * *after* bios already submitted have been completely processed.
1779 bio_trim(bio, io->sectors, ci.sector_count);
1780 trace_block_split(bio, bio->bi_iter.bi_sector);
1781 bio_inc_remaining(bio);
1782 submit_bio_noacct(bio);
1785 * Drop the extra reference count for non-POLLED bio, and hold one
1786 * reference for POLLED bio, which will be released in dm_poll_bio
1788 * Add every dm_io instance into the dm_io list head which is stored
1789 * in bio->bi_private, so that dm_poll_bio can poll them all.
1791 if (error || !ci.submit_as_polled) {
1793 * In case of submission failure, the extra reference for
1794 * submitting io isn't consumed yet
1797 atomic_dec(&io->io_count);
1798 dm_io_dec_pending(io, error);
1800 dm_queue_poll_io(bio, io);
1803 static void dm_submit_bio(struct bio *bio)
1805 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1807 struct dm_table *map;
1808 blk_opf_t bio_opf = bio->bi_opf;
1810 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1812 /* If suspended, or map not yet available, queue this IO for later */
1813 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1815 if (bio->bi_opf & REQ_NOWAIT)
1816 bio_wouldblock_error(bio);
1817 else if (bio->bi_opf & REQ_RAHEAD)
1824 dm_split_and_process_bio(md, map, bio);
1826 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1829 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1832 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1834 /* don't poll if the mapped io is done */
1835 if (atomic_read(&io->io_count) > 1)
1836 bio_poll(&io->tio.clone, iob, flags);
1838 /* bio_poll holds the last reference */
1839 return atomic_read(&io->io_count) == 1;
1842 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1845 struct dm_io **head = dm_poll_list_head(bio);
1846 struct dm_io *list = *head;
1847 struct dm_io *tmp = NULL;
1848 struct dm_io *curr, *next;
1850 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1851 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1854 WARN_ON_ONCE(!list);
1857 * Restore .bi_private before possibly completing dm_io.
1859 * bio_poll() is only possible once @bio has been completely
1860 * submitted via submit_bio_noacct()'s depth-first submission.
1861 * So there is no dm_queue_poll_io() race associated with
1862 * clearing REQ_DM_POLL_LIST here.
1864 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1865 bio->bi_private = list->data;
1867 for (curr = list, next = curr->next; curr; curr = next, next =
1868 curr ? curr->next : NULL) {
1869 if (dm_poll_dm_io(curr, iob, flags)) {
1871 * clone_endio() has already occurred, so no
1872 * error handling is needed here.
1874 __dm_io_dec_pending(curr);
1883 bio->bi_opf |= REQ_DM_POLL_LIST;
1884 /* Reset bio->bi_private to dm_io list head */
1891 /*-----------------------------------------------------------------
1892 * An IDR is used to keep track of allocated minor numbers.
1893 *---------------------------------------------------------------*/
1894 static void free_minor(int minor)
1896 spin_lock(&_minor_lock);
1897 idr_remove(&_minor_idr, minor);
1898 spin_unlock(&_minor_lock);
1902 * See if the device with a specific minor # is free.
1904 static int specific_minor(int minor)
1908 if (minor >= (1 << MINORBITS))
1911 idr_preload(GFP_KERNEL);
1912 spin_lock(&_minor_lock);
1914 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1916 spin_unlock(&_minor_lock);
1919 return r == -ENOSPC ? -EBUSY : r;
1923 static int next_free_minor(int *minor)
1927 idr_preload(GFP_KERNEL);
1928 spin_lock(&_minor_lock);
1930 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1932 spin_unlock(&_minor_lock);
1940 static const struct block_device_operations dm_blk_dops;
1941 static const struct block_device_operations dm_rq_blk_dops;
1942 static const struct dax_operations dm_dax_ops;
1944 static void dm_wq_work(struct work_struct *work);
1946 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1947 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1949 dm_destroy_crypto_profile(q->crypto_profile);
1952 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1954 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1957 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1959 static void cleanup_mapped_device(struct mapped_device *md)
1962 destroy_workqueue(md->wq);
1963 dm_free_md_mempools(md->mempools);
1966 dax_remove_host(md->disk);
1967 kill_dax(md->dax_dev);
1968 put_dax(md->dax_dev);
1972 dm_cleanup_zoned_dev(md);
1974 spin_lock(&_minor_lock);
1975 md->disk->private_data = NULL;
1976 spin_unlock(&_minor_lock);
1977 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1978 struct table_device *td;
1981 list_for_each_entry(td, &md->table_devices, list) {
1982 bd_unlink_disk_holder(td->dm_dev.bdev,
1987 * Hold lock to make sure del_gendisk() won't concurrent
1988 * with open/close_table_device().
1990 mutex_lock(&md->table_devices_lock);
1991 del_gendisk(md->disk);
1992 mutex_unlock(&md->table_devices_lock);
1994 dm_queue_destroy_crypto_profile(md->queue);
1998 if (md->pending_io) {
1999 free_percpu(md->pending_io);
2000 md->pending_io = NULL;
2003 cleanup_srcu_struct(&md->io_barrier);
2005 mutex_destroy(&md->suspend_lock);
2006 mutex_destroy(&md->type_lock);
2007 mutex_destroy(&md->table_devices_lock);
2008 mutex_destroy(&md->swap_bios_lock);
2010 dm_mq_cleanup_mapped_device(md);
2014 * Allocate and initialise a blank device with a given minor.
2016 static struct mapped_device *alloc_dev(int minor)
2018 int r, numa_node_id = dm_get_numa_node();
2019 struct mapped_device *md;
2022 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2024 DMERR("unable to allocate device, out of memory.");
2028 if (!try_module_get(THIS_MODULE))
2029 goto bad_module_get;
2031 /* get a minor number for the dev */
2032 if (minor == DM_ANY_MINOR)
2033 r = next_free_minor(&minor);
2035 r = specific_minor(minor);
2039 r = init_srcu_struct(&md->io_barrier);
2041 goto bad_io_barrier;
2043 md->numa_node_id = numa_node_id;
2044 md->init_tio_pdu = false;
2045 md->type = DM_TYPE_NONE;
2046 mutex_init(&md->suspend_lock);
2047 mutex_init(&md->type_lock);
2048 mutex_init(&md->table_devices_lock);
2049 spin_lock_init(&md->deferred_lock);
2050 atomic_set(&md->holders, 1);
2051 atomic_set(&md->open_count, 0);
2052 atomic_set(&md->event_nr, 0);
2053 atomic_set(&md->uevent_seq, 0);
2054 INIT_LIST_HEAD(&md->uevent_list);
2055 INIT_LIST_HEAD(&md->table_devices);
2056 spin_lock_init(&md->uevent_lock);
2059 * default to bio-based until DM table is loaded and md->type
2060 * established. If request-based table is loaded: blk-mq will
2061 * override accordingly.
2063 md->disk = blk_alloc_disk(md->numa_node_id);
2066 md->queue = md->disk->queue;
2068 init_waitqueue_head(&md->wait);
2069 INIT_WORK(&md->work, dm_wq_work);
2070 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2071 init_waitqueue_head(&md->eventq);
2072 init_completion(&md->kobj_holder.completion);
2074 md->requeue_list = NULL;
2075 md->swap_bios = get_swap_bios();
2076 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2077 mutex_init(&md->swap_bios_lock);
2079 md->disk->major = _major;
2080 md->disk->first_minor = minor;
2081 md->disk->minors = 1;
2082 md->disk->flags |= GENHD_FL_NO_PART;
2083 md->disk->fops = &dm_blk_dops;
2084 md->disk->private_data = md;
2085 sprintf(md->disk->disk_name, "dm-%d", minor);
2087 if (IS_ENABLED(CONFIG_FS_DAX)) {
2088 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2089 if (IS_ERR(md->dax_dev)) {
2093 set_dax_nocache(md->dax_dev);
2094 set_dax_nomc(md->dax_dev);
2095 if (dax_add_host(md->dax_dev, md->disk))
2099 format_dev_t(md->name, MKDEV(_major, minor));
2101 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2105 md->pending_io = alloc_percpu(unsigned long);
2106 if (!md->pending_io)
2109 r = dm_stats_init(&md->stats);
2113 /* Populate the mapping, nobody knows we exist yet */
2114 spin_lock(&_minor_lock);
2115 old_md = idr_replace(&_minor_idr, md, minor);
2116 spin_unlock(&_minor_lock);
2118 BUG_ON(old_md != MINOR_ALLOCED);
2123 cleanup_mapped_device(md);
2127 module_put(THIS_MODULE);
2133 static void unlock_fs(struct mapped_device *md);
2135 static void free_dev(struct mapped_device *md)
2137 int minor = MINOR(disk_devt(md->disk));
2141 cleanup_mapped_device(md);
2143 free_table_devices(&md->table_devices);
2144 dm_stats_cleanup(&md->stats);
2147 module_put(THIS_MODULE);
2152 * Bind a table to the device.
2154 static void event_callback(void *context)
2156 unsigned long flags;
2158 struct mapped_device *md = (struct mapped_device *) context;
2160 spin_lock_irqsave(&md->uevent_lock, flags);
2161 list_splice_init(&md->uevent_list, &uevents);
2162 spin_unlock_irqrestore(&md->uevent_lock, flags);
2164 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2166 atomic_inc(&md->event_nr);
2167 wake_up(&md->eventq);
2168 dm_issue_global_event();
2172 * Returns old map, which caller must destroy.
2174 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2175 struct queue_limits *limits)
2177 struct dm_table *old_map;
2181 lockdep_assert_held(&md->suspend_lock);
2183 size = dm_table_get_size(t);
2186 * Wipe any geometry if the size of the table changed.
2188 if (size != dm_get_size(md))
2189 memset(&md->geometry, 0, sizeof(md->geometry));
2191 set_capacity(md->disk, size);
2193 dm_table_event_callback(t, event_callback, md);
2195 if (dm_table_request_based(t)) {
2197 * Leverage the fact that request-based DM targets are
2198 * immutable singletons - used to optimize dm_mq_queue_rq.
2200 md->immutable_target = dm_table_get_immutable_target(t);
2203 * There is no need to reload with request-based dm because the
2204 * size of front_pad doesn't change.
2206 * Note for future: If you are to reload bioset, prep-ed
2207 * requests in the queue may refer to bio from the old bioset,
2208 * so you must walk through the queue to unprep.
2210 if (!md->mempools) {
2211 md->mempools = t->mempools;
2216 * The md may already have mempools that need changing.
2217 * If so, reload bioset because front_pad may have changed
2218 * because a different table was loaded.
2220 dm_free_md_mempools(md->mempools);
2221 md->mempools = t->mempools;
2225 ret = dm_table_set_restrictions(t, md->queue, limits);
2227 old_map = ERR_PTR(ret);
2231 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2232 rcu_assign_pointer(md->map, (void *)t);
2233 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2242 * Returns unbound table for the caller to free.
2244 static struct dm_table *__unbind(struct mapped_device *md)
2246 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2251 dm_table_event_callback(map, NULL, NULL);
2252 RCU_INIT_POINTER(md->map, NULL);
2259 * Constructor for a new device.
2261 int dm_create(int minor, struct mapped_device **result)
2263 struct mapped_device *md;
2265 md = alloc_dev(minor);
2269 dm_ima_reset_data(md);
2276 * Functions to manage md->type.
2277 * All are required to hold md->type_lock.
2279 void dm_lock_md_type(struct mapped_device *md)
2281 mutex_lock(&md->type_lock);
2284 void dm_unlock_md_type(struct mapped_device *md)
2286 mutex_unlock(&md->type_lock);
2289 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2291 BUG_ON(!mutex_is_locked(&md->type_lock));
2295 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2300 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2302 return md->immutable_target_type;
2306 * The queue_limits are only valid as long as you have a reference
2309 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2311 BUG_ON(!atomic_read(&md->holders));
2312 return &md->queue->limits;
2314 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2317 * Setup the DM device's queue based on md's type
2319 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2321 enum dm_queue_mode type = dm_table_get_type(t);
2322 struct queue_limits limits;
2323 struct table_device *td;
2327 case DM_TYPE_REQUEST_BASED:
2328 md->disk->fops = &dm_rq_blk_dops;
2329 r = dm_mq_init_request_queue(md, t);
2331 DMERR("Cannot initialize queue for request-based dm mapped device");
2335 case DM_TYPE_BIO_BASED:
2336 case DM_TYPE_DAX_BIO_BASED:
2343 r = dm_calculate_queue_limits(t, &limits);
2345 DMERR("Cannot calculate initial queue limits");
2348 r = dm_table_set_restrictions(t, md->queue, &limits);
2353 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2354 * with open_table_device() and close_table_device().
2356 mutex_lock(&md->table_devices_lock);
2357 r = add_disk(md->disk);
2358 mutex_unlock(&md->table_devices_lock);
2363 * Register the holder relationship for devices added before the disk
2366 list_for_each_entry(td, &md->table_devices, list) {
2367 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2369 goto out_undo_holders;
2372 r = dm_sysfs_init(md);
2374 goto out_undo_holders;
2380 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2381 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2382 mutex_lock(&md->table_devices_lock);
2383 del_gendisk(md->disk);
2384 mutex_unlock(&md->table_devices_lock);
2388 struct mapped_device *dm_get_md(dev_t dev)
2390 struct mapped_device *md;
2391 unsigned int minor = MINOR(dev);
2393 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2396 spin_lock(&_minor_lock);
2398 md = idr_find(&_minor_idr, minor);
2399 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2400 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2406 spin_unlock(&_minor_lock);
2410 EXPORT_SYMBOL_GPL(dm_get_md);
2412 void *dm_get_mdptr(struct mapped_device *md)
2414 return md->interface_ptr;
2417 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2419 md->interface_ptr = ptr;
2422 void dm_get(struct mapped_device *md)
2424 atomic_inc(&md->holders);
2425 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2428 int dm_hold(struct mapped_device *md)
2430 spin_lock(&_minor_lock);
2431 if (test_bit(DMF_FREEING, &md->flags)) {
2432 spin_unlock(&_minor_lock);
2436 spin_unlock(&_minor_lock);
2439 EXPORT_SYMBOL_GPL(dm_hold);
2441 const char *dm_device_name(struct mapped_device *md)
2445 EXPORT_SYMBOL_GPL(dm_device_name);
2447 static void __dm_destroy(struct mapped_device *md, bool wait)
2449 struct dm_table *map;
2454 spin_lock(&_minor_lock);
2455 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2456 set_bit(DMF_FREEING, &md->flags);
2457 spin_unlock(&_minor_lock);
2459 blk_mark_disk_dead(md->disk);
2462 * Take suspend_lock so that presuspend and postsuspend methods
2463 * do not race with internal suspend.
2465 mutex_lock(&md->suspend_lock);
2466 map = dm_get_live_table(md, &srcu_idx);
2467 if (!dm_suspended_md(md)) {
2468 dm_table_presuspend_targets(map);
2469 set_bit(DMF_SUSPENDED, &md->flags);
2470 set_bit(DMF_POST_SUSPENDING, &md->flags);
2471 dm_table_postsuspend_targets(map);
2473 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2474 dm_put_live_table(md, srcu_idx);
2475 mutex_unlock(&md->suspend_lock);
2478 * Rare, but there may be I/O requests still going to complete,
2479 * for example. Wait for all references to disappear.
2480 * No one should increment the reference count of the mapped_device,
2481 * after the mapped_device state becomes DMF_FREEING.
2484 while (atomic_read(&md->holders))
2486 else if (atomic_read(&md->holders))
2487 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2488 dm_device_name(md), atomic_read(&md->holders));
2490 dm_table_destroy(__unbind(md));
2494 void dm_destroy(struct mapped_device *md)
2496 __dm_destroy(md, true);
2499 void dm_destroy_immediate(struct mapped_device *md)
2501 __dm_destroy(md, false);
2504 void dm_put(struct mapped_device *md)
2506 atomic_dec(&md->holders);
2508 EXPORT_SYMBOL_GPL(dm_put);
2510 static bool dm_in_flight_bios(struct mapped_device *md)
2513 unsigned long sum = 0;
2515 for_each_possible_cpu(cpu)
2516 sum += *per_cpu_ptr(md->pending_io, cpu);
2521 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2527 prepare_to_wait(&md->wait, &wait, task_state);
2529 if (!dm_in_flight_bios(md))
2532 if (signal_pending_state(task_state, current)) {
2539 finish_wait(&md->wait, &wait);
2546 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2550 if (!queue_is_mq(md->queue))
2551 return dm_wait_for_bios_completion(md, task_state);
2554 if (!blk_mq_queue_inflight(md->queue))
2557 if (signal_pending_state(task_state, current)) {
2569 * Process the deferred bios
2571 static void dm_wq_work(struct work_struct *work)
2573 struct mapped_device *md = container_of(work, struct mapped_device, work);
2576 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2577 spin_lock_irq(&md->deferred_lock);
2578 bio = bio_list_pop(&md->deferred);
2579 spin_unlock_irq(&md->deferred_lock);
2584 submit_bio_noacct(bio);
2589 static void dm_queue_flush(struct mapped_device *md)
2591 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2592 smp_mb__after_atomic();
2593 queue_work(md->wq, &md->work);
2597 * Swap in a new table, returning the old one for the caller to destroy.
2599 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2601 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2602 struct queue_limits limits;
2605 mutex_lock(&md->suspend_lock);
2607 /* device must be suspended */
2608 if (!dm_suspended_md(md))
2612 * If the new table has no data devices, retain the existing limits.
2613 * This helps multipath with queue_if_no_path if all paths disappear,
2614 * then new I/O is queued based on these limits, and then some paths
2617 if (dm_table_has_no_data_devices(table)) {
2618 live_map = dm_get_live_table_fast(md);
2620 limits = md->queue->limits;
2621 dm_put_live_table_fast(md);
2625 r = dm_calculate_queue_limits(table, &limits);
2632 map = __bind(md, table, &limits);
2633 dm_issue_global_event();
2636 mutex_unlock(&md->suspend_lock);
2641 * Functions to lock and unlock any filesystem running on the
2644 static int lock_fs(struct mapped_device *md)
2648 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2650 r = freeze_bdev(md->disk->part0);
2652 set_bit(DMF_FROZEN, &md->flags);
2656 static void unlock_fs(struct mapped_device *md)
2658 if (!test_bit(DMF_FROZEN, &md->flags))
2660 thaw_bdev(md->disk->part0);
2661 clear_bit(DMF_FROZEN, &md->flags);
2665 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2666 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2667 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2669 * If __dm_suspend returns 0, the device is completely quiescent
2670 * now. There is no request-processing activity. All new requests
2671 * are being added to md->deferred list.
2673 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2674 unsigned int suspend_flags, unsigned int task_state,
2675 int dmf_suspended_flag)
2677 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2678 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2681 lockdep_assert_held(&md->suspend_lock);
2684 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2685 * This flag is cleared before dm_suspend returns.
2688 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2690 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2693 * This gets reverted if there's an error later and the targets
2694 * provide the .presuspend_undo hook.
2696 dm_table_presuspend_targets(map);
2699 * Flush I/O to the device.
2700 * Any I/O submitted after lock_fs() may not be flushed.
2701 * noflush takes precedence over do_lockfs.
2702 * (lock_fs() flushes I/Os and waits for them to complete.)
2704 if (!noflush && do_lockfs) {
2707 dm_table_presuspend_undo_targets(map);
2713 * Here we must make sure that no processes are submitting requests
2714 * to target drivers i.e. no one may be executing
2715 * dm_split_and_process_bio from dm_submit_bio.
2717 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2718 * we take the write lock. To prevent any process from reentering
2719 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2720 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2721 * flush_workqueue(md->wq).
2723 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2725 synchronize_srcu(&md->io_barrier);
2728 * Stop md->queue before flushing md->wq in case request-based
2729 * dm defers requests to md->wq from md->queue.
2731 if (dm_request_based(md))
2732 dm_stop_queue(md->queue);
2734 flush_workqueue(md->wq);
2737 * At this point no more requests are entering target request routines.
2738 * We call dm_wait_for_completion to wait for all existing requests
2741 r = dm_wait_for_completion(md, task_state);
2743 set_bit(dmf_suspended_flag, &md->flags);
2746 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2748 synchronize_srcu(&md->io_barrier);
2750 /* were we interrupted ? */
2754 if (dm_request_based(md))
2755 dm_start_queue(md->queue);
2758 dm_table_presuspend_undo_targets(map);
2759 /* pushback list is already flushed, so skip flush */
2766 * We need to be able to change a mapping table under a mounted
2767 * filesystem. For example we might want to move some data in
2768 * the background. Before the table can be swapped with
2769 * dm_bind_table, dm_suspend must be called to flush any in
2770 * flight bios and ensure that any further io gets deferred.
2773 * Suspend mechanism in request-based dm.
2775 * 1. Flush all I/Os by lock_fs() if needed.
2776 * 2. Stop dispatching any I/O by stopping the request_queue.
2777 * 3. Wait for all in-flight I/Os to be completed or requeued.
2779 * To abort suspend, start the request_queue.
2781 int dm_suspend(struct mapped_device *md, unsigned int suspend_flags)
2783 struct dm_table *map = NULL;
2787 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2789 if (dm_suspended_md(md)) {
2794 if (dm_suspended_internally_md(md)) {
2795 /* already internally suspended, wait for internal resume */
2796 mutex_unlock(&md->suspend_lock);
2797 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2803 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2805 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2809 set_bit(DMF_POST_SUSPENDING, &md->flags);
2810 dm_table_postsuspend_targets(map);
2811 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2814 mutex_unlock(&md->suspend_lock);
2818 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2821 int r = dm_table_resume_targets(map);
2829 * Flushing deferred I/Os must be done after targets are resumed
2830 * so that mapping of targets can work correctly.
2831 * Request-based dm is queueing the deferred I/Os in its request_queue.
2833 if (dm_request_based(md))
2834 dm_start_queue(md->queue);
2841 int dm_resume(struct mapped_device *md)
2844 struct dm_table *map = NULL;
2848 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2850 if (!dm_suspended_md(md))
2853 if (dm_suspended_internally_md(md)) {
2854 /* already internally suspended, wait for internal resume */
2855 mutex_unlock(&md->suspend_lock);
2856 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2862 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2863 if (!map || !dm_table_get_size(map))
2866 r = __dm_resume(md, map);
2870 clear_bit(DMF_SUSPENDED, &md->flags);
2872 mutex_unlock(&md->suspend_lock);
2878 * Internal suspend/resume works like userspace-driven suspend. It waits
2879 * until all bios finish and prevents issuing new bios to the target drivers.
2880 * It may be used only from the kernel.
2883 static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags)
2885 struct dm_table *map = NULL;
2887 lockdep_assert_held(&md->suspend_lock);
2889 if (md->internal_suspend_count++)
2890 return; /* nested internal suspend */
2892 if (dm_suspended_md(md)) {
2893 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2894 return; /* nest suspend */
2897 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2900 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2901 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2902 * would require changing .presuspend to return an error -- avoid this
2903 * until there is a need for more elaborate variants of internal suspend.
2905 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2906 DMF_SUSPENDED_INTERNALLY);
2908 set_bit(DMF_POST_SUSPENDING, &md->flags);
2909 dm_table_postsuspend_targets(map);
2910 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2913 static void __dm_internal_resume(struct mapped_device *md)
2915 BUG_ON(!md->internal_suspend_count);
2917 if (--md->internal_suspend_count)
2918 return; /* resume from nested internal suspend */
2920 if (dm_suspended_md(md))
2921 goto done; /* resume from nested suspend */
2924 * NOTE: existing callers don't need to call dm_table_resume_targets
2925 * (which may fail -- so best to avoid it for now by passing NULL map)
2927 (void) __dm_resume(md, NULL);
2930 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2931 smp_mb__after_atomic();
2932 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2935 void dm_internal_suspend_noflush(struct mapped_device *md)
2937 mutex_lock(&md->suspend_lock);
2938 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2939 mutex_unlock(&md->suspend_lock);
2941 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2943 void dm_internal_resume(struct mapped_device *md)
2945 mutex_lock(&md->suspend_lock);
2946 __dm_internal_resume(md);
2947 mutex_unlock(&md->suspend_lock);
2949 EXPORT_SYMBOL_GPL(dm_internal_resume);
2952 * Fast variants of internal suspend/resume hold md->suspend_lock,
2953 * which prevents interaction with userspace-driven suspend.
2956 void dm_internal_suspend_fast(struct mapped_device *md)
2958 mutex_lock(&md->suspend_lock);
2959 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2962 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2963 synchronize_srcu(&md->io_barrier);
2964 flush_workqueue(md->wq);
2965 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2967 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2969 void dm_internal_resume_fast(struct mapped_device *md)
2971 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2977 mutex_unlock(&md->suspend_lock);
2979 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2981 /*-----------------------------------------------------------------
2982 * Event notification.
2983 *---------------------------------------------------------------*/
2984 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2985 unsigned int cookie, bool need_resize_uevent)
2988 unsigned int noio_flag;
2989 char udev_cookie[DM_COOKIE_LENGTH];
2990 char *envp[3] = { NULL, NULL, NULL };
2991 char **envpp = envp;
2993 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2994 DM_COOKIE_ENV_VAR_NAME, cookie);
2995 *envpp++ = udev_cookie;
2997 if (need_resize_uevent) {
2998 *envpp++ = "RESIZE=1";
3001 noio_flag = memalloc_noio_save();
3003 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
3005 memalloc_noio_restore(noio_flag);
3010 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3012 return atomic_add_return(1, &md->uevent_seq);
3015 uint32_t dm_get_event_nr(struct mapped_device *md)
3017 return atomic_read(&md->event_nr);
3020 int dm_wait_event(struct mapped_device *md, int event_nr)
3022 return wait_event_interruptible(md->eventq,
3023 (event_nr != atomic_read(&md->event_nr)));
3026 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3028 unsigned long flags;
3030 spin_lock_irqsave(&md->uevent_lock, flags);
3031 list_add(elist, &md->uevent_list);
3032 spin_unlock_irqrestore(&md->uevent_lock, flags);
3036 * The gendisk is only valid as long as you have a reference
3039 struct gendisk *dm_disk(struct mapped_device *md)
3043 EXPORT_SYMBOL_GPL(dm_disk);
3045 struct kobject *dm_kobject(struct mapped_device *md)
3047 return &md->kobj_holder.kobj;
3050 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3052 struct mapped_device *md;
3054 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3056 spin_lock(&_minor_lock);
3057 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3063 spin_unlock(&_minor_lock);
3068 int dm_suspended_md(struct mapped_device *md)
3070 return test_bit(DMF_SUSPENDED, &md->flags);
3073 static int dm_post_suspending_md(struct mapped_device *md)
3075 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3078 int dm_suspended_internally_md(struct mapped_device *md)
3080 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3083 int dm_test_deferred_remove_flag(struct mapped_device *md)
3085 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3088 int dm_suspended(struct dm_target *ti)
3090 return dm_suspended_md(ti->table->md);
3092 EXPORT_SYMBOL_GPL(dm_suspended);
3094 int dm_post_suspending(struct dm_target *ti)
3096 return dm_post_suspending_md(ti->table->md);
3098 EXPORT_SYMBOL_GPL(dm_post_suspending);
3100 int dm_noflush_suspending(struct dm_target *ti)
3102 return __noflush_suspending(ti->table->md);
3104 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3106 void dm_free_md_mempools(struct dm_md_mempools *pools)
3111 bioset_exit(&pools->bs);
3112 bioset_exit(&pools->io_bs);
3127 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3130 struct mapped_device *md = bdev->bd_disk->private_data;
3131 struct dm_table *table;
3132 struct dm_target *ti;
3133 int ret = -ENOTTY, srcu_idx;
3135 table = dm_get_live_table(md, &srcu_idx);
3136 if (!table || !dm_table_get_size(table))
3139 /* We only support devices that have a single target */
3140 if (table->num_targets != 1)
3142 ti = dm_table_get_target(table, 0);
3144 if (dm_suspended_md(md)) {
3150 if (!ti->type->iterate_devices)
3153 ti->type->iterate_devices(ti, fn, pr);
3156 dm_put_live_table(md, srcu_idx);
3161 * For register / unregister we need to manually call out to every path.
3163 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3164 sector_t start, sector_t len, void *data)
3166 struct dm_pr *pr = data;
3167 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3170 if (!ops || !ops->pr_register) {
3171 pr->ret = -EOPNOTSUPP;
3175 ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3188 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3200 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3202 /* Didn't even get to register a path */
3213 /* unregister all paths if we failed to register any path */
3214 pr.old_key = new_key;
3217 pr.fail_early = false;
3218 (void) dm_call_pr(bdev, __dm_pr_register, &pr);
3223 static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3224 sector_t start, sector_t len, void *data)
3226 struct dm_pr *pr = data;
3227 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3229 if (!ops || !ops->pr_reserve) {
3230 pr->ret = -EOPNOTSUPP;
3234 pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3241 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3248 .fail_early = false,
3253 ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3261 * If there is a non-All Registrants type of reservation, the release must be
3262 * sent down the holding path. For the cases where there is no reservation or
3263 * the path is not the holder the device will also return success, so we must
3264 * try each path to make sure we got the correct path.
3266 static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3267 sector_t start, sector_t len, void *data)
3269 struct dm_pr *pr = data;
3270 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3272 if (!ops || !ops->pr_release) {
3273 pr->ret = -EOPNOTSUPP;
3277 pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3284 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3289 .fail_early = false,
3293 ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3300 static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3301 sector_t start, sector_t len, void *data)
3303 struct dm_pr *pr = data;
3304 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3306 if (!ops || !ops->pr_preempt) {
3307 pr->ret = -EOPNOTSUPP;
3311 pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3319 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3320 enum pr_type type, bool abort)
3326 .fail_early = false,
3330 ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3337 static int dm_pr_clear(struct block_device *bdev, u64 key)
3339 struct mapped_device *md = bdev->bd_disk->private_data;
3340 const struct pr_ops *ops;
3343 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3347 ops = bdev->bd_disk->fops->pr_ops;
3348 if (ops && ops->pr_clear)
3349 r = ops->pr_clear(bdev, key);
3353 dm_unprepare_ioctl(md, srcu_idx);
3357 static const struct pr_ops dm_pr_ops = {
3358 .pr_register = dm_pr_register,
3359 .pr_reserve = dm_pr_reserve,
3360 .pr_release = dm_pr_release,
3361 .pr_preempt = dm_pr_preempt,
3362 .pr_clear = dm_pr_clear,
3365 static const struct block_device_operations dm_blk_dops = {
3366 .submit_bio = dm_submit_bio,
3367 .poll_bio = dm_poll_bio,
3368 .open = dm_blk_open,
3369 .release = dm_blk_close,
3370 .ioctl = dm_blk_ioctl,
3371 .getgeo = dm_blk_getgeo,
3372 .report_zones = dm_blk_report_zones,
3373 .pr_ops = &dm_pr_ops,
3374 .owner = THIS_MODULE
3377 static const struct block_device_operations dm_rq_blk_dops = {
3378 .open = dm_blk_open,
3379 .release = dm_blk_close,
3380 .ioctl = dm_blk_ioctl,
3381 .getgeo = dm_blk_getgeo,
3382 .pr_ops = &dm_pr_ops,
3383 .owner = THIS_MODULE
3386 static const struct dax_operations dm_dax_ops = {
3387 .direct_access = dm_dax_direct_access,
3388 .zero_page_range = dm_dax_zero_page_range,
3389 .recovery_write = dm_dax_recovery_write,
3395 module_init(dm_init);
3396 module_exit(dm_exit);
3398 module_param(major, uint, 0);
3399 MODULE_PARM_DESC(major, "The major number of the device mapper");
3401 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3402 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3404 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3405 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3407 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3408 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3410 MODULE_DESCRIPTION(DM_NAME " driver");
3411 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3412 MODULE_LICENSE("GPL");