1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
6 * This file is released under the GPL.
11 #include "dm-uevent.h"
14 #include <linux/init.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/sched/mm.h>
18 #include <linux/sched/signal.h>
19 #include <linux/blkpg.h>
20 #include <linux/bio.h>
21 #include <linux/mempool.h>
22 #include <linux/dax.h>
23 #include <linux/slab.h>
24 #include <linux/idr.h>
25 #include <linux/uio.h>
26 #include <linux/hdreg.h>
27 #include <linux/delay.h>
28 #include <linux/wait.h>
30 #include <linux/refcount.h>
31 #include <linux/part_stat.h>
32 #include <linux/blk-crypto.h>
33 #include <linux/blk-crypto-profile.h>
35 #define DM_MSG_PREFIX "core"
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
45 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
46 * dm_io into one list, and reuse bio->bi_private as the list head. Before
47 * ending this fs bio, we will recover its ->bi_private.
49 #define REQ_DM_POLL_LIST REQ_DRV
51 static const char *_name = DM_NAME;
53 static unsigned int major;
54 static unsigned int _major;
56 static DEFINE_IDR(_minor_idr);
58 static DEFINE_SPINLOCK(_minor_lock);
60 static void do_deferred_remove(struct work_struct *w);
62 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
64 static struct workqueue_struct *deferred_remove_workqueue;
66 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
67 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
69 void dm_issue_global_event(void)
71 atomic_inc(&dm_global_event_nr);
72 wake_up(&dm_global_eventq);
75 DEFINE_STATIC_KEY_FALSE(stats_enabled);
76 DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
77 DEFINE_STATIC_KEY_FALSE(zoned_enabled);
80 * One of these is allocated (on-stack) per original bio.
87 unsigned int sector_count;
88 bool is_abnormal_io:1;
89 bool submit_as_polled:1;
92 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
94 return container_of(clone, struct dm_target_io, clone);
97 void *dm_per_bio_data(struct bio *bio, size_t data_size)
99 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
100 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
101 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
103 EXPORT_SYMBOL_GPL(dm_per_bio_data);
105 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
107 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
109 if (io->magic == DM_IO_MAGIC)
110 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
111 BUG_ON(io->magic != DM_TIO_MAGIC);
112 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
114 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
116 unsigned int dm_bio_get_target_bio_nr(const struct bio *bio)
118 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
120 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
122 #define MINOR_ALLOCED ((void *)-1)
124 #define DM_NUMA_NODE NUMA_NO_NODE
125 static int dm_numa_node = DM_NUMA_NODE;
127 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
128 static int swap_bios = DEFAULT_SWAP_BIOS;
129 static int get_swap_bios(void)
131 int latch = READ_ONCE(swap_bios);
133 if (unlikely(latch <= 0))
134 latch = DEFAULT_SWAP_BIOS;
138 struct table_device {
139 struct list_head list;
141 struct dm_dev dm_dev;
145 * Bio-based DM's mempools' reserved IOs set by the user.
147 #define RESERVED_BIO_BASED_IOS 16
148 static unsigned int reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
150 static int __dm_get_module_param_int(int *module_param, int min, int max)
152 int param = READ_ONCE(*module_param);
153 int modified_param = 0;
154 bool modified = true;
157 modified_param = min;
158 else if (param > max)
159 modified_param = max;
164 (void)cmpxchg(module_param, param, modified_param);
165 param = modified_param;
171 unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max)
173 unsigned int param = READ_ONCE(*module_param);
174 unsigned int modified_param = 0;
177 modified_param = def;
178 else if (param > max)
179 modified_param = max;
181 if (modified_param) {
182 (void)cmpxchg(module_param, param, modified_param);
183 param = modified_param;
189 unsigned int dm_get_reserved_bio_based_ios(void)
191 return __dm_get_module_param(&reserved_bio_based_ios,
192 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
194 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
196 static unsigned int dm_get_numa_node(void)
198 return __dm_get_module_param_int(&dm_numa_node,
199 DM_NUMA_NODE, num_online_nodes() - 1);
202 static int __init local_init(void)
206 r = dm_uevent_init();
210 deferred_remove_workqueue = alloc_ordered_workqueue("kdmremove", 0);
211 if (!deferred_remove_workqueue) {
213 goto out_uevent_exit;
217 r = register_blkdev(_major, _name);
219 goto out_free_workqueue;
227 destroy_workqueue(deferred_remove_workqueue);
234 static void local_exit(void)
236 destroy_workqueue(deferred_remove_workqueue);
238 unregister_blkdev(_major, _name);
243 DMINFO("cleaned up");
246 static int (*_inits[])(void) __initdata = {
257 static void (*_exits[])(void) = {
268 static int __init dm_init(void)
270 const int count = ARRAY_SIZE(_inits);
273 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
274 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
275 " Duplicate IMA measurements will not be recorded in the IMA log.");
278 for (i = 0; i < count; i++) {
292 static void __exit dm_exit(void)
294 int i = ARRAY_SIZE(_exits);
300 * Should be empty by this point.
302 idr_destroy(&_minor_idr);
306 * Block device functions
308 int dm_deleting_md(struct mapped_device *md)
310 return test_bit(DMF_DELETING, &md->flags);
313 static int dm_blk_open(struct gendisk *disk, blk_mode_t mode)
315 struct mapped_device *md;
317 spin_lock(&_minor_lock);
319 md = disk->private_data;
323 if (test_bit(DMF_FREEING, &md->flags) ||
324 dm_deleting_md(md)) {
330 atomic_inc(&md->open_count);
332 spin_unlock(&_minor_lock);
334 return md ? 0 : -ENXIO;
337 static void dm_blk_close(struct gendisk *disk)
339 struct mapped_device *md;
341 spin_lock(&_minor_lock);
343 md = disk->private_data;
347 if (atomic_dec_and_test(&md->open_count) &&
348 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
349 queue_work(deferred_remove_workqueue, &deferred_remove_work);
353 spin_unlock(&_minor_lock);
356 int dm_open_count(struct mapped_device *md)
358 return atomic_read(&md->open_count);
362 * Guarantees nothing is using the device before it's deleted.
364 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
368 spin_lock(&_minor_lock);
370 if (dm_open_count(md)) {
373 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
374 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
377 set_bit(DMF_DELETING, &md->flags);
379 spin_unlock(&_minor_lock);
384 int dm_cancel_deferred_remove(struct mapped_device *md)
388 spin_lock(&_minor_lock);
390 if (test_bit(DMF_DELETING, &md->flags))
393 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
395 spin_unlock(&_minor_lock);
400 static void do_deferred_remove(struct work_struct *w)
402 dm_deferred_remove();
405 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
407 struct mapped_device *md = bdev->bd_disk->private_data;
409 return dm_get_geometry(md, geo);
412 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
413 struct block_device **bdev)
415 struct dm_target *ti;
416 struct dm_table *map;
421 map = dm_get_live_table(md, srcu_idx);
422 if (!map || !dm_table_get_size(map))
425 /* We only support devices that have a single target */
426 if (map->num_targets != 1)
429 ti = dm_table_get_target(map, 0);
430 if (!ti->type->prepare_ioctl)
433 if (dm_suspended_md(md))
436 r = ti->type->prepare_ioctl(ti, bdev);
437 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
438 dm_put_live_table(md, *srcu_idx);
446 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
448 dm_put_live_table(md, srcu_idx);
451 static int dm_blk_ioctl(struct block_device *bdev, blk_mode_t mode,
452 unsigned int cmd, unsigned long arg)
454 struct mapped_device *md = bdev->bd_disk->private_data;
457 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
463 * Target determined this ioctl is being issued against a
464 * subset of the parent bdev; require extra privileges.
466 if (!capable(CAP_SYS_RAWIO)) {
468 "%s: sending ioctl %x to DM device without required privilege.",
475 if (!bdev->bd_disk->fops->ioctl)
478 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
480 dm_unprepare_ioctl(md, srcu_idx);
484 u64 dm_start_time_ns_from_clone(struct bio *bio)
486 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
488 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
490 static inline bool bio_is_flush_with_data(struct bio *bio)
492 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
495 static inline unsigned int dm_io_sectors(struct dm_io *io, struct bio *bio)
498 * If REQ_PREFLUSH set, don't account payload, it will be
499 * submitted (and accounted) after this flush completes.
501 if (bio_is_flush_with_data(bio))
503 if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
505 return bio_sectors(bio);
508 static void dm_io_acct(struct dm_io *io, bool end)
510 struct bio *bio = io->orig_bio;
512 if (dm_io_flagged(io, DM_IO_BLK_STAT)) {
514 bdev_start_io_acct(bio->bi_bdev, bio_op(bio),
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio),
518 dm_io_sectors(io, bio),
522 if (static_branch_unlikely(&stats_enabled) &&
523 unlikely(dm_stats_used(&io->md->stats))) {
526 if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT)))
527 sector = bio_end_sector(bio) - io->sector_offset;
529 sector = bio->bi_iter.bi_sector;
531 dm_stats_account_io(&io->md->stats, bio_data_dir(bio),
532 sector, dm_io_sectors(io, bio),
533 end, io->start_time, &io->stats_aux);
537 static void __dm_start_io_acct(struct dm_io *io)
539 dm_io_acct(io, false);
542 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
545 * Ensure IO accounting is only ever started once.
547 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
550 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
551 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
552 dm_io_set_flag(io, DM_IO_ACCOUNTED);
555 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
556 spin_lock_irqsave(&io->lock, flags);
557 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
558 spin_unlock_irqrestore(&io->lock, flags);
561 dm_io_set_flag(io, DM_IO_ACCOUNTED);
562 spin_unlock_irqrestore(&io->lock, flags);
565 __dm_start_io_acct(io);
568 static void dm_end_io_acct(struct dm_io *io)
570 dm_io_acct(io, true);
573 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
576 struct dm_target_io *tio;
579 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
580 tio = clone_to_tio(clone);
582 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
585 io = container_of(tio, struct dm_io, tio);
586 io->magic = DM_IO_MAGIC;
587 io->status = BLK_STS_OK;
589 /* one ref is for submission, the other is for completion */
590 atomic_set(&io->io_count, 2);
591 this_cpu_inc(*md->pending_io);
594 spin_lock_init(&io->lock);
595 io->start_time = jiffies;
597 if (blk_queue_io_stat(md->queue))
598 dm_io_set_flag(io, DM_IO_BLK_STAT);
600 if (static_branch_unlikely(&stats_enabled) &&
601 unlikely(dm_stats_used(&md->stats)))
602 dm_stats_record_start(&md->stats, &io->stats_aux);
607 static void free_io(struct dm_io *io)
609 bio_put(&io->tio.clone);
612 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
613 unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask)
615 struct mapped_device *md = ci->io->md;
616 struct dm_target_io *tio;
619 if (!ci->io->tio.io) {
620 /* the dm_target_io embedded in ci->io is available */
622 /* alloc_io() already initialized embedded clone */
625 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
630 /* REQ_DM_POLL_LIST shouldn't be inherited */
631 clone->bi_opf &= ~REQ_DM_POLL_LIST;
633 tio = clone_to_tio(clone);
634 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
637 tio->magic = DM_TIO_MAGIC;
640 tio->target_bio_nr = target_bio_nr;
644 /* Set default bdev, but target must bio_set_dev() before issuing IO */
645 clone->bi_bdev = md->disk->part0;
646 if (unlikely(ti->needs_bio_set_dev))
647 bio_set_dev(clone, md->disk->part0);
650 clone->bi_iter.bi_size = to_bytes(*len);
651 if (bio_integrity(clone))
652 bio_integrity_trim(clone);
658 static void free_tio(struct bio *clone)
660 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
666 * Add the bio to the list of deferred io.
668 static void queue_io(struct mapped_device *md, struct bio *bio)
672 spin_lock_irqsave(&md->deferred_lock, flags);
673 bio_list_add(&md->deferred, bio);
674 spin_unlock_irqrestore(&md->deferred_lock, flags);
675 queue_work(md->wq, &md->work);
679 * Everyone (including functions in this file), should use this
680 * function to access the md->map field, and make sure they call
681 * dm_put_live_table() when finished.
683 struct dm_table *dm_get_live_table(struct mapped_device *md,
684 int *srcu_idx) __acquires(md->io_barrier)
686 *srcu_idx = srcu_read_lock(&md->io_barrier);
688 return srcu_dereference(md->map, &md->io_barrier);
691 void dm_put_live_table(struct mapped_device *md,
692 int srcu_idx) __releases(md->io_barrier)
694 srcu_read_unlock(&md->io_barrier, srcu_idx);
697 void dm_sync_table(struct mapped_device *md)
699 synchronize_srcu(&md->io_barrier);
700 synchronize_rcu_expedited();
704 * A fast alternative to dm_get_live_table/dm_put_live_table.
705 * The caller must not block between these two functions.
707 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
710 return rcu_dereference(md->map);
713 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
718 static char *_dm_claim_ptr = "I belong to device-mapper";
721 * Open a table device so we can use it as a map destination.
723 static struct table_device *open_table_device(struct mapped_device *md,
724 dev_t dev, blk_mode_t mode)
726 struct table_device *td;
727 struct block_device *bdev;
731 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
733 return ERR_PTR(-ENOMEM);
734 refcount_set(&td->count, 1);
736 bdev = blkdev_get_by_dev(dev, mode, _dm_claim_ptr, NULL);
743 * We can be called before the dm disk is added. In that case we can't
744 * register the holder relation here. It will be done once add_disk was
747 if (md->disk->slave_dir) {
748 r = bd_link_disk_holder(bdev, md->disk);
753 td->dm_dev.mode = mode;
754 td->dm_dev.bdev = bdev;
755 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
756 format_dev_t(td->dm_dev.name, dev);
757 list_add(&td->list, &md->table_devices);
761 blkdev_put(bdev, _dm_claim_ptr);
768 * Close a table device that we've been using.
770 static void close_table_device(struct table_device *td, struct mapped_device *md)
772 if (md->disk->slave_dir)
773 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
774 blkdev_put(td->dm_dev.bdev, _dm_claim_ptr);
775 put_dax(td->dm_dev.dax_dev);
780 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
783 struct table_device *td;
785 list_for_each_entry(td, l, list)
786 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
792 int dm_get_table_device(struct mapped_device *md, dev_t dev, blk_mode_t mode,
793 struct dm_dev **result)
795 struct table_device *td;
797 mutex_lock(&md->table_devices_lock);
798 td = find_table_device(&md->table_devices, dev, mode);
800 td = open_table_device(md, dev, mode);
802 mutex_unlock(&md->table_devices_lock);
806 refcount_inc(&td->count);
808 mutex_unlock(&md->table_devices_lock);
810 *result = &td->dm_dev;
814 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
816 struct table_device *td = container_of(d, struct table_device, dm_dev);
818 mutex_lock(&md->table_devices_lock);
819 if (refcount_dec_and_test(&td->count))
820 close_table_device(td, md);
821 mutex_unlock(&md->table_devices_lock);
825 * Get the geometry associated with a dm device
827 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
835 * Set the geometry of a device.
837 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
839 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
841 if (geo->start > sz) {
842 DMERR("Start sector is beyond the geometry limits.");
851 static int __noflush_suspending(struct mapped_device *md)
853 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
856 static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
858 struct mapped_device *md = io->md;
861 struct dm_io *next = md->requeue_list;
863 md->requeue_list = io;
866 bio_list_add_head(&md->deferred, io->orig_bio);
870 static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
873 queue_work(md->wq, &md->requeue_work);
875 queue_work(md->wq, &md->work);
879 * Return true if the dm_io's original bio is requeued.
880 * io->status is updated with error if requeue disallowed.
882 static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
884 struct bio *bio = io->orig_bio;
885 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
886 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
887 (bio->bi_opf & REQ_POLLED));
888 struct mapped_device *md = io->md;
889 bool requeued = false;
891 if (handle_requeue || handle_polled_eagain) {
894 if (bio->bi_opf & REQ_POLLED) {
896 * Upper layer won't help us poll split bio
897 * (io->orig_bio may only reflect a subset of the
898 * pre-split original) so clear REQ_POLLED.
900 bio_clear_polled(bio);
904 * Target requested pushing back the I/O or
905 * polled IO hit BLK_STS_AGAIN.
907 spin_lock_irqsave(&md->deferred_lock, flags);
908 if ((__noflush_suspending(md) &&
909 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
910 handle_polled_eagain || first_stage) {
911 dm_requeue_add_io(io, first_stage);
915 * noflush suspend was interrupted or this is
916 * a write to a zoned target.
918 io->status = BLK_STS_IOERR;
920 spin_unlock_irqrestore(&md->deferred_lock, flags);
924 dm_kick_requeue(md, first_stage);
929 static void __dm_io_complete(struct dm_io *io, bool first_stage)
931 struct bio *bio = io->orig_bio;
932 struct mapped_device *md = io->md;
933 blk_status_t io_error;
936 requeued = dm_handle_requeue(io, first_stage);
937 if (requeued && first_stage)
940 io_error = io->status;
941 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
943 else if (!io_error) {
945 * Must handle target that DM_MAPIO_SUBMITTED only to
946 * then bio_endio() rather than dm_submit_bio_remap()
948 __dm_start_io_acct(io);
953 this_cpu_dec(*md->pending_io);
955 /* nudge anyone waiting on suspend queue */
956 if (unlikely(wq_has_sleeper(&md->wait)))
959 /* Return early if the original bio was requeued */
963 if (bio_is_flush_with_data(bio)) {
965 * Preflush done for flush with data, reissue
966 * without REQ_PREFLUSH.
968 bio->bi_opf &= ~REQ_PREFLUSH;
971 /* done with normal IO or empty flush */
973 bio->bi_status = io_error;
978 static void dm_wq_requeue_work(struct work_struct *work)
980 struct mapped_device *md = container_of(work, struct mapped_device,
985 /* reuse deferred lock to simplify dm_handle_requeue */
986 spin_lock_irqsave(&md->deferred_lock, flags);
987 io = md->requeue_list;
988 md->requeue_list = NULL;
989 spin_unlock_irqrestore(&md->deferred_lock, flags);
992 struct dm_io *next = io->next;
994 dm_io_rewind(io, &md->disk->bio_split);
997 __dm_io_complete(io, false);
1004 * Two staged requeue:
1006 * 1) io->orig_bio points to the real original bio, and the part mapped to
1007 * this io must be requeued, instead of other parts of the original bio.
1009 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1011 static void dm_io_complete(struct dm_io *io)
1016 * Only dm_io that has been split needs two stage requeue, otherwise
1017 * we may run into long bio clone chain during suspend and OOM could
1020 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1021 * also aren't handled via the first stage requeue.
1023 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1024 first_requeue = true;
1026 first_requeue = false;
1028 __dm_io_complete(io, first_requeue);
1032 * Decrements the number of outstanding ios that a bio has been
1033 * cloned into, completing the original io if necc.
1035 static inline void __dm_io_dec_pending(struct dm_io *io)
1037 if (atomic_dec_and_test(&io->io_count))
1041 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1043 unsigned long flags;
1045 /* Push-back supersedes any I/O errors */
1046 spin_lock_irqsave(&io->lock, flags);
1047 if (!(io->status == BLK_STS_DM_REQUEUE &&
1048 __noflush_suspending(io->md))) {
1051 spin_unlock_irqrestore(&io->lock, flags);
1054 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1056 if (unlikely(error))
1057 dm_io_set_error(io, error);
1059 __dm_io_dec_pending(io);
1063 * The queue_limits are only valid as long as you have a reference
1064 * count on 'md'. But _not_ imposing verification to avoid atomic_read(),
1066 static inline struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1068 return &md->queue->limits;
1071 void disable_discard(struct mapped_device *md)
1073 struct queue_limits *limits = dm_get_queue_limits(md);
1075 /* device doesn't really support DISCARD, disable it */
1076 limits->max_discard_sectors = 0;
1079 void disable_write_zeroes(struct mapped_device *md)
1081 struct queue_limits *limits = dm_get_queue_limits(md);
1083 /* device doesn't really support WRITE ZEROES, disable it */
1084 limits->max_write_zeroes_sectors = 0;
1087 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1089 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1092 static void clone_endio(struct bio *bio)
1094 blk_status_t error = bio->bi_status;
1095 struct dm_target_io *tio = clone_to_tio(bio);
1096 struct dm_target *ti = tio->ti;
1097 dm_endio_fn endio = ti->type->end_io;
1098 struct dm_io *io = tio->io;
1099 struct mapped_device *md = io->md;
1101 if (unlikely(error == BLK_STS_TARGET)) {
1102 if (bio_op(bio) == REQ_OP_DISCARD &&
1103 !bdev_max_discard_sectors(bio->bi_bdev))
1104 disable_discard(md);
1105 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1106 !bdev_write_zeroes_sectors(bio->bi_bdev))
1107 disable_write_zeroes(md);
1110 if (static_branch_unlikely(&zoned_enabled) &&
1111 unlikely(bdev_is_zoned(bio->bi_bdev)))
1112 dm_zone_endio(io, bio);
1115 int r = endio(ti, bio, &error);
1118 case DM_ENDIO_REQUEUE:
1119 if (static_branch_unlikely(&zoned_enabled)) {
1121 * Requeuing writes to a sequential zone of a zoned
1122 * target will break the sequential write pattern:
1125 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1126 error = BLK_STS_IOERR;
1128 error = BLK_STS_DM_REQUEUE;
1130 error = BLK_STS_DM_REQUEUE;
1134 case DM_ENDIO_INCOMPLETE:
1135 /* The target will handle the io */
1138 DMCRIT("unimplemented target endio return value: %d", r);
1143 if (static_branch_unlikely(&swap_bios_enabled) &&
1144 unlikely(swap_bios_limit(ti, bio)))
1145 up(&md->swap_bios_semaphore);
1148 dm_io_dec_pending(io, error);
1152 * Return maximum size of I/O possible at the supplied sector up to the current
1155 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1156 sector_t target_offset)
1158 return ti->len - target_offset;
1161 static sector_t __max_io_len(struct dm_target *ti, sector_t sector,
1162 unsigned int max_granularity,
1163 unsigned int max_sectors)
1165 sector_t target_offset = dm_target_offset(ti, sector);
1166 sector_t len = max_io_len_target_boundary(ti, target_offset);
1169 * Does the target need to split IO even further?
1170 * - varied (per target) IO splitting is a tenet of DM; this
1171 * explains why stacked chunk_sectors based splitting via
1172 * bio_split_to_limits() isn't possible here.
1174 if (!max_granularity)
1176 return min_t(sector_t, len,
1177 min(max_sectors ? : queue_max_sectors(ti->table->md->queue),
1178 blk_chunk_sectors_left(target_offset, max_granularity)));
1181 static inline sector_t max_io_len(struct dm_target *ti, sector_t sector)
1183 return __max_io_len(ti, sector, ti->max_io_len, 0);
1186 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1188 if (len > UINT_MAX) {
1189 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1190 (unsigned long long)len, UINT_MAX);
1191 ti->error = "Maximum size of target IO is too large";
1195 ti->max_io_len = (uint32_t) len;
1199 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1201 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1202 sector_t sector, int *srcu_idx)
1203 __acquires(md->io_barrier)
1205 struct dm_table *map;
1206 struct dm_target *ti;
1208 map = dm_get_live_table(md, srcu_idx);
1212 ti = dm_table_find_target(map, sector);
1219 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1220 long nr_pages, enum dax_access_mode mode, void **kaddr,
1223 struct mapped_device *md = dax_get_private(dax_dev);
1224 sector_t sector = pgoff * PAGE_SECTORS;
1225 struct dm_target *ti;
1226 long len, ret = -EIO;
1229 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1233 if (!ti->type->direct_access)
1235 len = max_io_len(ti, sector) / PAGE_SECTORS;
1238 nr_pages = min(len, nr_pages);
1239 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1242 dm_put_live_table(md, srcu_idx);
1247 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1250 struct mapped_device *md = dax_get_private(dax_dev);
1251 sector_t sector = pgoff * PAGE_SECTORS;
1252 struct dm_target *ti;
1256 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1260 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1262 * ->zero_page_range() is mandatory dax operation. If we are
1263 * here, something is wrong.
1267 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1269 dm_put_live_table(md, srcu_idx);
1274 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1275 void *addr, size_t bytes, struct iov_iter *i)
1277 struct mapped_device *md = dax_get_private(dax_dev);
1278 sector_t sector = pgoff * PAGE_SECTORS;
1279 struct dm_target *ti;
1283 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1284 if (!ti || !ti->type->dax_recovery_write)
1287 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1289 dm_put_live_table(md, srcu_idx);
1294 * A target may call dm_accept_partial_bio only from the map routine. It is
1295 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1296 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1297 * __send_duplicate_bios().
1299 * dm_accept_partial_bio informs the dm that the target only wants to process
1300 * additional n_sectors sectors of the bio and the rest of the data should be
1301 * sent in a next bio.
1303 * A diagram that explains the arithmetics:
1304 * +--------------------+---------------+-------+
1306 * +--------------------+---------------+-------+
1308 * <-------------- *tio->len_ptr --------------->
1309 * <----- bio_sectors ----->
1312 * Region 1 was already iterated over with bio_advance or similar function.
1313 * (it may be empty if the target doesn't use bio_advance)
1314 * Region 2 is the remaining bio size that the target wants to process.
1315 * (it may be empty if region 1 is non-empty, although there is no reason
1317 * The target requires that region 3 is to be sent in the next bio.
1319 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1320 * the partially processed part (the sum of regions 1+2) must be the same for all
1321 * copies of the bio.
1323 void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors)
1325 struct dm_target_io *tio = clone_to_tio(bio);
1326 struct dm_io *io = tio->io;
1327 unsigned int bio_sectors = bio_sectors(bio);
1329 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1330 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1331 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1332 BUG_ON(bio_sectors > *tio->len_ptr);
1333 BUG_ON(n_sectors > bio_sectors);
1335 *tio->len_ptr -= bio_sectors - n_sectors;
1336 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1339 * __split_and_process_bio() may have already saved mapped part
1340 * for accounting but it is being reduced so update accordingly.
1342 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1343 io->sectors = n_sectors;
1344 io->sector_offset = bio_sectors(io->orig_bio);
1346 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1349 * @clone: clone bio that DM core passed to target's .map function
1350 * @tgt_clone: clone of @clone bio that target needs submitted
1352 * Targets should use this interface to submit bios they take
1353 * ownership of when returning DM_MAPIO_SUBMITTED.
1355 * Target should also enable ti->accounts_remapped_io
1357 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1359 struct dm_target_io *tio = clone_to_tio(clone);
1360 struct dm_io *io = tio->io;
1362 /* establish bio that will get submitted */
1367 * Account io->origin_bio to DM dev on behalf of target
1368 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1370 dm_start_io_acct(io, clone);
1372 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1374 submit_bio_noacct(tgt_clone);
1376 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1378 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1380 mutex_lock(&md->swap_bios_lock);
1381 while (latch < md->swap_bios) {
1383 down(&md->swap_bios_semaphore);
1386 while (latch > md->swap_bios) {
1388 up(&md->swap_bios_semaphore);
1391 mutex_unlock(&md->swap_bios_lock);
1394 static void __map_bio(struct bio *clone)
1396 struct dm_target_io *tio = clone_to_tio(clone);
1397 struct dm_target *ti = tio->ti;
1398 struct dm_io *io = tio->io;
1399 struct mapped_device *md = io->md;
1402 clone->bi_end_io = clone_endio;
1407 tio->old_sector = clone->bi_iter.bi_sector;
1409 if (static_branch_unlikely(&swap_bios_enabled) &&
1410 unlikely(swap_bios_limit(ti, clone))) {
1411 int latch = get_swap_bios();
1413 if (unlikely(latch != md->swap_bios))
1414 __set_swap_bios_limit(md, latch);
1415 down(&md->swap_bios_semaphore);
1418 if (static_branch_unlikely(&zoned_enabled)) {
1420 * Check if the IO needs a special mapping due to zone append
1421 * emulation on zoned target. In this case, dm_zone_map_bio()
1422 * calls the target map operation.
1424 if (unlikely(dm_emulate_zone_append(md)))
1425 r = dm_zone_map_bio(tio);
1427 r = ti->type->map(ti, clone);
1429 r = ti->type->map(ti, clone);
1432 case DM_MAPIO_SUBMITTED:
1433 /* target has assumed ownership of this io */
1434 if (!ti->accounts_remapped_io)
1435 dm_start_io_acct(io, clone);
1437 case DM_MAPIO_REMAPPED:
1438 dm_submit_bio_remap(clone, NULL);
1441 case DM_MAPIO_REQUEUE:
1442 if (static_branch_unlikely(&swap_bios_enabled) &&
1443 unlikely(swap_bios_limit(ti, clone)))
1444 up(&md->swap_bios_semaphore);
1446 if (r == DM_MAPIO_KILL)
1447 dm_io_dec_pending(io, BLK_STS_IOERR);
1449 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1452 DMCRIT("unimplemented target map return value: %d", r);
1457 static void setup_split_accounting(struct clone_info *ci, unsigned int len)
1459 struct dm_io *io = ci->io;
1461 if (ci->sector_count > len) {
1463 * Split needed, save the mapped part for accounting.
1464 * NOTE: dm_accept_partial_bio() will update accordingly.
1466 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1468 io->sector_offset = bio_sectors(ci->bio);
1472 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1473 struct dm_target *ti, unsigned int num_bios,
1479 for (try = 0; try < 2; try++) {
1483 mutex_lock(&ci->io->md->table_devices_lock);
1484 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1485 bio = alloc_tio(ci, ti, bio_nr, len,
1486 try ? GFP_NOIO : GFP_NOWAIT);
1490 bio_list_add(blist, bio);
1493 mutex_unlock(&ci->io->md->table_devices_lock);
1494 if (bio_nr == num_bios)
1497 while ((bio = bio_list_pop(blist)))
1502 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1503 unsigned int num_bios, unsigned int *len)
1505 struct bio_list blist = BIO_EMPTY_LIST;
1507 unsigned int ret = 0;
1514 setup_split_accounting(ci, *len);
1515 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1521 setup_split_accounting(ci, *len);
1522 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1523 alloc_multiple_bios(&blist, ci, ti, num_bios, len);
1524 while ((clone = bio_list_pop(&blist))) {
1525 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1535 static void __send_empty_flush(struct clone_info *ci)
1537 struct dm_table *t = ci->map;
1538 struct bio flush_bio;
1541 * Use an on-stack bio for this, it's safe since we don't
1542 * need to reference it after submit. It's just used as
1543 * the basis for the clone(s).
1545 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1546 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1548 ci->bio = &flush_bio;
1549 ci->sector_count = 0;
1550 ci->io->tio.clone.bi_iter.bi_size = 0;
1552 for (unsigned int i = 0; i < t->num_targets; i++) {
1554 struct dm_target *ti = dm_table_get_target(t, i);
1556 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1557 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1558 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1562 * alloc_io() takes one extra reference for submission, so the
1563 * reference won't reach 0 without the following subtraction
1565 atomic_sub(1, &ci->io->io_count);
1567 bio_uninit(ci->bio);
1570 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1571 unsigned int num_bios,
1572 unsigned int max_granularity,
1573 unsigned int max_sectors)
1575 unsigned int len, bios;
1577 len = min_t(sector_t, ci->sector_count,
1578 __max_io_len(ti, ci->sector, max_granularity, max_sectors));
1580 atomic_add(num_bios, &ci->io->io_count);
1581 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1583 * alloc_io() takes one extra reference for submission, so the
1584 * reference won't reach 0 without the following (+1) subtraction
1586 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1589 ci->sector_count -= len;
1592 static bool is_abnormal_io(struct bio *bio)
1594 enum req_op op = bio_op(bio);
1596 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1598 case REQ_OP_DISCARD:
1599 case REQ_OP_SECURE_ERASE:
1600 case REQ_OP_WRITE_ZEROES:
1610 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1611 struct dm_target *ti)
1613 unsigned int num_bios = 0;
1614 unsigned int max_granularity = 0;
1615 unsigned int max_sectors = 0;
1616 struct queue_limits *limits = dm_get_queue_limits(ti->table->md);
1618 switch (bio_op(ci->bio)) {
1619 case REQ_OP_DISCARD:
1620 num_bios = ti->num_discard_bios;
1621 max_sectors = limits->max_discard_sectors;
1622 if (ti->max_discard_granularity)
1623 max_granularity = max_sectors;
1625 case REQ_OP_SECURE_ERASE:
1626 num_bios = ti->num_secure_erase_bios;
1627 max_sectors = limits->max_secure_erase_sectors;
1628 if (ti->max_secure_erase_granularity)
1629 max_granularity = max_sectors;
1631 case REQ_OP_WRITE_ZEROES:
1632 num_bios = ti->num_write_zeroes_bios;
1633 max_sectors = limits->max_write_zeroes_sectors;
1634 if (ti->max_write_zeroes_granularity)
1635 max_granularity = max_sectors;
1642 * Even though the device advertised support for this type of
1643 * request, that does not mean every target supports it, and
1644 * reconfiguration might also have changed that since the
1645 * check was performed.
1647 if (unlikely(!num_bios))
1648 return BLK_STS_NOTSUPP;
1650 __send_changing_extent_only(ci, ti, num_bios,
1651 max_granularity, max_sectors);
1656 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1657 * associated with this bio, and this bio's bi_private needs to be
1658 * stored in dm_io->data before the reuse.
1660 * bio->bi_private is owned by fs or upper layer, so block layer won't
1661 * touch it after splitting. Meantime it won't be changed by anyone after
1662 * bio is submitted. So this reuse is safe.
1664 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1666 return (struct dm_io **)&bio->bi_private;
1669 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1671 struct dm_io **head = dm_poll_list_head(bio);
1673 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1674 bio->bi_opf |= REQ_DM_POLL_LIST;
1676 * Save .bi_private into dm_io, so that we can reuse
1677 * .bi_private as dm_io list head for storing dm_io list
1679 io->data = bio->bi_private;
1681 /* tell block layer to poll for completion */
1682 bio->bi_cookie = ~BLK_QC_T_NONE;
1687 * bio recursed due to split, reuse original poll list,
1688 * and save bio->bi_private too.
1690 io->data = (*head)->data;
1698 * Select the correct strategy for processing a non-flush bio.
1700 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1703 struct dm_target *ti;
1706 ti = dm_table_find_target(ci->map, ci->sector);
1708 return BLK_STS_IOERR;
1710 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1711 unlikely(!dm_target_supports_nowait(ti->type)))
1712 return BLK_STS_NOTSUPP;
1714 if (unlikely(ci->is_abnormal_io))
1715 return __process_abnormal_io(ci, ti);
1718 * Only support bio polling for normal IO, and the target io is
1719 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1721 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1723 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1724 setup_split_accounting(ci, len);
1725 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1729 ci->sector_count -= len;
1734 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1735 struct dm_table *map, struct bio *bio, bool is_abnormal)
1738 ci->io = alloc_io(md, bio);
1740 ci->is_abnormal_io = is_abnormal;
1741 ci->submit_as_polled = false;
1742 ci->sector = bio->bi_iter.bi_sector;
1743 ci->sector_count = bio_sectors(bio);
1745 /* Shouldn't happen but sector_count was being set to 0 so... */
1746 if (static_branch_unlikely(&zoned_enabled) &&
1747 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1748 ci->sector_count = 0;
1752 * Entry point to split a bio into clones and submit them to the targets.
1754 static void dm_split_and_process_bio(struct mapped_device *md,
1755 struct dm_table *map, struct bio *bio)
1757 struct clone_info ci;
1759 blk_status_t error = BLK_STS_OK;
1762 is_abnormal = is_abnormal_io(bio);
1763 if (unlikely(is_abnormal)) {
1765 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1766 * otherwise associated queue_limits won't be imposed.
1768 bio = bio_split_to_limits(bio);
1773 init_clone_info(&ci, md, map, bio, is_abnormal);
1776 if (bio->bi_opf & REQ_PREFLUSH) {
1777 __send_empty_flush(&ci);
1778 /* dm_io_complete submits any data associated with flush */
1782 error = __split_and_process_bio(&ci);
1783 if (error || !ci.sector_count)
1786 * Remainder must be passed to submit_bio_noacct() so it gets handled
1787 * *after* bios already submitted have been completely processed.
1789 bio_trim(bio, io->sectors, ci.sector_count);
1790 trace_block_split(bio, bio->bi_iter.bi_sector);
1791 bio_inc_remaining(bio);
1792 submit_bio_noacct(bio);
1795 * Drop the extra reference count for non-POLLED bio, and hold one
1796 * reference for POLLED bio, which will be released in dm_poll_bio
1798 * Add every dm_io instance into the dm_io list head which is stored
1799 * in bio->bi_private, so that dm_poll_bio can poll them all.
1801 if (error || !ci.submit_as_polled) {
1803 * In case of submission failure, the extra reference for
1804 * submitting io isn't consumed yet
1807 atomic_dec(&io->io_count);
1808 dm_io_dec_pending(io, error);
1810 dm_queue_poll_io(bio, io);
1813 static void dm_submit_bio(struct bio *bio)
1815 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1817 struct dm_table *map;
1819 map = dm_get_live_table(md, &srcu_idx);
1821 /* If suspended, or map not yet available, queue this IO for later */
1822 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1824 if (bio->bi_opf & REQ_NOWAIT)
1825 bio_wouldblock_error(bio);
1826 else if (bio->bi_opf & REQ_RAHEAD)
1833 dm_split_and_process_bio(md, map, bio);
1835 dm_put_live_table(md, srcu_idx);
1838 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1841 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1843 /* don't poll if the mapped io is done */
1844 if (atomic_read(&io->io_count) > 1)
1845 bio_poll(&io->tio.clone, iob, flags);
1847 /* bio_poll holds the last reference */
1848 return atomic_read(&io->io_count) == 1;
1851 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1854 struct dm_io **head = dm_poll_list_head(bio);
1855 struct dm_io *list = *head;
1856 struct dm_io *tmp = NULL;
1857 struct dm_io *curr, *next;
1859 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1860 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1863 WARN_ON_ONCE(!list);
1866 * Restore .bi_private before possibly completing dm_io.
1868 * bio_poll() is only possible once @bio has been completely
1869 * submitted via submit_bio_noacct()'s depth-first submission.
1870 * So there is no dm_queue_poll_io() race associated with
1871 * clearing REQ_DM_POLL_LIST here.
1873 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1874 bio->bi_private = list->data;
1876 for (curr = list, next = curr->next; curr; curr = next, next =
1877 curr ? curr->next : NULL) {
1878 if (dm_poll_dm_io(curr, iob, flags)) {
1880 * clone_endio() has already occurred, so no
1881 * error handling is needed here.
1883 __dm_io_dec_pending(curr);
1892 bio->bi_opf |= REQ_DM_POLL_LIST;
1893 /* Reset bio->bi_private to dm_io list head */
1901 *---------------------------------------------------------------
1902 * An IDR is used to keep track of allocated minor numbers.
1903 *---------------------------------------------------------------
1905 static void free_minor(int minor)
1907 spin_lock(&_minor_lock);
1908 idr_remove(&_minor_idr, minor);
1909 spin_unlock(&_minor_lock);
1913 * See if the device with a specific minor # is free.
1915 static int specific_minor(int minor)
1919 if (minor >= (1 << MINORBITS))
1922 idr_preload(GFP_KERNEL);
1923 spin_lock(&_minor_lock);
1925 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1927 spin_unlock(&_minor_lock);
1930 return r == -ENOSPC ? -EBUSY : r;
1934 static int next_free_minor(int *minor)
1938 idr_preload(GFP_KERNEL);
1939 spin_lock(&_minor_lock);
1941 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1943 spin_unlock(&_minor_lock);
1951 static const struct block_device_operations dm_blk_dops;
1952 static const struct block_device_operations dm_rq_blk_dops;
1953 static const struct dax_operations dm_dax_ops;
1955 static void dm_wq_work(struct work_struct *work);
1957 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1958 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1960 dm_destroy_crypto_profile(q->crypto_profile);
1963 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1965 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1968 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1970 static void cleanup_mapped_device(struct mapped_device *md)
1973 destroy_workqueue(md->wq);
1974 dm_free_md_mempools(md->mempools);
1977 dax_remove_host(md->disk);
1978 kill_dax(md->dax_dev);
1979 put_dax(md->dax_dev);
1983 dm_cleanup_zoned_dev(md);
1985 spin_lock(&_minor_lock);
1986 md->disk->private_data = NULL;
1987 spin_unlock(&_minor_lock);
1988 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1989 struct table_device *td;
1992 list_for_each_entry(td, &md->table_devices, list) {
1993 bd_unlink_disk_holder(td->dm_dev.bdev,
1998 * Hold lock to make sure del_gendisk() won't concurrent
1999 * with open/close_table_device().
2001 mutex_lock(&md->table_devices_lock);
2002 del_gendisk(md->disk);
2003 mutex_unlock(&md->table_devices_lock);
2005 dm_queue_destroy_crypto_profile(md->queue);
2009 if (md->pending_io) {
2010 free_percpu(md->pending_io);
2011 md->pending_io = NULL;
2014 cleanup_srcu_struct(&md->io_barrier);
2016 mutex_destroy(&md->suspend_lock);
2017 mutex_destroy(&md->type_lock);
2018 mutex_destroy(&md->table_devices_lock);
2019 mutex_destroy(&md->swap_bios_lock);
2021 dm_mq_cleanup_mapped_device(md);
2025 * Allocate and initialise a blank device with a given minor.
2027 static struct mapped_device *alloc_dev(int minor)
2029 int r, numa_node_id = dm_get_numa_node();
2030 struct mapped_device *md;
2033 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2035 DMERR("unable to allocate device, out of memory.");
2039 if (!try_module_get(THIS_MODULE))
2040 goto bad_module_get;
2042 /* get a minor number for the dev */
2043 if (minor == DM_ANY_MINOR)
2044 r = next_free_minor(&minor);
2046 r = specific_minor(minor);
2050 r = init_srcu_struct(&md->io_barrier);
2052 goto bad_io_barrier;
2054 md->numa_node_id = numa_node_id;
2055 md->init_tio_pdu = false;
2056 md->type = DM_TYPE_NONE;
2057 mutex_init(&md->suspend_lock);
2058 mutex_init(&md->type_lock);
2059 mutex_init(&md->table_devices_lock);
2060 spin_lock_init(&md->deferred_lock);
2061 atomic_set(&md->holders, 1);
2062 atomic_set(&md->open_count, 0);
2063 atomic_set(&md->event_nr, 0);
2064 atomic_set(&md->uevent_seq, 0);
2065 INIT_LIST_HEAD(&md->uevent_list);
2066 INIT_LIST_HEAD(&md->table_devices);
2067 spin_lock_init(&md->uevent_lock);
2070 * default to bio-based until DM table is loaded and md->type
2071 * established. If request-based table is loaded: blk-mq will
2072 * override accordingly.
2074 md->disk = blk_alloc_disk(md->numa_node_id);
2077 md->queue = md->disk->queue;
2079 init_waitqueue_head(&md->wait);
2080 INIT_WORK(&md->work, dm_wq_work);
2081 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2082 init_waitqueue_head(&md->eventq);
2083 init_completion(&md->kobj_holder.completion);
2085 md->requeue_list = NULL;
2086 md->swap_bios = get_swap_bios();
2087 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2088 mutex_init(&md->swap_bios_lock);
2090 md->disk->major = _major;
2091 md->disk->first_minor = minor;
2092 md->disk->minors = 1;
2093 md->disk->flags |= GENHD_FL_NO_PART;
2094 md->disk->fops = &dm_blk_dops;
2095 md->disk->private_data = md;
2096 sprintf(md->disk->disk_name, "dm-%d", minor);
2098 if (IS_ENABLED(CONFIG_FS_DAX)) {
2099 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2100 if (IS_ERR(md->dax_dev)) {
2104 set_dax_nocache(md->dax_dev);
2105 set_dax_nomc(md->dax_dev);
2106 if (dax_add_host(md->dax_dev, md->disk))
2110 format_dev_t(md->name, MKDEV(_major, minor));
2112 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2116 md->pending_io = alloc_percpu(unsigned long);
2117 if (!md->pending_io)
2120 r = dm_stats_init(&md->stats);
2124 /* Populate the mapping, nobody knows we exist yet */
2125 spin_lock(&_minor_lock);
2126 old_md = idr_replace(&_minor_idr, md, minor);
2127 spin_unlock(&_minor_lock);
2129 BUG_ON(old_md != MINOR_ALLOCED);
2134 cleanup_mapped_device(md);
2138 module_put(THIS_MODULE);
2144 static void unlock_fs(struct mapped_device *md);
2146 static void free_dev(struct mapped_device *md)
2148 int minor = MINOR(disk_devt(md->disk));
2152 cleanup_mapped_device(md);
2154 WARN_ON_ONCE(!list_empty(&md->table_devices));
2155 dm_stats_cleanup(&md->stats);
2158 module_put(THIS_MODULE);
2163 * Bind a table to the device.
2165 static void event_callback(void *context)
2167 unsigned long flags;
2169 struct mapped_device *md = context;
2171 spin_lock_irqsave(&md->uevent_lock, flags);
2172 list_splice_init(&md->uevent_list, &uevents);
2173 spin_unlock_irqrestore(&md->uevent_lock, flags);
2175 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2177 atomic_inc(&md->event_nr);
2178 wake_up(&md->eventq);
2179 dm_issue_global_event();
2183 * Returns old map, which caller must destroy.
2185 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2186 struct queue_limits *limits)
2188 struct dm_table *old_map;
2192 lockdep_assert_held(&md->suspend_lock);
2194 size = dm_table_get_size(t);
2197 * Wipe any geometry if the size of the table changed.
2199 if (size != dm_get_size(md))
2200 memset(&md->geometry, 0, sizeof(md->geometry));
2202 set_capacity(md->disk, size);
2204 dm_table_event_callback(t, event_callback, md);
2206 if (dm_table_request_based(t)) {
2208 * Leverage the fact that request-based DM targets are
2209 * immutable singletons - used to optimize dm_mq_queue_rq.
2211 md->immutable_target = dm_table_get_immutable_target(t);
2214 * There is no need to reload with request-based dm because the
2215 * size of front_pad doesn't change.
2217 * Note for future: If you are to reload bioset, prep-ed
2218 * requests in the queue may refer to bio from the old bioset,
2219 * so you must walk through the queue to unprep.
2221 if (!md->mempools) {
2222 md->mempools = t->mempools;
2227 * The md may already have mempools that need changing.
2228 * If so, reload bioset because front_pad may have changed
2229 * because a different table was loaded.
2231 dm_free_md_mempools(md->mempools);
2232 md->mempools = t->mempools;
2236 ret = dm_table_set_restrictions(t, md->queue, limits);
2238 old_map = ERR_PTR(ret);
2242 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2243 rcu_assign_pointer(md->map, (void *)t);
2244 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2253 * Returns unbound table for the caller to free.
2255 static struct dm_table *__unbind(struct mapped_device *md)
2257 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2262 dm_table_event_callback(map, NULL, NULL);
2263 RCU_INIT_POINTER(md->map, NULL);
2270 * Constructor for a new device.
2272 int dm_create(int minor, struct mapped_device **result)
2274 struct mapped_device *md;
2276 md = alloc_dev(minor);
2280 dm_ima_reset_data(md);
2287 * Functions to manage md->type.
2288 * All are required to hold md->type_lock.
2290 void dm_lock_md_type(struct mapped_device *md)
2292 mutex_lock(&md->type_lock);
2295 void dm_unlock_md_type(struct mapped_device *md)
2297 mutex_unlock(&md->type_lock);
2300 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2302 BUG_ON(!mutex_is_locked(&md->type_lock));
2306 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2311 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2313 return md->immutable_target_type;
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:
2337 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, md->queue);
2344 r = dm_calculate_queue_limits(t, &limits);
2346 DMERR("Cannot calculate initial queue limits");
2349 r = dm_table_set_restrictions(t, md->queue, &limits);
2354 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2355 * with open_table_device() and close_table_device().
2357 mutex_lock(&md->table_devices_lock);
2358 r = add_disk(md->disk);
2359 mutex_unlock(&md->table_devices_lock);
2364 * Register the holder relationship for devices added before the disk
2367 list_for_each_entry(td, &md->table_devices, list) {
2368 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2370 goto out_undo_holders;
2373 r = dm_sysfs_init(md);
2375 goto out_undo_holders;
2381 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2382 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2383 mutex_lock(&md->table_devices_lock);
2384 del_gendisk(md->disk);
2385 mutex_unlock(&md->table_devices_lock);
2389 struct mapped_device *dm_get_md(dev_t dev)
2391 struct mapped_device *md;
2392 unsigned int minor = MINOR(dev);
2394 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2397 spin_lock(&_minor_lock);
2399 md = idr_find(&_minor_idr, minor);
2400 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2401 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2407 spin_unlock(&_minor_lock);
2411 EXPORT_SYMBOL_GPL(dm_get_md);
2413 void *dm_get_mdptr(struct mapped_device *md)
2415 return md->interface_ptr;
2418 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2420 md->interface_ptr = ptr;
2423 void dm_get(struct mapped_device *md)
2425 atomic_inc(&md->holders);
2426 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2429 int dm_hold(struct mapped_device *md)
2431 spin_lock(&_minor_lock);
2432 if (test_bit(DMF_FREEING, &md->flags)) {
2433 spin_unlock(&_minor_lock);
2437 spin_unlock(&_minor_lock);
2440 EXPORT_SYMBOL_GPL(dm_hold);
2442 const char *dm_device_name(struct mapped_device *md)
2446 EXPORT_SYMBOL_GPL(dm_device_name);
2448 static void __dm_destroy(struct mapped_device *md, bool wait)
2450 struct dm_table *map;
2455 spin_lock(&_minor_lock);
2456 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2457 set_bit(DMF_FREEING, &md->flags);
2458 spin_unlock(&_minor_lock);
2460 blk_mark_disk_dead(md->disk);
2463 * Take suspend_lock so that presuspend and postsuspend methods
2464 * do not race with internal suspend.
2466 mutex_lock(&md->suspend_lock);
2467 map = dm_get_live_table(md, &srcu_idx);
2468 if (!dm_suspended_md(md)) {
2469 dm_table_presuspend_targets(map);
2470 set_bit(DMF_SUSPENDED, &md->flags);
2471 set_bit(DMF_POST_SUSPENDING, &md->flags);
2472 dm_table_postsuspend_targets(map);
2474 /* dm_put_live_table must be before fsleep, otherwise deadlock is possible */
2475 dm_put_live_table(md, srcu_idx);
2476 mutex_unlock(&md->suspend_lock);
2479 * Rare, but there may be I/O requests still going to complete,
2480 * for example. Wait for all references to disappear.
2481 * No one should increment the reference count of the mapped_device,
2482 * after the mapped_device state becomes DMF_FREEING.
2485 while (atomic_read(&md->holders))
2487 else if (atomic_read(&md->holders))
2488 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2489 dm_device_name(md), atomic_read(&md->holders));
2491 dm_table_destroy(__unbind(md));
2495 void dm_destroy(struct mapped_device *md)
2497 __dm_destroy(md, true);
2500 void dm_destroy_immediate(struct mapped_device *md)
2502 __dm_destroy(md, false);
2505 void dm_put(struct mapped_device *md)
2507 atomic_dec(&md->holders);
2509 EXPORT_SYMBOL_GPL(dm_put);
2511 static bool dm_in_flight_bios(struct mapped_device *md)
2514 unsigned long sum = 0;
2516 for_each_possible_cpu(cpu)
2517 sum += *per_cpu_ptr(md->pending_io, cpu);
2522 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2528 prepare_to_wait(&md->wait, &wait, task_state);
2530 if (!dm_in_flight_bios(md))
2533 if (signal_pending_state(task_state, current)) {
2540 finish_wait(&md->wait, &wait);
2547 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2551 if (!queue_is_mq(md->queue))
2552 return dm_wait_for_bios_completion(md, task_state);
2555 if (!blk_mq_queue_inflight(md->queue))
2558 if (signal_pending_state(task_state, current)) {
2570 * Process the deferred bios
2572 static void dm_wq_work(struct work_struct *work)
2574 struct mapped_device *md = container_of(work, struct mapped_device, work);
2577 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2578 spin_lock_irq(&md->deferred_lock);
2579 bio = bio_list_pop(&md->deferred);
2580 spin_unlock_irq(&md->deferred_lock);
2585 submit_bio_noacct(bio);
2590 static void dm_queue_flush(struct mapped_device *md)
2592 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2593 smp_mb__after_atomic();
2594 queue_work(md->wq, &md->work);
2598 * Swap in a new table, returning the old one for the caller to destroy.
2600 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2602 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2603 struct queue_limits limits;
2606 mutex_lock(&md->suspend_lock);
2608 /* device must be suspended */
2609 if (!dm_suspended_md(md))
2613 * If the new table has no data devices, retain the existing limits.
2614 * This helps multipath with queue_if_no_path if all paths disappear,
2615 * then new I/O is queued based on these limits, and then some paths
2618 if (dm_table_has_no_data_devices(table)) {
2619 live_map = dm_get_live_table_fast(md);
2621 limits = md->queue->limits;
2622 dm_put_live_table_fast(md);
2626 r = dm_calculate_queue_limits(table, &limits);
2633 map = __bind(md, table, &limits);
2634 dm_issue_global_event();
2637 mutex_unlock(&md->suspend_lock);
2642 * Functions to lock and unlock any filesystem running on the
2645 static int lock_fs(struct mapped_device *md)
2649 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2651 r = freeze_bdev(md->disk->part0);
2653 set_bit(DMF_FROZEN, &md->flags);
2657 static void unlock_fs(struct mapped_device *md)
2659 if (!test_bit(DMF_FROZEN, &md->flags))
2661 thaw_bdev(md->disk->part0);
2662 clear_bit(DMF_FROZEN, &md->flags);
2666 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2667 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2668 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2670 * If __dm_suspend returns 0, the device is completely quiescent
2671 * now. There is no request-processing activity. All new requests
2672 * are being added to md->deferred list.
2674 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2675 unsigned int suspend_flags, unsigned int task_state,
2676 int dmf_suspended_flag)
2678 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2679 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2682 lockdep_assert_held(&md->suspend_lock);
2685 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2686 * This flag is cleared before dm_suspend returns.
2689 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2691 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2694 * This gets reverted if there's an error later and the targets
2695 * provide the .presuspend_undo hook.
2697 dm_table_presuspend_targets(map);
2700 * Flush I/O to the device.
2701 * Any I/O submitted after lock_fs() may not be flushed.
2702 * noflush takes precedence over do_lockfs.
2703 * (lock_fs() flushes I/Os and waits for them to complete.)
2705 if (!noflush && do_lockfs) {
2708 dm_table_presuspend_undo_targets(map);
2714 * Here we must make sure that no processes are submitting requests
2715 * to target drivers i.e. no one may be executing
2716 * dm_split_and_process_bio from dm_submit_bio.
2718 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2719 * we take the write lock. To prevent any process from reentering
2720 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2721 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2722 * flush_workqueue(md->wq).
2724 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2726 synchronize_srcu(&md->io_barrier);
2729 * Stop md->queue before flushing md->wq in case request-based
2730 * dm defers requests to md->wq from md->queue.
2732 if (dm_request_based(md))
2733 dm_stop_queue(md->queue);
2735 flush_workqueue(md->wq);
2738 * At this point no more requests are entering target request routines.
2739 * We call dm_wait_for_completion to wait for all existing requests
2742 r = dm_wait_for_completion(md, task_state);
2744 set_bit(dmf_suspended_flag, &md->flags);
2747 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2749 synchronize_srcu(&md->io_barrier);
2751 /* were we interrupted ? */
2755 if (dm_request_based(md))
2756 dm_start_queue(md->queue);
2759 dm_table_presuspend_undo_targets(map);
2760 /* pushback list is already flushed, so skip flush */
2767 * We need to be able to change a mapping table under a mounted
2768 * filesystem. For example we might want to move some data in
2769 * the background. Before the table can be swapped with
2770 * dm_bind_table, dm_suspend must be called to flush any in
2771 * flight bios and ensure that any further io gets deferred.
2774 * Suspend mechanism in request-based dm.
2776 * 1. Flush all I/Os by lock_fs() if needed.
2777 * 2. Stop dispatching any I/O by stopping the request_queue.
2778 * 3. Wait for all in-flight I/Os to be completed or requeued.
2780 * To abort suspend, start the request_queue.
2782 int dm_suspend(struct mapped_device *md, unsigned int suspend_flags)
2784 struct dm_table *map = NULL;
2788 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2790 if (dm_suspended_md(md)) {
2795 if (dm_suspended_internally_md(md)) {
2796 /* already internally suspended, wait for internal resume */
2797 mutex_unlock(&md->suspend_lock);
2798 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2804 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2806 /* avoid deadlock with fs/namespace.c:do_mount() */
2807 suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG;
2810 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2814 set_bit(DMF_POST_SUSPENDING, &md->flags);
2815 dm_table_postsuspend_targets(map);
2816 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2819 mutex_unlock(&md->suspend_lock);
2823 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2826 int r = dm_table_resume_targets(map);
2835 * Flushing deferred I/Os must be done after targets are resumed
2836 * so that mapping of targets can work correctly.
2837 * Request-based dm is queueing the deferred I/Os in its request_queue.
2839 if (dm_request_based(md))
2840 dm_start_queue(md->queue);
2847 int dm_resume(struct mapped_device *md)
2850 struct dm_table *map = NULL;
2854 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2856 if (!dm_suspended_md(md))
2859 if (dm_suspended_internally_md(md)) {
2860 /* already internally suspended, wait for internal resume */
2861 mutex_unlock(&md->suspend_lock);
2862 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2868 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2869 if (!map || !dm_table_get_size(map))
2872 r = __dm_resume(md, map);
2876 clear_bit(DMF_SUSPENDED, &md->flags);
2878 mutex_unlock(&md->suspend_lock);
2884 * Internal suspend/resume works like userspace-driven suspend. It waits
2885 * until all bios finish and prevents issuing new bios to the target drivers.
2886 * It may be used only from the kernel.
2889 static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags)
2891 struct dm_table *map = NULL;
2893 lockdep_assert_held(&md->suspend_lock);
2895 if (md->internal_suspend_count++)
2896 return; /* nested internal suspend */
2898 if (dm_suspended_md(md)) {
2899 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2900 return; /* nest suspend */
2903 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2906 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2907 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2908 * would require changing .presuspend to return an error -- avoid this
2909 * until there is a need for more elaborate variants of internal suspend.
2911 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2912 DMF_SUSPENDED_INTERNALLY);
2914 set_bit(DMF_POST_SUSPENDING, &md->flags);
2915 dm_table_postsuspend_targets(map);
2916 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2919 static void __dm_internal_resume(struct mapped_device *md)
2921 BUG_ON(!md->internal_suspend_count);
2923 if (--md->internal_suspend_count)
2924 return; /* resume from nested internal suspend */
2926 if (dm_suspended_md(md))
2927 goto done; /* resume from nested suspend */
2930 * NOTE: existing callers don't need to call dm_table_resume_targets
2931 * (which may fail -- so best to avoid it for now by passing NULL map)
2933 (void) __dm_resume(md, NULL);
2936 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2937 smp_mb__after_atomic();
2938 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2941 void dm_internal_suspend_noflush(struct mapped_device *md)
2943 mutex_lock(&md->suspend_lock);
2944 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2945 mutex_unlock(&md->suspend_lock);
2947 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2949 void dm_internal_resume(struct mapped_device *md)
2951 mutex_lock(&md->suspend_lock);
2952 __dm_internal_resume(md);
2953 mutex_unlock(&md->suspend_lock);
2955 EXPORT_SYMBOL_GPL(dm_internal_resume);
2958 * Fast variants of internal suspend/resume hold md->suspend_lock,
2959 * which prevents interaction with userspace-driven suspend.
2962 void dm_internal_suspend_fast(struct mapped_device *md)
2964 mutex_lock(&md->suspend_lock);
2965 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2968 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2969 synchronize_srcu(&md->io_barrier);
2970 flush_workqueue(md->wq);
2971 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2973 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2975 void dm_internal_resume_fast(struct mapped_device *md)
2977 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2983 mutex_unlock(&md->suspend_lock);
2985 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2988 *---------------------------------------------------------------
2989 * Event notification.
2990 *---------------------------------------------------------------
2992 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2993 unsigned int cookie, bool need_resize_uevent)
2996 unsigned int noio_flag;
2997 char udev_cookie[DM_COOKIE_LENGTH];
2998 char *envp[3] = { NULL, NULL, NULL };
2999 char **envpp = envp;
3001 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3002 DM_COOKIE_ENV_VAR_NAME, cookie);
3003 *envpp++ = udev_cookie;
3005 if (need_resize_uevent) {
3006 *envpp++ = "RESIZE=1";
3009 noio_flag = memalloc_noio_save();
3011 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
3013 memalloc_noio_restore(noio_flag);
3018 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3020 return atomic_add_return(1, &md->uevent_seq);
3023 uint32_t dm_get_event_nr(struct mapped_device *md)
3025 return atomic_read(&md->event_nr);
3028 int dm_wait_event(struct mapped_device *md, int event_nr)
3030 return wait_event_interruptible(md->eventq,
3031 (event_nr != atomic_read(&md->event_nr)));
3034 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3036 unsigned long flags;
3038 spin_lock_irqsave(&md->uevent_lock, flags);
3039 list_add(elist, &md->uevent_list);
3040 spin_unlock_irqrestore(&md->uevent_lock, flags);
3044 * The gendisk is only valid as long as you have a reference
3047 struct gendisk *dm_disk(struct mapped_device *md)
3051 EXPORT_SYMBOL_GPL(dm_disk);
3053 struct kobject *dm_kobject(struct mapped_device *md)
3055 return &md->kobj_holder.kobj;
3058 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3060 struct mapped_device *md;
3062 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3064 spin_lock(&_minor_lock);
3065 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3071 spin_unlock(&_minor_lock);
3076 int dm_suspended_md(struct mapped_device *md)
3078 return test_bit(DMF_SUSPENDED, &md->flags);
3081 static int dm_post_suspending_md(struct mapped_device *md)
3083 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3086 int dm_suspended_internally_md(struct mapped_device *md)
3088 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3091 int dm_test_deferred_remove_flag(struct mapped_device *md)
3093 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3096 int dm_suspended(struct dm_target *ti)
3098 return dm_suspended_md(ti->table->md);
3100 EXPORT_SYMBOL_GPL(dm_suspended);
3102 int dm_post_suspending(struct dm_target *ti)
3104 return dm_post_suspending_md(ti->table->md);
3106 EXPORT_SYMBOL_GPL(dm_post_suspending);
3108 int dm_noflush_suspending(struct dm_target *ti)
3110 return __noflush_suspending(ti->table->md);
3112 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3114 void dm_free_md_mempools(struct dm_md_mempools *pools)
3119 bioset_exit(&pools->bs);
3120 bioset_exit(&pools->io_bs);
3133 struct pr_keys *read_keys;
3134 struct pr_held_reservation *rsv;
3137 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3140 struct mapped_device *md = bdev->bd_disk->private_data;
3141 struct dm_table *table;
3142 struct dm_target *ti;
3143 int ret = -ENOTTY, srcu_idx;
3145 table = dm_get_live_table(md, &srcu_idx);
3146 if (!table || !dm_table_get_size(table))
3149 /* We only support devices that have a single target */
3150 if (table->num_targets != 1)
3152 ti = dm_table_get_target(table, 0);
3154 if (dm_suspended_md(md)) {
3160 if (!ti->type->iterate_devices)
3163 ti->type->iterate_devices(ti, fn, pr);
3166 dm_put_live_table(md, srcu_idx);
3171 * For register / unregister we need to manually call out to every path.
3173 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3174 sector_t start, sector_t len, void *data)
3176 struct dm_pr *pr = data;
3177 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3180 if (!ops || !ops->pr_register) {
3181 pr->ret = -EOPNOTSUPP;
3185 ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3198 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3210 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3212 /* Didn't even get to register a path */
3223 /* unregister all paths if we failed to register any path */
3224 pr.old_key = new_key;
3227 pr.fail_early = false;
3228 (void) dm_call_pr(bdev, __dm_pr_register, &pr);
3233 static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3234 sector_t start, sector_t len, void *data)
3236 struct dm_pr *pr = data;
3237 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3239 if (!ops || !ops->pr_reserve) {
3240 pr->ret = -EOPNOTSUPP;
3244 pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3251 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3258 .fail_early = false,
3263 ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3271 * If there is a non-All Registrants type of reservation, the release must be
3272 * sent down the holding path. For the cases where there is no reservation or
3273 * the path is not the holder the device will also return success, so we must
3274 * try each path to make sure we got the correct path.
3276 static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3277 sector_t start, sector_t len, void *data)
3279 struct dm_pr *pr = data;
3280 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3282 if (!ops || !ops->pr_release) {
3283 pr->ret = -EOPNOTSUPP;
3287 pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3294 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3299 .fail_early = false,
3303 ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3310 static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3311 sector_t start, sector_t len, void *data)
3313 struct dm_pr *pr = data;
3314 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3316 if (!ops || !ops->pr_preempt) {
3317 pr->ret = -EOPNOTSUPP;
3321 pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3329 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3330 enum pr_type type, bool abort)
3336 .fail_early = false,
3340 ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3347 static int dm_pr_clear(struct block_device *bdev, u64 key)
3349 struct mapped_device *md = bdev->bd_disk->private_data;
3350 const struct pr_ops *ops;
3353 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3357 ops = bdev->bd_disk->fops->pr_ops;
3358 if (ops && ops->pr_clear)
3359 r = ops->pr_clear(bdev, key);
3363 dm_unprepare_ioctl(md, srcu_idx);
3367 static int __dm_pr_read_keys(struct dm_target *ti, struct dm_dev *dev,
3368 sector_t start, sector_t len, void *data)
3370 struct dm_pr *pr = data;
3371 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3373 if (!ops || !ops->pr_read_keys) {
3374 pr->ret = -EOPNOTSUPP;
3378 pr->ret = ops->pr_read_keys(dev->bdev, pr->read_keys);
3385 static int dm_pr_read_keys(struct block_device *bdev, struct pr_keys *keys)
3392 ret = dm_call_pr(bdev, __dm_pr_read_keys, &pr);
3399 static int __dm_pr_read_reservation(struct dm_target *ti, struct dm_dev *dev,
3400 sector_t start, sector_t len, void *data)
3402 struct dm_pr *pr = data;
3403 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3405 if (!ops || !ops->pr_read_reservation) {
3406 pr->ret = -EOPNOTSUPP;
3410 pr->ret = ops->pr_read_reservation(dev->bdev, pr->rsv);
3417 static int dm_pr_read_reservation(struct block_device *bdev,
3418 struct pr_held_reservation *rsv)
3425 ret = dm_call_pr(bdev, __dm_pr_read_reservation, &pr);
3432 static const struct pr_ops dm_pr_ops = {
3433 .pr_register = dm_pr_register,
3434 .pr_reserve = dm_pr_reserve,
3435 .pr_release = dm_pr_release,
3436 .pr_preempt = dm_pr_preempt,
3437 .pr_clear = dm_pr_clear,
3438 .pr_read_keys = dm_pr_read_keys,
3439 .pr_read_reservation = dm_pr_read_reservation,
3442 static const struct block_device_operations dm_blk_dops = {
3443 .submit_bio = dm_submit_bio,
3444 .poll_bio = dm_poll_bio,
3445 .open = dm_blk_open,
3446 .release = dm_blk_close,
3447 .ioctl = dm_blk_ioctl,
3448 .getgeo = dm_blk_getgeo,
3449 .report_zones = dm_blk_report_zones,
3450 .pr_ops = &dm_pr_ops,
3451 .owner = THIS_MODULE
3454 static const struct block_device_operations dm_rq_blk_dops = {
3455 .open = dm_blk_open,
3456 .release = dm_blk_close,
3457 .ioctl = dm_blk_ioctl,
3458 .getgeo = dm_blk_getgeo,
3459 .pr_ops = &dm_pr_ops,
3460 .owner = THIS_MODULE
3463 static const struct dax_operations dm_dax_ops = {
3464 .direct_access = dm_dax_direct_access,
3465 .zero_page_range = dm_dax_zero_page_range,
3466 .recovery_write = dm_dax_recovery_write,
3472 module_init(dm_init);
3473 module_exit(dm_exit);
3475 module_param(major, uint, 0);
3476 MODULE_PARM_DESC(major, "The major number of the device mapper");
3478 module_param(reserved_bio_based_ios, uint, 0644);
3479 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3481 module_param(dm_numa_node, int, 0644);
3482 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3484 module_param(swap_bios, int, 0644);
3485 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3487 MODULE_DESCRIPTION(DM_NAME " driver");
3488 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3489 MODULE_LICENSE("GPL");