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_workqueue("kdmremove", WQ_UNBOUND, 1);
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 block_device *bdev, fmode_t mode)
315 struct mapped_device *md;
317 spin_lock(&_minor_lock);
319 md = bdev->bd_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, fmode_t mode)
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, fmode_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 bool bio_is_flush_with_data(struct bio *bio)
492 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
495 static void dm_io_acct(struct dm_io *io, bool end)
497 struct dm_stats_aux *stats_aux = &io->stats_aux;
498 unsigned long start_time = io->start_time;
499 struct mapped_device *md = io->md;
500 struct bio *bio = io->orig_bio;
501 unsigned int sectors;
504 * If REQ_PREFLUSH set, don't account payload, it will be
505 * submitted (and accounted) after this flush completes.
507 if (bio_is_flush_with_data(bio))
509 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
510 sectors = bio_sectors(bio);
512 sectors = io->sectors;
515 bdev_start_io_acct(bio->bi_bdev, bio_op(bio), start_time);
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), sectors,
520 if (static_branch_unlikely(&stats_enabled) &&
521 unlikely(dm_stats_used(&md->stats))) {
524 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
525 sector = bio->bi_iter.bi_sector;
527 sector = bio_end_sector(bio) - io->sector_offset;
529 dm_stats_account_io(&md->stats, bio_data_dir(bio),
531 end, start_time, stats_aux);
535 static void __dm_start_io_acct(struct dm_io *io)
537 dm_io_acct(io, false);
540 static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
543 * Ensure IO accounting is only ever started once.
545 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
548 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
549 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
550 dm_io_set_flag(io, DM_IO_ACCOUNTED);
553 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
554 spin_lock_irqsave(&io->lock, flags);
555 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
556 spin_unlock_irqrestore(&io->lock, flags);
559 dm_io_set_flag(io, DM_IO_ACCOUNTED);
560 spin_unlock_irqrestore(&io->lock, flags);
563 __dm_start_io_acct(io);
566 static void dm_end_io_acct(struct dm_io *io)
568 dm_io_acct(io, true);
571 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
574 struct dm_target_io *tio;
577 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
578 tio = clone_to_tio(clone);
580 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
583 io = container_of(tio, struct dm_io, tio);
584 io->magic = DM_IO_MAGIC;
585 io->status = BLK_STS_OK;
587 /* one ref is for submission, the other is for completion */
588 atomic_set(&io->io_count, 2);
589 this_cpu_inc(*md->pending_io);
592 spin_lock_init(&io->lock);
593 io->start_time = jiffies;
596 if (static_branch_unlikely(&stats_enabled))
597 dm_stats_record_start(&md->stats, &io->stats_aux);
602 static void free_io(struct dm_io *io)
604 bio_put(&io->tio.clone);
607 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
608 unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask)
610 struct mapped_device *md = ci->io->md;
611 struct dm_target_io *tio;
614 if (!ci->io->tio.io) {
615 /* the dm_target_io embedded in ci->io is available */
617 /* alloc_io() already initialized embedded clone */
620 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
625 /* REQ_DM_POLL_LIST shouldn't be inherited */
626 clone->bi_opf &= ~REQ_DM_POLL_LIST;
628 tio = clone_to_tio(clone);
629 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
632 tio->magic = DM_TIO_MAGIC;
635 tio->target_bio_nr = target_bio_nr;
639 /* Set default bdev, but target must bio_set_dev() before issuing IO */
640 clone->bi_bdev = md->disk->part0;
641 if (unlikely(ti->needs_bio_set_dev))
642 bio_set_dev(clone, md->disk->part0);
645 clone->bi_iter.bi_size = to_bytes(*len);
646 if (bio_integrity(clone))
647 bio_integrity_trim(clone);
653 static void free_tio(struct bio *clone)
655 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
661 * Add the bio to the list of deferred io.
663 static void queue_io(struct mapped_device *md, struct bio *bio)
667 spin_lock_irqsave(&md->deferred_lock, flags);
668 bio_list_add(&md->deferred, bio);
669 spin_unlock_irqrestore(&md->deferred_lock, flags);
670 queue_work(md->wq, &md->work);
674 * Everyone (including functions in this file), should use this
675 * function to access the md->map field, and make sure they call
676 * dm_put_live_table() when finished.
678 struct dm_table *dm_get_live_table(struct mapped_device *md,
679 int *srcu_idx) __acquires(md->io_barrier)
681 *srcu_idx = srcu_read_lock(&md->io_barrier);
683 return srcu_dereference(md->map, &md->io_barrier);
686 void dm_put_live_table(struct mapped_device *md,
687 int srcu_idx) __releases(md->io_barrier)
689 srcu_read_unlock(&md->io_barrier, srcu_idx);
692 void dm_sync_table(struct mapped_device *md)
694 synchronize_srcu(&md->io_barrier);
695 synchronize_rcu_expedited();
699 * A fast alternative to dm_get_live_table/dm_put_live_table.
700 * The caller must not block between these two functions.
702 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
705 return rcu_dereference(md->map);
708 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
713 static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
714 int *srcu_idx, blk_opf_t bio_opf)
716 if (bio_opf & REQ_NOWAIT)
717 return dm_get_live_table_fast(md);
719 return dm_get_live_table(md, srcu_idx);
722 static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
725 if (bio_opf & REQ_NOWAIT)
726 dm_put_live_table_fast(md);
728 dm_put_live_table(md, srcu_idx);
731 static char *_dm_claim_ptr = "I belong to device-mapper";
734 * Open a table device so we can use it as a map destination.
736 static struct table_device *open_table_device(struct mapped_device *md,
737 dev_t dev, fmode_t mode)
739 struct table_device *td;
740 struct block_device *bdev;
744 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
746 return ERR_PTR(-ENOMEM);
747 refcount_set(&td->count, 1);
749 bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr, NULL);
756 * We can be called before the dm disk is added. In that case we can't
757 * register the holder relation here. It will be done once add_disk was
760 if (md->disk->slave_dir) {
761 r = bd_link_disk_holder(bdev, md->disk);
766 td->dm_dev.mode = mode;
767 td->dm_dev.bdev = bdev;
768 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
769 format_dev_t(td->dm_dev.name, dev);
770 list_add(&td->list, &md->table_devices);
774 blkdev_put(bdev, mode | FMODE_EXCL);
781 * Close a table device that we've been using.
783 static void close_table_device(struct table_device *td, struct mapped_device *md)
785 if (md->disk->slave_dir)
786 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
787 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
788 put_dax(td->dm_dev.dax_dev);
793 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
796 struct table_device *td;
798 list_for_each_entry(td, l, list)
799 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
805 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
806 struct dm_dev **result)
808 struct table_device *td;
810 mutex_lock(&md->table_devices_lock);
811 td = find_table_device(&md->table_devices, dev, mode);
813 td = open_table_device(md, dev, mode);
815 mutex_unlock(&md->table_devices_lock);
819 refcount_inc(&td->count);
821 mutex_unlock(&md->table_devices_lock);
823 *result = &td->dm_dev;
827 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
829 struct table_device *td = container_of(d, struct table_device, dm_dev);
831 mutex_lock(&md->table_devices_lock);
832 if (refcount_dec_and_test(&td->count))
833 close_table_device(td, md);
834 mutex_unlock(&md->table_devices_lock);
838 * Get the geometry associated with a dm device
840 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
848 * Set the geometry of a device.
850 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
852 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
854 if (geo->start > sz) {
855 DMERR("Start sector is beyond the geometry limits.");
864 static int __noflush_suspending(struct mapped_device *md)
866 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
869 static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
871 struct mapped_device *md = io->md;
874 struct dm_io *next = md->requeue_list;
876 md->requeue_list = io;
879 bio_list_add_head(&md->deferred, io->orig_bio);
883 static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
886 queue_work(md->wq, &md->requeue_work);
888 queue_work(md->wq, &md->work);
892 * Return true if the dm_io's original bio is requeued.
893 * io->status is updated with error if requeue disallowed.
895 static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
897 struct bio *bio = io->orig_bio;
898 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
899 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
900 (bio->bi_opf & REQ_POLLED));
901 struct mapped_device *md = io->md;
902 bool requeued = false;
904 if (handle_requeue || handle_polled_eagain) {
907 if (bio->bi_opf & REQ_POLLED) {
909 * Upper layer won't help us poll split bio
910 * (io->orig_bio may only reflect a subset of the
911 * pre-split original) so clear REQ_POLLED.
913 bio_clear_polled(bio);
917 * Target requested pushing back the I/O or
918 * polled IO hit BLK_STS_AGAIN.
920 spin_lock_irqsave(&md->deferred_lock, flags);
921 if ((__noflush_suspending(md) &&
922 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
923 handle_polled_eagain || first_stage) {
924 dm_requeue_add_io(io, first_stage);
928 * noflush suspend was interrupted or this is
929 * a write to a zoned target.
931 io->status = BLK_STS_IOERR;
933 spin_unlock_irqrestore(&md->deferred_lock, flags);
937 dm_kick_requeue(md, first_stage);
942 static void __dm_io_complete(struct dm_io *io, bool first_stage)
944 struct bio *bio = io->orig_bio;
945 struct mapped_device *md = io->md;
946 blk_status_t io_error;
949 requeued = dm_handle_requeue(io, first_stage);
950 if (requeued && first_stage)
953 io_error = io->status;
954 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
956 else if (!io_error) {
958 * Must handle target that DM_MAPIO_SUBMITTED only to
959 * then bio_endio() rather than dm_submit_bio_remap()
961 __dm_start_io_acct(io);
966 this_cpu_dec(*md->pending_io);
968 /* nudge anyone waiting on suspend queue */
969 if (unlikely(wq_has_sleeper(&md->wait)))
972 /* Return early if the original bio was requeued */
976 if (bio_is_flush_with_data(bio)) {
978 * Preflush done for flush with data, reissue
979 * without REQ_PREFLUSH.
981 bio->bi_opf &= ~REQ_PREFLUSH;
984 /* done with normal IO or empty flush */
986 bio->bi_status = io_error;
991 static void dm_wq_requeue_work(struct work_struct *work)
993 struct mapped_device *md = container_of(work, struct mapped_device,
998 /* reuse deferred lock to simplify dm_handle_requeue */
999 spin_lock_irqsave(&md->deferred_lock, flags);
1000 io = md->requeue_list;
1001 md->requeue_list = NULL;
1002 spin_unlock_irqrestore(&md->deferred_lock, flags);
1005 struct dm_io *next = io->next;
1007 dm_io_rewind(io, &md->disk->bio_split);
1010 __dm_io_complete(io, false);
1017 * Two staged requeue:
1019 * 1) io->orig_bio points to the real original bio, and the part mapped to
1020 * this io must be requeued, instead of other parts of the original bio.
1022 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1024 static void dm_io_complete(struct dm_io *io)
1029 * Only dm_io that has been split needs two stage requeue, otherwise
1030 * we may run into long bio clone chain during suspend and OOM could
1033 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1034 * also aren't handled via the first stage requeue.
1036 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1037 first_requeue = true;
1039 first_requeue = false;
1041 __dm_io_complete(io, first_requeue);
1045 * Decrements the number of outstanding ios that a bio has been
1046 * cloned into, completing the original io if necc.
1048 static inline void __dm_io_dec_pending(struct dm_io *io)
1050 if (atomic_dec_and_test(&io->io_count))
1054 static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1056 unsigned long flags;
1058 /* Push-back supersedes any I/O errors */
1059 spin_lock_irqsave(&io->lock, flags);
1060 if (!(io->status == BLK_STS_DM_REQUEUE &&
1061 __noflush_suspending(io->md))) {
1064 spin_unlock_irqrestore(&io->lock, flags);
1067 static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1069 if (unlikely(error))
1070 dm_io_set_error(io, error);
1072 __dm_io_dec_pending(io);
1076 * The queue_limits are only valid as long as you have a reference
1077 * count on 'md'. But _not_ imposing verification to avoid atomic_read(),
1079 static inline struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1081 return &md->queue->limits;
1084 void disable_discard(struct mapped_device *md)
1086 struct queue_limits *limits = dm_get_queue_limits(md);
1088 /* device doesn't really support DISCARD, disable it */
1089 limits->max_discard_sectors = 0;
1092 void disable_write_zeroes(struct mapped_device *md)
1094 struct queue_limits *limits = dm_get_queue_limits(md);
1096 /* device doesn't really support WRITE ZEROES, disable it */
1097 limits->max_write_zeroes_sectors = 0;
1100 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1102 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1105 static void clone_endio(struct bio *bio)
1107 blk_status_t error = bio->bi_status;
1108 struct dm_target_io *tio = clone_to_tio(bio);
1109 struct dm_target *ti = tio->ti;
1110 dm_endio_fn endio = ti->type->end_io;
1111 struct dm_io *io = tio->io;
1112 struct mapped_device *md = io->md;
1114 if (unlikely(error == BLK_STS_TARGET)) {
1115 if (bio_op(bio) == REQ_OP_DISCARD &&
1116 !bdev_max_discard_sectors(bio->bi_bdev))
1117 disable_discard(md);
1118 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1119 !bdev_write_zeroes_sectors(bio->bi_bdev))
1120 disable_write_zeroes(md);
1123 if (static_branch_unlikely(&zoned_enabled) &&
1124 unlikely(bdev_is_zoned(bio->bi_bdev)))
1125 dm_zone_endio(io, bio);
1128 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,
1175 unsigned int max_granularity)
1177 sector_t target_offset = dm_target_offset(ti, sector);
1178 sector_t len = max_io_len_target_boundary(ti, target_offset);
1181 * Does the target need to split IO even further?
1182 * - varied (per target) IO splitting is a tenet of DM; this
1183 * explains why stacked chunk_sectors based splitting via
1184 * bio_split_to_limits() isn't possible here.
1186 if (!max_granularity)
1188 return min_t(sector_t, len,
1189 min(queue_max_sectors(ti->table->md->queue),
1190 blk_chunk_sectors_left(target_offset, max_granularity)));
1193 static inline sector_t max_io_len(struct dm_target *ti, sector_t sector)
1195 return __max_io_len(ti, sector, ti->max_io_len);
1198 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1200 if (len > UINT_MAX) {
1201 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1202 (unsigned long long)len, UINT_MAX);
1203 ti->error = "Maximum size of target IO is too large";
1207 ti->max_io_len = (uint32_t) len;
1211 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1213 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1214 sector_t sector, int *srcu_idx)
1215 __acquires(md->io_barrier)
1217 struct dm_table *map;
1218 struct dm_target *ti;
1220 map = dm_get_live_table(md, srcu_idx);
1224 ti = dm_table_find_target(map, sector);
1231 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1232 long nr_pages, enum dax_access_mode mode, void **kaddr,
1235 struct mapped_device *md = dax_get_private(dax_dev);
1236 sector_t sector = pgoff * PAGE_SECTORS;
1237 struct dm_target *ti;
1238 long len, ret = -EIO;
1241 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1245 if (!ti->type->direct_access)
1247 len = max_io_len(ti, sector) / PAGE_SECTORS;
1250 nr_pages = min(len, nr_pages);
1251 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1254 dm_put_live_table(md, srcu_idx);
1259 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1262 struct mapped_device *md = dax_get_private(dax_dev);
1263 sector_t sector = pgoff * PAGE_SECTORS;
1264 struct dm_target *ti;
1268 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1272 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1274 * ->zero_page_range() is mandatory dax operation. If we are
1275 * here, something is wrong.
1279 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1281 dm_put_live_table(md, srcu_idx);
1286 static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1287 void *addr, size_t bytes, struct iov_iter *i)
1289 struct mapped_device *md = dax_get_private(dax_dev);
1290 sector_t sector = pgoff * PAGE_SECTORS;
1291 struct dm_target *ti;
1295 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1296 if (!ti || !ti->type->dax_recovery_write)
1299 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1301 dm_put_live_table(md, srcu_idx);
1306 * A target may call dm_accept_partial_bio only from the map routine. It is
1307 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1308 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1309 * __send_duplicate_bios().
1311 * dm_accept_partial_bio informs the dm that the target only wants to process
1312 * additional n_sectors sectors of the bio and the rest of the data should be
1313 * sent in a next bio.
1315 * A diagram that explains the arithmetics:
1316 * +--------------------+---------------+-------+
1318 * +--------------------+---------------+-------+
1320 * <-------------- *tio->len_ptr --------------->
1321 * <----- bio_sectors ----->
1324 * Region 1 was already iterated over with bio_advance or similar function.
1325 * (it may be empty if the target doesn't use bio_advance)
1326 * Region 2 is the remaining bio size that the target wants to process.
1327 * (it may be empty if region 1 is non-empty, although there is no reason
1329 * The target requires that region 3 is to be sent in the next bio.
1331 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1332 * the partially processed part (the sum of regions 1+2) must be the same for all
1333 * copies of the bio.
1335 void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors)
1337 struct dm_target_io *tio = clone_to_tio(bio);
1338 struct dm_io *io = tio->io;
1339 unsigned int bio_sectors = bio_sectors(bio);
1341 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1342 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1343 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1344 BUG_ON(bio_sectors > *tio->len_ptr);
1345 BUG_ON(n_sectors > bio_sectors);
1347 *tio->len_ptr -= bio_sectors - n_sectors;
1348 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1351 * __split_and_process_bio() may have already saved mapped part
1352 * for accounting but it is being reduced so update accordingly.
1354 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1355 io->sectors = n_sectors;
1356 io->sector_offset = bio_sectors(io->orig_bio);
1358 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1361 * @clone: clone bio that DM core passed to target's .map function
1362 * @tgt_clone: clone of @clone bio that target needs submitted
1364 * Targets should use this interface to submit bios they take
1365 * ownership of when returning DM_MAPIO_SUBMITTED.
1367 * Target should also enable ti->accounts_remapped_io
1369 void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1371 struct dm_target_io *tio = clone_to_tio(clone);
1372 struct dm_io *io = tio->io;
1374 /* establish bio that will get submitted */
1379 * Account io->origin_bio to DM dev on behalf of target
1380 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1382 dm_start_io_acct(io, clone);
1384 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1386 submit_bio_noacct(tgt_clone);
1388 EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1390 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1392 mutex_lock(&md->swap_bios_lock);
1393 while (latch < md->swap_bios) {
1395 down(&md->swap_bios_semaphore);
1398 while (latch > md->swap_bios) {
1400 up(&md->swap_bios_semaphore);
1403 mutex_unlock(&md->swap_bios_lock);
1406 static void __map_bio(struct bio *clone)
1408 struct dm_target_io *tio = clone_to_tio(clone);
1409 struct dm_target *ti = tio->ti;
1410 struct dm_io *io = tio->io;
1411 struct mapped_device *md = io->md;
1414 clone->bi_end_io = clone_endio;
1419 tio->old_sector = clone->bi_iter.bi_sector;
1421 if (static_branch_unlikely(&swap_bios_enabled) &&
1422 unlikely(swap_bios_limit(ti, clone))) {
1423 int latch = get_swap_bios();
1425 if (unlikely(latch != md->swap_bios))
1426 __set_swap_bios_limit(md, latch);
1427 down(&md->swap_bios_semaphore);
1430 if (static_branch_unlikely(&zoned_enabled)) {
1432 * Check if the IO needs a special mapping due to zone append
1433 * emulation on zoned target. In this case, dm_zone_map_bio()
1434 * calls the target map operation.
1436 if (unlikely(dm_emulate_zone_append(md)))
1437 r = dm_zone_map_bio(tio);
1439 r = ti->type->map(ti, clone);
1441 r = ti->type->map(ti, clone);
1444 case DM_MAPIO_SUBMITTED:
1445 /* target has assumed ownership of this io */
1446 if (!ti->accounts_remapped_io)
1447 dm_start_io_acct(io, clone);
1449 case DM_MAPIO_REMAPPED:
1450 dm_submit_bio_remap(clone, NULL);
1453 case DM_MAPIO_REQUEUE:
1454 if (static_branch_unlikely(&swap_bios_enabled) &&
1455 unlikely(swap_bios_limit(ti, clone)))
1456 up(&md->swap_bios_semaphore);
1458 if (r == DM_MAPIO_KILL)
1459 dm_io_dec_pending(io, BLK_STS_IOERR);
1461 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1464 DMCRIT("unimplemented target map return value: %d", r);
1469 static void setup_split_accounting(struct clone_info *ci, unsigned int len)
1471 struct dm_io *io = ci->io;
1473 if (ci->sector_count > len) {
1475 * Split needed, save the mapped part for accounting.
1476 * NOTE: dm_accept_partial_bio() will update accordingly.
1478 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1480 io->sector_offset = bio_sectors(ci->bio);
1484 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1485 struct dm_target *ti, unsigned int num_bios,
1491 for (try = 0; try < 2; try++) {
1495 mutex_lock(&ci->io->md->table_devices_lock);
1496 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1497 bio = alloc_tio(ci, ti, bio_nr, len,
1498 try ? GFP_NOIO : GFP_NOWAIT);
1502 bio_list_add(blist, bio);
1505 mutex_unlock(&ci->io->md->table_devices_lock);
1506 if (bio_nr == num_bios)
1509 while ((bio = bio_list_pop(blist)))
1514 static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1515 unsigned int num_bios, unsigned int *len)
1517 struct bio_list blist = BIO_EMPTY_LIST;
1519 unsigned int ret = 0;
1526 setup_split_accounting(ci, *len);
1527 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1533 setup_split_accounting(ci, *len);
1534 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1535 alloc_multiple_bios(&blist, ci, ti, num_bios, len);
1536 while ((clone = bio_list_pop(&blist))) {
1537 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1547 static void __send_empty_flush(struct clone_info *ci)
1549 struct dm_table *t = ci->map;
1550 struct bio flush_bio;
1553 * Use an on-stack bio for this, it's safe since we don't
1554 * need to reference it after submit. It's just used as
1555 * the basis for the clone(s).
1557 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1558 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1560 ci->bio = &flush_bio;
1561 ci->sector_count = 0;
1562 ci->io->tio.clone.bi_iter.bi_size = 0;
1564 for (unsigned int i = 0; i < t->num_targets; i++) {
1566 struct dm_target *ti = dm_table_get_target(t, i);
1568 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1569 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1570 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1574 * alloc_io() takes one extra reference for submission, so the
1575 * reference won't reach 0 without the following subtraction
1577 atomic_sub(1, &ci->io->io_count);
1579 bio_uninit(ci->bio);
1582 static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1583 unsigned int num_bios,
1584 unsigned int max_granularity)
1586 unsigned int len, bios;
1588 len = min_t(sector_t, ci->sector_count,
1589 __max_io_len(ti, ci->sector, max_granularity));
1591 atomic_add(num_bios, &ci->io->io_count);
1592 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1594 * alloc_io() takes one extra reference for submission, so the
1595 * reference won't reach 0 without the following (+1) subtraction
1597 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1600 ci->sector_count -= len;
1603 static bool is_abnormal_io(struct bio *bio)
1605 enum req_op op = bio_op(bio);
1607 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1609 case REQ_OP_DISCARD:
1610 case REQ_OP_SECURE_ERASE:
1611 case REQ_OP_WRITE_ZEROES:
1621 static blk_status_t __process_abnormal_io(struct clone_info *ci,
1622 struct dm_target *ti)
1624 unsigned int num_bios = 0;
1625 unsigned int max_granularity = 0;
1626 struct queue_limits *limits = dm_get_queue_limits(ti->table->md);
1628 switch (bio_op(ci->bio)) {
1629 case REQ_OP_DISCARD:
1630 num_bios = ti->num_discard_bios;
1631 if (ti->max_discard_granularity)
1632 max_granularity = limits->max_discard_sectors;
1634 case REQ_OP_SECURE_ERASE:
1635 num_bios = ti->num_secure_erase_bios;
1636 if (ti->max_secure_erase_granularity)
1637 max_granularity = limits->max_secure_erase_sectors;
1639 case REQ_OP_WRITE_ZEROES:
1640 num_bios = ti->num_write_zeroes_bios;
1641 if (ti->max_write_zeroes_granularity)
1642 max_granularity = limits->max_write_zeroes_sectors;
1649 * Even though the device advertised support for this type of
1650 * request, that does not mean every target supports it, and
1651 * reconfiguration might also have changed that since the
1652 * check was performed.
1654 if (unlikely(!num_bios))
1655 return BLK_STS_NOTSUPP;
1657 __send_changing_extent_only(ci, ti, num_bios, max_granularity);
1662 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1663 * associated with this bio, and this bio's bi_private needs to be
1664 * stored in dm_io->data before the reuse.
1666 * bio->bi_private is owned by fs or upper layer, so block layer won't
1667 * touch it after splitting. Meantime it won't be changed by anyone after
1668 * bio is submitted. So this reuse is safe.
1670 static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1672 return (struct dm_io **)&bio->bi_private;
1675 static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1677 struct dm_io **head = dm_poll_list_head(bio);
1679 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1680 bio->bi_opf |= REQ_DM_POLL_LIST;
1682 * Save .bi_private into dm_io, so that we can reuse
1683 * .bi_private as dm_io list head for storing dm_io list
1685 io->data = bio->bi_private;
1687 /* tell block layer to poll for completion */
1688 bio->bi_cookie = ~BLK_QC_T_NONE;
1693 * bio recursed due to split, reuse original poll list,
1694 * and save bio->bi_private too.
1696 io->data = (*head)->data;
1704 * Select the correct strategy for processing a non-flush bio.
1706 static blk_status_t __split_and_process_bio(struct clone_info *ci)
1709 struct dm_target *ti;
1712 ti = dm_table_find_target(ci->map, ci->sector);
1714 return BLK_STS_IOERR;
1716 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1717 unlikely(!dm_target_supports_nowait(ti->type)))
1718 return BLK_STS_NOTSUPP;
1720 if (unlikely(ci->is_abnormal_io))
1721 return __process_abnormal_io(ci, ti);
1724 * Only support bio polling for normal IO, and the target io is
1725 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1727 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1729 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1730 setup_split_accounting(ci, len);
1731 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1735 ci->sector_count -= len;
1740 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1741 struct dm_table *map, struct bio *bio, bool is_abnormal)
1744 ci->io = alloc_io(md, bio);
1746 ci->is_abnormal_io = is_abnormal;
1747 ci->submit_as_polled = false;
1748 ci->sector = bio->bi_iter.bi_sector;
1749 ci->sector_count = bio_sectors(bio);
1751 /* Shouldn't happen but sector_count was being set to 0 so... */
1752 if (static_branch_unlikely(&zoned_enabled) &&
1753 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1754 ci->sector_count = 0;
1758 * Entry point to split a bio into clones and submit them to the targets.
1760 static void dm_split_and_process_bio(struct mapped_device *md,
1761 struct dm_table *map, struct bio *bio)
1763 struct clone_info ci;
1765 blk_status_t error = BLK_STS_OK;
1768 is_abnormal = is_abnormal_io(bio);
1769 if (unlikely(is_abnormal)) {
1771 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1772 * otherwise associated queue_limits won't be imposed.
1774 bio = bio_split_to_limits(bio);
1779 init_clone_info(&ci, md, map, bio, is_abnormal);
1782 if (bio->bi_opf & REQ_PREFLUSH) {
1783 __send_empty_flush(&ci);
1784 /* dm_io_complete submits any data associated with flush */
1788 error = __split_and_process_bio(&ci);
1789 if (error || !ci.sector_count)
1792 * Remainder must be passed to submit_bio_noacct() so it gets handled
1793 * *after* bios already submitted have been completely processed.
1795 bio_trim(bio, io->sectors, ci.sector_count);
1796 trace_block_split(bio, bio->bi_iter.bi_sector);
1797 bio_inc_remaining(bio);
1798 submit_bio_noacct(bio);
1801 * Drop the extra reference count for non-POLLED bio, and hold one
1802 * reference for POLLED bio, which will be released in dm_poll_bio
1804 * Add every dm_io instance into the dm_io list head which is stored
1805 * in bio->bi_private, so that dm_poll_bio can poll them all.
1807 if (error || !ci.submit_as_polled) {
1809 * In case of submission failure, the extra reference for
1810 * submitting io isn't consumed yet
1813 atomic_dec(&io->io_count);
1814 dm_io_dec_pending(io, error);
1816 dm_queue_poll_io(bio, io);
1819 static void dm_submit_bio(struct bio *bio)
1821 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1823 struct dm_table *map;
1824 blk_opf_t bio_opf = bio->bi_opf;
1826 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1828 /* If suspended, or map not yet available, queue this IO for later */
1829 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1831 if (bio->bi_opf & REQ_NOWAIT)
1832 bio_wouldblock_error(bio);
1833 else if (bio->bi_opf & REQ_RAHEAD)
1840 dm_split_and_process_bio(md, map, bio);
1842 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1845 static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1848 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1850 /* don't poll if the mapped io is done */
1851 if (atomic_read(&io->io_count) > 1)
1852 bio_poll(&io->tio.clone, iob, flags);
1854 /* bio_poll holds the last reference */
1855 return atomic_read(&io->io_count) == 1;
1858 static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1861 struct dm_io **head = dm_poll_list_head(bio);
1862 struct dm_io *list = *head;
1863 struct dm_io *tmp = NULL;
1864 struct dm_io *curr, *next;
1866 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1867 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1870 WARN_ON_ONCE(!list);
1873 * Restore .bi_private before possibly completing dm_io.
1875 * bio_poll() is only possible once @bio has been completely
1876 * submitted via submit_bio_noacct()'s depth-first submission.
1877 * So there is no dm_queue_poll_io() race associated with
1878 * clearing REQ_DM_POLL_LIST here.
1880 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1881 bio->bi_private = list->data;
1883 for (curr = list, next = curr->next; curr; curr = next, next =
1884 curr ? curr->next : NULL) {
1885 if (dm_poll_dm_io(curr, iob, flags)) {
1887 * clone_endio() has already occurred, so no
1888 * error handling is needed here.
1890 __dm_io_dec_pending(curr);
1899 bio->bi_opf |= REQ_DM_POLL_LIST;
1900 /* Reset bio->bi_private to dm_io list head */
1908 *---------------------------------------------------------------
1909 * An IDR is used to keep track of allocated minor numbers.
1910 *---------------------------------------------------------------
1912 static void free_minor(int minor)
1914 spin_lock(&_minor_lock);
1915 idr_remove(&_minor_idr, minor);
1916 spin_unlock(&_minor_lock);
1920 * See if the device with a specific minor # is free.
1922 static int specific_minor(int minor)
1926 if (minor >= (1 << MINORBITS))
1929 idr_preload(GFP_KERNEL);
1930 spin_lock(&_minor_lock);
1932 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1934 spin_unlock(&_minor_lock);
1937 return r == -ENOSPC ? -EBUSY : r;
1941 static int next_free_minor(int *minor)
1945 idr_preload(GFP_KERNEL);
1946 spin_lock(&_minor_lock);
1948 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1950 spin_unlock(&_minor_lock);
1958 static const struct block_device_operations dm_blk_dops;
1959 static const struct block_device_operations dm_rq_blk_dops;
1960 static const struct dax_operations dm_dax_ops;
1962 static void dm_wq_work(struct work_struct *work);
1964 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1965 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1967 dm_destroy_crypto_profile(q->crypto_profile);
1970 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1972 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1975 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1977 static void cleanup_mapped_device(struct mapped_device *md)
1980 destroy_workqueue(md->wq);
1981 dm_free_md_mempools(md->mempools);
1984 dax_remove_host(md->disk);
1985 kill_dax(md->dax_dev);
1986 put_dax(md->dax_dev);
1990 dm_cleanup_zoned_dev(md);
1992 spin_lock(&_minor_lock);
1993 md->disk->private_data = NULL;
1994 spin_unlock(&_minor_lock);
1995 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1996 struct table_device *td;
1999 list_for_each_entry(td, &md->table_devices, list) {
2000 bd_unlink_disk_holder(td->dm_dev.bdev,
2005 * Hold lock to make sure del_gendisk() won't concurrent
2006 * with open/close_table_device().
2008 mutex_lock(&md->table_devices_lock);
2009 del_gendisk(md->disk);
2010 mutex_unlock(&md->table_devices_lock);
2012 dm_queue_destroy_crypto_profile(md->queue);
2016 if (md->pending_io) {
2017 free_percpu(md->pending_io);
2018 md->pending_io = NULL;
2021 cleanup_srcu_struct(&md->io_barrier);
2023 mutex_destroy(&md->suspend_lock);
2024 mutex_destroy(&md->type_lock);
2025 mutex_destroy(&md->table_devices_lock);
2026 mutex_destroy(&md->swap_bios_lock);
2028 dm_mq_cleanup_mapped_device(md);
2032 * Allocate and initialise a blank device with a given minor.
2034 static struct mapped_device *alloc_dev(int minor)
2036 int r, numa_node_id = dm_get_numa_node();
2037 struct mapped_device *md;
2040 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2042 DMERR("unable to allocate device, out of memory.");
2046 if (!try_module_get(THIS_MODULE))
2047 goto bad_module_get;
2049 /* get a minor number for the dev */
2050 if (minor == DM_ANY_MINOR)
2051 r = next_free_minor(&minor);
2053 r = specific_minor(minor);
2057 r = init_srcu_struct(&md->io_barrier);
2059 goto bad_io_barrier;
2061 md->numa_node_id = numa_node_id;
2062 md->init_tio_pdu = false;
2063 md->type = DM_TYPE_NONE;
2064 mutex_init(&md->suspend_lock);
2065 mutex_init(&md->type_lock);
2066 mutex_init(&md->table_devices_lock);
2067 spin_lock_init(&md->deferred_lock);
2068 atomic_set(&md->holders, 1);
2069 atomic_set(&md->open_count, 0);
2070 atomic_set(&md->event_nr, 0);
2071 atomic_set(&md->uevent_seq, 0);
2072 INIT_LIST_HEAD(&md->uevent_list);
2073 INIT_LIST_HEAD(&md->table_devices);
2074 spin_lock_init(&md->uevent_lock);
2077 * default to bio-based until DM table is loaded and md->type
2078 * established. If request-based table is loaded: blk-mq will
2079 * override accordingly.
2081 md->disk = blk_alloc_disk(md->numa_node_id);
2084 md->queue = md->disk->queue;
2086 init_waitqueue_head(&md->wait);
2087 INIT_WORK(&md->work, dm_wq_work);
2088 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2089 init_waitqueue_head(&md->eventq);
2090 init_completion(&md->kobj_holder.completion);
2092 md->requeue_list = NULL;
2093 md->swap_bios = get_swap_bios();
2094 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2095 mutex_init(&md->swap_bios_lock);
2097 md->disk->major = _major;
2098 md->disk->first_minor = minor;
2099 md->disk->minors = 1;
2100 md->disk->flags |= GENHD_FL_NO_PART;
2101 md->disk->fops = &dm_blk_dops;
2102 md->disk->private_data = md;
2103 sprintf(md->disk->disk_name, "dm-%d", minor);
2105 if (IS_ENABLED(CONFIG_FS_DAX)) {
2106 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2107 if (IS_ERR(md->dax_dev)) {
2111 set_dax_nocache(md->dax_dev);
2112 set_dax_nomc(md->dax_dev);
2113 if (dax_add_host(md->dax_dev, md->disk))
2117 format_dev_t(md->name, MKDEV(_major, minor));
2119 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2123 md->pending_io = alloc_percpu(unsigned long);
2124 if (!md->pending_io)
2127 r = dm_stats_init(&md->stats);
2131 /* Populate the mapping, nobody knows we exist yet */
2132 spin_lock(&_minor_lock);
2133 old_md = idr_replace(&_minor_idr, md, minor);
2134 spin_unlock(&_minor_lock);
2136 BUG_ON(old_md != MINOR_ALLOCED);
2141 cleanup_mapped_device(md);
2145 module_put(THIS_MODULE);
2151 static void unlock_fs(struct mapped_device *md);
2153 static void free_dev(struct mapped_device *md)
2155 int minor = MINOR(disk_devt(md->disk));
2159 cleanup_mapped_device(md);
2161 WARN_ON_ONCE(!list_empty(&md->table_devices));
2162 dm_stats_cleanup(&md->stats);
2165 module_put(THIS_MODULE);
2170 * Bind a table to the device.
2172 static void event_callback(void *context)
2174 unsigned long flags;
2176 struct mapped_device *md = context;
2178 spin_lock_irqsave(&md->uevent_lock, flags);
2179 list_splice_init(&md->uevent_list, &uevents);
2180 spin_unlock_irqrestore(&md->uevent_lock, flags);
2182 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2184 atomic_inc(&md->event_nr);
2185 wake_up(&md->eventq);
2186 dm_issue_global_event();
2190 * Returns old map, which caller must destroy.
2192 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2193 struct queue_limits *limits)
2195 struct dm_table *old_map;
2199 lockdep_assert_held(&md->suspend_lock);
2201 size = dm_table_get_size(t);
2204 * Wipe any geometry if the size of the table changed.
2206 if (size != dm_get_size(md))
2207 memset(&md->geometry, 0, sizeof(md->geometry));
2209 set_capacity(md->disk, size);
2211 dm_table_event_callback(t, event_callback, md);
2213 if (dm_table_request_based(t)) {
2215 * Leverage the fact that request-based DM targets are
2216 * immutable singletons - used to optimize dm_mq_queue_rq.
2218 md->immutable_target = dm_table_get_immutable_target(t);
2221 * There is no need to reload with request-based dm because the
2222 * size of front_pad doesn't change.
2224 * Note for future: If you are to reload bioset, prep-ed
2225 * requests in the queue may refer to bio from the old bioset,
2226 * so you must walk through the queue to unprep.
2228 if (!md->mempools) {
2229 md->mempools = t->mempools;
2234 * The md may already have mempools that need changing.
2235 * If so, reload bioset because front_pad may have changed
2236 * because a different table was loaded.
2238 dm_free_md_mempools(md->mempools);
2239 md->mempools = t->mempools;
2243 ret = dm_table_set_restrictions(t, md->queue, limits);
2245 old_map = ERR_PTR(ret);
2249 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2250 rcu_assign_pointer(md->map, (void *)t);
2251 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2260 * Returns unbound table for the caller to free.
2262 static struct dm_table *__unbind(struct mapped_device *md)
2264 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2269 dm_table_event_callback(map, NULL, NULL);
2270 RCU_INIT_POINTER(md->map, NULL);
2277 * Constructor for a new device.
2279 int dm_create(int minor, struct mapped_device **result)
2281 struct mapped_device *md;
2283 md = alloc_dev(minor);
2287 dm_ima_reset_data(md);
2294 * Functions to manage md->type.
2295 * All are required to hold md->type_lock.
2297 void dm_lock_md_type(struct mapped_device *md)
2299 mutex_lock(&md->type_lock);
2302 void dm_unlock_md_type(struct mapped_device *md)
2304 mutex_unlock(&md->type_lock);
2307 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2309 BUG_ON(!mutex_is_locked(&md->type_lock));
2313 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2318 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2320 return md->immutable_target_type;
2324 * Setup the DM device's queue based on md's type
2326 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2328 enum dm_queue_mode type = dm_table_get_type(t);
2329 struct queue_limits limits;
2330 struct table_device *td;
2334 case DM_TYPE_REQUEST_BASED:
2335 md->disk->fops = &dm_rq_blk_dops;
2336 r = dm_mq_init_request_queue(md, t);
2338 DMERR("Cannot initialize queue for request-based dm mapped device");
2342 case DM_TYPE_BIO_BASED:
2343 case DM_TYPE_DAX_BIO_BASED:
2350 r = dm_calculate_queue_limits(t, &limits);
2352 DMERR("Cannot calculate initial queue limits");
2355 r = dm_table_set_restrictions(t, md->queue, &limits);
2360 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2361 * with open_table_device() and close_table_device().
2363 mutex_lock(&md->table_devices_lock);
2364 r = add_disk(md->disk);
2365 mutex_unlock(&md->table_devices_lock);
2370 * Register the holder relationship for devices added before the disk
2373 list_for_each_entry(td, &md->table_devices, list) {
2374 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2376 goto out_undo_holders;
2379 r = dm_sysfs_init(md);
2381 goto out_undo_holders;
2387 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2388 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2389 mutex_lock(&md->table_devices_lock);
2390 del_gendisk(md->disk);
2391 mutex_unlock(&md->table_devices_lock);
2395 struct mapped_device *dm_get_md(dev_t dev)
2397 struct mapped_device *md;
2398 unsigned int minor = MINOR(dev);
2400 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2403 spin_lock(&_minor_lock);
2405 md = idr_find(&_minor_idr, minor);
2406 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2407 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2413 spin_unlock(&_minor_lock);
2417 EXPORT_SYMBOL_GPL(dm_get_md);
2419 void *dm_get_mdptr(struct mapped_device *md)
2421 return md->interface_ptr;
2424 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2426 md->interface_ptr = ptr;
2429 void dm_get(struct mapped_device *md)
2431 atomic_inc(&md->holders);
2432 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2435 int dm_hold(struct mapped_device *md)
2437 spin_lock(&_minor_lock);
2438 if (test_bit(DMF_FREEING, &md->flags)) {
2439 spin_unlock(&_minor_lock);
2443 spin_unlock(&_minor_lock);
2446 EXPORT_SYMBOL_GPL(dm_hold);
2448 const char *dm_device_name(struct mapped_device *md)
2452 EXPORT_SYMBOL_GPL(dm_device_name);
2454 static void __dm_destroy(struct mapped_device *md, bool wait)
2456 struct dm_table *map;
2461 spin_lock(&_minor_lock);
2462 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2463 set_bit(DMF_FREEING, &md->flags);
2464 spin_unlock(&_minor_lock);
2466 blk_mark_disk_dead(md->disk);
2469 * Take suspend_lock so that presuspend and postsuspend methods
2470 * do not race with internal suspend.
2472 mutex_lock(&md->suspend_lock);
2473 map = dm_get_live_table(md, &srcu_idx);
2474 if (!dm_suspended_md(md)) {
2475 dm_table_presuspend_targets(map);
2476 set_bit(DMF_SUSPENDED, &md->flags);
2477 set_bit(DMF_POST_SUSPENDING, &md->flags);
2478 dm_table_postsuspend_targets(map);
2480 /* dm_put_live_table must be before fsleep, otherwise deadlock is possible */
2481 dm_put_live_table(md, srcu_idx);
2482 mutex_unlock(&md->suspend_lock);
2485 * Rare, but there may be I/O requests still going to complete,
2486 * for example. Wait for all references to disappear.
2487 * No one should increment the reference count of the mapped_device,
2488 * after the mapped_device state becomes DMF_FREEING.
2491 while (atomic_read(&md->holders))
2493 else if (atomic_read(&md->holders))
2494 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2495 dm_device_name(md), atomic_read(&md->holders));
2497 dm_table_destroy(__unbind(md));
2501 void dm_destroy(struct mapped_device *md)
2503 __dm_destroy(md, true);
2506 void dm_destroy_immediate(struct mapped_device *md)
2508 __dm_destroy(md, false);
2511 void dm_put(struct mapped_device *md)
2513 atomic_dec(&md->holders);
2515 EXPORT_SYMBOL_GPL(dm_put);
2517 static bool dm_in_flight_bios(struct mapped_device *md)
2520 unsigned long sum = 0;
2522 for_each_possible_cpu(cpu)
2523 sum += *per_cpu_ptr(md->pending_io, cpu);
2528 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2534 prepare_to_wait(&md->wait, &wait, task_state);
2536 if (!dm_in_flight_bios(md))
2539 if (signal_pending_state(task_state, current)) {
2546 finish_wait(&md->wait, &wait);
2553 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2557 if (!queue_is_mq(md->queue))
2558 return dm_wait_for_bios_completion(md, task_state);
2561 if (!blk_mq_queue_inflight(md->queue))
2564 if (signal_pending_state(task_state, current)) {
2576 * Process the deferred bios
2578 static void dm_wq_work(struct work_struct *work)
2580 struct mapped_device *md = container_of(work, struct mapped_device, work);
2583 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2584 spin_lock_irq(&md->deferred_lock);
2585 bio = bio_list_pop(&md->deferred);
2586 spin_unlock_irq(&md->deferred_lock);
2591 submit_bio_noacct(bio);
2596 static void dm_queue_flush(struct mapped_device *md)
2598 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2599 smp_mb__after_atomic();
2600 queue_work(md->wq, &md->work);
2604 * Swap in a new table, returning the old one for the caller to destroy.
2606 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2608 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2609 struct queue_limits limits;
2612 mutex_lock(&md->suspend_lock);
2614 /* device must be suspended */
2615 if (!dm_suspended_md(md))
2619 * If the new table has no data devices, retain the existing limits.
2620 * This helps multipath with queue_if_no_path if all paths disappear,
2621 * then new I/O is queued based on these limits, and then some paths
2624 if (dm_table_has_no_data_devices(table)) {
2625 live_map = dm_get_live_table_fast(md);
2627 limits = md->queue->limits;
2628 dm_put_live_table_fast(md);
2632 r = dm_calculate_queue_limits(table, &limits);
2639 map = __bind(md, table, &limits);
2640 dm_issue_global_event();
2643 mutex_unlock(&md->suspend_lock);
2648 * Functions to lock and unlock any filesystem running on the
2651 static int lock_fs(struct mapped_device *md)
2655 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2657 r = freeze_bdev(md->disk->part0);
2659 set_bit(DMF_FROZEN, &md->flags);
2663 static void unlock_fs(struct mapped_device *md)
2665 if (!test_bit(DMF_FROZEN, &md->flags))
2667 thaw_bdev(md->disk->part0);
2668 clear_bit(DMF_FROZEN, &md->flags);
2672 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2673 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2674 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2676 * If __dm_suspend returns 0, the device is completely quiescent
2677 * now. There is no request-processing activity. All new requests
2678 * are being added to md->deferred list.
2680 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2681 unsigned int suspend_flags, unsigned int task_state,
2682 int dmf_suspended_flag)
2684 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2685 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2688 lockdep_assert_held(&md->suspend_lock);
2691 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2692 * This flag is cleared before dm_suspend returns.
2695 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2697 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2700 * This gets reverted if there's an error later and the targets
2701 * provide the .presuspend_undo hook.
2703 dm_table_presuspend_targets(map);
2706 * Flush I/O to the device.
2707 * Any I/O submitted after lock_fs() may not be flushed.
2708 * noflush takes precedence over do_lockfs.
2709 * (lock_fs() flushes I/Os and waits for them to complete.)
2711 if (!noflush && do_lockfs) {
2714 dm_table_presuspend_undo_targets(map);
2720 * Here we must make sure that no processes are submitting requests
2721 * to target drivers i.e. no one may be executing
2722 * dm_split_and_process_bio from dm_submit_bio.
2724 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2725 * we take the write lock. To prevent any process from reentering
2726 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2727 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2728 * flush_workqueue(md->wq).
2730 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2732 synchronize_srcu(&md->io_barrier);
2735 * Stop md->queue before flushing md->wq in case request-based
2736 * dm defers requests to md->wq from md->queue.
2738 if (dm_request_based(md))
2739 dm_stop_queue(md->queue);
2741 flush_workqueue(md->wq);
2744 * At this point no more requests are entering target request routines.
2745 * We call dm_wait_for_completion to wait for all existing requests
2748 r = dm_wait_for_completion(md, task_state);
2750 set_bit(dmf_suspended_flag, &md->flags);
2753 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2755 synchronize_srcu(&md->io_barrier);
2757 /* were we interrupted ? */
2761 if (dm_request_based(md))
2762 dm_start_queue(md->queue);
2765 dm_table_presuspend_undo_targets(map);
2766 /* pushback list is already flushed, so skip flush */
2773 * We need to be able to change a mapping table under a mounted
2774 * filesystem. For example we might want to move some data in
2775 * the background. Before the table can be swapped with
2776 * dm_bind_table, dm_suspend must be called to flush any in
2777 * flight bios and ensure that any further io gets deferred.
2780 * Suspend mechanism in request-based dm.
2782 * 1. Flush all I/Os by lock_fs() if needed.
2783 * 2. Stop dispatching any I/O by stopping the request_queue.
2784 * 3. Wait for all in-flight I/Os to be completed or requeued.
2786 * To abort suspend, start the request_queue.
2788 int dm_suspend(struct mapped_device *md, unsigned int suspend_flags)
2790 struct dm_table *map = NULL;
2794 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2796 if (dm_suspended_md(md)) {
2801 if (dm_suspended_internally_md(md)) {
2802 /* already internally suspended, wait for internal resume */
2803 mutex_unlock(&md->suspend_lock);
2804 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2810 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2812 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2816 set_bit(DMF_POST_SUSPENDING, &md->flags);
2817 dm_table_postsuspend_targets(map);
2818 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2821 mutex_unlock(&md->suspend_lock);
2825 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2828 int r = dm_table_resume_targets(map);
2837 * Flushing deferred I/Os must be done after targets are resumed
2838 * so that mapping of targets can work correctly.
2839 * Request-based dm is queueing the deferred I/Os in its request_queue.
2841 if (dm_request_based(md))
2842 dm_start_queue(md->queue);
2849 int dm_resume(struct mapped_device *md)
2852 struct dm_table *map = NULL;
2856 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2858 if (!dm_suspended_md(md))
2861 if (dm_suspended_internally_md(md)) {
2862 /* already internally suspended, wait for internal resume */
2863 mutex_unlock(&md->suspend_lock);
2864 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2870 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2871 if (!map || !dm_table_get_size(map))
2874 r = __dm_resume(md, map);
2878 clear_bit(DMF_SUSPENDED, &md->flags);
2880 mutex_unlock(&md->suspend_lock);
2886 * Internal suspend/resume works like userspace-driven suspend. It waits
2887 * until all bios finish and prevents issuing new bios to the target drivers.
2888 * It may be used only from the kernel.
2891 static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags)
2893 struct dm_table *map = NULL;
2895 lockdep_assert_held(&md->suspend_lock);
2897 if (md->internal_suspend_count++)
2898 return; /* nested internal suspend */
2900 if (dm_suspended_md(md)) {
2901 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2902 return; /* nest suspend */
2905 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2908 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2909 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2910 * would require changing .presuspend to return an error -- avoid this
2911 * until there is a need for more elaborate variants of internal suspend.
2913 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2914 DMF_SUSPENDED_INTERNALLY);
2916 set_bit(DMF_POST_SUSPENDING, &md->flags);
2917 dm_table_postsuspend_targets(map);
2918 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2921 static void __dm_internal_resume(struct mapped_device *md)
2923 BUG_ON(!md->internal_suspend_count);
2925 if (--md->internal_suspend_count)
2926 return; /* resume from nested internal suspend */
2928 if (dm_suspended_md(md))
2929 goto done; /* resume from nested suspend */
2932 * NOTE: existing callers don't need to call dm_table_resume_targets
2933 * (which may fail -- so best to avoid it for now by passing NULL map)
2935 (void) __dm_resume(md, NULL);
2938 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2939 smp_mb__after_atomic();
2940 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2943 void dm_internal_suspend_noflush(struct mapped_device *md)
2945 mutex_lock(&md->suspend_lock);
2946 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2947 mutex_unlock(&md->suspend_lock);
2949 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2951 void dm_internal_resume(struct mapped_device *md)
2953 mutex_lock(&md->suspend_lock);
2954 __dm_internal_resume(md);
2955 mutex_unlock(&md->suspend_lock);
2957 EXPORT_SYMBOL_GPL(dm_internal_resume);
2960 * Fast variants of internal suspend/resume hold md->suspend_lock,
2961 * which prevents interaction with userspace-driven suspend.
2964 void dm_internal_suspend_fast(struct mapped_device *md)
2966 mutex_lock(&md->suspend_lock);
2967 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2970 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2971 synchronize_srcu(&md->io_barrier);
2972 flush_workqueue(md->wq);
2973 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2975 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2977 void dm_internal_resume_fast(struct mapped_device *md)
2979 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2985 mutex_unlock(&md->suspend_lock);
2987 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2990 *---------------------------------------------------------------
2991 * Event notification.
2992 *---------------------------------------------------------------
2994 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2995 unsigned int cookie, bool need_resize_uevent)
2998 unsigned int noio_flag;
2999 char udev_cookie[DM_COOKIE_LENGTH];
3000 char *envp[3] = { NULL, NULL, NULL };
3001 char **envpp = envp;
3003 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
3004 DM_COOKIE_ENV_VAR_NAME, cookie);
3005 *envpp++ = udev_cookie;
3007 if (need_resize_uevent) {
3008 *envpp++ = "RESIZE=1";
3011 noio_flag = memalloc_noio_save();
3013 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
3015 memalloc_noio_restore(noio_flag);
3020 uint32_t dm_next_uevent_seq(struct mapped_device *md)
3022 return atomic_add_return(1, &md->uevent_seq);
3025 uint32_t dm_get_event_nr(struct mapped_device *md)
3027 return atomic_read(&md->event_nr);
3030 int dm_wait_event(struct mapped_device *md, int event_nr)
3032 return wait_event_interruptible(md->eventq,
3033 (event_nr != atomic_read(&md->event_nr)));
3036 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3038 unsigned long flags;
3040 spin_lock_irqsave(&md->uevent_lock, flags);
3041 list_add(elist, &md->uevent_list);
3042 spin_unlock_irqrestore(&md->uevent_lock, flags);
3046 * The gendisk is only valid as long as you have a reference
3049 struct gendisk *dm_disk(struct mapped_device *md)
3053 EXPORT_SYMBOL_GPL(dm_disk);
3055 struct kobject *dm_kobject(struct mapped_device *md)
3057 return &md->kobj_holder.kobj;
3060 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3062 struct mapped_device *md;
3064 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3066 spin_lock(&_minor_lock);
3067 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3073 spin_unlock(&_minor_lock);
3078 int dm_suspended_md(struct mapped_device *md)
3080 return test_bit(DMF_SUSPENDED, &md->flags);
3083 static int dm_post_suspending_md(struct mapped_device *md)
3085 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3088 int dm_suspended_internally_md(struct mapped_device *md)
3090 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3093 int dm_test_deferred_remove_flag(struct mapped_device *md)
3095 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3098 int dm_suspended(struct dm_target *ti)
3100 return dm_suspended_md(ti->table->md);
3102 EXPORT_SYMBOL_GPL(dm_suspended);
3104 int dm_post_suspending(struct dm_target *ti)
3106 return dm_post_suspending_md(ti->table->md);
3108 EXPORT_SYMBOL_GPL(dm_post_suspending);
3110 int dm_noflush_suspending(struct dm_target *ti)
3112 return __noflush_suspending(ti->table->md);
3114 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3116 void dm_free_md_mempools(struct dm_md_mempools *pools)
3121 bioset_exit(&pools->bs);
3122 bioset_exit(&pools->io_bs);
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 const struct pr_ops dm_pr_ops = {
3368 .pr_register = dm_pr_register,
3369 .pr_reserve = dm_pr_reserve,
3370 .pr_release = dm_pr_release,
3371 .pr_preempt = dm_pr_preempt,
3372 .pr_clear = dm_pr_clear,
3375 static const struct block_device_operations dm_blk_dops = {
3376 .submit_bio = dm_submit_bio,
3377 .poll_bio = dm_poll_bio,
3378 .open = dm_blk_open,
3379 .release = dm_blk_close,
3380 .ioctl = dm_blk_ioctl,
3381 .getgeo = dm_blk_getgeo,
3382 .report_zones = dm_blk_report_zones,
3383 .pr_ops = &dm_pr_ops,
3384 .owner = THIS_MODULE
3387 static const struct block_device_operations dm_rq_blk_dops = {
3388 .open = dm_blk_open,
3389 .release = dm_blk_close,
3390 .ioctl = dm_blk_ioctl,
3391 .getgeo = dm_blk_getgeo,
3392 .pr_ops = &dm_pr_ops,
3393 .owner = THIS_MODULE
3396 static const struct dax_operations dm_dax_ops = {
3397 .direct_access = dm_dax_direct_access,
3398 .zero_page_range = dm_dax_zero_page_range,
3399 .recovery_write = dm_dax_recovery_write,
3405 module_init(dm_init);
3406 module_exit(dm_exit);
3408 module_param(major, uint, 0);
3409 MODULE_PARM_DESC(major, "The major number of the device mapper");
3411 module_param(reserved_bio_based_ios, uint, 0644);
3412 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3414 module_param(dm_numa_node, int, 0644);
3415 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3417 module_param(swap_bios, int, 0644);
3418 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3420 MODULE_DESCRIPTION(DM_NAME " driver");
3421 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3422 MODULE_LICENSE("GPL");