2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/mutex.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/signal.h>
18 #include <linux/blkpg.h>
19 #include <linux/bio.h>
20 #include <linux/mempool.h>
21 #include <linux/dax.h>
22 #include <linux/slab.h>
23 #include <linux/idr.h>
24 #include <linux/uio.h>
25 #include <linux/hdreg.h>
26 #include <linux/delay.h>
27 #include <linux/wait.h>
29 #include <linux/refcount.h>
30 #include <linux/part_stat.h>
31 #include <linux/blk-crypto.h>
32 #include <linux/blk-crypto-profile.h>
34 #define DM_MSG_PREFIX "core"
37 * Cookies are numeric values sent with CHANGE and REMOVE
38 * uevents while resuming, removing or renaming the device.
40 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41 #define DM_COOKIE_LENGTH 24
43 static const char *_name = DM_NAME;
45 static unsigned int major = 0;
46 static unsigned int _major = 0;
48 static DEFINE_IDR(_minor_idr);
50 static DEFINE_SPINLOCK(_minor_lock);
52 static void do_deferred_remove(struct work_struct *w);
54 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
56 static struct workqueue_struct *deferred_remove_workqueue;
58 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
59 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
61 void dm_issue_global_event(void)
63 atomic_inc(&dm_global_event_nr);
64 wake_up(&dm_global_eventq);
68 * One of these is allocated (on-stack) per original bio.
75 unsigned sector_count;
78 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
79 #define DM_IO_BIO_OFFSET \
80 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
82 static inline struct dm_target_io *clone_to_tio(struct bio *clone)
84 return container_of(clone, struct dm_target_io, clone);
87 void *dm_per_bio_data(struct bio *bio, size_t data_size)
89 if (!clone_to_tio(bio)->inside_dm_io)
90 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
91 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
93 EXPORT_SYMBOL_GPL(dm_per_bio_data);
95 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
97 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
98 if (io->magic == DM_IO_MAGIC)
99 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
100 BUG_ON(io->magic != DM_TIO_MAGIC);
101 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
103 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
105 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
107 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
109 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
111 #define MINOR_ALLOCED ((void *)-1)
113 #define DM_NUMA_NODE NUMA_NO_NODE
114 static int dm_numa_node = DM_NUMA_NODE;
116 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
117 static int swap_bios = DEFAULT_SWAP_BIOS;
118 static int get_swap_bios(void)
120 int latch = READ_ONCE(swap_bios);
121 if (unlikely(latch <= 0))
122 latch = DEFAULT_SWAP_BIOS;
127 * For mempools pre-allocation at the table loading time.
129 struct dm_md_mempools {
131 struct bio_set io_bs;
134 struct table_device {
135 struct list_head list;
137 struct dm_dev dm_dev;
141 * Bio-based DM's mempools' reserved IOs set by the user.
143 #define RESERVED_BIO_BASED_IOS 16
144 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
146 static int __dm_get_module_param_int(int *module_param, int min, int max)
148 int param = READ_ONCE(*module_param);
149 int modified_param = 0;
150 bool modified = true;
153 modified_param = min;
154 else if (param > max)
155 modified_param = max;
160 (void)cmpxchg(module_param, param, modified_param);
161 param = modified_param;
167 unsigned __dm_get_module_param(unsigned *module_param,
168 unsigned def, unsigned max)
170 unsigned param = READ_ONCE(*module_param);
171 unsigned modified_param = 0;
174 modified_param = def;
175 else if (param > max)
176 modified_param = max;
178 if (modified_param) {
179 (void)cmpxchg(module_param, param, modified_param);
180 param = modified_param;
186 unsigned dm_get_reserved_bio_based_ios(void)
188 return __dm_get_module_param(&reserved_bio_based_ios,
189 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
191 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
193 static unsigned dm_get_numa_node(void)
195 return __dm_get_module_param_int(&dm_numa_node,
196 DM_NUMA_NODE, num_online_nodes() - 1);
199 static int __init local_init(void)
203 r = dm_uevent_init();
207 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
208 if (!deferred_remove_workqueue) {
210 goto out_uevent_exit;
214 r = register_blkdev(_major, _name);
216 goto out_free_workqueue;
224 destroy_workqueue(deferred_remove_workqueue);
231 static void local_exit(void)
233 flush_scheduled_work();
234 destroy_workqueue(deferred_remove_workqueue);
236 unregister_blkdev(_major, _name);
241 DMINFO("cleaned up");
244 static int (*_inits[])(void) __initdata = {
255 static void (*_exits[])(void) = {
266 static int __init dm_init(void)
268 const int count = ARRAY_SIZE(_inits);
271 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
272 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
273 " Duplicate IMA measurements will not be recorded in the IMA log.");
276 for (i = 0; i < count; i++) {
290 static void __exit dm_exit(void)
292 int i = ARRAY_SIZE(_exits);
298 * Should be empty by this point.
300 idr_destroy(&_minor_idr);
304 * Block device functions
306 int dm_deleting_md(struct mapped_device *md)
308 return test_bit(DMF_DELETING, &md->flags);
311 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
313 struct mapped_device *md;
315 spin_lock(&_minor_lock);
317 md = bdev->bd_disk->private_data;
321 if (test_bit(DMF_FREEING, &md->flags) ||
322 dm_deleting_md(md)) {
328 atomic_inc(&md->open_count);
330 spin_unlock(&_minor_lock);
332 return md ? 0 : -ENXIO;
335 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
337 struct mapped_device *md;
339 spin_lock(&_minor_lock);
341 md = disk->private_data;
345 if (atomic_dec_and_test(&md->open_count) &&
346 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
347 queue_work(deferred_remove_workqueue, &deferred_remove_work);
351 spin_unlock(&_minor_lock);
354 int dm_open_count(struct mapped_device *md)
356 return atomic_read(&md->open_count);
360 * Guarantees nothing is using the device before it's deleted.
362 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
366 spin_lock(&_minor_lock);
368 if (dm_open_count(md)) {
371 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
372 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
375 set_bit(DMF_DELETING, &md->flags);
377 spin_unlock(&_minor_lock);
382 int dm_cancel_deferred_remove(struct mapped_device *md)
386 spin_lock(&_minor_lock);
388 if (test_bit(DMF_DELETING, &md->flags))
391 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
393 spin_unlock(&_minor_lock);
398 static void do_deferred_remove(struct work_struct *w)
400 dm_deferred_remove();
403 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
405 struct mapped_device *md = bdev->bd_disk->private_data;
407 return dm_get_geometry(md, geo);
410 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
411 struct block_device **bdev)
413 struct dm_target *tgt;
414 struct dm_table *map;
419 map = dm_get_live_table(md, srcu_idx);
420 if (!map || !dm_table_get_size(map))
423 /* We only support devices that have a single target */
424 if (dm_table_get_num_targets(map) != 1)
427 tgt = dm_table_get_target(map, 0);
428 if (!tgt->type->prepare_ioctl)
431 if (dm_suspended_md(md))
434 r = tgt->type->prepare_ioctl(tgt, bdev);
435 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
436 dm_put_live_table(md, *srcu_idx);
444 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
446 dm_put_live_table(md, srcu_idx);
449 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
450 unsigned int cmd, unsigned long arg)
452 struct mapped_device *md = bdev->bd_disk->private_data;
455 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
461 * Target determined this ioctl is being issued against a
462 * subset of the parent bdev; require extra privileges.
464 if (!capable(CAP_SYS_RAWIO)) {
466 "%s: sending ioctl %x to DM device without required privilege.",
473 if (!bdev->bd_disk->fops->ioctl)
476 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
478 dm_unprepare_ioctl(md, srcu_idx);
482 u64 dm_start_time_ns_from_clone(struct bio *bio)
484 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
486 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
488 static void start_io_acct(struct dm_io *io)
490 struct mapped_device *md = io->md;
491 struct bio *bio = io->orig_bio;
493 bio_start_io_acct_time(bio, io->start_time);
494 if (unlikely(dm_stats_used(&md->stats)))
495 dm_stats_account_io(&md->stats, bio_data_dir(bio),
496 bio->bi_iter.bi_sector, bio_sectors(bio),
497 false, 0, &io->stats_aux);
500 static void end_io_acct(struct mapped_device *md, struct bio *bio,
501 unsigned long start_time, struct dm_stats_aux *stats_aux)
503 unsigned long duration = jiffies - start_time;
505 bio_end_io_acct(bio, start_time);
507 if (unlikely(dm_stats_used(&md->stats)))
508 dm_stats_account_io(&md->stats, bio_data_dir(bio),
509 bio->bi_iter.bi_sector, bio_sectors(bio),
510 true, duration, stats_aux);
512 /* nudge anyone waiting on suspend queue */
513 if (unlikely(wq_has_sleeper(&md->wait)))
517 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
520 struct dm_target_io *tio;
523 clone = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, &md->io_bs);
525 tio = clone_to_tio(clone);
526 tio->inside_dm_io = true;
529 io = container_of(tio, struct dm_io, tio);
530 io->magic = DM_IO_MAGIC;
532 atomic_set(&io->io_count, 1);
535 spin_lock_init(&io->endio_lock);
537 io->start_time = jiffies;
542 static void free_io(struct mapped_device *md, struct dm_io *io)
544 bio_put(&io->tio.clone);
547 static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
548 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
550 struct dm_target_io *tio;
552 if (!ci->io->tio.io) {
553 /* the dm_target_io embedded in ci->io is available */
556 struct bio *clone = bio_alloc_clone(ci->bio->bi_bdev, ci->bio,
557 gfp_mask, &ci->io->md->bs);
561 tio = clone_to_tio(clone);
562 tio->inside_dm_io = false;
565 tio->magic = DM_TIO_MAGIC;
568 tio->target_bio_nr = target_bio_nr;
574 static void free_tio(struct bio *clone)
576 if (clone_to_tio(clone)->inside_dm_io)
582 * Add the bio to the list of deferred io.
584 static void queue_io(struct mapped_device *md, struct bio *bio)
588 spin_lock_irqsave(&md->deferred_lock, flags);
589 bio_list_add(&md->deferred, bio);
590 spin_unlock_irqrestore(&md->deferred_lock, flags);
591 queue_work(md->wq, &md->work);
595 * Everyone (including functions in this file), should use this
596 * function to access the md->map field, and make sure they call
597 * dm_put_live_table() when finished.
599 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
601 *srcu_idx = srcu_read_lock(&md->io_barrier);
603 return srcu_dereference(md->map, &md->io_barrier);
606 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
608 srcu_read_unlock(&md->io_barrier, srcu_idx);
611 void dm_sync_table(struct mapped_device *md)
613 synchronize_srcu(&md->io_barrier);
614 synchronize_rcu_expedited();
618 * A fast alternative to dm_get_live_table/dm_put_live_table.
619 * The caller must not block between these two functions.
621 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
624 return rcu_dereference(md->map);
627 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
632 static char *_dm_claim_ptr = "I belong to device-mapper";
635 * Open a table device so we can use it as a map destination.
637 static int open_table_device(struct table_device *td, dev_t dev,
638 struct mapped_device *md)
640 struct block_device *bdev;
644 BUG_ON(td->dm_dev.bdev);
646 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
648 return PTR_ERR(bdev);
650 r = bd_link_disk_holder(bdev, dm_disk(md));
652 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
656 td->dm_dev.bdev = bdev;
657 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off);
662 * Close a table device that we've been using.
664 static void close_table_device(struct table_device *td, struct mapped_device *md)
666 if (!td->dm_dev.bdev)
669 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
670 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
671 put_dax(td->dm_dev.dax_dev);
672 td->dm_dev.bdev = NULL;
673 td->dm_dev.dax_dev = NULL;
676 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
679 struct table_device *td;
681 list_for_each_entry(td, l, list)
682 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
688 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
689 struct dm_dev **result)
692 struct table_device *td;
694 mutex_lock(&md->table_devices_lock);
695 td = find_table_device(&md->table_devices, dev, mode);
697 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
699 mutex_unlock(&md->table_devices_lock);
703 td->dm_dev.mode = mode;
704 td->dm_dev.bdev = NULL;
706 if ((r = open_table_device(td, dev, md))) {
707 mutex_unlock(&md->table_devices_lock);
712 format_dev_t(td->dm_dev.name, dev);
714 refcount_set(&td->count, 1);
715 list_add(&td->list, &md->table_devices);
717 refcount_inc(&td->count);
719 mutex_unlock(&md->table_devices_lock);
721 *result = &td->dm_dev;
725 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
727 struct table_device *td = container_of(d, struct table_device, dm_dev);
729 mutex_lock(&md->table_devices_lock);
730 if (refcount_dec_and_test(&td->count)) {
731 close_table_device(td, md);
735 mutex_unlock(&md->table_devices_lock);
738 static void free_table_devices(struct list_head *devices)
740 struct list_head *tmp, *next;
742 list_for_each_safe(tmp, next, devices) {
743 struct table_device *td = list_entry(tmp, struct table_device, list);
745 DMWARN("dm_destroy: %s still exists with %d references",
746 td->dm_dev.name, refcount_read(&td->count));
752 * Get the geometry associated with a dm device
754 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
762 * Set the geometry of a device.
764 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
766 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
768 if (geo->start > sz) {
769 DMWARN("Start sector is beyond the geometry limits.");
778 static int __noflush_suspending(struct mapped_device *md)
780 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
784 * Decrements the number of outstanding ios that a bio has been
785 * cloned into, completing the original io if necc.
787 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
790 blk_status_t io_error;
792 struct mapped_device *md = io->md;
793 unsigned long start_time = 0;
794 struct dm_stats_aux stats_aux;
796 /* Push-back supersedes any I/O errors */
797 if (unlikely(error)) {
798 spin_lock_irqsave(&io->endio_lock, flags);
799 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
801 spin_unlock_irqrestore(&io->endio_lock, flags);
804 if (atomic_dec_and_test(&io->io_count)) {
806 if (io->status == BLK_STS_DM_REQUEUE) {
808 * Target requested pushing back the I/O.
810 spin_lock_irqsave(&md->deferred_lock, flags);
811 if (__noflush_suspending(md) &&
812 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
813 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
814 bio_list_add_head(&md->deferred, bio);
817 * noflush suspend was interrupted or this is
818 * a write to a zoned target.
820 io->status = BLK_STS_IOERR;
822 spin_unlock_irqrestore(&md->deferred_lock, flags);
825 io_error = io->status;
826 start_time = io->start_time;
827 stats_aux = io->stats_aux;
829 end_io_acct(md, bio, start_time, &stats_aux);
831 if (io_error == BLK_STS_DM_REQUEUE)
834 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
836 * Preflush done for flush with data, reissue
837 * without REQ_PREFLUSH.
839 bio->bi_opf &= ~REQ_PREFLUSH;
842 /* done with normal IO or empty flush */
844 bio->bi_status = io_error;
850 void disable_discard(struct mapped_device *md)
852 struct queue_limits *limits = dm_get_queue_limits(md);
854 /* device doesn't really support DISCARD, disable it */
855 limits->max_discard_sectors = 0;
856 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
859 void disable_write_same(struct mapped_device *md)
861 struct queue_limits *limits = dm_get_queue_limits(md);
863 /* device doesn't really support WRITE SAME, disable it */
864 limits->max_write_same_sectors = 0;
867 void disable_write_zeroes(struct mapped_device *md)
869 struct queue_limits *limits = dm_get_queue_limits(md);
871 /* device doesn't really support WRITE ZEROES, disable it */
872 limits->max_write_zeroes_sectors = 0;
875 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
877 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
880 static void clone_endio(struct bio *bio)
882 blk_status_t error = bio->bi_status;
883 struct dm_target_io *tio = clone_to_tio(bio);
884 struct dm_io *io = tio->io;
885 struct mapped_device *md = tio->io->md;
886 dm_endio_fn endio = tio->ti->type->end_io;
887 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
889 if (unlikely(error == BLK_STS_TARGET)) {
890 if (bio_op(bio) == REQ_OP_DISCARD &&
891 !q->limits.max_discard_sectors)
893 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
894 !q->limits.max_write_same_sectors)
895 disable_write_same(md);
896 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
897 !q->limits.max_write_zeroes_sectors)
898 disable_write_zeroes(md);
901 if (blk_queue_is_zoned(q))
902 dm_zone_endio(io, bio);
905 int r = endio(tio->ti, bio, &error);
907 case DM_ENDIO_REQUEUE:
909 * Requeuing writes to a sequential zone of a zoned
910 * target will break the sequential write pattern:
913 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
914 error = BLK_STS_IOERR;
916 error = BLK_STS_DM_REQUEUE;
920 case DM_ENDIO_INCOMPLETE:
921 /* The target will handle the io */
924 DMWARN("unimplemented target endio return value: %d", r);
929 if (unlikely(swap_bios_limit(tio->ti, bio))) {
930 struct mapped_device *md = io->md;
931 up(&md->swap_bios_semaphore);
935 dm_io_dec_pending(io, error);
939 * Return maximum size of I/O possible at the supplied sector up to the current
942 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
943 sector_t target_offset)
945 return ti->len - target_offset;
948 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
950 sector_t target_offset = dm_target_offset(ti, sector);
951 sector_t len = max_io_len_target_boundary(ti, target_offset);
955 * Does the target need to split IO even further?
956 * - varied (per target) IO splitting is a tenet of DM; this
957 * explains why stacked chunk_sectors based splitting via
958 * blk_max_size_offset() isn't possible here. So pass in
959 * ti->max_io_len to override stacked chunk_sectors.
961 if (ti->max_io_len) {
962 max_len = blk_max_size_offset(ti->table->md->queue,
963 target_offset, ti->max_io_len);
971 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
973 if (len > UINT_MAX) {
974 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
975 (unsigned long long)len, UINT_MAX);
976 ti->error = "Maximum size of target IO is too large";
980 ti->max_io_len = (uint32_t) len;
984 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
986 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
987 sector_t sector, int *srcu_idx)
988 __acquires(md->io_barrier)
990 struct dm_table *map;
991 struct dm_target *ti;
993 map = dm_get_live_table(md, srcu_idx);
997 ti = dm_table_find_target(map, sector);
1004 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1005 long nr_pages, void **kaddr, pfn_t *pfn)
1007 struct mapped_device *md = dax_get_private(dax_dev);
1008 sector_t sector = pgoff * PAGE_SECTORS;
1009 struct dm_target *ti;
1010 long len, ret = -EIO;
1013 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1017 if (!ti->type->direct_access)
1019 len = max_io_len(ti, sector) / PAGE_SECTORS;
1022 nr_pages = min(len, nr_pages);
1023 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1026 dm_put_live_table(md, srcu_idx);
1031 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1034 struct mapped_device *md = dax_get_private(dax_dev);
1035 sector_t sector = pgoff * PAGE_SECTORS;
1036 struct dm_target *ti;
1040 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1044 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1046 * ->zero_page_range() is mandatory dax operation. If we are
1047 * here, something is wrong.
1051 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1053 dm_put_live_table(md, srcu_idx);
1059 * A target may call dm_accept_partial_bio only from the map routine. It is
1060 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1061 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1063 * dm_accept_partial_bio informs the dm that the target only wants to process
1064 * additional n_sectors sectors of the bio and the rest of the data should be
1065 * sent in a next bio.
1067 * A diagram that explains the arithmetics:
1068 * +--------------------+---------------+-------+
1070 * +--------------------+---------------+-------+
1072 * <-------------- *tio->len_ptr --------------->
1073 * <------- bi_size ------->
1076 * Region 1 was already iterated over with bio_advance or similar function.
1077 * (it may be empty if the target doesn't use bio_advance)
1078 * Region 2 is the remaining bio size that the target wants to process.
1079 * (it may be empty if region 1 is non-empty, although there is no reason
1081 * The target requires that region 3 is to be sent in the next bio.
1083 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1084 * the partially processed part (the sum of regions 1+2) must be the same for all
1085 * copies of the bio.
1087 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1089 struct dm_target_io *tio = clone_to_tio(bio);
1090 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1092 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1093 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1094 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1095 BUG_ON(bi_size > *tio->len_ptr);
1096 BUG_ON(n_sectors > bi_size);
1098 *tio->len_ptr -= bi_size - n_sectors;
1099 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1101 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1103 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1105 mutex_lock(&md->swap_bios_lock);
1106 while (latch < md->swap_bios) {
1108 down(&md->swap_bios_semaphore);
1111 while (latch > md->swap_bios) {
1113 up(&md->swap_bios_semaphore);
1116 mutex_unlock(&md->swap_bios_lock);
1119 static void __map_bio(struct bio *clone)
1121 struct dm_target_io *tio = clone_to_tio(clone);
1124 struct dm_io *io = tio->io;
1125 struct dm_target *ti = tio->ti;
1127 clone->bi_end_io = clone_endio;
1130 * Map the clone. If r == 0 we don't need to do
1131 * anything, the target has assumed ownership of
1134 dm_io_inc_pending(io);
1135 sector = clone->bi_iter.bi_sector;
1137 if (unlikely(swap_bios_limit(ti, clone))) {
1138 struct mapped_device *md = io->md;
1139 int latch = get_swap_bios();
1140 if (unlikely(latch != md->swap_bios))
1141 __set_swap_bios_limit(md, latch);
1142 down(&md->swap_bios_semaphore);
1146 * Check if the IO needs a special mapping due to zone append emulation
1147 * on zoned target. In this case, dm_zone_map_bio() calls the target
1150 if (dm_emulate_zone_append(io->md))
1151 r = dm_zone_map_bio(tio);
1153 r = ti->type->map(ti, clone);
1156 case DM_MAPIO_SUBMITTED:
1158 case DM_MAPIO_REMAPPED:
1159 /* the bio has been remapped so dispatch it */
1160 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1161 submit_bio_noacct(clone);
1164 if (unlikely(swap_bios_limit(ti, clone))) {
1165 struct mapped_device *md = io->md;
1166 up(&md->swap_bios_semaphore);
1169 dm_io_dec_pending(io, BLK_STS_IOERR);
1171 case DM_MAPIO_REQUEUE:
1172 if (unlikely(swap_bios_limit(ti, clone))) {
1173 struct mapped_device *md = io->md;
1174 up(&md->swap_bios_semaphore);
1177 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1180 DMWARN("unimplemented target map return value: %d", r);
1185 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1187 bio->bi_iter.bi_sector = sector;
1188 bio->bi_iter.bi_size = to_bytes(len);
1192 * Creates a bio that consists of range of complete bvecs.
1194 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1195 sector_t sector, unsigned *len)
1197 struct bio *bio = ci->bio, *clone;
1199 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1200 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1201 clone->bi_iter.bi_size = to_bytes(*len);
1203 if (bio_integrity(bio))
1204 bio_integrity_trim(clone);
1210 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1211 struct dm_target *ti, unsigned num_bios,
1217 for (try = 0; try < 2; try++) {
1221 mutex_lock(&ci->io->md->table_devices_lock);
1222 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1223 bio = alloc_tio(ci, ti, bio_nr, len,
1224 try ? GFP_NOIO : GFP_NOWAIT);
1228 bio_list_add(blist, bio);
1231 mutex_unlock(&ci->io->md->table_devices_lock);
1232 if (bio_nr == num_bios)
1235 while ((bio = bio_list_pop(blist)))
1240 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1241 unsigned num_bios, unsigned *len)
1243 struct bio_list blist = BIO_EMPTY_LIST;
1250 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1252 bio_setup_sector(clone, ci->sector, *len);
1256 alloc_multiple_bios(&blist, ci, ti, num_bios, len);
1257 while ((clone = bio_list_pop(&blist))) {
1259 bio_setup_sector(clone, ci->sector, *len);
1266 static int __send_empty_flush(struct clone_info *ci)
1268 unsigned target_nr = 0;
1269 struct dm_target *ti;
1270 struct bio flush_bio;
1273 * Use an on-stack bio for this, it's safe since we don't
1274 * need to reference it after submit. It's just used as
1275 * the basis for the clone(s).
1277 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1278 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1280 ci->bio = &flush_bio;
1281 ci->sector_count = 0;
1283 BUG_ON(bio_has_data(ci->bio));
1284 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1285 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1287 bio_uninit(ci->bio);
1291 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1297 * Even though the device advertised support for this type of
1298 * request, that does not mean every target supports it, and
1299 * reconfiguration might also have changed that since the
1300 * check was performed.
1305 len = min_t(sector_t, ci->sector_count,
1306 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1308 __send_duplicate_bios(ci, ti, num_bios, &len);
1311 ci->sector_count -= len;
1316 static bool is_abnormal_io(struct bio *bio)
1320 switch (bio_op(bio)) {
1321 case REQ_OP_DISCARD:
1322 case REQ_OP_SECURE_ERASE:
1323 case REQ_OP_WRITE_SAME:
1324 case REQ_OP_WRITE_ZEROES:
1332 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1335 struct bio *bio = ci->bio;
1336 unsigned num_bios = 0;
1338 switch (bio_op(bio)) {
1339 case REQ_OP_DISCARD:
1340 num_bios = ti->num_discard_bios;
1342 case REQ_OP_SECURE_ERASE:
1343 num_bios = ti->num_secure_erase_bios;
1345 case REQ_OP_WRITE_SAME:
1346 num_bios = ti->num_write_same_bios;
1348 case REQ_OP_WRITE_ZEROES:
1349 num_bios = ti->num_write_zeroes_bios;
1355 *result = __send_changing_extent_only(ci, ti, num_bios);
1360 * Select the correct strategy for processing a non-flush bio.
1362 static int __split_and_process_non_flush(struct clone_info *ci)
1364 struct dm_target *ti;
1368 ti = dm_table_find_target(ci->map, ci->sector);
1372 if (__process_abnormal_io(ci, ti, &r))
1375 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1377 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1382 ci->sector_count -= len;
1387 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1388 struct dm_table *map, struct bio *bio)
1391 ci->io = alloc_io(md, bio);
1392 ci->sector = bio->bi_iter.bi_sector;
1396 * Entry point to split a bio into clones and submit them to the targets.
1398 static void __split_and_process_bio(struct mapped_device *md,
1399 struct dm_table *map, struct bio *bio)
1401 struct clone_info ci;
1404 init_clone_info(&ci, md, map, bio);
1406 if (bio->bi_opf & REQ_PREFLUSH) {
1407 error = __send_empty_flush(&ci);
1408 /* dm_io_dec_pending submits any data associated with flush */
1409 } else if (op_is_zone_mgmt(bio_op(bio))) {
1411 ci.sector_count = 0;
1412 error = __split_and_process_non_flush(&ci);
1415 ci.sector_count = bio_sectors(bio);
1416 error = __split_and_process_non_flush(&ci);
1417 if (ci.sector_count && !error) {
1419 * Remainder must be passed to submit_bio_noacct()
1420 * so that it gets handled *after* bios already submitted
1421 * have been completely processed.
1422 * We take a clone of the original to store in
1423 * ci.io->orig_bio to be used by end_io_acct() and
1424 * for dec_pending to use for completion handling.
1426 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1427 GFP_NOIO, &md->queue->bio_split);
1428 ci.io->orig_bio = b;
1431 trace_block_split(b, bio->bi_iter.bi_sector);
1432 submit_bio_noacct(bio);
1435 start_io_acct(ci.io);
1437 /* drop the extra reference count */
1438 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1441 static void dm_submit_bio(struct bio *bio)
1443 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1445 struct dm_table *map;
1447 map = dm_get_live_table(md, &srcu_idx);
1448 if (unlikely(!map)) {
1449 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1450 dm_device_name(md));
1455 /* If suspended, queue this IO for later */
1456 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1457 if (bio->bi_opf & REQ_NOWAIT)
1458 bio_wouldblock_error(bio);
1459 else if (bio->bi_opf & REQ_RAHEAD)
1467 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1468 * otherwise associated queue_limits won't be imposed.
1470 if (is_abnormal_io(bio))
1471 blk_queue_split(&bio);
1473 __split_and_process_bio(md, map, bio);
1475 dm_put_live_table(md, srcu_idx);
1478 /*-----------------------------------------------------------------
1479 * An IDR is used to keep track of allocated minor numbers.
1480 *---------------------------------------------------------------*/
1481 static void free_minor(int minor)
1483 spin_lock(&_minor_lock);
1484 idr_remove(&_minor_idr, minor);
1485 spin_unlock(&_minor_lock);
1489 * See if the device with a specific minor # is free.
1491 static int specific_minor(int minor)
1495 if (minor >= (1 << MINORBITS))
1498 idr_preload(GFP_KERNEL);
1499 spin_lock(&_minor_lock);
1501 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1503 spin_unlock(&_minor_lock);
1506 return r == -ENOSPC ? -EBUSY : r;
1510 static int next_free_minor(int *minor)
1514 idr_preload(GFP_KERNEL);
1515 spin_lock(&_minor_lock);
1517 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1519 spin_unlock(&_minor_lock);
1527 static const struct block_device_operations dm_blk_dops;
1528 static const struct block_device_operations dm_rq_blk_dops;
1529 static const struct dax_operations dm_dax_ops;
1531 static void dm_wq_work(struct work_struct *work);
1533 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1534 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1536 dm_destroy_crypto_profile(q->crypto_profile);
1539 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1541 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1544 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1546 static void cleanup_mapped_device(struct mapped_device *md)
1549 destroy_workqueue(md->wq);
1550 bioset_exit(&md->bs);
1551 bioset_exit(&md->io_bs);
1554 dax_remove_host(md->disk);
1555 kill_dax(md->dax_dev);
1556 put_dax(md->dax_dev);
1561 spin_lock(&_minor_lock);
1562 md->disk->private_data = NULL;
1563 spin_unlock(&_minor_lock);
1564 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1566 del_gendisk(md->disk);
1568 dm_queue_destroy_crypto_profile(md->queue);
1569 blk_cleanup_disk(md->disk);
1572 cleanup_srcu_struct(&md->io_barrier);
1574 mutex_destroy(&md->suspend_lock);
1575 mutex_destroy(&md->type_lock);
1576 mutex_destroy(&md->table_devices_lock);
1577 mutex_destroy(&md->swap_bios_lock);
1579 dm_mq_cleanup_mapped_device(md);
1580 dm_cleanup_zoned_dev(md);
1584 * Allocate and initialise a blank device with a given minor.
1586 static struct mapped_device *alloc_dev(int minor)
1588 int r, numa_node_id = dm_get_numa_node();
1589 struct mapped_device *md;
1592 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1594 DMWARN("unable to allocate device, out of memory.");
1598 if (!try_module_get(THIS_MODULE))
1599 goto bad_module_get;
1601 /* get a minor number for the dev */
1602 if (minor == DM_ANY_MINOR)
1603 r = next_free_minor(&minor);
1605 r = specific_minor(minor);
1609 r = init_srcu_struct(&md->io_barrier);
1611 goto bad_io_barrier;
1613 md->numa_node_id = numa_node_id;
1614 md->init_tio_pdu = false;
1615 md->type = DM_TYPE_NONE;
1616 mutex_init(&md->suspend_lock);
1617 mutex_init(&md->type_lock);
1618 mutex_init(&md->table_devices_lock);
1619 spin_lock_init(&md->deferred_lock);
1620 atomic_set(&md->holders, 1);
1621 atomic_set(&md->open_count, 0);
1622 atomic_set(&md->event_nr, 0);
1623 atomic_set(&md->uevent_seq, 0);
1624 INIT_LIST_HEAD(&md->uevent_list);
1625 INIT_LIST_HEAD(&md->table_devices);
1626 spin_lock_init(&md->uevent_lock);
1629 * default to bio-based until DM table is loaded and md->type
1630 * established. If request-based table is loaded: blk-mq will
1631 * override accordingly.
1633 md->disk = blk_alloc_disk(md->numa_node_id);
1636 md->queue = md->disk->queue;
1638 init_waitqueue_head(&md->wait);
1639 INIT_WORK(&md->work, dm_wq_work);
1640 init_waitqueue_head(&md->eventq);
1641 init_completion(&md->kobj_holder.completion);
1643 md->swap_bios = get_swap_bios();
1644 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1645 mutex_init(&md->swap_bios_lock);
1647 md->disk->major = _major;
1648 md->disk->first_minor = minor;
1649 md->disk->minors = 1;
1650 md->disk->flags |= GENHD_FL_NO_PART;
1651 md->disk->fops = &dm_blk_dops;
1652 md->disk->queue = md->queue;
1653 md->disk->private_data = md;
1654 sprintf(md->disk->disk_name, "dm-%d", minor);
1656 if (IS_ENABLED(CONFIG_FS_DAX)) {
1657 md->dax_dev = alloc_dax(md, &dm_dax_ops);
1658 if (IS_ERR(md->dax_dev)) {
1662 set_dax_nocache(md->dax_dev);
1663 set_dax_nomc(md->dax_dev);
1664 if (dax_add_host(md->dax_dev, md->disk))
1668 format_dev_t(md->name, MKDEV(_major, minor));
1670 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
1674 dm_stats_init(&md->stats);
1676 /* Populate the mapping, nobody knows we exist yet */
1677 spin_lock(&_minor_lock);
1678 old_md = idr_replace(&_minor_idr, md, minor);
1679 spin_unlock(&_minor_lock);
1681 BUG_ON(old_md != MINOR_ALLOCED);
1686 cleanup_mapped_device(md);
1690 module_put(THIS_MODULE);
1696 static void unlock_fs(struct mapped_device *md);
1698 static void free_dev(struct mapped_device *md)
1700 int minor = MINOR(disk_devt(md->disk));
1704 cleanup_mapped_device(md);
1706 free_table_devices(&md->table_devices);
1707 dm_stats_cleanup(&md->stats);
1710 module_put(THIS_MODULE);
1714 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1716 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1719 if (dm_table_bio_based(t)) {
1721 * The md may already have mempools that need changing.
1722 * If so, reload bioset because front_pad may have changed
1723 * because a different table was loaded.
1725 bioset_exit(&md->bs);
1726 bioset_exit(&md->io_bs);
1728 } else if (bioset_initialized(&md->bs)) {
1730 * There's no need to reload with request-based dm
1731 * because the size of front_pad doesn't change.
1732 * Note for future: If you are to reload bioset,
1733 * prep-ed requests in the queue may refer
1734 * to bio from the old bioset, so you must walk
1735 * through the queue to unprep.
1741 bioset_initialized(&md->bs) ||
1742 bioset_initialized(&md->io_bs));
1744 ret = bioset_init_from_src(&md->bs, &p->bs);
1747 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1749 bioset_exit(&md->bs);
1751 /* mempool bind completed, no longer need any mempools in the table */
1752 dm_table_free_md_mempools(t);
1757 * Bind a table to the device.
1759 static void event_callback(void *context)
1761 unsigned long flags;
1763 struct mapped_device *md = (struct mapped_device *) context;
1765 spin_lock_irqsave(&md->uevent_lock, flags);
1766 list_splice_init(&md->uevent_list, &uevents);
1767 spin_unlock_irqrestore(&md->uevent_lock, flags);
1769 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1771 atomic_inc(&md->event_nr);
1772 wake_up(&md->eventq);
1773 dm_issue_global_event();
1777 * Returns old map, which caller must destroy.
1779 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1780 struct queue_limits *limits)
1782 struct dm_table *old_map;
1783 struct request_queue *q = md->queue;
1784 bool request_based = dm_table_request_based(t);
1788 lockdep_assert_held(&md->suspend_lock);
1790 size = dm_table_get_size(t);
1793 * Wipe any geometry if the size of the table changed.
1795 if (size != dm_get_size(md))
1796 memset(&md->geometry, 0, sizeof(md->geometry));
1798 if (!get_capacity(md->disk))
1799 set_capacity(md->disk, size);
1801 set_capacity_and_notify(md->disk, size);
1803 dm_table_event_callback(t, event_callback, md);
1805 if (request_based) {
1807 * Leverage the fact that request-based DM targets are
1808 * immutable singletons - used to optimize dm_mq_queue_rq.
1810 md->immutable_target = dm_table_get_immutable_target(t);
1813 ret = __bind_mempools(md, t);
1815 old_map = ERR_PTR(ret);
1819 ret = dm_table_set_restrictions(t, q, limits);
1821 old_map = ERR_PTR(ret);
1825 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1826 rcu_assign_pointer(md->map, (void *)t);
1827 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1837 * Returns unbound table for the caller to free.
1839 static struct dm_table *__unbind(struct mapped_device *md)
1841 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1846 dm_table_event_callback(map, NULL, NULL);
1847 RCU_INIT_POINTER(md->map, NULL);
1854 * Constructor for a new device.
1856 int dm_create(int minor, struct mapped_device **result)
1858 struct mapped_device *md;
1860 md = alloc_dev(minor);
1864 dm_ima_reset_data(md);
1871 * Functions to manage md->type.
1872 * All are required to hold md->type_lock.
1874 void dm_lock_md_type(struct mapped_device *md)
1876 mutex_lock(&md->type_lock);
1879 void dm_unlock_md_type(struct mapped_device *md)
1881 mutex_unlock(&md->type_lock);
1884 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1886 BUG_ON(!mutex_is_locked(&md->type_lock));
1890 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1895 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1897 return md->immutable_target_type;
1901 * The queue_limits are only valid as long as you have a reference
1904 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1906 BUG_ON(!atomic_read(&md->holders));
1907 return &md->queue->limits;
1909 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1912 * Setup the DM device's queue based on md's type
1914 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1916 enum dm_queue_mode type = dm_table_get_type(t);
1917 struct queue_limits limits;
1921 case DM_TYPE_REQUEST_BASED:
1922 md->disk->fops = &dm_rq_blk_dops;
1923 r = dm_mq_init_request_queue(md, t);
1925 DMERR("Cannot initialize queue for request-based dm mapped device");
1929 case DM_TYPE_BIO_BASED:
1930 case DM_TYPE_DAX_BIO_BASED:
1937 r = dm_calculate_queue_limits(t, &limits);
1939 DMERR("Cannot calculate initial queue limits");
1942 r = dm_table_set_restrictions(t, md->queue, &limits);
1946 r = add_disk(md->disk);
1950 r = dm_sysfs_init(md);
1952 del_gendisk(md->disk);
1959 struct mapped_device *dm_get_md(dev_t dev)
1961 struct mapped_device *md;
1962 unsigned minor = MINOR(dev);
1964 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1967 spin_lock(&_minor_lock);
1969 md = idr_find(&_minor_idr, minor);
1970 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
1971 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
1977 spin_unlock(&_minor_lock);
1981 EXPORT_SYMBOL_GPL(dm_get_md);
1983 void *dm_get_mdptr(struct mapped_device *md)
1985 return md->interface_ptr;
1988 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1990 md->interface_ptr = ptr;
1993 void dm_get(struct mapped_device *md)
1995 atomic_inc(&md->holders);
1996 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1999 int dm_hold(struct mapped_device *md)
2001 spin_lock(&_minor_lock);
2002 if (test_bit(DMF_FREEING, &md->flags)) {
2003 spin_unlock(&_minor_lock);
2007 spin_unlock(&_minor_lock);
2010 EXPORT_SYMBOL_GPL(dm_hold);
2012 const char *dm_device_name(struct mapped_device *md)
2016 EXPORT_SYMBOL_GPL(dm_device_name);
2018 static void __dm_destroy(struct mapped_device *md, bool wait)
2020 struct dm_table *map;
2025 spin_lock(&_minor_lock);
2026 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2027 set_bit(DMF_FREEING, &md->flags);
2028 spin_unlock(&_minor_lock);
2030 blk_mark_disk_dead(md->disk);
2033 * Take suspend_lock so that presuspend and postsuspend methods
2034 * do not race with internal suspend.
2036 mutex_lock(&md->suspend_lock);
2037 map = dm_get_live_table(md, &srcu_idx);
2038 if (!dm_suspended_md(md)) {
2039 dm_table_presuspend_targets(map);
2040 set_bit(DMF_SUSPENDED, &md->flags);
2041 set_bit(DMF_POST_SUSPENDING, &md->flags);
2042 dm_table_postsuspend_targets(map);
2044 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2045 dm_put_live_table(md, srcu_idx);
2046 mutex_unlock(&md->suspend_lock);
2049 * Rare, but there may be I/O requests still going to complete,
2050 * for example. Wait for all references to disappear.
2051 * No one should increment the reference count of the mapped_device,
2052 * after the mapped_device state becomes DMF_FREEING.
2055 while (atomic_read(&md->holders))
2057 else if (atomic_read(&md->holders))
2058 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2059 dm_device_name(md), atomic_read(&md->holders));
2061 dm_table_destroy(__unbind(md));
2065 void dm_destroy(struct mapped_device *md)
2067 __dm_destroy(md, true);
2070 void dm_destroy_immediate(struct mapped_device *md)
2072 __dm_destroy(md, false);
2075 void dm_put(struct mapped_device *md)
2077 atomic_dec(&md->holders);
2079 EXPORT_SYMBOL_GPL(dm_put);
2081 static bool md_in_flight_bios(struct mapped_device *md)
2084 struct block_device *part = dm_disk(md)->part0;
2087 for_each_possible_cpu(cpu) {
2088 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2089 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2095 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2101 prepare_to_wait(&md->wait, &wait, task_state);
2103 if (!md_in_flight_bios(md))
2106 if (signal_pending_state(task_state, current)) {
2113 finish_wait(&md->wait, &wait);
2118 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2122 if (!queue_is_mq(md->queue))
2123 return dm_wait_for_bios_completion(md, task_state);
2126 if (!blk_mq_queue_inflight(md->queue))
2129 if (signal_pending_state(task_state, current)) {
2141 * Process the deferred bios
2143 static void dm_wq_work(struct work_struct *work)
2145 struct mapped_device *md = container_of(work, struct mapped_device, work);
2148 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2149 spin_lock_irq(&md->deferred_lock);
2150 bio = bio_list_pop(&md->deferred);
2151 spin_unlock_irq(&md->deferred_lock);
2156 submit_bio_noacct(bio);
2160 static void dm_queue_flush(struct mapped_device *md)
2162 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2163 smp_mb__after_atomic();
2164 queue_work(md->wq, &md->work);
2168 * Swap in a new table, returning the old one for the caller to destroy.
2170 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2172 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2173 struct queue_limits limits;
2176 mutex_lock(&md->suspend_lock);
2178 /* device must be suspended */
2179 if (!dm_suspended_md(md))
2183 * If the new table has no data devices, retain the existing limits.
2184 * This helps multipath with queue_if_no_path if all paths disappear,
2185 * then new I/O is queued based on these limits, and then some paths
2188 if (dm_table_has_no_data_devices(table)) {
2189 live_map = dm_get_live_table_fast(md);
2191 limits = md->queue->limits;
2192 dm_put_live_table_fast(md);
2196 r = dm_calculate_queue_limits(table, &limits);
2203 map = __bind(md, table, &limits);
2204 dm_issue_global_event();
2207 mutex_unlock(&md->suspend_lock);
2212 * Functions to lock and unlock any filesystem running on the
2215 static int lock_fs(struct mapped_device *md)
2219 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2221 r = freeze_bdev(md->disk->part0);
2223 set_bit(DMF_FROZEN, &md->flags);
2227 static void unlock_fs(struct mapped_device *md)
2229 if (!test_bit(DMF_FROZEN, &md->flags))
2231 thaw_bdev(md->disk->part0);
2232 clear_bit(DMF_FROZEN, &md->flags);
2236 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2237 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2238 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2240 * If __dm_suspend returns 0, the device is completely quiescent
2241 * now. There is no request-processing activity. All new requests
2242 * are being added to md->deferred list.
2244 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2245 unsigned suspend_flags, unsigned int task_state,
2246 int dmf_suspended_flag)
2248 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2249 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2252 lockdep_assert_held(&md->suspend_lock);
2255 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2256 * This flag is cleared before dm_suspend returns.
2259 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2261 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2264 * This gets reverted if there's an error later and the targets
2265 * provide the .presuspend_undo hook.
2267 dm_table_presuspend_targets(map);
2270 * Flush I/O to the device.
2271 * Any I/O submitted after lock_fs() may not be flushed.
2272 * noflush takes precedence over do_lockfs.
2273 * (lock_fs() flushes I/Os and waits for them to complete.)
2275 if (!noflush && do_lockfs) {
2278 dm_table_presuspend_undo_targets(map);
2284 * Here we must make sure that no processes are submitting requests
2285 * to target drivers i.e. no one may be executing
2286 * __split_and_process_bio from dm_submit_bio.
2288 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2289 * we take the write lock. To prevent any process from reentering
2290 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2291 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2292 * flush_workqueue(md->wq).
2294 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2296 synchronize_srcu(&md->io_barrier);
2299 * Stop md->queue before flushing md->wq in case request-based
2300 * dm defers requests to md->wq from md->queue.
2302 if (dm_request_based(md))
2303 dm_stop_queue(md->queue);
2305 flush_workqueue(md->wq);
2308 * At this point no more requests are entering target request routines.
2309 * We call dm_wait_for_completion to wait for all existing requests
2312 r = dm_wait_for_completion(md, task_state);
2314 set_bit(dmf_suspended_flag, &md->flags);
2317 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2319 synchronize_srcu(&md->io_barrier);
2321 /* were we interrupted ? */
2325 if (dm_request_based(md))
2326 dm_start_queue(md->queue);
2329 dm_table_presuspend_undo_targets(map);
2330 /* pushback list is already flushed, so skip flush */
2337 * We need to be able to change a mapping table under a mounted
2338 * filesystem. For example we might want to move some data in
2339 * the background. Before the table can be swapped with
2340 * dm_bind_table, dm_suspend must be called to flush any in
2341 * flight bios and ensure that any further io gets deferred.
2344 * Suspend mechanism in request-based dm.
2346 * 1. Flush all I/Os by lock_fs() if needed.
2347 * 2. Stop dispatching any I/O by stopping the request_queue.
2348 * 3. Wait for all in-flight I/Os to be completed or requeued.
2350 * To abort suspend, start the request_queue.
2352 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2354 struct dm_table *map = NULL;
2358 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2360 if (dm_suspended_md(md)) {
2365 if (dm_suspended_internally_md(md)) {
2366 /* already internally suspended, wait for internal resume */
2367 mutex_unlock(&md->suspend_lock);
2368 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2374 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2376 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2380 set_bit(DMF_POST_SUSPENDING, &md->flags);
2381 dm_table_postsuspend_targets(map);
2382 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2385 mutex_unlock(&md->suspend_lock);
2389 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2392 int r = dm_table_resume_targets(map);
2400 * Flushing deferred I/Os must be done after targets are resumed
2401 * so that mapping of targets can work correctly.
2402 * Request-based dm is queueing the deferred I/Os in its request_queue.
2404 if (dm_request_based(md))
2405 dm_start_queue(md->queue);
2412 int dm_resume(struct mapped_device *md)
2415 struct dm_table *map = NULL;
2419 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2421 if (!dm_suspended_md(md))
2424 if (dm_suspended_internally_md(md)) {
2425 /* already internally suspended, wait for internal resume */
2426 mutex_unlock(&md->suspend_lock);
2427 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2433 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2434 if (!map || !dm_table_get_size(map))
2437 r = __dm_resume(md, map);
2441 clear_bit(DMF_SUSPENDED, &md->flags);
2443 mutex_unlock(&md->suspend_lock);
2449 * Internal suspend/resume works like userspace-driven suspend. It waits
2450 * until all bios finish and prevents issuing new bios to the target drivers.
2451 * It may be used only from the kernel.
2454 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2456 struct dm_table *map = NULL;
2458 lockdep_assert_held(&md->suspend_lock);
2460 if (md->internal_suspend_count++)
2461 return; /* nested internal suspend */
2463 if (dm_suspended_md(md)) {
2464 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2465 return; /* nest suspend */
2468 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2471 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2472 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2473 * would require changing .presuspend to return an error -- avoid this
2474 * until there is a need for more elaborate variants of internal suspend.
2476 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2477 DMF_SUSPENDED_INTERNALLY);
2479 set_bit(DMF_POST_SUSPENDING, &md->flags);
2480 dm_table_postsuspend_targets(map);
2481 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2484 static void __dm_internal_resume(struct mapped_device *md)
2486 BUG_ON(!md->internal_suspend_count);
2488 if (--md->internal_suspend_count)
2489 return; /* resume from nested internal suspend */
2491 if (dm_suspended_md(md))
2492 goto done; /* resume from nested suspend */
2495 * NOTE: existing callers don't need to call dm_table_resume_targets
2496 * (which may fail -- so best to avoid it for now by passing NULL map)
2498 (void) __dm_resume(md, NULL);
2501 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2502 smp_mb__after_atomic();
2503 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2506 void dm_internal_suspend_noflush(struct mapped_device *md)
2508 mutex_lock(&md->suspend_lock);
2509 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2510 mutex_unlock(&md->suspend_lock);
2512 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2514 void dm_internal_resume(struct mapped_device *md)
2516 mutex_lock(&md->suspend_lock);
2517 __dm_internal_resume(md);
2518 mutex_unlock(&md->suspend_lock);
2520 EXPORT_SYMBOL_GPL(dm_internal_resume);
2523 * Fast variants of internal suspend/resume hold md->suspend_lock,
2524 * which prevents interaction with userspace-driven suspend.
2527 void dm_internal_suspend_fast(struct mapped_device *md)
2529 mutex_lock(&md->suspend_lock);
2530 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2533 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2534 synchronize_srcu(&md->io_barrier);
2535 flush_workqueue(md->wq);
2536 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2538 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2540 void dm_internal_resume_fast(struct mapped_device *md)
2542 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2548 mutex_unlock(&md->suspend_lock);
2550 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2552 /*-----------------------------------------------------------------
2553 * Event notification.
2554 *---------------------------------------------------------------*/
2555 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2560 char udev_cookie[DM_COOKIE_LENGTH];
2561 char *envp[] = { udev_cookie, NULL };
2563 noio_flag = memalloc_noio_save();
2566 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2568 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2569 DM_COOKIE_ENV_VAR_NAME, cookie);
2570 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2574 memalloc_noio_restore(noio_flag);
2579 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2581 return atomic_add_return(1, &md->uevent_seq);
2584 uint32_t dm_get_event_nr(struct mapped_device *md)
2586 return atomic_read(&md->event_nr);
2589 int dm_wait_event(struct mapped_device *md, int event_nr)
2591 return wait_event_interruptible(md->eventq,
2592 (event_nr != atomic_read(&md->event_nr)));
2595 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2597 unsigned long flags;
2599 spin_lock_irqsave(&md->uevent_lock, flags);
2600 list_add(elist, &md->uevent_list);
2601 spin_unlock_irqrestore(&md->uevent_lock, flags);
2605 * The gendisk is only valid as long as you have a reference
2608 struct gendisk *dm_disk(struct mapped_device *md)
2612 EXPORT_SYMBOL_GPL(dm_disk);
2614 struct kobject *dm_kobject(struct mapped_device *md)
2616 return &md->kobj_holder.kobj;
2619 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2621 struct mapped_device *md;
2623 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2625 spin_lock(&_minor_lock);
2626 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2632 spin_unlock(&_minor_lock);
2637 int dm_suspended_md(struct mapped_device *md)
2639 return test_bit(DMF_SUSPENDED, &md->flags);
2642 static int dm_post_suspending_md(struct mapped_device *md)
2644 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2647 int dm_suspended_internally_md(struct mapped_device *md)
2649 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2652 int dm_test_deferred_remove_flag(struct mapped_device *md)
2654 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2657 int dm_suspended(struct dm_target *ti)
2659 return dm_suspended_md(ti->table->md);
2661 EXPORT_SYMBOL_GPL(dm_suspended);
2663 int dm_post_suspending(struct dm_target *ti)
2665 return dm_post_suspending_md(ti->table->md);
2667 EXPORT_SYMBOL_GPL(dm_post_suspending);
2669 int dm_noflush_suspending(struct dm_target *ti)
2671 return __noflush_suspending(ti->table->md);
2673 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2675 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2676 unsigned integrity, unsigned per_io_data_size,
2677 unsigned min_pool_size)
2679 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2680 unsigned int pool_size = 0;
2681 unsigned int front_pad, io_front_pad;
2688 case DM_TYPE_BIO_BASED:
2689 case DM_TYPE_DAX_BIO_BASED:
2690 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2691 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2692 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2693 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2696 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2699 case DM_TYPE_REQUEST_BASED:
2700 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2701 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2702 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2708 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2712 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2718 dm_free_md_mempools(pools);
2723 void dm_free_md_mempools(struct dm_md_mempools *pools)
2728 bioset_exit(&pools->bs);
2729 bioset_exit(&pools->io_bs);
2741 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2744 struct mapped_device *md = bdev->bd_disk->private_data;
2745 struct dm_table *table;
2746 struct dm_target *ti;
2747 int ret = -ENOTTY, srcu_idx;
2749 table = dm_get_live_table(md, &srcu_idx);
2750 if (!table || !dm_table_get_size(table))
2753 /* We only support devices that have a single target */
2754 if (dm_table_get_num_targets(table) != 1)
2756 ti = dm_table_get_target(table, 0);
2759 if (!ti->type->iterate_devices)
2762 ret = ti->type->iterate_devices(ti, fn, data);
2764 dm_put_live_table(md, srcu_idx);
2769 * For register / unregister we need to manually call out to every path.
2771 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2772 sector_t start, sector_t len, void *data)
2774 struct dm_pr *pr = data;
2775 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2777 if (!ops || !ops->pr_register)
2779 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2782 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2793 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2794 if (ret && new_key) {
2795 /* unregister all paths if we failed to register any path */
2796 pr.old_key = new_key;
2799 pr.fail_early = false;
2800 dm_call_pr(bdev, __dm_pr_register, &pr);
2806 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2809 struct mapped_device *md = bdev->bd_disk->private_data;
2810 const struct pr_ops *ops;
2813 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2817 ops = bdev->bd_disk->fops->pr_ops;
2818 if (ops && ops->pr_reserve)
2819 r = ops->pr_reserve(bdev, key, type, flags);
2823 dm_unprepare_ioctl(md, srcu_idx);
2827 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2829 struct mapped_device *md = bdev->bd_disk->private_data;
2830 const struct pr_ops *ops;
2833 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2837 ops = bdev->bd_disk->fops->pr_ops;
2838 if (ops && ops->pr_release)
2839 r = ops->pr_release(bdev, key, type);
2843 dm_unprepare_ioctl(md, srcu_idx);
2847 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2848 enum pr_type type, bool abort)
2850 struct mapped_device *md = bdev->bd_disk->private_data;
2851 const struct pr_ops *ops;
2854 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2858 ops = bdev->bd_disk->fops->pr_ops;
2859 if (ops && ops->pr_preempt)
2860 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2864 dm_unprepare_ioctl(md, srcu_idx);
2868 static int dm_pr_clear(struct block_device *bdev, u64 key)
2870 struct mapped_device *md = bdev->bd_disk->private_data;
2871 const struct pr_ops *ops;
2874 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2878 ops = bdev->bd_disk->fops->pr_ops;
2879 if (ops && ops->pr_clear)
2880 r = ops->pr_clear(bdev, key);
2884 dm_unprepare_ioctl(md, srcu_idx);
2888 static const struct pr_ops dm_pr_ops = {
2889 .pr_register = dm_pr_register,
2890 .pr_reserve = dm_pr_reserve,
2891 .pr_release = dm_pr_release,
2892 .pr_preempt = dm_pr_preempt,
2893 .pr_clear = dm_pr_clear,
2896 static const struct block_device_operations dm_blk_dops = {
2897 .submit_bio = dm_submit_bio,
2898 .open = dm_blk_open,
2899 .release = dm_blk_close,
2900 .ioctl = dm_blk_ioctl,
2901 .getgeo = dm_blk_getgeo,
2902 .report_zones = dm_blk_report_zones,
2903 .pr_ops = &dm_pr_ops,
2904 .owner = THIS_MODULE
2907 static const struct block_device_operations dm_rq_blk_dops = {
2908 .open = dm_blk_open,
2909 .release = dm_blk_close,
2910 .ioctl = dm_blk_ioctl,
2911 .getgeo = dm_blk_getgeo,
2912 .pr_ops = &dm_pr_ops,
2913 .owner = THIS_MODULE
2916 static const struct dax_operations dm_dax_ops = {
2917 .direct_access = dm_dax_direct_access,
2918 .zero_page_range = dm_dax_zero_page_range,
2924 module_init(dm_init);
2925 module_exit(dm_exit);
2927 module_param(major, uint, 0);
2928 MODULE_PARM_DESC(major, "The major number of the device mapper");
2930 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2931 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2933 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2934 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2936 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
2937 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
2939 MODULE_DESCRIPTION(DM_NAME " driver");
2940 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2941 MODULE_LICENSE("GPL");