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 void *dm_per_bio_data(struct bio *bio, size_t data_size)
84 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
85 if (!tio->inside_dm_io)
86 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
87 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
89 EXPORT_SYMBOL_GPL(dm_per_bio_data);
91 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
93 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
94 if (io->magic == DM_IO_MAGIC)
95 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
96 BUG_ON(io->magic != DM_TIO_MAGIC);
97 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
99 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
101 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
103 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
105 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
107 #define MINOR_ALLOCED ((void *)-1)
109 #define DM_NUMA_NODE NUMA_NO_NODE
110 static int dm_numa_node = DM_NUMA_NODE;
112 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
113 static int swap_bios = DEFAULT_SWAP_BIOS;
114 static int get_swap_bios(void)
116 int latch = READ_ONCE(swap_bios);
117 if (unlikely(latch <= 0))
118 latch = DEFAULT_SWAP_BIOS;
123 * For mempools pre-allocation at the table loading time.
125 struct dm_md_mempools {
127 struct bio_set io_bs;
130 struct table_device {
131 struct list_head list;
133 struct dm_dev dm_dev;
137 * Bio-based DM's mempools' reserved IOs set by the user.
139 #define RESERVED_BIO_BASED_IOS 16
140 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
142 static int __dm_get_module_param_int(int *module_param, int min, int max)
144 int param = READ_ONCE(*module_param);
145 int modified_param = 0;
146 bool modified = true;
149 modified_param = min;
150 else if (param > max)
151 modified_param = max;
156 (void)cmpxchg(module_param, param, modified_param);
157 param = modified_param;
163 unsigned __dm_get_module_param(unsigned *module_param,
164 unsigned def, unsigned max)
166 unsigned param = READ_ONCE(*module_param);
167 unsigned modified_param = 0;
170 modified_param = def;
171 else if (param > max)
172 modified_param = max;
174 if (modified_param) {
175 (void)cmpxchg(module_param, param, modified_param);
176 param = modified_param;
182 unsigned dm_get_reserved_bio_based_ios(void)
184 return __dm_get_module_param(&reserved_bio_based_ios,
185 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
187 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
189 static unsigned dm_get_numa_node(void)
191 return __dm_get_module_param_int(&dm_numa_node,
192 DM_NUMA_NODE, num_online_nodes() - 1);
195 static int __init local_init(void)
199 r = dm_uevent_init();
203 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
204 if (!deferred_remove_workqueue) {
206 goto out_uevent_exit;
210 r = register_blkdev(_major, _name);
212 goto out_free_workqueue;
220 destroy_workqueue(deferred_remove_workqueue);
227 static void local_exit(void)
229 flush_scheduled_work();
230 destroy_workqueue(deferred_remove_workqueue);
232 unregister_blkdev(_major, _name);
237 DMINFO("cleaned up");
240 static int (*_inits[])(void) __initdata = {
251 static void (*_exits[])(void) = {
262 static int __init dm_init(void)
264 const int count = ARRAY_SIZE(_inits);
267 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
268 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
269 " Duplicate IMA measurements will not be recorded in the IMA log.");
272 for (i = 0; i < count; i++) {
286 static void __exit dm_exit(void)
288 int i = ARRAY_SIZE(_exits);
294 * Should be empty by this point.
296 idr_destroy(&_minor_idr);
300 * Block device functions
302 int dm_deleting_md(struct mapped_device *md)
304 return test_bit(DMF_DELETING, &md->flags);
307 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
309 struct mapped_device *md;
311 spin_lock(&_minor_lock);
313 md = bdev->bd_disk->private_data;
317 if (test_bit(DMF_FREEING, &md->flags) ||
318 dm_deleting_md(md)) {
324 atomic_inc(&md->open_count);
326 spin_unlock(&_minor_lock);
328 return md ? 0 : -ENXIO;
331 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
333 struct mapped_device *md;
335 spin_lock(&_minor_lock);
337 md = disk->private_data;
341 if (atomic_dec_and_test(&md->open_count) &&
342 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
343 queue_work(deferred_remove_workqueue, &deferred_remove_work);
347 spin_unlock(&_minor_lock);
350 int dm_open_count(struct mapped_device *md)
352 return atomic_read(&md->open_count);
356 * Guarantees nothing is using the device before it's deleted.
358 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
362 spin_lock(&_minor_lock);
364 if (dm_open_count(md)) {
367 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
368 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
371 set_bit(DMF_DELETING, &md->flags);
373 spin_unlock(&_minor_lock);
378 int dm_cancel_deferred_remove(struct mapped_device *md)
382 spin_lock(&_minor_lock);
384 if (test_bit(DMF_DELETING, &md->flags))
387 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
389 spin_unlock(&_minor_lock);
394 static void do_deferred_remove(struct work_struct *w)
396 dm_deferred_remove();
399 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
401 struct mapped_device *md = bdev->bd_disk->private_data;
403 return dm_get_geometry(md, geo);
406 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
407 struct block_device **bdev)
409 struct dm_target *tgt;
410 struct dm_table *map;
415 map = dm_get_live_table(md, srcu_idx);
416 if (!map || !dm_table_get_size(map))
419 /* We only support devices that have a single target */
420 if (dm_table_get_num_targets(map) != 1)
423 tgt = dm_table_get_target(map, 0);
424 if (!tgt->type->prepare_ioctl)
427 if (dm_suspended_md(md))
430 r = tgt->type->prepare_ioctl(tgt, bdev);
431 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
432 dm_put_live_table(md, *srcu_idx);
440 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
442 dm_put_live_table(md, srcu_idx);
445 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
446 unsigned int cmd, unsigned long arg)
448 struct mapped_device *md = bdev->bd_disk->private_data;
451 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
457 * Target determined this ioctl is being issued against a
458 * subset of the parent bdev; require extra privileges.
460 if (!capable(CAP_SYS_RAWIO)) {
462 "%s: sending ioctl %x to DM device without required privilege.",
469 if (!bdev->bd_disk->fops->ioctl)
472 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
474 dm_unprepare_ioctl(md, srcu_idx);
478 u64 dm_start_time_ns_from_clone(struct bio *bio)
480 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
481 struct dm_io *io = tio->io;
483 return jiffies_to_nsecs(io->start_time);
485 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
487 static void start_io_acct(struct dm_io *io)
489 struct mapped_device *md = io->md;
490 struct bio *bio = io->orig_bio;
492 bio_start_io_acct_time(bio, io->start_time);
493 if (unlikely(dm_stats_used(&md->stats)))
494 dm_stats_account_io(&md->stats, bio_data_dir(bio),
495 bio->bi_iter.bi_sector, bio_sectors(bio),
496 false, 0, &io->stats_aux);
499 static void end_io_acct(struct mapped_device *md, struct bio *bio,
500 unsigned long start_time, struct dm_stats_aux *stats_aux)
502 unsigned long duration = jiffies - start_time;
504 bio_end_io_acct(bio, start_time);
506 if (unlikely(dm_stats_used(&md->stats)))
507 dm_stats_account_io(&md->stats, bio_data_dir(bio),
508 bio->bi_iter.bi_sector, bio_sectors(bio),
509 true, duration, stats_aux);
511 /* nudge anyone waiting on suspend queue */
512 if (unlikely(wq_has_sleeper(&md->wait)))
516 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
519 struct dm_target_io *tio;
522 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
526 tio = container_of(clone, struct dm_target_io, clone);
527 tio->inside_dm_io = true;
530 io = container_of(tio, struct dm_io, tio);
531 io->magic = DM_IO_MAGIC;
533 atomic_set(&io->io_count, 1);
536 spin_lock_init(&io->endio_lock);
538 io->start_time = jiffies;
543 static void free_io(struct mapped_device *md, struct dm_io *io)
545 bio_put(&io->tio.clone);
548 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
549 unsigned target_bio_nr, gfp_t gfp_mask)
551 struct dm_target_io *tio;
553 if (!ci->io->tio.io) {
554 /* the dm_target_io embedded in ci->io is available */
557 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
561 tio = container_of(clone, struct dm_target_io, clone);
562 tio->inside_dm_io = false;
565 tio->magic = DM_TIO_MAGIC;
568 tio->target_bio_nr = target_bio_nr;
573 static void free_tio(struct dm_target_io *tio)
575 if (tio->inside_dm_io)
577 bio_put(&tio->clone);
581 * Add the bio to the list of deferred io.
583 static void queue_io(struct mapped_device *md, struct bio *bio)
587 spin_lock_irqsave(&md->deferred_lock, flags);
588 bio_list_add(&md->deferred, bio);
589 spin_unlock_irqrestore(&md->deferred_lock, flags);
590 queue_work(md->wq, &md->work);
594 * Everyone (including functions in this file), should use this
595 * function to access the md->map field, and make sure they call
596 * dm_put_live_table() when finished.
598 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
600 *srcu_idx = srcu_read_lock(&md->io_barrier);
602 return srcu_dereference(md->map, &md->io_barrier);
605 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
607 srcu_read_unlock(&md->io_barrier, srcu_idx);
610 void dm_sync_table(struct mapped_device *md)
612 synchronize_srcu(&md->io_barrier);
613 synchronize_rcu_expedited();
617 * A fast alternative to dm_get_live_table/dm_put_live_table.
618 * The caller must not block between these two functions.
620 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
623 return rcu_dereference(md->map);
626 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
631 static char *_dm_claim_ptr = "I belong to device-mapper";
634 * Open a table device so we can use it as a map destination.
636 static int open_table_device(struct table_device *td, dev_t dev,
637 struct mapped_device *md)
639 struct block_device *bdev;
643 BUG_ON(td->dm_dev.bdev);
645 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
647 return PTR_ERR(bdev);
649 r = bd_link_disk_holder(bdev, dm_disk(md));
651 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
655 td->dm_dev.bdev = bdev;
656 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off);
661 * Close a table device that we've been using.
663 static void close_table_device(struct table_device *td, struct mapped_device *md)
665 if (!td->dm_dev.bdev)
668 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
669 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
670 put_dax(td->dm_dev.dax_dev);
671 td->dm_dev.bdev = NULL;
672 td->dm_dev.dax_dev = NULL;
675 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
678 struct table_device *td;
680 list_for_each_entry(td, l, list)
681 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
687 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
688 struct dm_dev **result)
691 struct table_device *td;
693 mutex_lock(&md->table_devices_lock);
694 td = find_table_device(&md->table_devices, dev, mode);
696 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
698 mutex_unlock(&md->table_devices_lock);
702 td->dm_dev.mode = mode;
703 td->dm_dev.bdev = NULL;
705 if ((r = open_table_device(td, dev, md))) {
706 mutex_unlock(&md->table_devices_lock);
711 format_dev_t(td->dm_dev.name, dev);
713 refcount_set(&td->count, 1);
714 list_add(&td->list, &md->table_devices);
716 refcount_inc(&td->count);
718 mutex_unlock(&md->table_devices_lock);
720 *result = &td->dm_dev;
724 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
726 struct table_device *td = container_of(d, struct table_device, dm_dev);
728 mutex_lock(&md->table_devices_lock);
729 if (refcount_dec_and_test(&td->count)) {
730 close_table_device(td, md);
734 mutex_unlock(&md->table_devices_lock);
737 static void free_table_devices(struct list_head *devices)
739 struct list_head *tmp, *next;
741 list_for_each_safe(tmp, next, devices) {
742 struct table_device *td = list_entry(tmp, struct table_device, list);
744 DMWARN("dm_destroy: %s still exists with %d references",
745 td->dm_dev.name, refcount_read(&td->count));
751 * Get the geometry associated with a dm device
753 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
761 * Set the geometry of a device.
763 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
765 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
767 if (geo->start > sz) {
768 DMWARN("Start sector is beyond the geometry limits.");
777 static int __noflush_suspending(struct mapped_device *md)
779 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
783 * Decrements the number of outstanding ios that a bio has been
784 * cloned into, completing the original io if necc.
786 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
789 blk_status_t io_error;
791 struct mapped_device *md = io->md;
792 unsigned long start_time = 0;
793 struct dm_stats_aux stats_aux;
795 /* Push-back supersedes any I/O errors */
796 if (unlikely(error)) {
797 spin_lock_irqsave(&io->endio_lock, flags);
798 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
800 spin_unlock_irqrestore(&io->endio_lock, flags);
803 if (atomic_dec_and_test(&io->io_count)) {
805 if (io->status == BLK_STS_DM_REQUEUE) {
807 * Target requested pushing back the I/O.
809 spin_lock_irqsave(&md->deferred_lock, flags);
810 if (__noflush_suspending(md) &&
811 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
812 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
813 bio_list_add_head(&md->deferred, bio);
816 * noflush suspend was interrupted or this is
817 * a write to a zoned target.
819 io->status = BLK_STS_IOERR;
821 spin_unlock_irqrestore(&md->deferred_lock, flags);
824 io_error = io->status;
825 start_time = io->start_time;
826 stats_aux = io->stats_aux;
828 end_io_acct(md, bio, start_time, &stats_aux);
830 if (io_error == BLK_STS_DM_REQUEUE)
833 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
835 * Preflush done for flush with data, reissue
836 * without REQ_PREFLUSH.
838 bio->bi_opf &= ~REQ_PREFLUSH;
841 /* done with normal IO or empty flush */
843 bio->bi_status = io_error;
849 void disable_discard(struct mapped_device *md)
851 struct queue_limits *limits = dm_get_queue_limits(md);
853 /* device doesn't really support DISCARD, disable it */
854 limits->max_discard_sectors = 0;
855 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
858 void disable_write_same(struct mapped_device *md)
860 struct queue_limits *limits = dm_get_queue_limits(md);
862 /* device doesn't really support WRITE SAME, disable it */
863 limits->max_write_same_sectors = 0;
866 void disable_write_zeroes(struct mapped_device *md)
868 struct queue_limits *limits = dm_get_queue_limits(md);
870 /* device doesn't really support WRITE ZEROES, disable it */
871 limits->max_write_zeroes_sectors = 0;
874 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
876 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
879 static void clone_endio(struct bio *bio)
881 blk_status_t error = bio->bi_status;
882 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
883 struct dm_io *io = tio->io;
884 struct mapped_device *md = tio->io->md;
885 dm_endio_fn endio = tio->ti->type->end_io;
886 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
888 if (unlikely(error == BLK_STS_TARGET)) {
889 if (bio_op(bio) == REQ_OP_DISCARD &&
890 !q->limits.max_discard_sectors)
892 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
893 !q->limits.max_write_same_sectors)
894 disable_write_same(md);
895 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
896 !q->limits.max_write_zeroes_sectors)
897 disable_write_zeroes(md);
900 if (blk_queue_is_zoned(q))
901 dm_zone_endio(io, bio);
904 int r = endio(tio->ti, bio, &error);
906 case DM_ENDIO_REQUEUE:
908 * Requeuing writes to a sequential zone of a zoned
909 * target will break the sequential write pattern:
912 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
913 error = BLK_STS_IOERR;
915 error = BLK_STS_DM_REQUEUE;
919 case DM_ENDIO_INCOMPLETE:
920 /* The target will handle the io */
923 DMWARN("unimplemented target endio return value: %d", r);
928 if (unlikely(swap_bios_limit(tio->ti, bio))) {
929 struct mapped_device *md = io->md;
930 up(&md->swap_bios_semaphore);
934 dm_io_dec_pending(io, error);
938 * Return maximum size of I/O possible at the supplied sector up to the current
941 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
942 sector_t target_offset)
944 return ti->len - target_offset;
947 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
949 sector_t target_offset = dm_target_offset(ti, sector);
950 sector_t len = max_io_len_target_boundary(ti, target_offset);
954 * Does the target need to split IO even further?
955 * - varied (per target) IO splitting is a tenet of DM; this
956 * explains why stacked chunk_sectors based splitting via
957 * blk_max_size_offset() isn't possible here. So pass in
958 * ti->max_io_len to override stacked chunk_sectors.
960 if (ti->max_io_len) {
961 max_len = blk_max_size_offset(ti->table->md->queue,
962 target_offset, ti->max_io_len);
970 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
972 if (len > UINT_MAX) {
973 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
974 (unsigned long long)len, UINT_MAX);
975 ti->error = "Maximum size of target IO is too large";
979 ti->max_io_len = (uint32_t) len;
983 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
985 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
986 sector_t sector, int *srcu_idx)
987 __acquires(md->io_barrier)
989 struct dm_table *map;
990 struct dm_target *ti;
992 map = dm_get_live_table(md, srcu_idx);
996 ti = dm_table_find_target(map, sector);
1003 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1004 long nr_pages, void **kaddr, pfn_t *pfn)
1006 struct mapped_device *md = dax_get_private(dax_dev);
1007 sector_t sector = pgoff * PAGE_SECTORS;
1008 struct dm_target *ti;
1009 long len, ret = -EIO;
1012 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1016 if (!ti->type->direct_access)
1018 len = max_io_len(ti, sector) / PAGE_SECTORS;
1021 nr_pages = min(len, nr_pages);
1022 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1025 dm_put_live_table(md, srcu_idx);
1030 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1033 struct mapped_device *md = dax_get_private(dax_dev);
1034 sector_t sector = pgoff * PAGE_SECTORS;
1035 struct dm_target *ti;
1039 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1043 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1045 * ->zero_page_range() is mandatory dax operation. If we are
1046 * here, something is wrong.
1050 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1052 dm_put_live_table(md, srcu_idx);
1058 * A target may call dm_accept_partial_bio only from the map routine. It is
1059 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1060 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1062 * dm_accept_partial_bio informs the dm that the target only wants to process
1063 * additional n_sectors sectors of the bio and the rest of the data should be
1064 * sent in a next bio.
1066 * A diagram that explains the arithmetics:
1067 * +--------------------+---------------+-------+
1069 * +--------------------+---------------+-------+
1071 * <-------------- *tio->len_ptr --------------->
1072 * <------- bi_size ------->
1075 * Region 1 was already iterated over with bio_advance or similar function.
1076 * (it may be empty if the target doesn't use bio_advance)
1077 * Region 2 is the remaining bio size that the target wants to process.
1078 * (it may be empty if region 1 is non-empty, although there is no reason
1080 * The target requires that region 3 is to be sent in the next bio.
1082 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1083 * the partially processed part (the sum of regions 1+2) must be the same for all
1084 * copies of the bio.
1086 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1088 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1089 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1091 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1092 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1093 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1094 BUG_ON(bi_size > *tio->len_ptr);
1095 BUG_ON(n_sectors > bi_size);
1097 *tio->len_ptr -= bi_size - n_sectors;
1098 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1100 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1102 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1104 mutex_lock(&md->swap_bios_lock);
1105 while (latch < md->swap_bios) {
1107 down(&md->swap_bios_semaphore);
1110 while (latch > md->swap_bios) {
1112 up(&md->swap_bios_semaphore);
1115 mutex_unlock(&md->swap_bios_lock);
1118 static void __map_bio(struct dm_target_io *tio)
1122 struct bio *clone = &tio->clone;
1123 struct dm_io *io = tio->io;
1124 struct dm_target *ti = tio->ti;
1126 clone->bi_end_io = clone_endio;
1129 * Map the clone. If r == 0 we don't need to do
1130 * anything, the target has assumed ownership of
1133 dm_io_inc_pending(io);
1134 sector = clone->bi_iter.bi_sector;
1136 if (unlikely(swap_bios_limit(ti, clone))) {
1137 struct mapped_device *md = io->md;
1138 int latch = get_swap_bios();
1139 if (unlikely(latch != md->swap_bios))
1140 __set_swap_bios_limit(md, latch);
1141 down(&md->swap_bios_semaphore);
1145 * Check if the IO needs a special mapping due to zone append emulation
1146 * on zoned target. In this case, dm_zone_map_bio() calls the target
1149 if (dm_emulate_zone_append(io->md))
1150 r = dm_zone_map_bio(tio);
1152 r = ti->type->map(ti, clone);
1155 case DM_MAPIO_SUBMITTED:
1157 case DM_MAPIO_REMAPPED:
1158 /* the bio has been remapped so dispatch it */
1159 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1160 submit_bio_noacct(clone);
1163 if (unlikely(swap_bios_limit(ti, clone))) {
1164 struct mapped_device *md = io->md;
1165 up(&md->swap_bios_semaphore);
1168 dm_io_dec_pending(io, BLK_STS_IOERR);
1170 case DM_MAPIO_REQUEUE:
1171 if (unlikely(swap_bios_limit(ti, clone))) {
1172 struct mapped_device *md = io->md;
1173 up(&md->swap_bios_semaphore);
1176 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1179 DMWARN("unimplemented target map return value: %d", r);
1184 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1186 bio->bi_iter.bi_sector = sector;
1187 bio->bi_iter.bi_size = to_bytes(len);
1191 * Creates a bio that consists of range of complete bvecs.
1193 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1194 sector_t sector, unsigned len)
1196 struct bio *clone = &tio->clone;
1199 __bio_clone_fast(clone, bio);
1201 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1205 if (bio_integrity(bio)) {
1206 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1207 !dm_target_passes_integrity(tio->ti->type))) {
1208 DMWARN("%s: the target %s doesn't support integrity data.",
1209 dm_device_name(tio->io->md),
1210 tio->ti->type->name);
1214 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1219 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1220 clone->bi_iter.bi_size = to_bytes(len);
1222 if (bio_integrity(bio))
1223 bio_integrity_trim(clone);
1228 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1229 struct dm_target *ti, unsigned num_bios)
1231 struct dm_target_io *tio;
1237 if (num_bios == 1) {
1238 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1239 bio_list_add(blist, &tio->clone);
1243 for (try = 0; try < 2; try++) {
1248 mutex_lock(&ci->io->md->table_devices_lock);
1249 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1250 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1254 bio_list_add(blist, &tio->clone);
1257 mutex_unlock(&ci->io->md->table_devices_lock);
1258 if (bio_nr == num_bios)
1261 while ((bio = bio_list_pop(blist))) {
1262 tio = container_of(bio, struct dm_target_io, clone);
1268 static void __clone_and_map_simple_bio(struct clone_info *ci,
1269 struct dm_target_io *tio, unsigned *len)
1271 struct bio *clone = &tio->clone;
1275 __bio_clone_fast(clone, ci->bio);
1277 bio_setup_sector(clone, ci->sector, *len);
1281 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1282 unsigned num_bios, unsigned *len)
1284 struct bio_list blist = BIO_EMPTY_LIST;
1286 struct dm_target_io *tio;
1288 alloc_multiple_bios(&blist, ci, ti, num_bios);
1290 while ((bio = bio_list_pop(&blist))) {
1291 tio = container_of(bio, struct dm_target_io, clone);
1292 __clone_and_map_simple_bio(ci, tio, len);
1296 static int __send_empty_flush(struct clone_info *ci)
1298 unsigned target_nr = 0;
1299 struct dm_target *ti;
1300 struct bio flush_bio;
1303 * Use an on-stack bio for this, it's safe since we don't
1304 * need to reference it after submit. It's just used as
1305 * the basis for the clone(s).
1307 bio_init(&flush_bio, NULL, 0);
1308 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1309 bio_set_dev(&flush_bio, ci->io->md->disk->part0);
1311 ci->bio = &flush_bio;
1312 ci->sector_count = 0;
1314 BUG_ON(bio_has_data(ci->bio));
1315 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1316 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1318 bio_uninit(ci->bio);
1322 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1323 sector_t sector, unsigned *len)
1325 struct bio *bio = ci->bio;
1326 struct dm_target_io *tio;
1329 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1331 r = clone_bio(tio, bio, sector, *len);
1341 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1347 * Even though the device advertised support for this type of
1348 * request, that does not mean every target supports it, and
1349 * reconfiguration might also have changed that since the
1350 * check was performed.
1355 len = min_t(sector_t, ci->sector_count,
1356 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1358 __send_duplicate_bios(ci, ti, num_bios, &len);
1361 ci->sector_count -= len;
1366 static bool is_abnormal_io(struct bio *bio)
1370 switch (bio_op(bio)) {
1371 case REQ_OP_DISCARD:
1372 case REQ_OP_SECURE_ERASE:
1373 case REQ_OP_WRITE_SAME:
1374 case REQ_OP_WRITE_ZEROES:
1382 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1385 struct bio *bio = ci->bio;
1386 unsigned num_bios = 0;
1388 switch (bio_op(bio)) {
1389 case REQ_OP_DISCARD:
1390 num_bios = ti->num_discard_bios;
1392 case REQ_OP_SECURE_ERASE:
1393 num_bios = ti->num_secure_erase_bios;
1395 case REQ_OP_WRITE_SAME:
1396 num_bios = ti->num_write_same_bios;
1398 case REQ_OP_WRITE_ZEROES:
1399 num_bios = ti->num_write_zeroes_bios;
1405 *result = __send_changing_extent_only(ci, ti, num_bios);
1410 * Select the correct strategy for processing a non-flush bio.
1412 static int __split_and_process_non_flush(struct clone_info *ci)
1414 struct dm_target *ti;
1418 ti = dm_table_find_target(ci->map, ci->sector);
1422 if (__process_abnormal_io(ci, ti, &r))
1425 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1427 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1432 ci->sector_count -= len;
1437 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1438 struct dm_table *map, struct bio *bio)
1441 ci->io = alloc_io(md, bio);
1442 ci->sector = bio->bi_iter.bi_sector;
1446 * Entry point to split a bio into clones and submit them to the targets.
1448 static void __split_and_process_bio(struct mapped_device *md,
1449 struct dm_table *map, struct bio *bio)
1451 struct clone_info ci;
1454 init_clone_info(&ci, md, map, bio);
1456 if (bio->bi_opf & REQ_PREFLUSH) {
1457 error = __send_empty_flush(&ci);
1458 /* dm_io_dec_pending submits any data associated with flush */
1459 } else if (op_is_zone_mgmt(bio_op(bio))) {
1461 ci.sector_count = 0;
1462 error = __split_and_process_non_flush(&ci);
1465 ci.sector_count = bio_sectors(bio);
1466 error = __split_and_process_non_flush(&ci);
1467 if (ci.sector_count && !error) {
1469 * Remainder must be passed to submit_bio_noacct()
1470 * so that it gets handled *after* bios already submitted
1471 * have been completely processed.
1472 * We take a clone of the original to store in
1473 * ci.io->orig_bio to be used by end_io_acct() and
1474 * for dec_pending to use for completion handling.
1476 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1477 GFP_NOIO, &md->queue->bio_split);
1478 ci.io->orig_bio = b;
1481 trace_block_split(b, bio->bi_iter.bi_sector);
1482 submit_bio_noacct(bio);
1485 start_io_acct(ci.io);
1487 /* drop the extra reference count */
1488 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1491 static void dm_submit_bio(struct bio *bio)
1493 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1495 struct dm_table *map;
1497 map = dm_get_live_table(md, &srcu_idx);
1498 if (unlikely(!map)) {
1499 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1500 dm_device_name(md));
1505 /* If suspended, queue this IO for later */
1506 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1507 if (bio->bi_opf & REQ_NOWAIT)
1508 bio_wouldblock_error(bio);
1509 else if (bio->bi_opf & REQ_RAHEAD)
1517 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1518 * otherwise associated queue_limits won't be imposed.
1520 if (is_abnormal_io(bio))
1521 blk_queue_split(&bio);
1523 __split_and_process_bio(md, map, bio);
1525 dm_put_live_table(md, srcu_idx);
1528 /*-----------------------------------------------------------------
1529 * An IDR is used to keep track of allocated minor numbers.
1530 *---------------------------------------------------------------*/
1531 static void free_minor(int minor)
1533 spin_lock(&_minor_lock);
1534 idr_remove(&_minor_idr, minor);
1535 spin_unlock(&_minor_lock);
1539 * See if the device with a specific minor # is free.
1541 static int specific_minor(int minor)
1545 if (minor >= (1 << MINORBITS))
1548 idr_preload(GFP_KERNEL);
1549 spin_lock(&_minor_lock);
1551 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1553 spin_unlock(&_minor_lock);
1556 return r == -ENOSPC ? -EBUSY : r;
1560 static int next_free_minor(int *minor)
1564 idr_preload(GFP_KERNEL);
1565 spin_lock(&_minor_lock);
1567 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1569 spin_unlock(&_minor_lock);
1577 static const struct block_device_operations dm_blk_dops;
1578 static const struct block_device_operations dm_rq_blk_dops;
1579 static const struct dax_operations dm_dax_ops;
1581 static void dm_wq_work(struct work_struct *work);
1583 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1584 static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1586 dm_destroy_crypto_profile(q->crypto_profile);
1589 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1591 static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1594 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1596 static void cleanup_mapped_device(struct mapped_device *md)
1599 destroy_workqueue(md->wq);
1600 bioset_exit(&md->bs);
1601 bioset_exit(&md->io_bs);
1604 dax_remove_host(md->disk);
1605 kill_dax(md->dax_dev);
1606 put_dax(md->dax_dev);
1611 spin_lock(&_minor_lock);
1612 md->disk->private_data = NULL;
1613 spin_unlock(&_minor_lock);
1614 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1616 del_gendisk(md->disk);
1618 dm_queue_destroy_crypto_profile(md->queue);
1619 blk_cleanup_disk(md->disk);
1622 cleanup_srcu_struct(&md->io_barrier);
1624 mutex_destroy(&md->suspend_lock);
1625 mutex_destroy(&md->type_lock);
1626 mutex_destroy(&md->table_devices_lock);
1627 mutex_destroy(&md->swap_bios_lock);
1629 dm_mq_cleanup_mapped_device(md);
1630 dm_cleanup_zoned_dev(md);
1634 * Allocate and initialise a blank device with a given minor.
1636 static struct mapped_device *alloc_dev(int minor)
1638 int r, numa_node_id = dm_get_numa_node();
1639 struct mapped_device *md;
1642 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1644 DMWARN("unable to allocate device, out of memory.");
1648 if (!try_module_get(THIS_MODULE))
1649 goto bad_module_get;
1651 /* get a minor number for the dev */
1652 if (minor == DM_ANY_MINOR)
1653 r = next_free_minor(&minor);
1655 r = specific_minor(minor);
1659 r = init_srcu_struct(&md->io_barrier);
1661 goto bad_io_barrier;
1663 md->numa_node_id = numa_node_id;
1664 md->init_tio_pdu = false;
1665 md->type = DM_TYPE_NONE;
1666 mutex_init(&md->suspend_lock);
1667 mutex_init(&md->type_lock);
1668 mutex_init(&md->table_devices_lock);
1669 spin_lock_init(&md->deferred_lock);
1670 atomic_set(&md->holders, 1);
1671 atomic_set(&md->open_count, 0);
1672 atomic_set(&md->event_nr, 0);
1673 atomic_set(&md->uevent_seq, 0);
1674 INIT_LIST_HEAD(&md->uevent_list);
1675 INIT_LIST_HEAD(&md->table_devices);
1676 spin_lock_init(&md->uevent_lock);
1679 * default to bio-based until DM table is loaded and md->type
1680 * established. If request-based table is loaded: blk-mq will
1681 * override accordingly.
1683 md->disk = blk_alloc_disk(md->numa_node_id);
1686 md->queue = md->disk->queue;
1688 init_waitqueue_head(&md->wait);
1689 INIT_WORK(&md->work, dm_wq_work);
1690 init_waitqueue_head(&md->eventq);
1691 init_completion(&md->kobj_holder.completion);
1693 md->swap_bios = get_swap_bios();
1694 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1695 mutex_init(&md->swap_bios_lock);
1697 md->disk->major = _major;
1698 md->disk->first_minor = minor;
1699 md->disk->minors = 1;
1700 md->disk->flags |= GENHD_FL_NO_PART;
1701 md->disk->fops = &dm_blk_dops;
1702 md->disk->queue = md->queue;
1703 md->disk->private_data = md;
1704 sprintf(md->disk->disk_name, "dm-%d", minor);
1706 if (IS_ENABLED(CONFIG_FS_DAX)) {
1707 md->dax_dev = alloc_dax(md, &dm_dax_ops);
1708 if (IS_ERR(md->dax_dev)) {
1712 set_dax_nocache(md->dax_dev);
1713 set_dax_nomc(md->dax_dev);
1714 if (dax_add_host(md->dax_dev, md->disk))
1718 format_dev_t(md->name, MKDEV(_major, minor));
1720 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
1724 dm_stats_init(&md->stats);
1726 /* Populate the mapping, nobody knows we exist yet */
1727 spin_lock(&_minor_lock);
1728 old_md = idr_replace(&_minor_idr, md, minor);
1729 spin_unlock(&_minor_lock);
1731 BUG_ON(old_md != MINOR_ALLOCED);
1736 cleanup_mapped_device(md);
1740 module_put(THIS_MODULE);
1746 static void unlock_fs(struct mapped_device *md);
1748 static void free_dev(struct mapped_device *md)
1750 int minor = MINOR(disk_devt(md->disk));
1754 cleanup_mapped_device(md);
1756 free_table_devices(&md->table_devices);
1757 dm_stats_cleanup(&md->stats);
1760 module_put(THIS_MODULE);
1764 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1766 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1769 if (dm_table_bio_based(t)) {
1771 * The md may already have mempools that need changing.
1772 * If so, reload bioset because front_pad may have changed
1773 * because a different table was loaded.
1775 bioset_exit(&md->bs);
1776 bioset_exit(&md->io_bs);
1778 } else if (bioset_initialized(&md->bs)) {
1780 * There's no need to reload with request-based dm
1781 * because the size of front_pad doesn't change.
1782 * Note for future: If you are to reload bioset,
1783 * prep-ed requests in the queue may refer
1784 * to bio from the old bioset, so you must walk
1785 * through the queue to unprep.
1791 bioset_initialized(&md->bs) ||
1792 bioset_initialized(&md->io_bs));
1794 ret = bioset_init_from_src(&md->bs, &p->bs);
1797 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1799 bioset_exit(&md->bs);
1801 /* mempool bind completed, no longer need any mempools in the table */
1802 dm_table_free_md_mempools(t);
1807 * Bind a table to the device.
1809 static void event_callback(void *context)
1811 unsigned long flags;
1813 struct mapped_device *md = (struct mapped_device *) context;
1815 spin_lock_irqsave(&md->uevent_lock, flags);
1816 list_splice_init(&md->uevent_list, &uevents);
1817 spin_unlock_irqrestore(&md->uevent_lock, flags);
1819 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1821 atomic_inc(&md->event_nr);
1822 wake_up(&md->eventq);
1823 dm_issue_global_event();
1827 * Returns old map, which caller must destroy.
1829 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1830 struct queue_limits *limits)
1832 struct dm_table *old_map;
1833 struct request_queue *q = md->queue;
1834 bool request_based = dm_table_request_based(t);
1838 lockdep_assert_held(&md->suspend_lock);
1840 size = dm_table_get_size(t);
1843 * Wipe any geometry if the size of the table changed.
1845 if (size != dm_get_size(md))
1846 memset(&md->geometry, 0, sizeof(md->geometry));
1848 if (!get_capacity(md->disk))
1849 set_capacity(md->disk, size);
1851 set_capacity_and_notify(md->disk, size);
1853 dm_table_event_callback(t, event_callback, md);
1855 if (request_based) {
1857 * Leverage the fact that request-based DM targets are
1858 * immutable singletons - used to optimize dm_mq_queue_rq.
1860 md->immutable_target = dm_table_get_immutable_target(t);
1863 ret = __bind_mempools(md, t);
1865 old_map = ERR_PTR(ret);
1869 ret = dm_table_set_restrictions(t, q, limits);
1871 old_map = ERR_PTR(ret);
1875 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1876 rcu_assign_pointer(md->map, (void *)t);
1877 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1887 * Returns unbound table for the caller to free.
1889 static struct dm_table *__unbind(struct mapped_device *md)
1891 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1896 dm_table_event_callback(map, NULL, NULL);
1897 RCU_INIT_POINTER(md->map, NULL);
1904 * Constructor for a new device.
1906 int dm_create(int minor, struct mapped_device **result)
1908 struct mapped_device *md;
1910 md = alloc_dev(minor);
1914 dm_ima_reset_data(md);
1921 * Functions to manage md->type.
1922 * All are required to hold md->type_lock.
1924 void dm_lock_md_type(struct mapped_device *md)
1926 mutex_lock(&md->type_lock);
1929 void dm_unlock_md_type(struct mapped_device *md)
1931 mutex_unlock(&md->type_lock);
1934 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1936 BUG_ON(!mutex_is_locked(&md->type_lock));
1940 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1945 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1947 return md->immutable_target_type;
1951 * The queue_limits are only valid as long as you have a reference
1954 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1956 BUG_ON(!atomic_read(&md->holders));
1957 return &md->queue->limits;
1959 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1962 * Setup the DM device's queue based on md's type
1964 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1966 enum dm_queue_mode type = dm_table_get_type(t);
1967 struct queue_limits limits;
1971 case DM_TYPE_REQUEST_BASED:
1972 md->disk->fops = &dm_rq_blk_dops;
1973 r = dm_mq_init_request_queue(md, t);
1975 DMERR("Cannot initialize queue for request-based dm mapped device");
1979 case DM_TYPE_BIO_BASED:
1980 case DM_TYPE_DAX_BIO_BASED:
1987 r = dm_calculate_queue_limits(t, &limits);
1989 DMERR("Cannot calculate initial queue limits");
1992 r = dm_table_set_restrictions(t, md->queue, &limits);
1996 r = add_disk(md->disk);
2000 r = dm_sysfs_init(md);
2002 del_gendisk(md->disk);
2009 struct mapped_device *dm_get_md(dev_t dev)
2011 struct mapped_device *md;
2012 unsigned minor = MINOR(dev);
2014 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2017 spin_lock(&_minor_lock);
2019 md = idr_find(&_minor_idr, minor);
2020 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2021 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2027 spin_unlock(&_minor_lock);
2031 EXPORT_SYMBOL_GPL(dm_get_md);
2033 void *dm_get_mdptr(struct mapped_device *md)
2035 return md->interface_ptr;
2038 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2040 md->interface_ptr = ptr;
2043 void dm_get(struct mapped_device *md)
2045 atomic_inc(&md->holders);
2046 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2049 int dm_hold(struct mapped_device *md)
2051 spin_lock(&_minor_lock);
2052 if (test_bit(DMF_FREEING, &md->flags)) {
2053 spin_unlock(&_minor_lock);
2057 spin_unlock(&_minor_lock);
2060 EXPORT_SYMBOL_GPL(dm_hold);
2062 const char *dm_device_name(struct mapped_device *md)
2066 EXPORT_SYMBOL_GPL(dm_device_name);
2068 static void __dm_destroy(struct mapped_device *md, bool wait)
2070 struct dm_table *map;
2075 spin_lock(&_minor_lock);
2076 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2077 set_bit(DMF_FREEING, &md->flags);
2078 spin_unlock(&_minor_lock);
2080 blk_mark_disk_dead(md->disk);
2083 * Take suspend_lock so that presuspend and postsuspend methods
2084 * do not race with internal suspend.
2086 mutex_lock(&md->suspend_lock);
2087 map = dm_get_live_table(md, &srcu_idx);
2088 if (!dm_suspended_md(md)) {
2089 dm_table_presuspend_targets(map);
2090 set_bit(DMF_SUSPENDED, &md->flags);
2091 set_bit(DMF_POST_SUSPENDING, &md->flags);
2092 dm_table_postsuspend_targets(map);
2094 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2095 dm_put_live_table(md, srcu_idx);
2096 mutex_unlock(&md->suspend_lock);
2099 * Rare, but there may be I/O requests still going to complete,
2100 * for example. Wait for all references to disappear.
2101 * No one should increment the reference count of the mapped_device,
2102 * after the mapped_device state becomes DMF_FREEING.
2105 while (atomic_read(&md->holders))
2107 else if (atomic_read(&md->holders))
2108 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2109 dm_device_name(md), atomic_read(&md->holders));
2111 dm_table_destroy(__unbind(md));
2115 void dm_destroy(struct mapped_device *md)
2117 __dm_destroy(md, true);
2120 void dm_destroy_immediate(struct mapped_device *md)
2122 __dm_destroy(md, false);
2125 void dm_put(struct mapped_device *md)
2127 atomic_dec(&md->holders);
2129 EXPORT_SYMBOL_GPL(dm_put);
2131 static bool md_in_flight_bios(struct mapped_device *md)
2134 struct block_device *part = dm_disk(md)->part0;
2137 for_each_possible_cpu(cpu) {
2138 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2139 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2145 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2151 prepare_to_wait(&md->wait, &wait, task_state);
2153 if (!md_in_flight_bios(md))
2156 if (signal_pending_state(task_state, current)) {
2163 finish_wait(&md->wait, &wait);
2168 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2172 if (!queue_is_mq(md->queue))
2173 return dm_wait_for_bios_completion(md, task_state);
2176 if (!blk_mq_queue_inflight(md->queue))
2179 if (signal_pending_state(task_state, current)) {
2191 * Process the deferred bios
2193 static void dm_wq_work(struct work_struct *work)
2195 struct mapped_device *md = container_of(work, struct mapped_device, work);
2198 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2199 spin_lock_irq(&md->deferred_lock);
2200 bio = bio_list_pop(&md->deferred);
2201 spin_unlock_irq(&md->deferred_lock);
2206 submit_bio_noacct(bio);
2210 static void dm_queue_flush(struct mapped_device *md)
2212 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2213 smp_mb__after_atomic();
2214 queue_work(md->wq, &md->work);
2218 * Swap in a new table, returning the old one for the caller to destroy.
2220 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2222 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2223 struct queue_limits limits;
2226 mutex_lock(&md->suspend_lock);
2228 /* device must be suspended */
2229 if (!dm_suspended_md(md))
2233 * If the new table has no data devices, retain the existing limits.
2234 * This helps multipath with queue_if_no_path if all paths disappear,
2235 * then new I/O is queued based on these limits, and then some paths
2238 if (dm_table_has_no_data_devices(table)) {
2239 live_map = dm_get_live_table_fast(md);
2241 limits = md->queue->limits;
2242 dm_put_live_table_fast(md);
2246 r = dm_calculate_queue_limits(table, &limits);
2253 map = __bind(md, table, &limits);
2254 dm_issue_global_event();
2257 mutex_unlock(&md->suspend_lock);
2262 * Functions to lock and unlock any filesystem running on the
2265 static int lock_fs(struct mapped_device *md)
2269 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2271 r = freeze_bdev(md->disk->part0);
2273 set_bit(DMF_FROZEN, &md->flags);
2277 static void unlock_fs(struct mapped_device *md)
2279 if (!test_bit(DMF_FROZEN, &md->flags))
2281 thaw_bdev(md->disk->part0);
2282 clear_bit(DMF_FROZEN, &md->flags);
2286 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2287 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2288 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2290 * If __dm_suspend returns 0, the device is completely quiescent
2291 * now. There is no request-processing activity. All new requests
2292 * are being added to md->deferred list.
2294 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2295 unsigned suspend_flags, unsigned int task_state,
2296 int dmf_suspended_flag)
2298 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2299 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2302 lockdep_assert_held(&md->suspend_lock);
2305 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2306 * This flag is cleared before dm_suspend returns.
2309 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2311 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2314 * This gets reverted if there's an error later and the targets
2315 * provide the .presuspend_undo hook.
2317 dm_table_presuspend_targets(map);
2320 * Flush I/O to the device.
2321 * Any I/O submitted after lock_fs() may not be flushed.
2322 * noflush takes precedence over do_lockfs.
2323 * (lock_fs() flushes I/Os and waits for them to complete.)
2325 if (!noflush && do_lockfs) {
2328 dm_table_presuspend_undo_targets(map);
2334 * Here we must make sure that no processes are submitting requests
2335 * to target drivers i.e. no one may be executing
2336 * __split_and_process_bio from dm_submit_bio.
2338 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2339 * we take the write lock. To prevent any process from reentering
2340 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2341 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2342 * flush_workqueue(md->wq).
2344 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2346 synchronize_srcu(&md->io_barrier);
2349 * Stop md->queue before flushing md->wq in case request-based
2350 * dm defers requests to md->wq from md->queue.
2352 if (dm_request_based(md))
2353 dm_stop_queue(md->queue);
2355 flush_workqueue(md->wq);
2358 * At this point no more requests are entering target request routines.
2359 * We call dm_wait_for_completion to wait for all existing requests
2362 r = dm_wait_for_completion(md, task_state);
2364 set_bit(dmf_suspended_flag, &md->flags);
2367 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2369 synchronize_srcu(&md->io_barrier);
2371 /* were we interrupted ? */
2375 if (dm_request_based(md))
2376 dm_start_queue(md->queue);
2379 dm_table_presuspend_undo_targets(map);
2380 /* pushback list is already flushed, so skip flush */
2387 * We need to be able to change a mapping table under a mounted
2388 * filesystem. For example we might want to move some data in
2389 * the background. Before the table can be swapped with
2390 * dm_bind_table, dm_suspend must be called to flush any in
2391 * flight bios and ensure that any further io gets deferred.
2394 * Suspend mechanism in request-based dm.
2396 * 1. Flush all I/Os by lock_fs() if needed.
2397 * 2. Stop dispatching any I/O by stopping the request_queue.
2398 * 3. Wait for all in-flight I/Os to be completed or requeued.
2400 * To abort suspend, start the request_queue.
2402 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2404 struct dm_table *map = NULL;
2408 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2410 if (dm_suspended_md(md)) {
2415 if (dm_suspended_internally_md(md)) {
2416 /* already internally suspended, wait for internal resume */
2417 mutex_unlock(&md->suspend_lock);
2418 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2424 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2426 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2430 set_bit(DMF_POST_SUSPENDING, &md->flags);
2431 dm_table_postsuspend_targets(map);
2432 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2435 mutex_unlock(&md->suspend_lock);
2439 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2442 int r = dm_table_resume_targets(map);
2450 * Flushing deferred I/Os must be done after targets are resumed
2451 * so that mapping of targets can work correctly.
2452 * Request-based dm is queueing the deferred I/Os in its request_queue.
2454 if (dm_request_based(md))
2455 dm_start_queue(md->queue);
2462 int dm_resume(struct mapped_device *md)
2465 struct dm_table *map = NULL;
2469 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2471 if (!dm_suspended_md(md))
2474 if (dm_suspended_internally_md(md)) {
2475 /* already internally suspended, wait for internal resume */
2476 mutex_unlock(&md->suspend_lock);
2477 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2483 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2484 if (!map || !dm_table_get_size(map))
2487 r = __dm_resume(md, map);
2491 clear_bit(DMF_SUSPENDED, &md->flags);
2493 mutex_unlock(&md->suspend_lock);
2499 * Internal suspend/resume works like userspace-driven suspend. It waits
2500 * until all bios finish and prevents issuing new bios to the target drivers.
2501 * It may be used only from the kernel.
2504 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2506 struct dm_table *map = NULL;
2508 lockdep_assert_held(&md->suspend_lock);
2510 if (md->internal_suspend_count++)
2511 return; /* nested internal suspend */
2513 if (dm_suspended_md(md)) {
2514 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2515 return; /* nest suspend */
2518 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2521 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2522 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2523 * would require changing .presuspend to return an error -- avoid this
2524 * until there is a need for more elaborate variants of internal suspend.
2526 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2527 DMF_SUSPENDED_INTERNALLY);
2529 set_bit(DMF_POST_SUSPENDING, &md->flags);
2530 dm_table_postsuspend_targets(map);
2531 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2534 static void __dm_internal_resume(struct mapped_device *md)
2536 BUG_ON(!md->internal_suspend_count);
2538 if (--md->internal_suspend_count)
2539 return; /* resume from nested internal suspend */
2541 if (dm_suspended_md(md))
2542 goto done; /* resume from nested suspend */
2545 * NOTE: existing callers don't need to call dm_table_resume_targets
2546 * (which may fail -- so best to avoid it for now by passing NULL map)
2548 (void) __dm_resume(md, NULL);
2551 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2552 smp_mb__after_atomic();
2553 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2556 void dm_internal_suspend_noflush(struct mapped_device *md)
2558 mutex_lock(&md->suspend_lock);
2559 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2560 mutex_unlock(&md->suspend_lock);
2562 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2564 void dm_internal_resume(struct mapped_device *md)
2566 mutex_lock(&md->suspend_lock);
2567 __dm_internal_resume(md);
2568 mutex_unlock(&md->suspend_lock);
2570 EXPORT_SYMBOL_GPL(dm_internal_resume);
2573 * Fast variants of internal suspend/resume hold md->suspend_lock,
2574 * which prevents interaction with userspace-driven suspend.
2577 void dm_internal_suspend_fast(struct mapped_device *md)
2579 mutex_lock(&md->suspend_lock);
2580 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2583 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2584 synchronize_srcu(&md->io_barrier);
2585 flush_workqueue(md->wq);
2586 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2588 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2590 void dm_internal_resume_fast(struct mapped_device *md)
2592 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2598 mutex_unlock(&md->suspend_lock);
2600 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2602 /*-----------------------------------------------------------------
2603 * Event notification.
2604 *---------------------------------------------------------------*/
2605 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2610 char udev_cookie[DM_COOKIE_LENGTH];
2611 char *envp[] = { udev_cookie, NULL };
2613 noio_flag = memalloc_noio_save();
2616 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2618 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2619 DM_COOKIE_ENV_VAR_NAME, cookie);
2620 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2624 memalloc_noio_restore(noio_flag);
2629 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2631 return atomic_add_return(1, &md->uevent_seq);
2634 uint32_t dm_get_event_nr(struct mapped_device *md)
2636 return atomic_read(&md->event_nr);
2639 int dm_wait_event(struct mapped_device *md, int event_nr)
2641 return wait_event_interruptible(md->eventq,
2642 (event_nr != atomic_read(&md->event_nr)));
2645 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2647 unsigned long flags;
2649 spin_lock_irqsave(&md->uevent_lock, flags);
2650 list_add(elist, &md->uevent_list);
2651 spin_unlock_irqrestore(&md->uevent_lock, flags);
2655 * The gendisk is only valid as long as you have a reference
2658 struct gendisk *dm_disk(struct mapped_device *md)
2662 EXPORT_SYMBOL_GPL(dm_disk);
2664 struct kobject *dm_kobject(struct mapped_device *md)
2666 return &md->kobj_holder.kobj;
2669 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2671 struct mapped_device *md;
2673 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2675 spin_lock(&_minor_lock);
2676 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2682 spin_unlock(&_minor_lock);
2687 int dm_suspended_md(struct mapped_device *md)
2689 return test_bit(DMF_SUSPENDED, &md->flags);
2692 static int dm_post_suspending_md(struct mapped_device *md)
2694 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2697 int dm_suspended_internally_md(struct mapped_device *md)
2699 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2702 int dm_test_deferred_remove_flag(struct mapped_device *md)
2704 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2707 int dm_suspended(struct dm_target *ti)
2709 return dm_suspended_md(ti->table->md);
2711 EXPORT_SYMBOL_GPL(dm_suspended);
2713 int dm_post_suspending(struct dm_target *ti)
2715 return dm_post_suspending_md(ti->table->md);
2717 EXPORT_SYMBOL_GPL(dm_post_suspending);
2719 int dm_noflush_suspending(struct dm_target *ti)
2721 return __noflush_suspending(ti->table->md);
2723 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2725 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2726 unsigned integrity, unsigned per_io_data_size,
2727 unsigned min_pool_size)
2729 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2730 unsigned int pool_size = 0;
2731 unsigned int front_pad, io_front_pad;
2738 case DM_TYPE_BIO_BASED:
2739 case DM_TYPE_DAX_BIO_BASED:
2740 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2741 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2742 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2743 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2746 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2749 case DM_TYPE_REQUEST_BASED:
2750 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2751 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2752 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2758 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2762 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2768 dm_free_md_mempools(pools);
2773 void dm_free_md_mempools(struct dm_md_mempools *pools)
2778 bioset_exit(&pools->bs);
2779 bioset_exit(&pools->io_bs);
2791 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2794 struct mapped_device *md = bdev->bd_disk->private_data;
2795 struct dm_table *table;
2796 struct dm_target *ti;
2797 int ret = -ENOTTY, srcu_idx;
2799 table = dm_get_live_table(md, &srcu_idx);
2800 if (!table || !dm_table_get_size(table))
2803 /* We only support devices that have a single target */
2804 if (dm_table_get_num_targets(table) != 1)
2806 ti = dm_table_get_target(table, 0);
2809 if (!ti->type->iterate_devices)
2812 ret = ti->type->iterate_devices(ti, fn, data);
2814 dm_put_live_table(md, srcu_idx);
2819 * For register / unregister we need to manually call out to every path.
2821 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2822 sector_t start, sector_t len, void *data)
2824 struct dm_pr *pr = data;
2825 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2827 if (!ops || !ops->pr_register)
2829 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2832 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2843 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2844 if (ret && new_key) {
2845 /* unregister all paths if we failed to register any path */
2846 pr.old_key = new_key;
2849 pr.fail_early = false;
2850 dm_call_pr(bdev, __dm_pr_register, &pr);
2856 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2859 struct mapped_device *md = bdev->bd_disk->private_data;
2860 const struct pr_ops *ops;
2863 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2867 ops = bdev->bd_disk->fops->pr_ops;
2868 if (ops && ops->pr_reserve)
2869 r = ops->pr_reserve(bdev, key, type, flags);
2873 dm_unprepare_ioctl(md, srcu_idx);
2877 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2879 struct mapped_device *md = bdev->bd_disk->private_data;
2880 const struct pr_ops *ops;
2883 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2887 ops = bdev->bd_disk->fops->pr_ops;
2888 if (ops && ops->pr_release)
2889 r = ops->pr_release(bdev, key, type);
2893 dm_unprepare_ioctl(md, srcu_idx);
2897 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2898 enum pr_type type, bool abort)
2900 struct mapped_device *md = bdev->bd_disk->private_data;
2901 const struct pr_ops *ops;
2904 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2908 ops = bdev->bd_disk->fops->pr_ops;
2909 if (ops && ops->pr_preempt)
2910 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2914 dm_unprepare_ioctl(md, srcu_idx);
2918 static int dm_pr_clear(struct block_device *bdev, u64 key)
2920 struct mapped_device *md = bdev->bd_disk->private_data;
2921 const struct pr_ops *ops;
2924 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2928 ops = bdev->bd_disk->fops->pr_ops;
2929 if (ops && ops->pr_clear)
2930 r = ops->pr_clear(bdev, key);
2934 dm_unprepare_ioctl(md, srcu_idx);
2938 static const struct pr_ops dm_pr_ops = {
2939 .pr_register = dm_pr_register,
2940 .pr_reserve = dm_pr_reserve,
2941 .pr_release = dm_pr_release,
2942 .pr_preempt = dm_pr_preempt,
2943 .pr_clear = dm_pr_clear,
2946 static const struct block_device_operations dm_blk_dops = {
2947 .submit_bio = dm_submit_bio,
2948 .open = dm_blk_open,
2949 .release = dm_blk_close,
2950 .ioctl = dm_blk_ioctl,
2951 .getgeo = dm_blk_getgeo,
2952 .report_zones = dm_blk_report_zones,
2953 .pr_ops = &dm_pr_ops,
2954 .owner = THIS_MODULE
2957 static const struct block_device_operations dm_rq_blk_dops = {
2958 .open = dm_blk_open,
2959 .release = dm_blk_close,
2960 .ioctl = dm_blk_ioctl,
2961 .getgeo = dm_blk_getgeo,
2962 .pr_ops = &dm_pr_ops,
2963 .owner = THIS_MODULE
2966 static const struct dax_operations dm_dax_ops = {
2967 .direct_access = dm_dax_direct_access,
2968 .zero_page_range = dm_dax_zero_page_range,
2974 module_init(dm_init);
2975 module_exit(dm_exit);
2977 module_param(major, uint, 0);
2978 MODULE_PARM_DESC(major, "The major number of the device mapper");
2980 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2981 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2983 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2984 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2986 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
2987 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
2989 MODULE_DESCRIPTION(DM_NAME " driver");
2990 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2991 MODULE_LICENSE("GPL");