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"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/mm.h>
16 #include <linux/sched/signal.h>
17 #include <linux/blkpg.h>
18 #include <linux/bio.h>
19 #include <linux/mempool.h>
20 #include <linux/dax.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/uio.h>
24 #include <linux/hdreg.h>
25 #include <linux/delay.h>
26 #include <linux/wait.h>
28 #include <linux/refcount.h>
29 #include <linux/part_stat.h>
30 #include <linux/blk-crypto.h>
32 #define DM_MSG_PREFIX "core"
35 * Cookies are numeric values sent with CHANGE and REMOVE
36 * uevents while resuming, removing or renaming the device.
38 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
39 #define DM_COOKIE_LENGTH 24
41 static const char *_name = DM_NAME;
43 static unsigned int major = 0;
44 static unsigned int _major = 0;
46 static DEFINE_IDR(_minor_idr);
48 static DEFINE_SPINLOCK(_minor_lock);
50 static void do_deferred_remove(struct work_struct *w);
52 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
54 static struct workqueue_struct *deferred_remove_workqueue;
56 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
57 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
59 void dm_issue_global_event(void)
61 atomic_inc(&dm_global_event_nr);
62 wake_up(&dm_global_eventq);
66 * One of these is allocated (on-stack) per original bio.
73 unsigned sector_count;
77 * One of these is allocated per clone bio.
79 #define DM_TIO_MAGIC 7282014
84 unsigned target_bio_nr;
91 * One of these is allocated per original bio.
92 * It contains the first clone used for that original.
94 #define DM_IO_MAGIC 5191977
97 struct mapped_device *md;
100 struct bio *orig_bio;
101 unsigned long start_time;
102 spinlock_t endio_lock;
103 struct dm_stats_aux stats_aux;
104 /* last member of dm_target_io is 'struct bio' */
105 struct dm_target_io tio;
108 void *dm_per_bio_data(struct bio *bio, size_t data_size)
110 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
111 if (!tio->inside_dm_io)
112 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
113 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
115 EXPORT_SYMBOL_GPL(dm_per_bio_data);
117 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
119 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
120 if (io->magic == DM_IO_MAGIC)
121 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
122 BUG_ON(io->magic != DM_TIO_MAGIC);
123 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
125 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
127 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
129 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
131 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
133 #define MINOR_ALLOCED ((void *)-1)
136 * Bits for the md->flags field.
138 #define DMF_BLOCK_IO_FOR_SUSPEND 0
139 #define DMF_SUSPENDED 1
141 #define DMF_FREEING 3
142 #define DMF_DELETING 4
143 #define DMF_NOFLUSH_SUSPENDING 5
144 #define DMF_DEFERRED_REMOVE 6
145 #define DMF_SUSPENDED_INTERNALLY 7
146 #define DMF_POST_SUSPENDING 8
148 #define DM_NUMA_NODE NUMA_NO_NODE
149 static int dm_numa_node = DM_NUMA_NODE;
152 * For mempools pre-allocation at the table loading time.
154 struct dm_md_mempools {
156 struct bio_set io_bs;
159 struct table_device {
160 struct list_head list;
162 struct dm_dev dm_dev;
166 * Bio-based DM's mempools' reserved IOs set by the user.
168 #define RESERVED_BIO_BASED_IOS 16
169 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
171 static int __dm_get_module_param_int(int *module_param, int min, int max)
173 int param = READ_ONCE(*module_param);
174 int modified_param = 0;
175 bool modified = true;
178 modified_param = min;
179 else if (param > max)
180 modified_param = max;
185 (void)cmpxchg(module_param, param, modified_param);
186 param = modified_param;
192 unsigned __dm_get_module_param(unsigned *module_param,
193 unsigned def, unsigned max)
195 unsigned param = READ_ONCE(*module_param);
196 unsigned modified_param = 0;
199 modified_param = def;
200 else if (param > max)
201 modified_param = max;
203 if (modified_param) {
204 (void)cmpxchg(module_param, param, modified_param);
205 param = modified_param;
211 unsigned dm_get_reserved_bio_based_ios(void)
213 return __dm_get_module_param(&reserved_bio_based_ios,
214 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
216 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
218 static unsigned dm_get_numa_node(void)
220 return __dm_get_module_param_int(&dm_numa_node,
221 DM_NUMA_NODE, num_online_nodes() - 1);
224 static int __init local_init(void)
228 r = dm_uevent_init();
232 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
233 if (!deferred_remove_workqueue) {
235 goto out_uevent_exit;
239 r = register_blkdev(_major, _name);
241 goto out_free_workqueue;
249 destroy_workqueue(deferred_remove_workqueue);
256 static void local_exit(void)
258 flush_scheduled_work();
259 destroy_workqueue(deferred_remove_workqueue);
261 unregister_blkdev(_major, _name);
266 DMINFO("cleaned up");
269 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
291 static int __init dm_init(void)
293 const int count = ARRAY_SIZE(_inits);
297 for (i = 0; i < count; i++) {
312 static void __exit dm_exit(void)
314 int i = ARRAY_SIZE(_exits);
320 * Should be empty by this point.
322 idr_destroy(&_minor_idr);
326 * Block device functions
328 int dm_deleting_md(struct mapped_device *md)
330 return test_bit(DMF_DELETING, &md->flags);
333 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
335 struct mapped_device *md;
337 spin_lock(&_minor_lock);
339 md = bdev->bd_disk->private_data;
343 if (test_bit(DMF_FREEING, &md->flags) ||
344 dm_deleting_md(md)) {
350 atomic_inc(&md->open_count);
352 spin_unlock(&_minor_lock);
354 return md ? 0 : -ENXIO;
357 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
359 struct mapped_device *md;
361 spin_lock(&_minor_lock);
363 md = disk->private_data;
367 if (atomic_dec_and_test(&md->open_count) &&
368 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
369 queue_work(deferred_remove_workqueue, &deferred_remove_work);
373 spin_unlock(&_minor_lock);
376 int dm_open_count(struct mapped_device *md)
378 return atomic_read(&md->open_count);
382 * Guarantees nothing is using the device before it's deleted.
384 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
388 spin_lock(&_minor_lock);
390 if (dm_open_count(md)) {
393 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
394 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
397 set_bit(DMF_DELETING, &md->flags);
399 spin_unlock(&_minor_lock);
404 int dm_cancel_deferred_remove(struct mapped_device *md)
408 spin_lock(&_minor_lock);
410 if (test_bit(DMF_DELETING, &md->flags))
413 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
415 spin_unlock(&_minor_lock);
420 static void do_deferred_remove(struct work_struct *w)
422 dm_deferred_remove();
425 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
427 struct mapped_device *md = bdev->bd_disk->private_data;
429 return dm_get_geometry(md, geo);
432 #ifdef CONFIG_BLK_DEV_ZONED
433 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
435 struct dm_report_zones_args *args = data;
436 sector_t sector_diff = args->tgt->begin - args->start;
439 * Ignore zones beyond the target range.
441 if (zone->start >= args->start + args->tgt->len)
445 * Remap the start sector and write pointer position of the zone
446 * to match its position in the target range.
448 zone->start += sector_diff;
449 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
450 if (zone->cond == BLK_ZONE_COND_FULL)
451 zone->wp = zone->start + zone->len;
452 else if (zone->cond == BLK_ZONE_COND_EMPTY)
453 zone->wp = zone->start;
455 zone->wp += sector_diff;
458 args->next_sector = zone->start + zone->len;
459 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
461 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
463 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
464 unsigned int nr_zones, report_zones_cb cb, void *data)
466 struct mapped_device *md = disk->private_data;
467 struct dm_table *map;
469 struct dm_report_zones_args args = {
470 .next_sector = sector,
475 if (dm_suspended_md(md))
478 map = dm_get_live_table(md, &srcu_idx);
483 struct dm_target *tgt;
485 tgt = dm_table_find_target(map, args.next_sector);
486 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
492 ret = tgt->type->report_zones(tgt, &args,
493 nr_zones - args.zone_idx);
496 } while (args.zone_idx < nr_zones &&
497 args.next_sector < get_capacity(disk));
501 dm_put_live_table(md, srcu_idx);
505 #define dm_blk_report_zones NULL
506 #endif /* CONFIG_BLK_DEV_ZONED */
508 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
509 struct block_device **bdev)
510 __acquires(md->io_barrier)
512 struct dm_target *tgt;
513 struct dm_table *map;
518 map = dm_get_live_table(md, srcu_idx);
519 if (!map || !dm_table_get_size(map))
522 /* We only support devices that have a single target */
523 if (dm_table_get_num_targets(map) != 1)
526 tgt = dm_table_get_target(map, 0);
527 if (!tgt->type->prepare_ioctl)
530 if (dm_suspended_md(md))
533 r = tgt->type->prepare_ioctl(tgt, bdev);
534 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
535 dm_put_live_table(md, *srcu_idx);
543 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
544 __releases(md->io_barrier)
546 dm_put_live_table(md, srcu_idx);
549 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
550 unsigned int cmd, unsigned long arg)
552 struct mapped_device *md = bdev->bd_disk->private_data;
555 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
561 * Target determined this ioctl is being issued against a
562 * subset of the parent bdev; require extra privileges.
564 if (!capable(CAP_SYS_RAWIO)) {
566 "%s: sending ioctl %x to DM device without required privilege.",
573 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
575 dm_unprepare_ioctl(md, srcu_idx);
579 u64 dm_start_time_ns_from_clone(struct bio *bio)
581 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
582 struct dm_io *io = tio->io;
584 return jiffies_to_nsecs(io->start_time);
586 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
588 static void start_io_acct(struct dm_io *io)
590 struct mapped_device *md = io->md;
591 struct bio *bio = io->orig_bio;
593 io->start_time = bio_start_io_acct(bio);
594 if (unlikely(dm_stats_used(&md->stats)))
595 dm_stats_account_io(&md->stats, bio_data_dir(bio),
596 bio->bi_iter.bi_sector, bio_sectors(bio),
597 false, 0, &io->stats_aux);
600 static void end_io_acct(struct dm_io *io)
602 struct mapped_device *md = io->md;
603 struct bio *bio = io->orig_bio;
604 unsigned long duration = jiffies - io->start_time;
606 bio_end_io_acct(bio, io->start_time);
608 if (unlikely(dm_stats_used(&md->stats)))
609 dm_stats_account_io(&md->stats, bio_data_dir(bio),
610 bio->bi_iter.bi_sector, bio_sectors(bio),
611 true, duration, &io->stats_aux);
613 /* nudge anyone waiting on suspend queue */
614 if (unlikely(wq_has_sleeper(&md->wait)))
618 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
621 struct dm_target_io *tio;
624 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
628 tio = container_of(clone, struct dm_target_io, clone);
629 tio->inside_dm_io = true;
632 io = container_of(tio, struct dm_io, tio);
633 io->magic = DM_IO_MAGIC;
635 atomic_set(&io->io_count, 1);
638 spin_lock_init(&io->endio_lock);
645 static void free_io(struct mapped_device *md, struct dm_io *io)
647 bio_put(&io->tio.clone);
650 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
651 unsigned target_bio_nr, gfp_t gfp_mask)
653 struct dm_target_io *tio;
655 if (!ci->io->tio.io) {
656 /* the dm_target_io embedded in ci->io is available */
659 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
663 tio = container_of(clone, struct dm_target_io, clone);
664 tio->inside_dm_io = false;
667 tio->magic = DM_TIO_MAGIC;
670 tio->target_bio_nr = target_bio_nr;
675 static void free_tio(struct dm_target_io *tio)
677 if (tio->inside_dm_io)
679 bio_put(&tio->clone);
683 * Add the bio to the list of deferred io.
685 static void queue_io(struct mapped_device *md, struct bio *bio)
689 spin_lock_irqsave(&md->deferred_lock, flags);
690 bio_list_add(&md->deferred, bio);
691 spin_unlock_irqrestore(&md->deferred_lock, flags);
692 queue_work(md->wq, &md->work);
696 * Everyone (including functions in this file), should use this
697 * function to access the md->map field, and make sure they call
698 * dm_put_live_table() when finished.
700 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
702 *srcu_idx = srcu_read_lock(&md->io_barrier);
704 return srcu_dereference(md->map, &md->io_barrier);
707 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
709 srcu_read_unlock(&md->io_barrier, srcu_idx);
712 void dm_sync_table(struct mapped_device *md)
714 synchronize_srcu(&md->io_barrier);
715 synchronize_rcu_expedited();
719 * A fast alternative to dm_get_live_table/dm_put_live_table.
720 * The caller must not block between these two functions.
722 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
725 return rcu_dereference(md->map);
728 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
733 static char *_dm_claim_ptr = "I belong to device-mapper";
736 * Open a table device so we can use it as a map destination.
738 static int open_table_device(struct table_device *td, dev_t dev,
739 struct mapped_device *md)
741 struct block_device *bdev;
745 BUG_ON(td->dm_dev.bdev);
747 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
749 return PTR_ERR(bdev);
751 r = bd_link_disk_holder(bdev, dm_disk(md));
753 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
757 td->dm_dev.bdev = bdev;
758 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
763 * Close a table device that we've been using.
765 static void close_table_device(struct table_device *td, struct mapped_device *md)
767 if (!td->dm_dev.bdev)
770 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
771 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
772 put_dax(td->dm_dev.dax_dev);
773 td->dm_dev.bdev = NULL;
774 td->dm_dev.dax_dev = NULL;
777 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
780 struct table_device *td;
782 list_for_each_entry(td, l, list)
783 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
789 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
790 struct dm_dev **result)
793 struct table_device *td;
795 mutex_lock(&md->table_devices_lock);
796 td = find_table_device(&md->table_devices, dev, mode);
798 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
800 mutex_unlock(&md->table_devices_lock);
804 td->dm_dev.mode = mode;
805 td->dm_dev.bdev = NULL;
807 if ((r = open_table_device(td, dev, md))) {
808 mutex_unlock(&md->table_devices_lock);
813 format_dev_t(td->dm_dev.name, dev);
815 refcount_set(&td->count, 1);
816 list_add(&td->list, &md->table_devices);
818 refcount_inc(&td->count);
820 mutex_unlock(&md->table_devices_lock);
822 *result = &td->dm_dev;
825 EXPORT_SYMBOL_GPL(dm_get_table_device);
827 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
829 struct table_device *td = container_of(d, struct table_device, dm_dev);
831 mutex_lock(&md->table_devices_lock);
832 if (refcount_dec_and_test(&td->count)) {
833 close_table_device(td, md);
837 mutex_unlock(&md->table_devices_lock);
839 EXPORT_SYMBOL(dm_put_table_device);
841 static void free_table_devices(struct list_head *devices)
843 struct list_head *tmp, *next;
845 list_for_each_safe(tmp, next, devices) {
846 struct table_device *td = list_entry(tmp, struct table_device, list);
848 DMWARN("dm_destroy: %s still exists with %d references",
849 td->dm_dev.name, refcount_read(&td->count));
855 * Get the geometry associated with a dm device
857 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
865 * Set the geometry of a device.
867 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
869 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
871 if (geo->start > sz) {
872 DMWARN("Start sector is beyond the geometry limits.");
881 static int __noflush_suspending(struct mapped_device *md)
883 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
887 * Decrements the number of outstanding ios that a bio has been
888 * cloned into, completing the original io if necc.
890 static void dec_pending(struct dm_io *io, blk_status_t error)
893 blk_status_t io_error;
895 struct mapped_device *md = io->md;
897 /* Push-back supersedes any I/O errors */
898 if (unlikely(error)) {
899 spin_lock_irqsave(&io->endio_lock, flags);
900 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
902 spin_unlock_irqrestore(&io->endio_lock, flags);
905 if (atomic_dec_and_test(&io->io_count)) {
906 if (io->status == BLK_STS_DM_REQUEUE) {
908 * Target requested pushing back the I/O.
910 spin_lock_irqsave(&md->deferred_lock, flags);
911 if (__noflush_suspending(md))
912 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
913 bio_list_add_head(&md->deferred, io->orig_bio);
915 /* noflush suspend was interrupted. */
916 io->status = BLK_STS_IOERR;
917 spin_unlock_irqrestore(&md->deferred_lock, flags);
920 io_error = io->status;
925 if (io_error == BLK_STS_DM_REQUEUE)
928 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
930 * Preflush done for flush with data, reissue
931 * without REQ_PREFLUSH.
933 bio->bi_opf &= ~REQ_PREFLUSH;
936 /* done with normal IO or empty flush */
938 bio->bi_status = io_error;
944 void disable_discard(struct mapped_device *md)
946 struct queue_limits *limits = dm_get_queue_limits(md);
948 /* device doesn't really support DISCARD, disable it */
949 limits->max_discard_sectors = 0;
950 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
953 void disable_write_same(struct mapped_device *md)
955 struct queue_limits *limits = dm_get_queue_limits(md);
957 /* device doesn't really support WRITE SAME, disable it */
958 limits->max_write_same_sectors = 0;
961 void disable_write_zeroes(struct mapped_device *md)
963 struct queue_limits *limits = dm_get_queue_limits(md);
965 /* device doesn't really support WRITE ZEROES, disable it */
966 limits->max_write_zeroes_sectors = 0;
969 static void clone_endio(struct bio *bio)
971 blk_status_t error = bio->bi_status;
972 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
973 struct dm_io *io = tio->io;
974 struct mapped_device *md = tio->io->md;
975 dm_endio_fn endio = tio->ti->type->end_io;
976 struct bio *orig_bio = io->orig_bio;
978 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
979 if (bio_op(bio) == REQ_OP_DISCARD &&
980 !bio->bi_disk->queue->limits.max_discard_sectors)
982 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
983 !bio->bi_disk->queue->limits.max_write_same_sectors)
984 disable_write_same(md);
985 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
986 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
987 disable_write_zeroes(md);
991 * For zone-append bios get offset in zone of the written
992 * sector and add that to the original bio sector pos.
994 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
995 sector_t written_sector = bio->bi_iter.bi_sector;
996 struct request_queue *q = orig_bio->bi_disk->queue;
997 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
999 orig_bio->bi_iter.bi_sector += written_sector & mask;
1003 int r = endio(tio->ti, bio, &error);
1005 case DM_ENDIO_REQUEUE:
1006 error = BLK_STS_DM_REQUEUE;
1010 case DM_ENDIO_INCOMPLETE:
1011 /* The target will handle the io */
1014 DMWARN("unimplemented target endio return value: %d", r);
1020 dec_pending(io, error);
1024 * Return maximum size of I/O possible at the supplied sector up to the current
1027 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1028 sector_t target_offset)
1030 return ti->len - target_offset;
1033 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1035 sector_t target_offset = dm_target_offset(ti, sector);
1036 sector_t len = max_io_len_target_boundary(ti, target_offset);
1040 * Does the target need to split even further?
1041 * - q->limits.chunk_sectors reflects ti->max_io_len so
1042 * blk_max_size_offset() provides required splitting.
1043 * - blk_max_size_offset() also respects q->limits.max_sectors
1045 max_len = blk_max_size_offset(ti->table->md->queue,
1053 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1055 if (len > UINT_MAX) {
1056 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1057 (unsigned long long)len, UINT_MAX);
1058 ti->error = "Maximum size of target IO is too large";
1062 ti->max_io_len = (uint32_t) len;
1066 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1068 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1069 sector_t sector, int *srcu_idx)
1070 __acquires(md->io_barrier)
1072 struct dm_table *map;
1073 struct dm_target *ti;
1075 map = dm_get_live_table(md, srcu_idx);
1079 ti = dm_table_find_target(map, sector);
1086 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1087 long nr_pages, void **kaddr, pfn_t *pfn)
1089 struct mapped_device *md = dax_get_private(dax_dev);
1090 sector_t sector = pgoff * PAGE_SECTORS;
1091 struct dm_target *ti;
1092 long len, ret = -EIO;
1095 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1099 if (!ti->type->direct_access)
1101 len = max_io_len(ti, sector) / PAGE_SECTORS;
1104 nr_pages = min(len, nr_pages);
1105 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1108 dm_put_live_table(md, srcu_idx);
1113 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1114 int blocksize, sector_t start, sector_t len)
1116 struct mapped_device *md = dax_get_private(dax_dev);
1117 struct dm_table *map;
1121 map = dm_get_live_table(md, &srcu_idx);
1125 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1128 dm_put_live_table(md, srcu_idx);
1133 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1134 void *addr, size_t bytes, struct iov_iter *i)
1136 struct mapped_device *md = dax_get_private(dax_dev);
1137 sector_t sector = pgoff * PAGE_SECTORS;
1138 struct dm_target *ti;
1142 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1146 if (!ti->type->dax_copy_from_iter) {
1147 ret = copy_from_iter(addr, bytes, i);
1150 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1152 dm_put_live_table(md, srcu_idx);
1157 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1158 void *addr, size_t bytes, struct iov_iter *i)
1160 struct mapped_device *md = dax_get_private(dax_dev);
1161 sector_t sector = pgoff * PAGE_SECTORS;
1162 struct dm_target *ti;
1166 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1170 if (!ti->type->dax_copy_to_iter) {
1171 ret = copy_to_iter(addr, bytes, i);
1174 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1176 dm_put_live_table(md, srcu_idx);
1181 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1184 struct mapped_device *md = dax_get_private(dax_dev);
1185 sector_t sector = pgoff * PAGE_SECTORS;
1186 struct dm_target *ti;
1190 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1194 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1196 * ->zero_page_range() is mandatory dax operation. If we are
1197 * here, something is wrong.
1199 dm_put_live_table(md, srcu_idx);
1202 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1205 dm_put_live_table(md, srcu_idx);
1211 * A target may call dm_accept_partial_bio only from the map routine. It is
1212 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1213 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1215 * dm_accept_partial_bio informs the dm that the target only wants to process
1216 * additional n_sectors sectors of the bio and the rest of the data should be
1217 * sent in a next bio.
1219 * A diagram that explains the arithmetics:
1220 * +--------------------+---------------+-------+
1222 * +--------------------+---------------+-------+
1224 * <-------------- *tio->len_ptr --------------->
1225 * <------- bi_size ------->
1228 * Region 1 was already iterated over with bio_advance or similar function.
1229 * (it may be empty if the target doesn't use bio_advance)
1230 * Region 2 is the remaining bio size that the target wants to process.
1231 * (it may be empty if region 1 is non-empty, although there is no reason
1233 * The target requires that region 3 is to be sent in the next bio.
1235 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1236 * the partially processed part (the sum of regions 1+2) must be the same for all
1237 * copies of the bio.
1239 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1241 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1242 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1243 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1244 BUG_ON(bi_size > *tio->len_ptr);
1245 BUG_ON(n_sectors > bi_size);
1246 *tio->len_ptr -= bi_size - n_sectors;
1247 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1249 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1251 static blk_qc_t __map_bio(struct dm_target_io *tio)
1255 struct bio *clone = &tio->clone;
1256 struct dm_io *io = tio->io;
1257 struct dm_target *ti = tio->ti;
1258 blk_qc_t ret = BLK_QC_T_NONE;
1260 clone->bi_end_io = clone_endio;
1263 * Map the clone. If r == 0 we don't need to do
1264 * anything, the target has assumed ownership of
1267 atomic_inc(&io->io_count);
1268 sector = clone->bi_iter.bi_sector;
1270 r = ti->type->map(ti, clone);
1272 case DM_MAPIO_SUBMITTED:
1274 case DM_MAPIO_REMAPPED:
1275 /* the bio has been remapped so dispatch it */
1276 trace_block_bio_remap(clone->bi_disk->queue, clone,
1277 bio_dev(io->orig_bio), sector);
1278 ret = submit_bio_noacct(clone);
1282 dec_pending(io, BLK_STS_IOERR);
1284 case DM_MAPIO_REQUEUE:
1286 dec_pending(io, BLK_STS_DM_REQUEUE);
1289 DMWARN("unimplemented target map return value: %d", r);
1296 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1298 bio->bi_iter.bi_sector = sector;
1299 bio->bi_iter.bi_size = to_bytes(len);
1303 * Creates a bio that consists of range of complete bvecs.
1305 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1306 sector_t sector, unsigned len)
1308 struct bio *clone = &tio->clone;
1310 __bio_clone_fast(clone, bio);
1312 bio_crypt_clone(clone, bio, GFP_NOIO);
1314 if (bio_integrity(bio)) {
1317 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1318 !dm_target_passes_integrity(tio->ti->type))) {
1319 DMWARN("%s: the target %s doesn't support integrity data.",
1320 dm_device_name(tio->io->md),
1321 tio->ti->type->name);
1325 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1330 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1331 clone->bi_iter.bi_size = to_bytes(len);
1333 if (bio_integrity(bio))
1334 bio_integrity_trim(clone);
1339 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1340 struct dm_target *ti, unsigned num_bios)
1342 struct dm_target_io *tio;
1348 if (num_bios == 1) {
1349 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1350 bio_list_add(blist, &tio->clone);
1354 for (try = 0; try < 2; try++) {
1359 mutex_lock(&ci->io->md->table_devices_lock);
1360 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1361 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1365 bio_list_add(blist, &tio->clone);
1368 mutex_unlock(&ci->io->md->table_devices_lock);
1369 if (bio_nr == num_bios)
1372 while ((bio = bio_list_pop(blist))) {
1373 tio = container_of(bio, struct dm_target_io, clone);
1379 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1380 struct dm_target_io *tio, unsigned *len)
1382 struct bio *clone = &tio->clone;
1386 __bio_clone_fast(clone, ci->bio);
1388 bio_setup_sector(clone, ci->sector, *len);
1390 return __map_bio(tio);
1393 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1394 unsigned num_bios, unsigned *len)
1396 struct bio_list blist = BIO_EMPTY_LIST;
1398 struct dm_target_io *tio;
1400 alloc_multiple_bios(&blist, ci, ti, num_bios);
1402 while ((bio = bio_list_pop(&blist))) {
1403 tio = container_of(bio, struct dm_target_io, clone);
1404 (void) __clone_and_map_simple_bio(ci, tio, len);
1408 static int __send_empty_flush(struct clone_info *ci)
1410 unsigned target_nr = 0;
1411 struct dm_target *ti;
1412 struct bio flush_bio;
1415 * Use an on-stack bio for this, it's safe since we don't
1416 * need to reference it after submit. It's just used as
1417 * the basis for the clone(s).
1419 bio_init(&flush_bio, NULL, 0);
1420 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1421 ci->bio = &flush_bio;
1422 ci->sector_count = 0;
1425 * Empty flush uses a statically initialized bio, as the base for
1426 * cloning. However, blkg association requires that a bdev is
1427 * associated with a gendisk, which doesn't happen until the bdev is
1428 * opened. So, blkg association is done at issue time of the flush
1429 * rather than when the device is created in alloc_dev().
1431 bio_set_dev(ci->bio, ci->io->md->bdev);
1433 BUG_ON(bio_has_data(ci->bio));
1434 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1435 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1437 bio_uninit(ci->bio);
1441 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1442 sector_t sector, unsigned *len)
1444 struct bio *bio = ci->bio;
1445 struct dm_target_io *tio;
1448 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1450 r = clone_bio(tio, bio, sector, *len);
1455 (void) __map_bio(tio);
1460 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1466 * Even though the device advertised support for this type of
1467 * request, that does not mean every target supports it, and
1468 * reconfiguration might also have changed that since the
1469 * check was performed.
1474 len = min_t(sector_t, ci->sector_count,
1475 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1477 __send_duplicate_bios(ci, ti, num_bios, &len);
1480 ci->sector_count -= len;
1485 static bool is_abnormal_io(struct bio *bio)
1489 switch (bio_op(bio)) {
1490 case REQ_OP_DISCARD:
1491 case REQ_OP_SECURE_ERASE:
1492 case REQ_OP_WRITE_SAME:
1493 case REQ_OP_WRITE_ZEROES:
1501 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1504 struct bio *bio = ci->bio;
1505 unsigned num_bios = 0;
1507 switch (bio_op(bio)) {
1508 case REQ_OP_DISCARD:
1509 num_bios = ti->num_discard_bios;
1511 case REQ_OP_SECURE_ERASE:
1512 num_bios = ti->num_secure_erase_bios;
1514 case REQ_OP_WRITE_SAME:
1515 num_bios = ti->num_write_same_bios;
1517 case REQ_OP_WRITE_ZEROES:
1518 num_bios = ti->num_write_zeroes_bios;
1524 *result = __send_changing_extent_only(ci, ti, num_bios);
1529 * Select the correct strategy for processing a non-flush bio.
1531 static int __split_and_process_non_flush(struct clone_info *ci)
1533 struct dm_target *ti;
1537 ti = dm_table_find_target(ci->map, ci->sector);
1541 if (__process_abnormal_io(ci, ti, &r))
1544 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1546 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1551 ci->sector_count -= len;
1556 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1557 struct dm_table *map, struct bio *bio)
1560 ci->io = alloc_io(md, bio);
1561 ci->sector = bio->bi_iter.bi_sector;
1564 #define __dm_part_stat_sub(part, field, subnd) \
1565 (part_stat_get(part, field) -= (subnd))
1568 * Entry point to split a bio into clones and submit them to the targets.
1570 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1571 struct dm_table *map, struct bio *bio)
1573 struct clone_info ci;
1574 blk_qc_t ret = BLK_QC_T_NONE;
1577 init_clone_info(&ci, md, map, bio);
1579 if (bio->bi_opf & REQ_PREFLUSH) {
1580 error = __send_empty_flush(&ci);
1581 /* dec_pending submits any data associated with flush */
1582 } else if (op_is_zone_mgmt(bio_op(bio))) {
1584 ci.sector_count = 0;
1585 error = __split_and_process_non_flush(&ci);
1588 ci.sector_count = bio_sectors(bio);
1589 while (ci.sector_count && !error) {
1590 error = __split_and_process_non_flush(&ci);
1591 if (current->bio_list && ci.sector_count && !error) {
1593 * Remainder must be passed to submit_bio_noacct()
1594 * so that it gets handled *after* bios already submitted
1595 * have been completely processed.
1596 * We take a clone of the original to store in
1597 * ci.io->orig_bio to be used by end_io_acct() and
1598 * for dec_pending to use for completion handling.
1600 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1601 GFP_NOIO, &md->queue->bio_split);
1602 ci.io->orig_bio = b;
1605 * Adjust IO stats for each split, otherwise upon queue
1606 * reentry there will be redundant IO accounting.
1607 * NOTE: this is a stop-gap fix, a proper fix involves
1608 * significant refactoring of DM core's bio splitting
1609 * (by eliminating DM's splitting and just using bio_split)
1612 __dm_part_stat_sub(&dm_disk(md)->part0,
1613 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1617 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1618 ret = submit_bio_noacct(bio);
1624 /* drop the extra reference count */
1625 dec_pending(ci.io, errno_to_blk_status(error));
1630 * Optimized variant of __split_and_process_bio that leverages the
1631 * fact that targets that use it do _not_ have a need to split bios.
1633 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1636 struct clone_info ci;
1637 blk_qc_t ret = BLK_QC_T_NONE;
1640 init_clone_info(&ci, md, map, bio);
1642 if (bio->bi_opf & REQ_PREFLUSH) {
1643 error = __send_empty_flush(&ci);
1644 /* dec_pending submits any data associated with flush */
1646 struct dm_target_io *tio;
1647 struct dm_target *ti = md->immutable_target;
1649 if (WARN_ON_ONCE(!ti)) {
1655 ci.sector_count = bio_sectors(bio);
1656 if (__process_abnormal_io(&ci, ti, &error))
1659 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1660 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1663 /* drop the extra reference count */
1664 dec_pending(ci.io, errno_to_blk_status(error));
1668 static blk_qc_t dm_process_bio(struct mapped_device *md,
1669 struct dm_table *map, struct bio *bio)
1671 blk_qc_t ret = BLK_QC_T_NONE;
1673 if (unlikely(!map)) {
1679 * If in ->submit_bio we need to use blk_queue_split(), otherwise
1680 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1682 * If called from dm_wq_work() for deferred bio processing, bio
1683 * was already handled by following code with previous ->submit_bio.
1685 if (current->bio_list) {
1686 if (is_abnormal_io(bio))
1687 blk_queue_split(&bio);
1688 /* regular IO is split by __split_and_process_bio */
1691 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1692 return __process_bio(md, map, bio);
1693 return __split_and_process_bio(md, map, bio);
1696 static blk_qc_t dm_submit_bio(struct bio *bio)
1698 struct mapped_device *md = bio->bi_disk->private_data;
1699 blk_qc_t ret = BLK_QC_T_NONE;
1701 struct dm_table *map;
1703 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
1705 * We are called with a live reference on q_usage_counter, but
1706 * that one will be released as soon as we return. Grab an
1707 * extra one as blk_mq_submit_bio expects to be able to consume
1708 * a reference (which lives until the request is freed in case a
1709 * request is allocated).
1711 percpu_ref_get(&bio->bi_disk->queue->q_usage_counter);
1712 return blk_mq_submit_bio(bio);
1715 map = dm_get_live_table(md, &srcu_idx);
1717 /* if we're suspended, we have to queue this io for later */
1718 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1719 dm_put_live_table(md, srcu_idx);
1721 if (bio->bi_opf & REQ_NOWAIT)
1722 bio_wouldblock_error(bio);
1723 else if (!(bio->bi_opf & REQ_RAHEAD))
1730 ret = dm_process_bio(md, map, bio);
1732 dm_put_live_table(md, srcu_idx);
1736 /*-----------------------------------------------------------------
1737 * An IDR is used to keep track of allocated minor numbers.
1738 *---------------------------------------------------------------*/
1739 static void free_minor(int minor)
1741 spin_lock(&_minor_lock);
1742 idr_remove(&_minor_idr, minor);
1743 spin_unlock(&_minor_lock);
1747 * See if the device with a specific minor # is free.
1749 static int specific_minor(int minor)
1753 if (minor >= (1 << MINORBITS))
1756 idr_preload(GFP_KERNEL);
1757 spin_lock(&_minor_lock);
1759 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1761 spin_unlock(&_minor_lock);
1764 return r == -ENOSPC ? -EBUSY : r;
1768 static int next_free_minor(int *minor)
1772 idr_preload(GFP_KERNEL);
1773 spin_lock(&_minor_lock);
1775 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1777 spin_unlock(&_minor_lock);
1785 static const struct block_device_operations dm_blk_dops;
1786 static const struct dax_operations dm_dax_ops;
1788 static void dm_wq_work(struct work_struct *work);
1790 static void cleanup_mapped_device(struct mapped_device *md)
1793 destroy_workqueue(md->wq);
1794 bioset_exit(&md->bs);
1795 bioset_exit(&md->io_bs);
1798 kill_dax(md->dax_dev);
1799 put_dax(md->dax_dev);
1804 spin_lock(&_minor_lock);
1805 md->disk->private_data = NULL;
1806 spin_unlock(&_minor_lock);
1807 del_gendisk(md->disk);
1812 blk_cleanup_queue(md->queue);
1814 cleanup_srcu_struct(&md->io_barrier);
1821 mutex_destroy(&md->suspend_lock);
1822 mutex_destroy(&md->type_lock);
1823 mutex_destroy(&md->table_devices_lock);
1825 dm_mq_cleanup_mapped_device(md);
1829 * Allocate and initialise a blank device with a given minor.
1831 static struct mapped_device *alloc_dev(int minor)
1833 int r, numa_node_id = dm_get_numa_node();
1834 struct mapped_device *md;
1837 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1839 DMWARN("unable to allocate device, out of memory.");
1843 if (!try_module_get(THIS_MODULE))
1844 goto bad_module_get;
1846 /* get a minor number for the dev */
1847 if (minor == DM_ANY_MINOR)
1848 r = next_free_minor(&minor);
1850 r = specific_minor(minor);
1854 r = init_srcu_struct(&md->io_barrier);
1856 goto bad_io_barrier;
1858 md->numa_node_id = numa_node_id;
1859 md->init_tio_pdu = false;
1860 md->type = DM_TYPE_NONE;
1861 mutex_init(&md->suspend_lock);
1862 mutex_init(&md->type_lock);
1863 mutex_init(&md->table_devices_lock);
1864 spin_lock_init(&md->deferred_lock);
1865 atomic_set(&md->holders, 1);
1866 atomic_set(&md->open_count, 0);
1867 atomic_set(&md->event_nr, 0);
1868 atomic_set(&md->uevent_seq, 0);
1869 INIT_LIST_HEAD(&md->uevent_list);
1870 INIT_LIST_HEAD(&md->table_devices);
1871 spin_lock_init(&md->uevent_lock);
1874 * default to bio-based until DM table is loaded and md->type
1875 * established. If request-based table is loaded: blk-mq will
1876 * override accordingly.
1878 md->queue = blk_alloc_queue(numa_node_id);
1882 md->disk = alloc_disk_node(1, md->numa_node_id);
1886 init_waitqueue_head(&md->wait);
1887 INIT_WORK(&md->work, dm_wq_work);
1888 init_waitqueue_head(&md->eventq);
1889 init_completion(&md->kobj_holder.completion);
1891 md->disk->major = _major;
1892 md->disk->first_minor = minor;
1893 md->disk->fops = &dm_blk_dops;
1894 md->disk->queue = md->queue;
1895 md->disk->private_data = md;
1896 sprintf(md->disk->disk_name, "dm-%d", minor);
1898 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1899 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1901 if (IS_ERR(md->dax_dev))
1905 add_disk_no_queue_reg(md->disk);
1906 format_dev_t(md->name, MKDEV(_major, minor));
1908 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1912 md->bdev = bdget_disk(md->disk, 0);
1916 dm_stats_init(&md->stats);
1918 /* Populate the mapping, nobody knows we exist yet */
1919 spin_lock(&_minor_lock);
1920 old_md = idr_replace(&_minor_idr, md, minor);
1921 spin_unlock(&_minor_lock);
1923 BUG_ON(old_md != MINOR_ALLOCED);
1928 cleanup_mapped_device(md);
1932 module_put(THIS_MODULE);
1938 static void unlock_fs(struct mapped_device *md);
1940 static void free_dev(struct mapped_device *md)
1942 int minor = MINOR(disk_devt(md->disk));
1946 cleanup_mapped_device(md);
1948 free_table_devices(&md->table_devices);
1949 dm_stats_cleanup(&md->stats);
1952 module_put(THIS_MODULE);
1956 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1958 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1961 if (dm_table_bio_based(t)) {
1963 * The md may already have mempools that need changing.
1964 * If so, reload bioset because front_pad may have changed
1965 * because a different table was loaded.
1967 bioset_exit(&md->bs);
1968 bioset_exit(&md->io_bs);
1970 } else if (bioset_initialized(&md->bs)) {
1972 * There's no need to reload with request-based dm
1973 * because the size of front_pad doesn't change.
1974 * Note for future: If you are to reload bioset,
1975 * prep-ed requests in the queue may refer
1976 * to bio from the old bioset, so you must walk
1977 * through the queue to unprep.
1983 bioset_initialized(&md->bs) ||
1984 bioset_initialized(&md->io_bs));
1986 ret = bioset_init_from_src(&md->bs, &p->bs);
1989 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1991 bioset_exit(&md->bs);
1993 /* mempool bind completed, no longer need any mempools in the table */
1994 dm_table_free_md_mempools(t);
1999 * Bind a table to the device.
2001 static void event_callback(void *context)
2003 unsigned long flags;
2005 struct mapped_device *md = (struct mapped_device *) context;
2007 spin_lock_irqsave(&md->uevent_lock, flags);
2008 list_splice_init(&md->uevent_list, &uevents);
2009 spin_unlock_irqrestore(&md->uevent_lock, flags);
2011 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2013 atomic_inc(&md->event_nr);
2014 wake_up(&md->eventq);
2015 dm_issue_global_event();
2019 * Returns old map, which caller must destroy.
2021 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2022 struct queue_limits *limits)
2024 struct dm_table *old_map;
2025 struct request_queue *q = md->queue;
2026 bool request_based = dm_table_request_based(t);
2030 lockdep_assert_held(&md->suspend_lock);
2032 size = dm_table_get_size(t);
2035 * Wipe any geometry if the size of the table changed.
2037 if (size != dm_get_size(md))
2038 memset(&md->geometry, 0, sizeof(md->geometry));
2040 set_capacity(md->disk, size);
2041 bd_set_nr_sectors(md->bdev, size);
2043 dm_table_event_callback(t, event_callback, md);
2046 * The queue hasn't been stopped yet, if the old table type wasn't
2047 * for request-based during suspension. So stop it to prevent
2048 * I/O mapping before resume.
2049 * This must be done before setting the queue restrictions,
2050 * because request-based dm may be run just after the setting.
2055 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2057 * Leverage the fact that request-based DM targets and
2058 * NVMe bio based targets are immutable singletons
2059 * - used to optimize both __process_bio and dm_mq_queue_rq
2061 md->immutable_target = dm_table_get_immutable_target(t);
2064 ret = __bind_mempools(md, t);
2066 old_map = ERR_PTR(ret);
2070 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2071 rcu_assign_pointer(md->map, (void *)t);
2072 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2074 dm_table_set_restrictions(t, q, limits);
2083 * Returns unbound table for the caller to free.
2085 static struct dm_table *__unbind(struct mapped_device *md)
2087 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2092 dm_table_event_callback(map, NULL, NULL);
2093 RCU_INIT_POINTER(md->map, NULL);
2100 * Constructor for a new device.
2102 int dm_create(int minor, struct mapped_device **result)
2105 struct mapped_device *md;
2107 md = alloc_dev(minor);
2111 r = dm_sysfs_init(md);
2122 * Functions to manage md->type.
2123 * All are required to hold md->type_lock.
2125 void dm_lock_md_type(struct mapped_device *md)
2127 mutex_lock(&md->type_lock);
2130 void dm_unlock_md_type(struct mapped_device *md)
2132 mutex_unlock(&md->type_lock);
2135 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2137 BUG_ON(!mutex_is_locked(&md->type_lock));
2141 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2146 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2148 return md->immutable_target_type;
2152 * The queue_limits are only valid as long as you have a reference
2155 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2157 BUG_ON(!atomic_read(&md->holders));
2158 return &md->queue->limits;
2160 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2163 * Setup the DM device's queue based on md's type
2165 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2168 struct queue_limits limits;
2169 enum dm_queue_mode type = dm_get_md_type(md);
2172 case DM_TYPE_REQUEST_BASED:
2173 r = dm_mq_init_request_queue(md, t);
2175 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2179 case DM_TYPE_BIO_BASED:
2180 case DM_TYPE_DAX_BIO_BASED:
2181 case DM_TYPE_NVME_BIO_BASED:
2188 r = dm_calculate_queue_limits(t, &limits);
2190 DMERR("Cannot calculate initial queue limits");
2193 dm_table_set_restrictions(t, md->queue, &limits);
2194 blk_register_queue(md->disk);
2199 struct mapped_device *dm_get_md(dev_t dev)
2201 struct mapped_device *md;
2202 unsigned minor = MINOR(dev);
2204 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2207 spin_lock(&_minor_lock);
2209 md = idr_find(&_minor_idr, minor);
2210 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2211 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2217 spin_unlock(&_minor_lock);
2221 EXPORT_SYMBOL_GPL(dm_get_md);
2223 void *dm_get_mdptr(struct mapped_device *md)
2225 return md->interface_ptr;
2228 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2230 md->interface_ptr = ptr;
2233 void dm_get(struct mapped_device *md)
2235 atomic_inc(&md->holders);
2236 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2239 int dm_hold(struct mapped_device *md)
2241 spin_lock(&_minor_lock);
2242 if (test_bit(DMF_FREEING, &md->flags)) {
2243 spin_unlock(&_minor_lock);
2247 spin_unlock(&_minor_lock);
2250 EXPORT_SYMBOL_GPL(dm_hold);
2252 const char *dm_device_name(struct mapped_device *md)
2256 EXPORT_SYMBOL_GPL(dm_device_name);
2258 static void __dm_destroy(struct mapped_device *md, bool wait)
2260 struct dm_table *map;
2265 spin_lock(&_minor_lock);
2266 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2267 set_bit(DMF_FREEING, &md->flags);
2268 spin_unlock(&_minor_lock);
2270 blk_set_queue_dying(md->queue);
2273 * Take suspend_lock so that presuspend and postsuspend methods
2274 * do not race with internal suspend.
2276 mutex_lock(&md->suspend_lock);
2277 map = dm_get_live_table(md, &srcu_idx);
2278 if (!dm_suspended_md(md)) {
2279 dm_table_presuspend_targets(map);
2280 set_bit(DMF_SUSPENDED, &md->flags);
2281 set_bit(DMF_POST_SUSPENDING, &md->flags);
2282 dm_table_postsuspend_targets(map);
2284 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2285 dm_put_live_table(md, srcu_idx);
2286 mutex_unlock(&md->suspend_lock);
2289 * Rare, but there may be I/O requests still going to complete,
2290 * for example. Wait for all references to disappear.
2291 * No one should increment the reference count of the mapped_device,
2292 * after the mapped_device state becomes DMF_FREEING.
2295 while (atomic_read(&md->holders))
2297 else if (atomic_read(&md->holders))
2298 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2299 dm_device_name(md), atomic_read(&md->holders));
2302 dm_table_destroy(__unbind(md));
2306 void dm_destroy(struct mapped_device *md)
2308 __dm_destroy(md, true);
2311 void dm_destroy_immediate(struct mapped_device *md)
2313 __dm_destroy(md, false);
2316 void dm_put(struct mapped_device *md)
2318 atomic_dec(&md->holders);
2320 EXPORT_SYMBOL_GPL(dm_put);
2322 static bool md_in_flight_bios(struct mapped_device *md)
2325 struct hd_struct *part = &dm_disk(md)->part0;
2328 for_each_possible_cpu(cpu) {
2329 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2330 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2336 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2342 prepare_to_wait(&md->wait, &wait, task_state);
2344 if (!md_in_flight_bios(md))
2347 if (signal_pending_state(task_state, current)) {
2354 finish_wait(&md->wait, &wait);
2359 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2363 if (!queue_is_mq(md->queue))
2364 return dm_wait_for_bios_completion(md, task_state);
2367 if (!blk_mq_queue_inflight(md->queue))
2370 if (signal_pending_state(task_state, current)) {
2382 * Process the deferred bios
2384 static void dm_wq_work(struct work_struct *work)
2386 struct mapped_device *md = container_of(work, struct mapped_device,
2390 struct dm_table *map;
2392 map = dm_get_live_table(md, &srcu_idx);
2394 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2395 spin_lock_irq(&md->deferred_lock);
2396 c = bio_list_pop(&md->deferred);
2397 spin_unlock_irq(&md->deferred_lock);
2402 if (dm_request_based(md))
2403 (void) submit_bio_noacct(c);
2405 (void) dm_process_bio(md, map, c);
2408 dm_put_live_table(md, srcu_idx);
2411 static void dm_queue_flush(struct mapped_device *md)
2413 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2414 smp_mb__after_atomic();
2415 queue_work(md->wq, &md->work);
2419 * Swap in a new table, returning the old one for the caller to destroy.
2421 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2423 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2424 struct queue_limits limits;
2427 mutex_lock(&md->suspend_lock);
2429 /* device must be suspended */
2430 if (!dm_suspended_md(md))
2434 * If the new table has no data devices, retain the existing limits.
2435 * This helps multipath with queue_if_no_path if all paths disappear,
2436 * then new I/O is queued based on these limits, and then some paths
2439 if (dm_table_has_no_data_devices(table)) {
2440 live_map = dm_get_live_table_fast(md);
2442 limits = md->queue->limits;
2443 dm_put_live_table_fast(md);
2447 r = dm_calculate_queue_limits(table, &limits);
2454 map = __bind(md, table, &limits);
2455 dm_issue_global_event();
2458 mutex_unlock(&md->suspend_lock);
2463 * Functions to lock and unlock any filesystem running on the
2466 static int lock_fs(struct mapped_device *md)
2470 WARN_ON(md->frozen_sb);
2472 md->frozen_sb = freeze_bdev(md->bdev);
2473 if (IS_ERR(md->frozen_sb)) {
2474 r = PTR_ERR(md->frozen_sb);
2475 md->frozen_sb = NULL;
2479 set_bit(DMF_FROZEN, &md->flags);
2484 static void unlock_fs(struct mapped_device *md)
2486 if (!test_bit(DMF_FROZEN, &md->flags))
2489 thaw_bdev(md->bdev, md->frozen_sb);
2490 md->frozen_sb = NULL;
2491 clear_bit(DMF_FROZEN, &md->flags);
2495 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2496 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2497 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2499 * If __dm_suspend returns 0, the device is completely quiescent
2500 * now. There is no request-processing activity. All new requests
2501 * are being added to md->deferred list.
2503 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2504 unsigned suspend_flags, long task_state,
2505 int dmf_suspended_flag)
2507 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2508 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2511 lockdep_assert_held(&md->suspend_lock);
2514 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2515 * This flag is cleared before dm_suspend returns.
2518 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2520 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2523 * This gets reverted if there's an error later and the targets
2524 * provide the .presuspend_undo hook.
2526 dm_table_presuspend_targets(map);
2529 * Flush I/O to the device.
2530 * Any I/O submitted after lock_fs() may not be flushed.
2531 * noflush takes precedence over do_lockfs.
2532 * (lock_fs() flushes I/Os and waits for them to complete.)
2534 if (!noflush && do_lockfs) {
2537 dm_table_presuspend_undo_targets(map);
2543 * Here we must make sure that no processes are submitting requests
2544 * to target drivers i.e. no one may be executing
2545 * __split_and_process_bio. This is called from dm_request and
2548 * To get all processes out of __split_and_process_bio in dm_request,
2549 * we take the write lock. To prevent any process from reentering
2550 * __split_and_process_bio from dm_request and quiesce the thread
2551 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2552 * flush_workqueue(md->wq).
2554 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2556 synchronize_srcu(&md->io_barrier);
2559 * Stop md->queue before flushing md->wq in case request-based
2560 * dm defers requests to md->wq from md->queue.
2562 if (dm_request_based(md))
2563 dm_stop_queue(md->queue);
2565 flush_workqueue(md->wq);
2568 * At this point no more requests are entering target request routines.
2569 * We call dm_wait_for_completion to wait for all existing requests
2572 r = dm_wait_for_completion(md, task_state);
2574 set_bit(dmf_suspended_flag, &md->flags);
2577 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2579 synchronize_srcu(&md->io_barrier);
2581 /* were we interrupted ? */
2585 if (dm_request_based(md))
2586 dm_start_queue(md->queue);
2589 dm_table_presuspend_undo_targets(map);
2590 /* pushback list is already flushed, so skip flush */
2597 * We need to be able to change a mapping table under a mounted
2598 * filesystem. For example we might want to move some data in
2599 * the background. Before the table can be swapped with
2600 * dm_bind_table, dm_suspend must be called to flush any in
2601 * flight bios and ensure that any further io gets deferred.
2604 * Suspend mechanism in request-based dm.
2606 * 1. Flush all I/Os by lock_fs() if needed.
2607 * 2. Stop dispatching any I/O by stopping the request_queue.
2608 * 3. Wait for all in-flight I/Os to be completed or requeued.
2610 * To abort suspend, start the request_queue.
2612 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2614 struct dm_table *map = NULL;
2618 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2620 if (dm_suspended_md(md)) {
2625 if (dm_suspended_internally_md(md)) {
2626 /* already internally suspended, wait for internal resume */
2627 mutex_unlock(&md->suspend_lock);
2628 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2634 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2636 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2640 set_bit(DMF_POST_SUSPENDING, &md->flags);
2641 dm_table_postsuspend_targets(map);
2642 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2645 mutex_unlock(&md->suspend_lock);
2649 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2652 int r = dm_table_resume_targets(map);
2660 * Flushing deferred I/Os must be done after targets are resumed
2661 * so that mapping of targets can work correctly.
2662 * Request-based dm is queueing the deferred I/Os in its request_queue.
2664 if (dm_request_based(md))
2665 dm_start_queue(md->queue);
2672 int dm_resume(struct mapped_device *md)
2675 struct dm_table *map = NULL;
2679 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2681 if (!dm_suspended_md(md))
2684 if (dm_suspended_internally_md(md)) {
2685 /* already internally suspended, wait for internal resume */
2686 mutex_unlock(&md->suspend_lock);
2687 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2693 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2694 if (!map || !dm_table_get_size(map))
2697 r = __dm_resume(md, map);
2701 clear_bit(DMF_SUSPENDED, &md->flags);
2703 mutex_unlock(&md->suspend_lock);
2709 * Internal suspend/resume works like userspace-driven suspend. It waits
2710 * until all bios finish and prevents issuing new bios to the target drivers.
2711 * It may be used only from the kernel.
2714 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2716 struct dm_table *map = NULL;
2718 lockdep_assert_held(&md->suspend_lock);
2720 if (md->internal_suspend_count++)
2721 return; /* nested internal suspend */
2723 if (dm_suspended_md(md)) {
2724 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2725 return; /* nest suspend */
2728 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2731 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2732 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2733 * would require changing .presuspend to return an error -- avoid this
2734 * until there is a need for more elaborate variants of internal suspend.
2736 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2737 DMF_SUSPENDED_INTERNALLY);
2739 set_bit(DMF_POST_SUSPENDING, &md->flags);
2740 dm_table_postsuspend_targets(map);
2741 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2744 static void __dm_internal_resume(struct mapped_device *md)
2746 BUG_ON(!md->internal_suspend_count);
2748 if (--md->internal_suspend_count)
2749 return; /* resume from nested internal suspend */
2751 if (dm_suspended_md(md))
2752 goto done; /* resume from nested suspend */
2755 * NOTE: existing callers don't need to call dm_table_resume_targets
2756 * (which may fail -- so best to avoid it for now by passing NULL map)
2758 (void) __dm_resume(md, NULL);
2761 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2762 smp_mb__after_atomic();
2763 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2766 void dm_internal_suspend_noflush(struct mapped_device *md)
2768 mutex_lock(&md->suspend_lock);
2769 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2770 mutex_unlock(&md->suspend_lock);
2772 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2774 void dm_internal_resume(struct mapped_device *md)
2776 mutex_lock(&md->suspend_lock);
2777 __dm_internal_resume(md);
2778 mutex_unlock(&md->suspend_lock);
2780 EXPORT_SYMBOL_GPL(dm_internal_resume);
2783 * Fast variants of internal suspend/resume hold md->suspend_lock,
2784 * which prevents interaction with userspace-driven suspend.
2787 void dm_internal_suspend_fast(struct mapped_device *md)
2789 mutex_lock(&md->suspend_lock);
2790 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2793 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2794 synchronize_srcu(&md->io_barrier);
2795 flush_workqueue(md->wq);
2796 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2798 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2800 void dm_internal_resume_fast(struct mapped_device *md)
2802 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2808 mutex_unlock(&md->suspend_lock);
2810 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2812 /*-----------------------------------------------------------------
2813 * Event notification.
2814 *---------------------------------------------------------------*/
2815 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2820 char udev_cookie[DM_COOKIE_LENGTH];
2821 char *envp[] = { udev_cookie, NULL };
2823 noio_flag = memalloc_noio_save();
2826 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2828 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2829 DM_COOKIE_ENV_VAR_NAME, cookie);
2830 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2834 memalloc_noio_restore(noio_flag);
2839 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2841 return atomic_add_return(1, &md->uevent_seq);
2844 uint32_t dm_get_event_nr(struct mapped_device *md)
2846 return atomic_read(&md->event_nr);
2849 int dm_wait_event(struct mapped_device *md, int event_nr)
2851 return wait_event_interruptible(md->eventq,
2852 (event_nr != atomic_read(&md->event_nr)));
2855 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2857 unsigned long flags;
2859 spin_lock_irqsave(&md->uevent_lock, flags);
2860 list_add(elist, &md->uevent_list);
2861 spin_unlock_irqrestore(&md->uevent_lock, flags);
2865 * The gendisk is only valid as long as you have a reference
2868 struct gendisk *dm_disk(struct mapped_device *md)
2872 EXPORT_SYMBOL_GPL(dm_disk);
2874 struct kobject *dm_kobject(struct mapped_device *md)
2876 return &md->kobj_holder.kobj;
2879 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2881 struct mapped_device *md;
2883 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2885 spin_lock(&_minor_lock);
2886 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2892 spin_unlock(&_minor_lock);
2897 int dm_suspended_md(struct mapped_device *md)
2899 return test_bit(DMF_SUSPENDED, &md->flags);
2902 static int dm_post_suspending_md(struct mapped_device *md)
2904 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2907 int dm_suspended_internally_md(struct mapped_device *md)
2909 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2912 int dm_test_deferred_remove_flag(struct mapped_device *md)
2914 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2917 int dm_suspended(struct dm_target *ti)
2919 return dm_suspended_md(ti->table->md);
2921 EXPORT_SYMBOL_GPL(dm_suspended);
2923 int dm_post_suspending(struct dm_target *ti)
2925 return dm_post_suspending_md(ti->table->md);
2927 EXPORT_SYMBOL_GPL(dm_post_suspending);
2929 int dm_noflush_suspending(struct dm_target *ti)
2931 return __noflush_suspending(ti->table->md);
2933 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2935 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2936 unsigned integrity, unsigned per_io_data_size,
2937 unsigned min_pool_size)
2939 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2940 unsigned int pool_size = 0;
2941 unsigned int front_pad, io_front_pad;
2948 case DM_TYPE_BIO_BASED:
2949 case DM_TYPE_DAX_BIO_BASED:
2950 case DM_TYPE_NVME_BIO_BASED:
2951 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2952 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2953 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2954 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2957 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2960 case DM_TYPE_REQUEST_BASED:
2961 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2962 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2963 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2969 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2973 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2979 dm_free_md_mempools(pools);
2984 void dm_free_md_mempools(struct dm_md_mempools *pools)
2989 bioset_exit(&pools->bs);
2990 bioset_exit(&pools->io_bs);
3002 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3005 struct mapped_device *md = bdev->bd_disk->private_data;
3006 struct dm_table *table;
3007 struct dm_target *ti;
3008 int ret = -ENOTTY, srcu_idx;
3010 table = dm_get_live_table(md, &srcu_idx);
3011 if (!table || !dm_table_get_size(table))
3014 /* We only support devices that have a single target */
3015 if (dm_table_get_num_targets(table) != 1)
3017 ti = dm_table_get_target(table, 0);
3020 if (!ti->type->iterate_devices)
3023 ret = ti->type->iterate_devices(ti, fn, data);
3025 dm_put_live_table(md, srcu_idx);
3030 * For register / unregister we need to manually call out to every path.
3032 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3033 sector_t start, sector_t len, void *data)
3035 struct dm_pr *pr = data;
3036 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3038 if (!ops || !ops->pr_register)
3040 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3043 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3054 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3055 if (ret && new_key) {
3056 /* unregister all paths if we failed to register any path */
3057 pr.old_key = new_key;
3060 pr.fail_early = false;
3061 dm_call_pr(bdev, __dm_pr_register, &pr);
3067 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3070 struct mapped_device *md = bdev->bd_disk->private_data;
3071 const struct pr_ops *ops;
3074 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3078 ops = bdev->bd_disk->fops->pr_ops;
3079 if (ops && ops->pr_reserve)
3080 r = ops->pr_reserve(bdev, key, type, flags);
3084 dm_unprepare_ioctl(md, srcu_idx);
3088 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3090 struct mapped_device *md = bdev->bd_disk->private_data;
3091 const struct pr_ops *ops;
3094 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3098 ops = bdev->bd_disk->fops->pr_ops;
3099 if (ops && ops->pr_release)
3100 r = ops->pr_release(bdev, key, type);
3104 dm_unprepare_ioctl(md, srcu_idx);
3108 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3109 enum pr_type type, bool abort)
3111 struct mapped_device *md = bdev->bd_disk->private_data;
3112 const struct pr_ops *ops;
3115 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3119 ops = bdev->bd_disk->fops->pr_ops;
3120 if (ops && ops->pr_preempt)
3121 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3125 dm_unprepare_ioctl(md, srcu_idx);
3129 static int dm_pr_clear(struct block_device *bdev, u64 key)
3131 struct mapped_device *md = bdev->bd_disk->private_data;
3132 const struct pr_ops *ops;
3135 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3139 ops = bdev->bd_disk->fops->pr_ops;
3140 if (ops && ops->pr_clear)
3141 r = ops->pr_clear(bdev, key);
3145 dm_unprepare_ioctl(md, srcu_idx);
3149 static const struct pr_ops dm_pr_ops = {
3150 .pr_register = dm_pr_register,
3151 .pr_reserve = dm_pr_reserve,
3152 .pr_release = dm_pr_release,
3153 .pr_preempt = dm_pr_preempt,
3154 .pr_clear = dm_pr_clear,
3157 static const struct block_device_operations dm_blk_dops = {
3158 .submit_bio = dm_submit_bio,
3159 .open = dm_blk_open,
3160 .release = dm_blk_close,
3161 .ioctl = dm_blk_ioctl,
3162 .getgeo = dm_blk_getgeo,
3163 .report_zones = dm_blk_report_zones,
3164 .pr_ops = &dm_pr_ops,
3165 .owner = THIS_MODULE
3168 static const struct dax_operations dm_dax_ops = {
3169 .direct_access = dm_dax_direct_access,
3170 .dax_supported = dm_dax_supported,
3171 .copy_from_iter = dm_dax_copy_from_iter,
3172 .copy_to_iter = dm_dax_copy_to_iter,
3173 .zero_page_range = dm_dax_zero_page_range,
3179 module_init(dm_init);
3180 module_exit(dm_exit);
3182 module_param(major, uint, 0);
3183 MODULE_PARM_DESC(major, "The major number of the device mapper");
3185 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3186 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3188 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3189 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3191 MODULE_DESCRIPTION(DM_NAME " driver");
3192 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3193 MODULE_LICENSE("GPL");