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);
485 struct dm_target *tgt;
487 tgt = dm_table_find_target(map, args.next_sector);
488 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
494 ret = tgt->type->report_zones(tgt, &args,
495 nr_zones - args.zone_idx);
498 } while (args.zone_idx < nr_zones &&
499 args.next_sector < get_capacity(disk));
503 dm_put_live_table(md, srcu_idx);
507 #define dm_blk_report_zones NULL
508 #endif /* CONFIG_BLK_DEV_ZONED */
510 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
511 struct block_device **bdev)
513 struct dm_target *tgt;
514 struct dm_table *map;
519 map = dm_get_live_table(md, srcu_idx);
520 if (!map || !dm_table_get_size(map))
523 /* We only support devices that have a single target */
524 if (dm_table_get_num_targets(map) != 1)
527 tgt = dm_table_get_target(map, 0);
528 if (!tgt->type->prepare_ioctl)
531 if (dm_suspended_md(md))
534 r = tgt->type->prepare_ioctl(tgt, bdev);
535 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
536 dm_put_live_table(md, *srcu_idx);
544 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
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 if (!bdev->bd_disk->fops->ioctl)
576 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
578 dm_unprepare_ioctl(md, srcu_idx);
582 u64 dm_start_time_ns_from_clone(struct bio *bio)
584 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
585 struct dm_io *io = tio->io;
587 return jiffies_to_nsecs(io->start_time);
589 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
591 static void start_io_acct(struct dm_io *io)
593 struct mapped_device *md = io->md;
594 struct bio *bio = io->orig_bio;
596 io->start_time = bio_start_io_acct(bio);
597 if (unlikely(dm_stats_used(&md->stats)))
598 dm_stats_account_io(&md->stats, bio_data_dir(bio),
599 bio->bi_iter.bi_sector, bio_sectors(bio),
600 false, 0, &io->stats_aux);
603 static void end_io_acct(struct dm_io *io)
605 struct mapped_device *md = io->md;
606 struct bio *bio = io->orig_bio;
607 unsigned long duration = jiffies - io->start_time;
609 bio_end_io_acct(bio, io->start_time);
611 if (unlikely(dm_stats_used(&md->stats)))
612 dm_stats_account_io(&md->stats, bio_data_dir(bio),
613 bio->bi_iter.bi_sector, bio_sectors(bio),
614 true, duration, &io->stats_aux);
616 /* nudge anyone waiting on suspend queue */
617 if (unlikely(wq_has_sleeper(&md->wait)))
621 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
624 struct dm_target_io *tio;
627 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
631 tio = container_of(clone, struct dm_target_io, clone);
632 tio->inside_dm_io = true;
635 io = container_of(tio, struct dm_io, tio);
636 io->magic = DM_IO_MAGIC;
638 atomic_set(&io->io_count, 1);
641 spin_lock_init(&io->endio_lock);
648 static void free_io(struct mapped_device *md, struct dm_io *io)
650 bio_put(&io->tio.clone);
653 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
654 unsigned target_bio_nr, gfp_t gfp_mask)
656 struct dm_target_io *tio;
658 if (!ci->io->tio.io) {
659 /* the dm_target_io embedded in ci->io is available */
662 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
666 tio = container_of(clone, struct dm_target_io, clone);
667 tio->inside_dm_io = false;
670 tio->magic = DM_TIO_MAGIC;
673 tio->target_bio_nr = target_bio_nr;
678 static void free_tio(struct dm_target_io *tio)
680 if (tio->inside_dm_io)
682 bio_put(&tio->clone);
686 * Add the bio to the list of deferred io.
688 static void queue_io(struct mapped_device *md, struct bio *bio)
692 spin_lock_irqsave(&md->deferred_lock, flags);
693 bio_list_add(&md->deferred, bio);
694 spin_unlock_irqrestore(&md->deferred_lock, flags);
695 queue_work(md->wq, &md->work);
699 * Everyone (including functions in this file), should use this
700 * function to access the md->map field, and make sure they call
701 * dm_put_live_table() when finished.
703 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
705 *srcu_idx = srcu_read_lock(&md->io_barrier);
707 return srcu_dereference(md->map, &md->io_barrier);
710 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
712 srcu_read_unlock(&md->io_barrier, srcu_idx);
715 void dm_sync_table(struct mapped_device *md)
717 synchronize_srcu(&md->io_barrier);
718 synchronize_rcu_expedited();
722 * A fast alternative to dm_get_live_table/dm_put_live_table.
723 * The caller must not block between these two functions.
725 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
728 return rcu_dereference(md->map);
731 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
736 static char *_dm_claim_ptr = "I belong to device-mapper";
739 * Open a table device so we can use it as a map destination.
741 static int open_table_device(struct table_device *td, dev_t dev,
742 struct mapped_device *md)
744 struct block_device *bdev;
748 BUG_ON(td->dm_dev.bdev);
750 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
752 return PTR_ERR(bdev);
754 r = bd_link_disk_holder(bdev, dm_disk(md));
756 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
760 td->dm_dev.bdev = bdev;
761 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
766 * Close a table device that we've been using.
768 static void close_table_device(struct table_device *td, struct mapped_device *md)
770 if (!td->dm_dev.bdev)
773 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
774 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
775 put_dax(td->dm_dev.dax_dev);
776 td->dm_dev.bdev = NULL;
777 td->dm_dev.dax_dev = NULL;
780 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
783 struct table_device *td;
785 list_for_each_entry(td, l, list)
786 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
792 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
793 struct dm_dev **result)
796 struct table_device *td;
798 mutex_lock(&md->table_devices_lock);
799 td = find_table_device(&md->table_devices, dev, mode);
801 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
803 mutex_unlock(&md->table_devices_lock);
807 td->dm_dev.mode = mode;
808 td->dm_dev.bdev = NULL;
810 if ((r = open_table_device(td, dev, md))) {
811 mutex_unlock(&md->table_devices_lock);
816 format_dev_t(td->dm_dev.name, dev);
818 refcount_set(&td->count, 1);
819 list_add(&td->list, &md->table_devices);
821 refcount_inc(&td->count);
823 mutex_unlock(&md->table_devices_lock);
825 *result = &td->dm_dev;
828 EXPORT_SYMBOL_GPL(dm_get_table_device);
830 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
832 struct table_device *td = container_of(d, struct table_device, dm_dev);
834 mutex_lock(&md->table_devices_lock);
835 if (refcount_dec_and_test(&td->count)) {
836 close_table_device(td, md);
840 mutex_unlock(&md->table_devices_lock);
842 EXPORT_SYMBOL(dm_put_table_device);
844 static void free_table_devices(struct list_head *devices)
846 struct list_head *tmp, *next;
848 list_for_each_safe(tmp, next, devices) {
849 struct table_device *td = list_entry(tmp, struct table_device, list);
851 DMWARN("dm_destroy: %s still exists with %d references",
852 td->dm_dev.name, refcount_read(&td->count));
858 * Get the geometry associated with a dm device
860 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
868 * Set the geometry of a device.
870 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
872 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
874 if (geo->start > sz) {
875 DMWARN("Start sector is beyond the geometry limits.");
884 static int __noflush_suspending(struct mapped_device *md)
886 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
890 * Decrements the number of outstanding ios that a bio has been
891 * cloned into, completing the original io if necc.
893 static void dec_pending(struct dm_io *io, blk_status_t error)
896 blk_status_t io_error;
898 struct mapped_device *md = io->md;
900 /* Push-back supersedes any I/O errors */
901 if (unlikely(error)) {
902 spin_lock_irqsave(&io->endio_lock, flags);
903 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
905 spin_unlock_irqrestore(&io->endio_lock, flags);
908 if (atomic_dec_and_test(&io->io_count)) {
909 if (io->status == BLK_STS_DM_REQUEUE) {
911 * Target requested pushing back the I/O.
913 spin_lock_irqsave(&md->deferred_lock, flags);
914 if (__noflush_suspending(md))
915 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
916 bio_list_add_head(&md->deferred, io->orig_bio);
918 /* noflush suspend was interrupted. */
919 io->status = BLK_STS_IOERR;
920 spin_unlock_irqrestore(&md->deferred_lock, flags);
923 io_error = io->status;
928 if (io_error == BLK_STS_DM_REQUEUE)
931 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
933 * Preflush done for flush with data, reissue
934 * without REQ_PREFLUSH.
936 bio->bi_opf &= ~REQ_PREFLUSH;
939 /* done with normal IO or empty flush */
941 bio->bi_status = io_error;
947 void disable_discard(struct mapped_device *md)
949 struct queue_limits *limits = dm_get_queue_limits(md);
951 /* device doesn't really support DISCARD, disable it */
952 limits->max_discard_sectors = 0;
953 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
956 void disable_write_same(struct mapped_device *md)
958 struct queue_limits *limits = dm_get_queue_limits(md);
960 /* device doesn't really support WRITE SAME, disable it */
961 limits->max_write_same_sectors = 0;
964 void disable_write_zeroes(struct mapped_device *md)
966 struct queue_limits *limits = dm_get_queue_limits(md);
968 /* device doesn't really support WRITE ZEROES, disable it */
969 limits->max_write_zeroes_sectors = 0;
972 static void clone_endio(struct bio *bio)
974 blk_status_t error = bio->bi_status;
975 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
976 struct dm_io *io = tio->io;
977 struct mapped_device *md = tio->io->md;
978 dm_endio_fn endio = tio->ti->type->end_io;
979 struct bio *orig_bio = io->orig_bio;
981 if (unlikely(error == BLK_STS_TARGET)) {
982 if (bio_op(bio) == REQ_OP_DISCARD &&
983 !bio->bi_disk->queue->limits.max_discard_sectors)
985 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
986 !bio->bi_disk->queue->limits.max_write_same_sectors)
987 disable_write_same(md);
988 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
989 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
990 disable_write_zeroes(md);
994 * For zone-append bios get offset in zone of the written
995 * sector and add that to the original bio sector pos.
997 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
998 sector_t written_sector = bio->bi_iter.bi_sector;
999 struct request_queue *q = orig_bio->bi_disk->queue;
1000 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1002 orig_bio->bi_iter.bi_sector += written_sector & mask;
1006 int r = endio(tio->ti, bio, &error);
1008 case DM_ENDIO_REQUEUE:
1009 error = BLK_STS_DM_REQUEUE;
1013 case DM_ENDIO_INCOMPLETE:
1014 /* The target will handle the io */
1017 DMWARN("unimplemented target endio return value: %d", r);
1023 dec_pending(io, error);
1027 * Return maximum size of I/O possible at the supplied sector up to the current
1030 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1031 sector_t target_offset)
1033 return ti->len - target_offset;
1036 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1038 sector_t target_offset = dm_target_offset(ti, sector);
1039 sector_t len = max_io_len_target_boundary(ti, target_offset);
1043 * Does the target need to split IO even further?
1044 * - varied (per target) IO splitting is a tenet of DM; this
1045 * explains why stacked chunk_sectors based splitting via
1046 * blk_max_size_offset() isn't possible here. So pass in
1047 * ti->max_io_len to override stacked chunk_sectors.
1049 if (ti->max_io_len) {
1050 max_len = blk_max_size_offset(ti->table->md->queue,
1051 target_offset, ti->max_io_len);
1059 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1061 if (len > UINT_MAX) {
1062 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1063 (unsigned long long)len, UINT_MAX);
1064 ti->error = "Maximum size of target IO is too large";
1068 ti->max_io_len = (uint32_t) len;
1072 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1074 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1075 sector_t sector, int *srcu_idx)
1076 __acquires(md->io_barrier)
1078 struct dm_table *map;
1079 struct dm_target *ti;
1081 map = dm_get_live_table(md, srcu_idx);
1085 ti = dm_table_find_target(map, sector);
1092 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1093 long nr_pages, void **kaddr, pfn_t *pfn)
1095 struct mapped_device *md = dax_get_private(dax_dev);
1096 sector_t sector = pgoff * PAGE_SECTORS;
1097 struct dm_target *ti;
1098 long len, ret = -EIO;
1101 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1105 if (!ti->type->direct_access)
1107 len = max_io_len(ti, sector) / PAGE_SECTORS;
1110 nr_pages = min(len, nr_pages);
1111 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1114 dm_put_live_table(md, srcu_idx);
1119 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1120 int blocksize, sector_t start, sector_t len)
1122 struct mapped_device *md = dax_get_private(dax_dev);
1123 struct dm_table *map;
1127 map = dm_get_live_table(md, &srcu_idx);
1131 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1134 dm_put_live_table(md, srcu_idx);
1139 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1140 void *addr, size_t bytes, struct iov_iter *i)
1142 struct mapped_device *md = dax_get_private(dax_dev);
1143 sector_t sector = pgoff * PAGE_SECTORS;
1144 struct dm_target *ti;
1148 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1152 if (!ti->type->dax_copy_from_iter) {
1153 ret = copy_from_iter(addr, bytes, i);
1156 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1158 dm_put_live_table(md, srcu_idx);
1163 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1164 void *addr, size_t bytes, struct iov_iter *i)
1166 struct mapped_device *md = dax_get_private(dax_dev);
1167 sector_t sector = pgoff * PAGE_SECTORS;
1168 struct dm_target *ti;
1172 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1176 if (!ti->type->dax_copy_to_iter) {
1177 ret = copy_to_iter(addr, bytes, i);
1180 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1182 dm_put_live_table(md, srcu_idx);
1187 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1190 struct mapped_device *md = dax_get_private(dax_dev);
1191 sector_t sector = pgoff * PAGE_SECTORS;
1192 struct dm_target *ti;
1196 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1200 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1202 * ->zero_page_range() is mandatory dax operation. If we are
1203 * here, something is wrong.
1207 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1209 dm_put_live_table(md, srcu_idx);
1215 * A target may call dm_accept_partial_bio only from the map routine. It is
1216 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1217 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1219 * dm_accept_partial_bio informs the dm that the target only wants to process
1220 * additional n_sectors sectors of the bio and the rest of the data should be
1221 * sent in a next bio.
1223 * A diagram that explains the arithmetics:
1224 * +--------------------+---------------+-------+
1226 * +--------------------+---------------+-------+
1228 * <-------------- *tio->len_ptr --------------->
1229 * <------- bi_size ------->
1232 * Region 1 was already iterated over with bio_advance or similar function.
1233 * (it may be empty if the target doesn't use bio_advance)
1234 * Region 2 is the remaining bio size that the target wants to process.
1235 * (it may be empty if region 1 is non-empty, although there is no reason
1237 * The target requires that region 3 is to be sent in the next bio.
1239 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1240 * the partially processed part (the sum of regions 1+2) must be the same for all
1241 * copies of the bio.
1243 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1245 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1246 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1247 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1248 BUG_ON(bi_size > *tio->len_ptr);
1249 BUG_ON(n_sectors > bi_size);
1250 *tio->len_ptr -= bi_size - n_sectors;
1251 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1253 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1255 static blk_qc_t __map_bio(struct dm_target_io *tio)
1259 struct bio *clone = &tio->clone;
1260 struct dm_io *io = tio->io;
1261 struct dm_target *ti = tio->ti;
1262 blk_qc_t ret = BLK_QC_T_NONE;
1264 clone->bi_end_io = clone_endio;
1267 * Map the clone. If r == 0 we don't need to do
1268 * anything, the target has assumed ownership of
1271 atomic_inc(&io->io_count);
1272 sector = clone->bi_iter.bi_sector;
1274 r = ti->type->map(ti, clone);
1276 case DM_MAPIO_SUBMITTED:
1278 case DM_MAPIO_REMAPPED:
1279 /* the bio has been remapped so dispatch it */
1280 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1281 ret = submit_bio_noacct(clone);
1285 dec_pending(io, BLK_STS_IOERR);
1287 case DM_MAPIO_REQUEUE:
1289 dec_pending(io, BLK_STS_DM_REQUEUE);
1292 DMWARN("unimplemented target map return value: %d", r);
1299 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1301 bio->bi_iter.bi_sector = sector;
1302 bio->bi_iter.bi_size = to_bytes(len);
1306 * Creates a bio that consists of range of complete bvecs.
1308 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1309 sector_t sector, unsigned len)
1311 struct bio *clone = &tio->clone;
1314 __bio_clone_fast(clone, bio);
1316 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1320 if (bio_integrity(bio)) {
1321 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1322 !dm_target_passes_integrity(tio->ti->type))) {
1323 DMWARN("%s: the target %s doesn't support integrity data.",
1324 dm_device_name(tio->io->md),
1325 tio->ti->type->name);
1329 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1334 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1335 clone->bi_iter.bi_size = to_bytes(len);
1337 if (bio_integrity(bio))
1338 bio_integrity_trim(clone);
1343 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1344 struct dm_target *ti, unsigned num_bios)
1346 struct dm_target_io *tio;
1352 if (num_bios == 1) {
1353 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1354 bio_list_add(blist, &tio->clone);
1358 for (try = 0; try < 2; try++) {
1363 mutex_lock(&ci->io->md->table_devices_lock);
1364 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1365 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1369 bio_list_add(blist, &tio->clone);
1372 mutex_unlock(&ci->io->md->table_devices_lock);
1373 if (bio_nr == num_bios)
1376 while ((bio = bio_list_pop(blist))) {
1377 tio = container_of(bio, struct dm_target_io, clone);
1383 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1384 struct dm_target_io *tio, unsigned *len)
1386 struct bio *clone = &tio->clone;
1390 __bio_clone_fast(clone, ci->bio);
1392 bio_setup_sector(clone, ci->sector, *len);
1394 return __map_bio(tio);
1397 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1398 unsigned num_bios, unsigned *len)
1400 struct bio_list blist = BIO_EMPTY_LIST;
1402 struct dm_target_io *tio;
1404 alloc_multiple_bios(&blist, ci, ti, num_bios);
1406 while ((bio = bio_list_pop(&blist))) {
1407 tio = container_of(bio, struct dm_target_io, clone);
1408 (void) __clone_and_map_simple_bio(ci, tio, len);
1412 static int __send_empty_flush(struct clone_info *ci)
1414 unsigned target_nr = 0;
1415 struct dm_target *ti;
1416 struct bio flush_bio;
1419 * Use an on-stack bio for this, it's safe since we don't
1420 * need to reference it after submit. It's just used as
1421 * the basis for the clone(s).
1423 bio_init(&flush_bio, NULL, 0);
1424 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1425 flush_bio.bi_disk = ci->io->md->disk;
1426 bio_associate_blkg(&flush_bio);
1428 ci->bio = &flush_bio;
1429 ci->sector_count = 0;
1431 BUG_ON(bio_has_data(ci->bio));
1432 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1433 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1435 bio_uninit(ci->bio);
1439 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1440 sector_t sector, unsigned *len)
1442 struct bio *bio = ci->bio;
1443 struct dm_target_io *tio;
1446 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1448 r = clone_bio(tio, bio, sector, *len);
1453 (void) __map_bio(tio);
1458 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1464 * Even though the device advertised support for this type of
1465 * request, that does not mean every target supports it, and
1466 * reconfiguration might also have changed that since the
1467 * check was performed.
1472 len = min_t(sector_t, ci->sector_count,
1473 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1475 __send_duplicate_bios(ci, ti, num_bios, &len);
1478 ci->sector_count -= len;
1483 static bool is_abnormal_io(struct bio *bio)
1487 switch (bio_op(bio)) {
1488 case REQ_OP_DISCARD:
1489 case REQ_OP_SECURE_ERASE:
1490 case REQ_OP_WRITE_SAME:
1491 case REQ_OP_WRITE_ZEROES:
1499 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1502 struct bio *bio = ci->bio;
1503 unsigned num_bios = 0;
1505 switch (bio_op(bio)) {
1506 case REQ_OP_DISCARD:
1507 num_bios = ti->num_discard_bios;
1509 case REQ_OP_SECURE_ERASE:
1510 num_bios = ti->num_secure_erase_bios;
1512 case REQ_OP_WRITE_SAME:
1513 num_bios = ti->num_write_same_bios;
1515 case REQ_OP_WRITE_ZEROES:
1516 num_bios = ti->num_write_zeroes_bios;
1522 *result = __send_changing_extent_only(ci, ti, num_bios);
1527 * Select the correct strategy for processing a non-flush bio.
1529 static int __split_and_process_non_flush(struct clone_info *ci)
1531 struct dm_target *ti;
1535 ti = dm_table_find_target(ci->map, ci->sector);
1539 if (__process_abnormal_io(ci, ti, &r))
1542 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1544 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1549 ci->sector_count -= len;
1554 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1555 struct dm_table *map, struct bio *bio)
1558 ci->io = alloc_io(md, bio);
1559 ci->sector = bio->bi_iter.bi_sector;
1562 #define __dm_part_stat_sub(part, field, subnd) \
1563 (part_stat_get(part, field) -= (subnd))
1566 * Entry point to split a bio into clones and submit them to the targets.
1568 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1569 struct dm_table *map, struct bio *bio)
1571 struct clone_info ci;
1572 blk_qc_t ret = BLK_QC_T_NONE;
1575 init_clone_info(&ci, md, map, bio);
1577 if (bio->bi_opf & REQ_PREFLUSH) {
1578 error = __send_empty_flush(&ci);
1579 /* dec_pending submits any data associated with flush */
1580 } else if (op_is_zone_mgmt(bio_op(bio))) {
1582 ci.sector_count = 0;
1583 error = __split_and_process_non_flush(&ci);
1586 ci.sector_count = bio_sectors(bio);
1587 while (ci.sector_count && !error) {
1588 error = __split_and_process_non_flush(&ci);
1589 if (ci.sector_count && !error) {
1591 * Remainder must be passed to submit_bio_noacct()
1592 * so that it gets handled *after* bios already submitted
1593 * have been completely processed.
1594 * We take a clone of the original to store in
1595 * ci.io->orig_bio to be used by end_io_acct() and
1596 * for dec_pending to use for completion handling.
1598 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1599 GFP_NOIO, &md->queue->bio_split);
1600 ci.io->orig_bio = b;
1603 * Adjust IO stats for each split, otherwise upon queue
1604 * reentry there will be redundant IO accounting.
1605 * NOTE: this is a stop-gap fix, a proper fix involves
1606 * significant refactoring of DM core's bio splitting
1607 * (by eliminating DM's splitting and just using bio_split)
1610 __dm_part_stat_sub(dm_disk(md)->part0,
1611 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1615 trace_block_split(b, bio->bi_iter.bi_sector);
1616 ret = submit_bio_noacct(bio);
1622 /* drop the extra reference count */
1623 dec_pending(ci.io, errno_to_blk_status(error));
1627 static blk_qc_t dm_submit_bio(struct bio *bio)
1629 struct mapped_device *md = bio->bi_disk->private_data;
1630 blk_qc_t ret = BLK_QC_T_NONE;
1632 struct dm_table *map;
1634 map = dm_get_live_table(md, &srcu_idx);
1635 if (unlikely(!map)) {
1636 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1637 dm_device_name(md));
1642 /* If suspended, queue this IO for later */
1643 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1644 if (bio->bi_opf & REQ_NOWAIT)
1645 bio_wouldblock_error(bio);
1646 else if (bio->bi_opf & REQ_RAHEAD)
1654 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1655 * otherwise associated queue_limits won't be imposed.
1657 if (is_abnormal_io(bio))
1658 blk_queue_split(&bio);
1660 ret = __split_and_process_bio(md, map, bio);
1662 dm_put_live_table(md, srcu_idx);
1666 /*-----------------------------------------------------------------
1667 * An IDR is used to keep track of allocated minor numbers.
1668 *---------------------------------------------------------------*/
1669 static void free_minor(int minor)
1671 spin_lock(&_minor_lock);
1672 idr_remove(&_minor_idr, minor);
1673 spin_unlock(&_minor_lock);
1677 * See if the device with a specific minor # is free.
1679 static int specific_minor(int minor)
1683 if (minor >= (1 << MINORBITS))
1686 idr_preload(GFP_KERNEL);
1687 spin_lock(&_minor_lock);
1689 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1691 spin_unlock(&_minor_lock);
1694 return r == -ENOSPC ? -EBUSY : r;
1698 static int next_free_minor(int *minor)
1702 idr_preload(GFP_KERNEL);
1703 spin_lock(&_minor_lock);
1705 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1707 spin_unlock(&_minor_lock);
1715 static const struct block_device_operations dm_blk_dops;
1716 static const struct block_device_operations dm_rq_blk_dops;
1717 static const struct dax_operations dm_dax_ops;
1719 static void dm_wq_work(struct work_struct *work);
1721 static void cleanup_mapped_device(struct mapped_device *md)
1724 destroy_workqueue(md->wq);
1725 bioset_exit(&md->bs);
1726 bioset_exit(&md->io_bs);
1729 kill_dax(md->dax_dev);
1730 put_dax(md->dax_dev);
1735 spin_lock(&_minor_lock);
1736 md->disk->private_data = NULL;
1737 spin_unlock(&_minor_lock);
1738 del_gendisk(md->disk);
1743 blk_cleanup_queue(md->queue);
1745 cleanup_srcu_struct(&md->io_barrier);
1747 mutex_destroy(&md->suspend_lock);
1748 mutex_destroy(&md->type_lock);
1749 mutex_destroy(&md->table_devices_lock);
1751 dm_mq_cleanup_mapped_device(md);
1755 * Allocate and initialise a blank device with a given minor.
1757 static struct mapped_device *alloc_dev(int minor)
1759 int r, numa_node_id = dm_get_numa_node();
1760 struct mapped_device *md;
1763 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1765 DMWARN("unable to allocate device, out of memory.");
1769 if (!try_module_get(THIS_MODULE))
1770 goto bad_module_get;
1772 /* get a minor number for the dev */
1773 if (minor == DM_ANY_MINOR)
1774 r = next_free_minor(&minor);
1776 r = specific_minor(minor);
1780 r = init_srcu_struct(&md->io_barrier);
1782 goto bad_io_barrier;
1784 md->numa_node_id = numa_node_id;
1785 md->init_tio_pdu = false;
1786 md->type = DM_TYPE_NONE;
1787 mutex_init(&md->suspend_lock);
1788 mutex_init(&md->type_lock);
1789 mutex_init(&md->table_devices_lock);
1790 spin_lock_init(&md->deferred_lock);
1791 atomic_set(&md->holders, 1);
1792 atomic_set(&md->open_count, 0);
1793 atomic_set(&md->event_nr, 0);
1794 atomic_set(&md->uevent_seq, 0);
1795 INIT_LIST_HEAD(&md->uevent_list);
1796 INIT_LIST_HEAD(&md->table_devices);
1797 spin_lock_init(&md->uevent_lock);
1800 * default to bio-based until DM table is loaded and md->type
1801 * established. If request-based table is loaded: blk-mq will
1802 * override accordingly.
1804 md->queue = blk_alloc_queue(numa_node_id);
1808 md->disk = alloc_disk_node(1, md->numa_node_id);
1812 init_waitqueue_head(&md->wait);
1813 INIT_WORK(&md->work, dm_wq_work);
1814 init_waitqueue_head(&md->eventq);
1815 init_completion(&md->kobj_holder.completion);
1817 md->disk->major = _major;
1818 md->disk->first_minor = minor;
1819 md->disk->fops = &dm_blk_dops;
1820 md->disk->queue = md->queue;
1821 md->disk->private_data = md;
1822 sprintf(md->disk->disk_name, "dm-%d", minor);
1824 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1825 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1827 if (IS_ERR(md->dax_dev))
1831 add_disk_no_queue_reg(md->disk);
1832 format_dev_t(md->name, MKDEV(_major, minor));
1834 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1838 dm_stats_init(&md->stats);
1840 /* Populate the mapping, nobody knows we exist yet */
1841 spin_lock(&_minor_lock);
1842 old_md = idr_replace(&_minor_idr, md, minor);
1843 spin_unlock(&_minor_lock);
1845 BUG_ON(old_md != MINOR_ALLOCED);
1850 cleanup_mapped_device(md);
1854 module_put(THIS_MODULE);
1860 static void unlock_fs(struct mapped_device *md);
1862 static void free_dev(struct mapped_device *md)
1864 int minor = MINOR(disk_devt(md->disk));
1868 cleanup_mapped_device(md);
1870 free_table_devices(&md->table_devices);
1871 dm_stats_cleanup(&md->stats);
1874 module_put(THIS_MODULE);
1878 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1880 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1883 if (dm_table_bio_based(t)) {
1885 * The md may already have mempools that need changing.
1886 * If so, reload bioset because front_pad may have changed
1887 * because a different table was loaded.
1889 bioset_exit(&md->bs);
1890 bioset_exit(&md->io_bs);
1892 } else if (bioset_initialized(&md->bs)) {
1894 * There's no need to reload with request-based dm
1895 * because the size of front_pad doesn't change.
1896 * Note for future: If you are to reload bioset,
1897 * prep-ed requests in the queue may refer
1898 * to bio from the old bioset, so you must walk
1899 * through the queue to unprep.
1905 bioset_initialized(&md->bs) ||
1906 bioset_initialized(&md->io_bs));
1908 ret = bioset_init_from_src(&md->bs, &p->bs);
1911 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1913 bioset_exit(&md->bs);
1915 /* mempool bind completed, no longer need any mempools in the table */
1916 dm_table_free_md_mempools(t);
1921 * Bind a table to the device.
1923 static void event_callback(void *context)
1925 unsigned long flags;
1927 struct mapped_device *md = (struct mapped_device *) context;
1929 spin_lock_irqsave(&md->uevent_lock, flags);
1930 list_splice_init(&md->uevent_list, &uevents);
1931 spin_unlock_irqrestore(&md->uevent_lock, flags);
1933 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1935 atomic_inc(&md->event_nr);
1936 wake_up(&md->eventq);
1937 dm_issue_global_event();
1941 * Returns old map, which caller must destroy.
1943 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1944 struct queue_limits *limits)
1946 struct dm_table *old_map;
1947 struct request_queue *q = md->queue;
1948 bool request_based = dm_table_request_based(t);
1952 lockdep_assert_held(&md->suspend_lock);
1954 size = dm_table_get_size(t);
1957 * Wipe any geometry if the size of the table changed.
1959 if (size != dm_get_size(md))
1960 memset(&md->geometry, 0, sizeof(md->geometry));
1962 set_capacity_and_notify(md->disk, size);
1964 dm_table_event_callback(t, event_callback, md);
1967 * The queue hasn't been stopped yet, if the old table type wasn't
1968 * for request-based during suspension. So stop it to prevent
1969 * I/O mapping before resume.
1970 * This must be done before setting the queue restrictions,
1971 * because request-based dm may be run just after the setting.
1976 if (request_based) {
1978 * Leverage the fact that request-based DM targets are
1979 * immutable singletons - used to optimize dm_mq_queue_rq.
1981 md->immutable_target = dm_table_get_immutable_target(t);
1984 ret = __bind_mempools(md, t);
1986 old_map = ERR_PTR(ret);
1990 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1991 rcu_assign_pointer(md->map, (void *)t);
1992 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1994 dm_table_set_restrictions(t, q, limits);
2003 * Returns unbound table for the caller to free.
2005 static struct dm_table *__unbind(struct mapped_device *md)
2007 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2012 dm_table_event_callback(map, NULL, NULL);
2013 RCU_INIT_POINTER(md->map, NULL);
2020 * Constructor for a new device.
2022 int dm_create(int minor, struct mapped_device **result)
2025 struct mapped_device *md;
2027 md = alloc_dev(minor);
2031 r = dm_sysfs_init(md);
2042 * Functions to manage md->type.
2043 * All are required to hold md->type_lock.
2045 void dm_lock_md_type(struct mapped_device *md)
2047 mutex_lock(&md->type_lock);
2050 void dm_unlock_md_type(struct mapped_device *md)
2052 mutex_unlock(&md->type_lock);
2055 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2057 BUG_ON(!mutex_is_locked(&md->type_lock));
2061 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2066 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2068 return md->immutable_target_type;
2072 * The queue_limits are only valid as long as you have a reference
2075 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2077 BUG_ON(!atomic_read(&md->holders));
2078 return &md->queue->limits;
2080 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2083 * Setup the DM device's queue based on md's type
2085 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2088 struct queue_limits limits;
2089 enum dm_queue_mode type = dm_get_md_type(md);
2092 case DM_TYPE_REQUEST_BASED:
2093 md->disk->fops = &dm_rq_blk_dops;
2094 r = dm_mq_init_request_queue(md, t);
2096 DMERR("Cannot initialize queue for request-based dm mapped device");
2100 case DM_TYPE_BIO_BASED:
2101 case DM_TYPE_DAX_BIO_BASED:
2108 r = dm_calculate_queue_limits(t, &limits);
2110 DMERR("Cannot calculate initial queue limits");
2113 dm_table_set_restrictions(t, md->queue, &limits);
2114 blk_register_queue(md->disk);
2119 struct mapped_device *dm_get_md(dev_t dev)
2121 struct mapped_device *md;
2122 unsigned minor = MINOR(dev);
2124 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2127 spin_lock(&_minor_lock);
2129 md = idr_find(&_minor_idr, minor);
2130 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2131 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2137 spin_unlock(&_minor_lock);
2141 EXPORT_SYMBOL_GPL(dm_get_md);
2143 void *dm_get_mdptr(struct mapped_device *md)
2145 return md->interface_ptr;
2148 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2150 md->interface_ptr = ptr;
2153 void dm_get(struct mapped_device *md)
2155 atomic_inc(&md->holders);
2156 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2159 int dm_hold(struct mapped_device *md)
2161 spin_lock(&_minor_lock);
2162 if (test_bit(DMF_FREEING, &md->flags)) {
2163 spin_unlock(&_minor_lock);
2167 spin_unlock(&_minor_lock);
2170 EXPORT_SYMBOL_GPL(dm_hold);
2172 const char *dm_device_name(struct mapped_device *md)
2176 EXPORT_SYMBOL_GPL(dm_device_name);
2178 static void __dm_destroy(struct mapped_device *md, bool wait)
2180 struct dm_table *map;
2185 spin_lock(&_minor_lock);
2186 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2187 set_bit(DMF_FREEING, &md->flags);
2188 spin_unlock(&_minor_lock);
2190 blk_set_queue_dying(md->queue);
2193 * Take suspend_lock so that presuspend and postsuspend methods
2194 * do not race with internal suspend.
2196 mutex_lock(&md->suspend_lock);
2197 map = dm_get_live_table(md, &srcu_idx);
2198 if (!dm_suspended_md(md)) {
2199 dm_table_presuspend_targets(map);
2200 set_bit(DMF_SUSPENDED, &md->flags);
2201 set_bit(DMF_POST_SUSPENDING, &md->flags);
2202 dm_table_postsuspend_targets(map);
2204 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2205 dm_put_live_table(md, srcu_idx);
2206 mutex_unlock(&md->suspend_lock);
2209 * Rare, but there may be I/O requests still going to complete,
2210 * for example. Wait for all references to disappear.
2211 * No one should increment the reference count of the mapped_device,
2212 * after the mapped_device state becomes DMF_FREEING.
2215 while (atomic_read(&md->holders))
2217 else if (atomic_read(&md->holders))
2218 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2219 dm_device_name(md), atomic_read(&md->holders));
2222 dm_table_destroy(__unbind(md));
2226 void dm_destroy(struct mapped_device *md)
2228 __dm_destroy(md, true);
2231 void dm_destroy_immediate(struct mapped_device *md)
2233 __dm_destroy(md, false);
2236 void dm_put(struct mapped_device *md)
2238 atomic_dec(&md->holders);
2240 EXPORT_SYMBOL_GPL(dm_put);
2242 static bool md_in_flight_bios(struct mapped_device *md)
2245 struct block_device *part = dm_disk(md)->part0;
2248 for_each_possible_cpu(cpu) {
2249 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2250 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2256 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2262 prepare_to_wait(&md->wait, &wait, task_state);
2264 if (!md_in_flight_bios(md))
2267 if (signal_pending_state(task_state, current)) {
2274 finish_wait(&md->wait, &wait);
2279 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2283 if (!queue_is_mq(md->queue))
2284 return dm_wait_for_bios_completion(md, task_state);
2287 if (!blk_mq_queue_inflight(md->queue))
2290 if (signal_pending_state(task_state, current)) {
2302 * Process the deferred bios
2304 static void dm_wq_work(struct work_struct *work)
2306 struct mapped_device *md = container_of(work, struct mapped_device, work);
2309 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2310 spin_lock_irq(&md->deferred_lock);
2311 bio = bio_list_pop(&md->deferred);
2312 spin_unlock_irq(&md->deferred_lock);
2317 submit_bio_noacct(bio);
2321 static void dm_queue_flush(struct mapped_device *md)
2323 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2324 smp_mb__after_atomic();
2325 queue_work(md->wq, &md->work);
2329 * Swap in a new table, returning the old one for the caller to destroy.
2331 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2333 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2334 struct queue_limits limits;
2337 mutex_lock(&md->suspend_lock);
2339 /* device must be suspended */
2340 if (!dm_suspended_md(md))
2344 * If the new table has no data devices, retain the existing limits.
2345 * This helps multipath with queue_if_no_path if all paths disappear,
2346 * then new I/O is queued based on these limits, and then some paths
2349 if (dm_table_has_no_data_devices(table)) {
2350 live_map = dm_get_live_table_fast(md);
2352 limits = md->queue->limits;
2353 dm_put_live_table_fast(md);
2357 r = dm_calculate_queue_limits(table, &limits);
2364 map = __bind(md, table, &limits);
2365 dm_issue_global_event();
2368 mutex_unlock(&md->suspend_lock);
2373 * Functions to lock and unlock any filesystem running on the
2376 static int lock_fs(struct mapped_device *md)
2380 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2382 r = freeze_bdev(md->disk->part0);
2384 set_bit(DMF_FROZEN, &md->flags);
2388 static void unlock_fs(struct mapped_device *md)
2390 if (!test_bit(DMF_FROZEN, &md->flags))
2392 thaw_bdev(md->disk->part0);
2393 clear_bit(DMF_FROZEN, &md->flags);
2397 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2398 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2399 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2401 * If __dm_suspend returns 0, the device is completely quiescent
2402 * now. There is no request-processing activity. All new requests
2403 * are being added to md->deferred list.
2405 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2406 unsigned suspend_flags, long task_state,
2407 int dmf_suspended_flag)
2409 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2410 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2413 lockdep_assert_held(&md->suspend_lock);
2416 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2417 * This flag is cleared before dm_suspend returns.
2420 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2422 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2425 * This gets reverted if there's an error later and the targets
2426 * provide the .presuspend_undo hook.
2428 dm_table_presuspend_targets(map);
2431 * Flush I/O to the device.
2432 * Any I/O submitted after lock_fs() may not be flushed.
2433 * noflush takes precedence over do_lockfs.
2434 * (lock_fs() flushes I/Os and waits for them to complete.)
2436 if (!noflush && do_lockfs) {
2439 dm_table_presuspend_undo_targets(map);
2445 * Here we must make sure that no processes are submitting requests
2446 * to target drivers i.e. no one may be executing
2447 * __split_and_process_bio from dm_submit_bio.
2449 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2450 * we take the write lock. To prevent any process from reentering
2451 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2452 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2453 * flush_workqueue(md->wq).
2455 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2457 synchronize_srcu(&md->io_barrier);
2460 * Stop md->queue before flushing md->wq in case request-based
2461 * dm defers requests to md->wq from md->queue.
2463 if (dm_request_based(md))
2464 dm_stop_queue(md->queue);
2466 flush_workqueue(md->wq);
2469 * At this point no more requests are entering target request routines.
2470 * We call dm_wait_for_completion to wait for all existing requests
2473 r = dm_wait_for_completion(md, task_state);
2475 set_bit(dmf_suspended_flag, &md->flags);
2478 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2480 synchronize_srcu(&md->io_barrier);
2482 /* were we interrupted ? */
2486 if (dm_request_based(md))
2487 dm_start_queue(md->queue);
2490 dm_table_presuspend_undo_targets(map);
2491 /* pushback list is already flushed, so skip flush */
2498 * We need to be able to change a mapping table under a mounted
2499 * filesystem. For example we might want to move some data in
2500 * the background. Before the table can be swapped with
2501 * dm_bind_table, dm_suspend must be called to flush any in
2502 * flight bios and ensure that any further io gets deferred.
2505 * Suspend mechanism in request-based dm.
2507 * 1. Flush all I/Os by lock_fs() if needed.
2508 * 2. Stop dispatching any I/O by stopping the request_queue.
2509 * 3. Wait for all in-flight I/Os to be completed or requeued.
2511 * To abort suspend, start the request_queue.
2513 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2515 struct dm_table *map = NULL;
2519 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2521 if (dm_suspended_md(md)) {
2526 if (dm_suspended_internally_md(md)) {
2527 /* already internally suspended, wait for internal resume */
2528 mutex_unlock(&md->suspend_lock);
2529 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2535 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2537 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2541 set_bit(DMF_POST_SUSPENDING, &md->flags);
2542 dm_table_postsuspend_targets(map);
2543 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2546 mutex_unlock(&md->suspend_lock);
2550 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2553 int r = dm_table_resume_targets(map);
2561 * Flushing deferred I/Os must be done after targets are resumed
2562 * so that mapping of targets can work correctly.
2563 * Request-based dm is queueing the deferred I/Os in its request_queue.
2565 if (dm_request_based(md))
2566 dm_start_queue(md->queue);
2573 int dm_resume(struct mapped_device *md)
2576 struct dm_table *map = NULL;
2580 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2582 if (!dm_suspended_md(md))
2585 if (dm_suspended_internally_md(md)) {
2586 /* already internally suspended, wait for internal resume */
2587 mutex_unlock(&md->suspend_lock);
2588 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2594 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2595 if (!map || !dm_table_get_size(map))
2598 r = __dm_resume(md, map);
2602 clear_bit(DMF_SUSPENDED, &md->flags);
2604 mutex_unlock(&md->suspend_lock);
2610 * Internal suspend/resume works like userspace-driven suspend. It waits
2611 * until all bios finish and prevents issuing new bios to the target drivers.
2612 * It may be used only from the kernel.
2615 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2617 struct dm_table *map = NULL;
2619 lockdep_assert_held(&md->suspend_lock);
2621 if (md->internal_suspend_count++)
2622 return; /* nested internal suspend */
2624 if (dm_suspended_md(md)) {
2625 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2626 return; /* nest suspend */
2629 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2632 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2633 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2634 * would require changing .presuspend to return an error -- avoid this
2635 * until there is a need for more elaborate variants of internal suspend.
2637 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2638 DMF_SUSPENDED_INTERNALLY);
2640 set_bit(DMF_POST_SUSPENDING, &md->flags);
2641 dm_table_postsuspend_targets(map);
2642 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2645 static void __dm_internal_resume(struct mapped_device *md)
2647 BUG_ON(!md->internal_suspend_count);
2649 if (--md->internal_suspend_count)
2650 return; /* resume from nested internal suspend */
2652 if (dm_suspended_md(md))
2653 goto done; /* resume from nested suspend */
2656 * NOTE: existing callers don't need to call dm_table_resume_targets
2657 * (which may fail -- so best to avoid it for now by passing NULL map)
2659 (void) __dm_resume(md, NULL);
2662 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2663 smp_mb__after_atomic();
2664 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2667 void dm_internal_suspend_noflush(struct mapped_device *md)
2669 mutex_lock(&md->suspend_lock);
2670 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2671 mutex_unlock(&md->suspend_lock);
2673 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2675 void dm_internal_resume(struct mapped_device *md)
2677 mutex_lock(&md->suspend_lock);
2678 __dm_internal_resume(md);
2679 mutex_unlock(&md->suspend_lock);
2681 EXPORT_SYMBOL_GPL(dm_internal_resume);
2684 * Fast variants of internal suspend/resume hold md->suspend_lock,
2685 * which prevents interaction with userspace-driven suspend.
2688 void dm_internal_suspend_fast(struct mapped_device *md)
2690 mutex_lock(&md->suspend_lock);
2691 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2694 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2695 synchronize_srcu(&md->io_barrier);
2696 flush_workqueue(md->wq);
2697 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2699 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2701 void dm_internal_resume_fast(struct mapped_device *md)
2703 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2709 mutex_unlock(&md->suspend_lock);
2711 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2713 /*-----------------------------------------------------------------
2714 * Event notification.
2715 *---------------------------------------------------------------*/
2716 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2721 char udev_cookie[DM_COOKIE_LENGTH];
2722 char *envp[] = { udev_cookie, NULL };
2724 noio_flag = memalloc_noio_save();
2727 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2729 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2730 DM_COOKIE_ENV_VAR_NAME, cookie);
2731 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2735 memalloc_noio_restore(noio_flag);
2740 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2742 return atomic_add_return(1, &md->uevent_seq);
2745 uint32_t dm_get_event_nr(struct mapped_device *md)
2747 return atomic_read(&md->event_nr);
2750 int dm_wait_event(struct mapped_device *md, int event_nr)
2752 return wait_event_interruptible(md->eventq,
2753 (event_nr != atomic_read(&md->event_nr)));
2756 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2758 unsigned long flags;
2760 spin_lock_irqsave(&md->uevent_lock, flags);
2761 list_add(elist, &md->uevent_list);
2762 spin_unlock_irqrestore(&md->uevent_lock, flags);
2766 * The gendisk is only valid as long as you have a reference
2769 struct gendisk *dm_disk(struct mapped_device *md)
2773 EXPORT_SYMBOL_GPL(dm_disk);
2775 struct kobject *dm_kobject(struct mapped_device *md)
2777 return &md->kobj_holder.kobj;
2780 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2782 struct mapped_device *md;
2784 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2786 spin_lock(&_minor_lock);
2787 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2793 spin_unlock(&_minor_lock);
2798 int dm_suspended_md(struct mapped_device *md)
2800 return test_bit(DMF_SUSPENDED, &md->flags);
2803 static int dm_post_suspending_md(struct mapped_device *md)
2805 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2808 int dm_suspended_internally_md(struct mapped_device *md)
2810 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2813 int dm_test_deferred_remove_flag(struct mapped_device *md)
2815 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2818 int dm_suspended(struct dm_target *ti)
2820 return dm_suspended_md(ti->table->md);
2822 EXPORT_SYMBOL_GPL(dm_suspended);
2824 int dm_post_suspending(struct dm_target *ti)
2826 return dm_post_suspending_md(ti->table->md);
2828 EXPORT_SYMBOL_GPL(dm_post_suspending);
2830 int dm_noflush_suspending(struct dm_target *ti)
2832 return __noflush_suspending(ti->table->md);
2834 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2836 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2837 unsigned integrity, unsigned per_io_data_size,
2838 unsigned min_pool_size)
2840 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2841 unsigned int pool_size = 0;
2842 unsigned int front_pad, io_front_pad;
2849 case DM_TYPE_BIO_BASED:
2850 case DM_TYPE_DAX_BIO_BASED:
2851 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2852 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2853 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2854 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2857 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2860 case DM_TYPE_REQUEST_BASED:
2861 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2862 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2863 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2869 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2873 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2879 dm_free_md_mempools(pools);
2884 void dm_free_md_mempools(struct dm_md_mempools *pools)
2889 bioset_exit(&pools->bs);
2890 bioset_exit(&pools->io_bs);
2902 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2905 struct mapped_device *md = bdev->bd_disk->private_data;
2906 struct dm_table *table;
2907 struct dm_target *ti;
2908 int ret = -ENOTTY, srcu_idx;
2910 table = dm_get_live_table(md, &srcu_idx);
2911 if (!table || !dm_table_get_size(table))
2914 /* We only support devices that have a single target */
2915 if (dm_table_get_num_targets(table) != 1)
2917 ti = dm_table_get_target(table, 0);
2920 if (!ti->type->iterate_devices)
2923 ret = ti->type->iterate_devices(ti, fn, data);
2925 dm_put_live_table(md, srcu_idx);
2930 * For register / unregister we need to manually call out to every path.
2932 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2933 sector_t start, sector_t len, void *data)
2935 struct dm_pr *pr = data;
2936 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2938 if (!ops || !ops->pr_register)
2940 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2943 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2954 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2955 if (ret && new_key) {
2956 /* unregister all paths if we failed to register any path */
2957 pr.old_key = new_key;
2960 pr.fail_early = false;
2961 dm_call_pr(bdev, __dm_pr_register, &pr);
2967 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2970 struct mapped_device *md = bdev->bd_disk->private_data;
2971 const struct pr_ops *ops;
2974 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2978 ops = bdev->bd_disk->fops->pr_ops;
2979 if (ops && ops->pr_reserve)
2980 r = ops->pr_reserve(bdev, key, type, flags);
2984 dm_unprepare_ioctl(md, srcu_idx);
2988 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2990 struct mapped_device *md = bdev->bd_disk->private_data;
2991 const struct pr_ops *ops;
2994 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2998 ops = bdev->bd_disk->fops->pr_ops;
2999 if (ops && ops->pr_release)
3000 r = ops->pr_release(bdev, key, type);
3004 dm_unprepare_ioctl(md, srcu_idx);
3008 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3009 enum pr_type type, bool abort)
3011 struct mapped_device *md = bdev->bd_disk->private_data;
3012 const struct pr_ops *ops;
3015 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3019 ops = bdev->bd_disk->fops->pr_ops;
3020 if (ops && ops->pr_preempt)
3021 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3025 dm_unprepare_ioctl(md, srcu_idx);
3029 static int dm_pr_clear(struct block_device *bdev, u64 key)
3031 struct mapped_device *md = bdev->bd_disk->private_data;
3032 const struct pr_ops *ops;
3035 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3039 ops = bdev->bd_disk->fops->pr_ops;
3040 if (ops && ops->pr_clear)
3041 r = ops->pr_clear(bdev, key);
3045 dm_unprepare_ioctl(md, srcu_idx);
3049 static const struct pr_ops dm_pr_ops = {
3050 .pr_register = dm_pr_register,
3051 .pr_reserve = dm_pr_reserve,
3052 .pr_release = dm_pr_release,
3053 .pr_preempt = dm_pr_preempt,
3054 .pr_clear = dm_pr_clear,
3057 static const struct block_device_operations dm_blk_dops = {
3058 .submit_bio = dm_submit_bio,
3059 .open = dm_blk_open,
3060 .release = dm_blk_close,
3061 .ioctl = dm_blk_ioctl,
3062 .getgeo = dm_blk_getgeo,
3063 .report_zones = dm_blk_report_zones,
3064 .pr_ops = &dm_pr_ops,
3065 .owner = THIS_MODULE
3068 static const struct block_device_operations dm_rq_blk_dops = {
3069 .open = dm_blk_open,
3070 .release = dm_blk_close,
3071 .ioctl = dm_blk_ioctl,
3072 .getgeo = dm_blk_getgeo,
3073 .pr_ops = &dm_pr_ops,
3074 .owner = THIS_MODULE
3077 static const struct dax_operations dm_dax_ops = {
3078 .direct_access = dm_dax_direct_access,
3079 .dax_supported = dm_dax_supported,
3080 .copy_from_iter = dm_dax_copy_from_iter,
3081 .copy_to_iter = dm_dax_copy_to_iter,
3082 .zero_page_range = dm_dax_zero_page_range,
3088 module_init(dm_init);
3089 module_exit(dm_exit);
3091 module_param(major, uint, 0);
3092 MODULE_PARM_DESC(major, "The major number of the device mapper");
3094 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3095 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3097 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3098 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3100 MODULE_DESCRIPTION(DM_NAME " driver");
3101 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3102 MODULE_LICENSE("GPL");