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 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 even further?
1044 * - q->limits.chunk_sectors reflects ti->max_io_len so
1045 * blk_max_size_offset() provides required splitting.
1046 * - blk_max_size_offset() also respects q->limits.max_sectors
1048 max_len = blk_max_size_offset(ti->table->md->queue,
1056 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1058 if (len > UINT_MAX) {
1059 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1060 (unsigned long long)len, UINT_MAX);
1061 ti->error = "Maximum size of target IO is too large";
1065 ti->max_io_len = (uint32_t) len;
1069 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1071 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1072 sector_t sector, int *srcu_idx)
1073 __acquires(md->io_barrier)
1075 struct dm_table *map;
1076 struct dm_target *ti;
1078 map = dm_get_live_table(md, srcu_idx);
1082 ti = dm_table_find_target(map, sector);
1089 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1090 long nr_pages, void **kaddr, pfn_t *pfn)
1092 struct mapped_device *md = dax_get_private(dax_dev);
1093 sector_t sector = pgoff * PAGE_SECTORS;
1094 struct dm_target *ti;
1095 long len, ret = -EIO;
1098 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1102 if (!ti->type->direct_access)
1104 len = max_io_len(ti, sector) / PAGE_SECTORS;
1107 nr_pages = min(len, nr_pages);
1108 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1111 dm_put_live_table(md, srcu_idx);
1116 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1117 int blocksize, sector_t start, sector_t len)
1119 struct mapped_device *md = dax_get_private(dax_dev);
1120 struct dm_table *map;
1124 map = dm_get_live_table(md, &srcu_idx);
1128 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1131 dm_put_live_table(md, srcu_idx);
1136 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1137 void *addr, size_t bytes, struct iov_iter *i)
1139 struct mapped_device *md = dax_get_private(dax_dev);
1140 sector_t sector = pgoff * PAGE_SECTORS;
1141 struct dm_target *ti;
1145 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1149 if (!ti->type->dax_copy_from_iter) {
1150 ret = copy_from_iter(addr, bytes, i);
1153 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1155 dm_put_live_table(md, srcu_idx);
1160 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1161 void *addr, size_t bytes, struct iov_iter *i)
1163 struct mapped_device *md = dax_get_private(dax_dev);
1164 sector_t sector = pgoff * PAGE_SECTORS;
1165 struct dm_target *ti;
1169 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1173 if (!ti->type->dax_copy_to_iter) {
1174 ret = copy_to_iter(addr, bytes, i);
1177 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1179 dm_put_live_table(md, srcu_idx);
1184 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1187 struct mapped_device *md = dax_get_private(dax_dev);
1188 sector_t sector = pgoff * PAGE_SECTORS;
1189 struct dm_target *ti;
1193 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1197 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1199 * ->zero_page_range() is mandatory dax operation. If we are
1200 * here, something is wrong.
1202 dm_put_live_table(md, srcu_idx);
1205 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1208 dm_put_live_table(md, srcu_idx);
1214 * A target may call dm_accept_partial_bio only from the map routine. It is
1215 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1216 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1218 * dm_accept_partial_bio informs the dm that the target only wants to process
1219 * additional n_sectors sectors of the bio and the rest of the data should be
1220 * sent in a next bio.
1222 * A diagram that explains the arithmetics:
1223 * +--------------------+---------------+-------+
1225 * +--------------------+---------------+-------+
1227 * <-------------- *tio->len_ptr --------------->
1228 * <------- bi_size ------->
1231 * Region 1 was already iterated over with bio_advance or similar function.
1232 * (it may be empty if the target doesn't use bio_advance)
1233 * Region 2 is the remaining bio size that the target wants to process.
1234 * (it may be empty if region 1 is non-empty, although there is no reason
1236 * The target requires that region 3 is to be sent in the next bio.
1238 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1239 * the partially processed part (the sum of regions 1+2) must be the same for all
1240 * copies of the bio.
1242 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1244 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1245 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1246 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1247 BUG_ON(bi_size > *tio->len_ptr);
1248 BUG_ON(n_sectors > bi_size);
1249 *tio->len_ptr -= bi_size - n_sectors;
1250 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1252 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1254 static blk_qc_t __map_bio(struct dm_target_io *tio)
1258 struct bio *clone = &tio->clone;
1259 struct dm_io *io = tio->io;
1260 struct dm_target *ti = tio->ti;
1261 blk_qc_t ret = BLK_QC_T_NONE;
1263 clone->bi_end_io = clone_endio;
1266 * Map the clone. If r == 0 we don't need to do
1267 * anything, the target has assumed ownership of
1270 atomic_inc(&io->io_count);
1271 sector = clone->bi_iter.bi_sector;
1273 r = ti->type->map(ti, clone);
1275 case DM_MAPIO_SUBMITTED:
1277 case DM_MAPIO_REMAPPED:
1278 /* the bio has been remapped so dispatch it */
1279 trace_block_bio_remap(clone->bi_disk->queue, clone,
1280 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 ci->bio = &flush_bio;
1426 ci->sector_count = 0;
1429 * Empty flush uses a statically initialized bio, as the base for
1430 * cloning. However, blkg association requires that a bdev is
1431 * associated with a gendisk, which doesn't happen until the bdev is
1432 * opened. So, blkg association is done at issue time of the flush
1433 * rather than when the device is created in alloc_dev().
1435 bio_set_dev(ci->bio, ci->io->md->bdev);
1437 BUG_ON(bio_has_data(ci->bio));
1438 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1439 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1441 bio_uninit(ci->bio);
1445 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1446 sector_t sector, unsigned *len)
1448 struct bio *bio = ci->bio;
1449 struct dm_target_io *tio;
1452 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1454 r = clone_bio(tio, bio, sector, *len);
1459 (void) __map_bio(tio);
1464 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1470 * Even though the device advertised support for this type of
1471 * request, that does not mean every target supports it, and
1472 * reconfiguration might also have changed that since the
1473 * check was performed.
1478 len = min_t(sector_t, ci->sector_count,
1479 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1481 __send_duplicate_bios(ci, ti, num_bios, &len);
1484 ci->sector_count -= len;
1489 static bool is_abnormal_io(struct bio *bio)
1493 switch (bio_op(bio)) {
1494 case REQ_OP_DISCARD:
1495 case REQ_OP_SECURE_ERASE:
1496 case REQ_OP_WRITE_SAME:
1497 case REQ_OP_WRITE_ZEROES:
1505 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1508 struct bio *bio = ci->bio;
1509 unsigned num_bios = 0;
1511 switch (bio_op(bio)) {
1512 case REQ_OP_DISCARD:
1513 num_bios = ti->num_discard_bios;
1515 case REQ_OP_SECURE_ERASE:
1516 num_bios = ti->num_secure_erase_bios;
1518 case REQ_OP_WRITE_SAME:
1519 num_bios = ti->num_write_same_bios;
1521 case REQ_OP_WRITE_ZEROES:
1522 num_bios = ti->num_write_zeroes_bios;
1528 *result = __send_changing_extent_only(ci, ti, num_bios);
1533 * Select the correct strategy for processing a non-flush bio.
1535 static int __split_and_process_non_flush(struct clone_info *ci)
1537 struct dm_target *ti;
1541 ti = dm_table_find_target(ci->map, ci->sector);
1545 if (__process_abnormal_io(ci, ti, &r))
1548 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1550 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1555 ci->sector_count -= len;
1560 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1561 struct dm_table *map, struct bio *bio)
1564 ci->io = alloc_io(md, bio);
1565 ci->sector = bio->bi_iter.bi_sector;
1568 #define __dm_part_stat_sub(part, field, subnd) \
1569 (part_stat_get(part, field) -= (subnd))
1572 * Entry point to split a bio into clones and submit them to the targets.
1574 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1575 struct dm_table *map, struct bio *bio)
1577 struct clone_info ci;
1578 blk_qc_t ret = BLK_QC_T_NONE;
1581 init_clone_info(&ci, md, map, bio);
1583 if (bio->bi_opf & REQ_PREFLUSH) {
1584 error = __send_empty_flush(&ci);
1585 /* dec_pending submits any data associated with flush */
1586 } else if (op_is_zone_mgmt(bio_op(bio))) {
1588 ci.sector_count = 0;
1589 error = __split_and_process_non_flush(&ci);
1592 ci.sector_count = bio_sectors(bio);
1593 while (ci.sector_count && !error) {
1594 error = __split_and_process_non_flush(&ci);
1595 if (current->bio_list && ci.sector_count && !error) {
1597 * Remainder must be passed to submit_bio_noacct()
1598 * so that it gets handled *after* bios already submitted
1599 * have been completely processed.
1600 * We take a clone of the original to store in
1601 * ci.io->orig_bio to be used by end_io_acct() and
1602 * for dec_pending to use for completion handling.
1604 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1605 GFP_NOIO, &md->queue->bio_split);
1606 ci.io->orig_bio = b;
1609 * Adjust IO stats for each split, otherwise upon queue
1610 * reentry there will be redundant IO accounting.
1611 * NOTE: this is a stop-gap fix, a proper fix involves
1612 * significant refactoring of DM core's bio splitting
1613 * (by eliminating DM's splitting and just using bio_split)
1616 __dm_part_stat_sub(&dm_disk(md)->part0,
1617 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1621 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1622 ret = submit_bio_noacct(bio);
1628 /* drop the extra reference count */
1629 dec_pending(ci.io, errno_to_blk_status(error));
1633 static blk_qc_t dm_submit_bio(struct bio *bio)
1635 struct mapped_device *md = bio->bi_disk->private_data;
1636 blk_qc_t ret = BLK_QC_T_NONE;
1638 struct dm_table *map;
1640 map = dm_get_live_table(md, &srcu_idx);
1641 if (unlikely(!map)) {
1642 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1643 dm_device_name(md));
1648 /* If suspended, queue this IO for later */
1649 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1650 if (bio->bi_opf & REQ_NOWAIT)
1651 bio_wouldblock_error(bio);
1652 else if (bio->bi_opf & REQ_RAHEAD)
1660 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1661 * otherwise associated queue_limits won't be imposed.
1663 if (is_abnormal_io(bio))
1664 blk_queue_split(&bio);
1666 ret = __split_and_process_bio(md, map, bio);
1668 dm_put_live_table(md, srcu_idx);
1672 /*-----------------------------------------------------------------
1673 * An IDR is used to keep track of allocated minor numbers.
1674 *---------------------------------------------------------------*/
1675 static void free_minor(int minor)
1677 spin_lock(&_minor_lock);
1678 idr_remove(&_minor_idr, minor);
1679 spin_unlock(&_minor_lock);
1683 * See if the device with a specific minor # is free.
1685 static int specific_minor(int minor)
1689 if (minor >= (1 << MINORBITS))
1692 idr_preload(GFP_KERNEL);
1693 spin_lock(&_minor_lock);
1695 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1697 spin_unlock(&_minor_lock);
1700 return r == -ENOSPC ? -EBUSY : r;
1704 static int next_free_minor(int *minor)
1708 idr_preload(GFP_KERNEL);
1709 spin_lock(&_minor_lock);
1711 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1713 spin_unlock(&_minor_lock);
1721 static const struct block_device_operations dm_blk_dops;
1722 static const struct block_device_operations dm_rq_blk_dops;
1723 static const struct dax_operations dm_dax_ops;
1725 static void dm_wq_work(struct work_struct *work);
1727 static void cleanup_mapped_device(struct mapped_device *md)
1730 destroy_workqueue(md->wq);
1731 bioset_exit(&md->bs);
1732 bioset_exit(&md->io_bs);
1735 kill_dax(md->dax_dev);
1736 put_dax(md->dax_dev);
1741 spin_lock(&_minor_lock);
1742 md->disk->private_data = NULL;
1743 spin_unlock(&_minor_lock);
1744 del_gendisk(md->disk);
1749 blk_cleanup_queue(md->queue);
1751 cleanup_srcu_struct(&md->io_barrier);
1758 mutex_destroy(&md->suspend_lock);
1759 mutex_destroy(&md->type_lock);
1760 mutex_destroy(&md->table_devices_lock);
1762 dm_mq_cleanup_mapped_device(md);
1766 * Allocate and initialise a blank device with a given minor.
1768 static struct mapped_device *alloc_dev(int minor)
1770 int r, numa_node_id = dm_get_numa_node();
1771 struct mapped_device *md;
1774 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1776 DMWARN("unable to allocate device, out of memory.");
1780 if (!try_module_get(THIS_MODULE))
1781 goto bad_module_get;
1783 /* get a minor number for the dev */
1784 if (minor == DM_ANY_MINOR)
1785 r = next_free_minor(&minor);
1787 r = specific_minor(minor);
1791 r = init_srcu_struct(&md->io_barrier);
1793 goto bad_io_barrier;
1795 md->numa_node_id = numa_node_id;
1796 md->init_tio_pdu = false;
1797 md->type = DM_TYPE_NONE;
1798 mutex_init(&md->suspend_lock);
1799 mutex_init(&md->type_lock);
1800 mutex_init(&md->table_devices_lock);
1801 spin_lock_init(&md->deferred_lock);
1802 atomic_set(&md->holders, 1);
1803 atomic_set(&md->open_count, 0);
1804 atomic_set(&md->event_nr, 0);
1805 atomic_set(&md->uevent_seq, 0);
1806 INIT_LIST_HEAD(&md->uevent_list);
1807 INIT_LIST_HEAD(&md->table_devices);
1808 spin_lock_init(&md->uevent_lock);
1811 * default to bio-based until DM table is loaded and md->type
1812 * established. If request-based table is loaded: blk-mq will
1813 * override accordingly.
1815 md->queue = blk_alloc_queue(numa_node_id);
1819 md->disk = alloc_disk_node(1, md->numa_node_id);
1823 init_waitqueue_head(&md->wait);
1824 INIT_WORK(&md->work, dm_wq_work);
1825 init_waitqueue_head(&md->eventq);
1826 init_completion(&md->kobj_holder.completion);
1828 md->disk->major = _major;
1829 md->disk->first_minor = minor;
1830 md->disk->fops = &dm_blk_dops;
1831 md->disk->queue = md->queue;
1832 md->disk->private_data = md;
1833 sprintf(md->disk->disk_name, "dm-%d", minor);
1835 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1836 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1838 if (IS_ERR(md->dax_dev))
1842 add_disk_no_queue_reg(md->disk);
1843 format_dev_t(md->name, MKDEV(_major, minor));
1845 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1849 md->bdev = bdget_disk(md->disk, 0);
1853 dm_stats_init(&md->stats);
1855 /* Populate the mapping, nobody knows we exist yet */
1856 spin_lock(&_minor_lock);
1857 old_md = idr_replace(&_minor_idr, md, minor);
1858 spin_unlock(&_minor_lock);
1860 BUG_ON(old_md != MINOR_ALLOCED);
1865 cleanup_mapped_device(md);
1869 module_put(THIS_MODULE);
1875 static void unlock_fs(struct mapped_device *md);
1877 static void free_dev(struct mapped_device *md)
1879 int minor = MINOR(disk_devt(md->disk));
1883 cleanup_mapped_device(md);
1885 free_table_devices(&md->table_devices);
1886 dm_stats_cleanup(&md->stats);
1889 module_put(THIS_MODULE);
1893 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1895 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1898 if (dm_table_bio_based(t)) {
1900 * The md may already have mempools that need changing.
1901 * If so, reload bioset because front_pad may have changed
1902 * because a different table was loaded.
1904 bioset_exit(&md->bs);
1905 bioset_exit(&md->io_bs);
1907 } else if (bioset_initialized(&md->bs)) {
1909 * There's no need to reload with request-based dm
1910 * because the size of front_pad doesn't change.
1911 * Note for future: If you are to reload bioset,
1912 * prep-ed requests in the queue may refer
1913 * to bio from the old bioset, so you must walk
1914 * through the queue to unprep.
1920 bioset_initialized(&md->bs) ||
1921 bioset_initialized(&md->io_bs));
1923 ret = bioset_init_from_src(&md->bs, &p->bs);
1926 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1928 bioset_exit(&md->bs);
1930 /* mempool bind completed, no longer need any mempools in the table */
1931 dm_table_free_md_mempools(t);
1936 * Bind a table to the device.
1938 static void event_callback(void *context)
1940 unsigned long flags;
1942 struct mapped_device *md = (struct mapped_device *) context;
1944 spin_lock_irqsave(&md->uevent_lock, flags);
1945 list_splice_init(&md->uevent_list, &uevents);
1946 spin_unlock_irqrestore(&md->uevent_lock, flags);
1948 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1950 atomic_inc(&md->event_nr);
1951 wake_up(&md->eventq);
1952 dm_issue_global_event();
1956 * Returns old map, which caller must destroy.
1958 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1959 struct queue_limits *limits)
1961 struct dm_table *old_map;
1962 struct request_queue *q = md->queue;
1963 bool request_based = dm_table_request_based(t);
1967 lockdep_assert_held(&md->suspend_lock);
1969 size = dm_table_get_size(t);
1972 * Wipe any geometry if the size of the table changed.
1974 if (size != dm_get_size(md))
1975 memset(&md->geometry, 0, sizeof(md->geometry));
1977 set_capacity(md->disk, size);
1978 bd_set_nr_sectors(md->bdev, size);
1980 dm_table_event_callback(t, event_callback, md);
1983 * The queue hasn't been stopped yet, if the old table type wasn't
1984 * for request-based during suspension. So stop it to prevent
1985 * I/O mapping before resume.
1986 * This must be done before setting the queue restrictions,
1987 * because request-based dm may be run just after the setting.
1992 if (request_based) {
1994 * Leverage the fact that request-based DM targets are
1995 * immutable singletons - used to optimize dm_mq_queue_rq.
1997 md->immutable_target = dm_table_get_immutable_target(t);
2000 ret = __bind_mempools(md, t);
2002 old_map = ERR_PTR(ret);
2006 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2007 rcu_assign_pointer(md->map, (void *)t);
2008 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2010 dm_table_set_restrictions(t, q, limits);
2019 * Returns unbound table for the caller to free.
2021 static struct dm_table *__unbind(struct mapped_device *md)
2023 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2028 dm_table_event_callback(map, NULL, NULL);
2029 RCU_INIT_POINTER(md->map, NULL);
2036 * Constructor for a new device.
2038 int dm_create(int minor, struct mapped_device **result)
2041 struct mapped_device *md;
2043 md = alloc_dev(minor);
2047 r = dm_sysfs_init(md);
2058 * Functions to manage md->type.
2059 * All are required to hold md->type_lock.
2061 void dm_lock_md_type(struct mapped_device *md)
2063 mutex_lock(&md->type_lock);
2066 void dm_unlock_md_type(struct mapped_device *md)
2068 mutex_unlock(&md->type_lock);
2071 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2073 BUG_ON(!mutex_is_locked(&md->type_lock));
2077 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2082 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2084 return md->immutable_target_type;
2088 * The queue_limits are only valid as long as you have a reference
2091 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2093 BUG_ON(!atomic_read(&md->holders));
2094 return &md->queue->limits;
2096 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2099 * Setup the DM device's queue based on md's type
2101 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2104 struct queue_limits limits;
2105 enum dm_queue_mode type = dm_get_md_type(md);
2108 case DM_TYPE_REQUEST_BASED:
2109 md->disk->fops = &dm_rq_blk_dops;
2110 r = dm_mq_init_request_queue(md, t);
2112 DMERR("Cannot initialize queue for request-based dm mapped device");
2116 case DM_TYPE_BIO_BASED:
2117 case DM_TYPE_DAX_BIO_BASED:
2124 r = dm_calculate_queue_limits(t, &limits);
2126 DMERR("Cannot calculate initial queue limits");
2129 dm_table_set_restrictions(t, md->queue, &limits);
2130 blk_register_queue(md->disk);
2135 struct mapped_device *dm_get_md(dev_t dev)
2137 struct mapped_device *md;
2138 unsigned minor = MINOR(dev);
2140 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2143 spin_lock(&_minor_lock);
2145 md = idr_find(&_minor_idr, minor);
2146 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2147 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2153 spin_unlock(&_minor_lock);
2157 EXPORT_SYMBOL_GPL(dm_get_md);
2159 void *dm_get_mdptr(struct mapped_device *md)
2161 return md->interface_ptr;
2164 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2166 md->interface_ptr = ptr;
2169 void dm_get(struct mapped_device *md)
2171 atomic_inc(&md->holders);
2172 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2175 int dm_hold(struct mapped_device *md)
2177 spin_lock(&_minor_lock);
2178 if (test_bit(DMF_FREEING, &md->flags)) {
2179 spin_unlock(&_minor_lock);
2183 spin_unlock(&_minor_lock);
2186 EXPORT_SYMBOL_GPL(dm_hold);
2188 const char *dm_device_name(struct mapped_device *md)
2192 EXPORT_SYMBOL_GPL(dm_device_name);
2194 static void __dm_destroy(struct mapped_device *md, bool wait)
2196 struct dm_table *map;
2201 spin_lock(&_minor_lock);
2202 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2203 set_bit(DMF_FREEING, &md->flags);
2204 spin_unlock(&_minor_lock);
2206 blk_set_queue_dying(md->queue);
2209 * Take suspend_lock so that presuspend and postsuspend methods
2210 * do not race with internal suspend.
2212 mutex_lock(&md->suspend_lock);
2213 map = dm_get_live_table(md, &srcu_idx);
2214 if (!dm_suspended_md(md)) {
2215 dm_table_presuspend_targets(map);
2216 set_bit(DMF_SUSPENDED, &md->flags);
2217 set_bit(DMF_POST_SUSPENDING, &md->flags);
2218 dm_table_postsuspend_targets(map);
2220 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2221 dm_put_live_table(md, srcu_idx);
2222 mutex_unlock(&md->suspend_lock);
2225 * Rare, but there may be I/O requests still going to complete,
2226 * for example. Wait for all references to disappear.
2227 * No one should increment the reference count of the mapped_device,
2228 * after the mapped_device state becomes DMF_FREEING.
2231 while (atomic_read(&md->holders))
2233 else if (atomic_read(&md->holders))
2234 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2235 dm_device_name(md), atomic_read(&md->holders));
2238 dm_table_destroy(__unbind(md));
2242 void dm_destroy(struct mapped_device *md)
2244 __dm_destroy(md, true);
2247 void dm_destroy_immediate(struct mapped_device *md)
2249 __dm_destroy(md, false);
2252 void dm_put(struct mapped_device *md)
2254 atomic_dec(&md->holders);
2256 EXPORT_SYMBOL_GPL(dm_put);
2258 static bool md_in_flight_bios(struct mapped_device *md)
2261 struct hd_struct *part = &dm_disk(md)->part0;
2264 for_each_possible_cpu(cpu) {
2265 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2266 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2272 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2278 prepare_to_wait(&md->wait, &wait, task_state);
2280 if (!md_in_flight_bios(md))
2283 if (signal_pending_state(task_state, current)) {
2290 finish_wait(&md->wait, &wait);
2295 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2299 if (!queue_is_mq(md->queue))
2300 return dm_wait_for_bios_completion(md, task_state);
2303 if (!blk_mq_queue_inflight(md->queue))
2306 if (signal_pending_state(task_state, current)) {
2318 * Process the deferred bios
2320 static void dm_wq_work(struct work_struct *work)
2322 struct mapped_device *md = container_of(work, struct mapped_device, work);
2325 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2326 spin_lock_irq(&md->deferred_lock);
2327 bio = bio_list_pop(&md->deferred);
2328 spin_unlock_irq(&md->deferred_lock);
2333 submit_bio_noacct(bio);
2337 static void dm_queue_flush(struct mapped_device *md)
2339 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2340 smp_mb__after_atomic();
2341 queue_work(md->wq, &md->work);
2345 * Swap in a new table, returning the old one for the caller to destroy.
2347 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2349 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2350 struct queue_limits limits;
2353 mutex_lock(&md->suspend_lock);
2355 /* device must be suspended */
2356 if (!dm_suspended_md(md))
2360 * If the new table has no data devices, retain the existing limits.
2361 * This helps multipath with queue_if_no_path if all paths disappear,
2362 * then new I/O is queued based on these limits, and then some paths
2365 if (dm_table_has_no_data_devices(table)) {
2366 live_map = dm_get_live_table_fast(md);
2368 limits = md->queue->limits;
2369 dm_put_live_table_fast(md);
2373 r = dm_calculate_queue_limits(table, &limits);
2380 map = __bind(md, table, &limits);
2381 dm_issue_global_event();
2384 mutex_unlock(&md->suspend_lock);
2389 * Functions to lock and unlock any filesystem running on the
2392 static int lock_fs(struct mapped_device *md)
2396 WARN_ON(md->frozen_sb);
2398 md->frozen_sb = freeze_bdev(md->bdev);
2399 if (IS_ERR(md->frozen_sb)) {
2400 r = PTR_ERR(md->frozen_sb);
2401 md->frozen_sb = NULL;
2405 set_bit(DMF_FROZEN, &md->flags);
2410 static void unlock_fs(struct mapped_device *md)
2412 if (!test_bit(DMF_FROZEN, &md->flags))
2415 thaw_bdev(md->bdev, md->frozen_sb);
2416 md->frozen_sb = NULL;
2417 clear_bit(DMF_FROZEN, &md->flags);
2421 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2422 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2423 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2425 * If __dm_suspend returns 0, the device is completely quiescent
2426 * now. There is no request-processing activity. All new requests
2427 * are being added to md->deferred list.
2429 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2430 unsigned suspend_flags, long task_state,
2431 int dmf_suspended_flag)
2433 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2434 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2437 lockdep_assert_held(&md->suspend_lock);
2440 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2441 * This flag is cleared before dm_suspend returns.
2444 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2446 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2449 * This gets reverted if there's an error later and the targets
2450 * provide the .presuspend_undo hook.
2452 dm_table_presuspend_targets(map);
2455 * Flush I/O to the device.
2456 * Any I/O submitted after lock_fs() may not be flushed.
2457 * noflush takes precedence over do_lockfs.
2458 * (lock_fs() flushes I/Os and waits for them to complete.)
2460 if (!noflush && do_lockfs) {
2463 dm_table_presuspend_undo_targets(map);
2469 * Here we must make sure that no processes are submitting requests
2470 * to target drivers i.e. no one may be executing
2471 * __split_and_process_bio from dm_submit_bio.
2473 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2474 * we take the write lock. To prevent any process from reentering
2475 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2476 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2477 * flush_workqueue(md->wq).
2479 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2481 synchronize_srcu(&md->io_barrier);
2484 * Stop md->queue before flushing md->wq in case request-based
2485 * dm defers requests to md->wq from md->queue.
2487 if (dm_request_based(md))
2488 dm_stop_queue(md->queue);
2490 flush_workqueue(md->wq);
2493 * At this point no more requests are entering target request routines.
2494 * We call dm_wait_for_completion to wait for all existing requests
2497 r = dm_wait_for_completion(md, task_state);
2499 set_bit(dmf_suspended_flag, &md->flags);
2502 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2504 synchronize_srcu(&md->io_barrier);
2506 /* were we interrupted ? */
2510 if (dm_request_based(md))
2511 dm_start_queue(md->queue);
2514 dm_table_presuspend_undo_targets(map);
2515 /* pushback list is already flushed, so skip flush */
2522 * We need to be able to change a mapping table under a mounted
2523 * filesystem. For example we might want to move some data in
2524 * the background. Before the table can be swapped with
2525 * dm_bind_table, dm_suspend must be called to flush any in
2526 * flight bios and ensure that any further io gets deferred.
2529 * Suspend mechanism in request-based dm.
2531 * 1. Flush all I/Os by lock_fs() if needed.
2532 * 2. Stop dispatching any I/O by stopping the request_queue.
2533 * 3. Wait for all in-flight I/Os to be completed or requeued.
2535 * To abort suspend, start the request_queue.
2537 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2539 struct dm_table *map = NULL;
2543 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2545 if (dm_suspended_md(md)) {
2550 if (dm_suspended_internally_md(md)) {
2551 /* already internally suspended, wait for internal resume */
2552 mutex_unlock(&md->suspend_lock);
2553 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2559 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2561 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2565 set_bit(DMF_POST_SUSPENDING, &md->flags);
2566 dm_table_postsuspend_targets(map);
2567 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2570 mutex_unlock(&md->suspend_lock);
2574 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2577 int r = dm_table_resume_targets(map);
2585 * Flushing deferred I/Os must be done after targets are resumed
2586 * so that mapping of targets can work correctly.
2587 * Request-based dm is queueing the deferred I/Os in its request_queue.
2589 if (dm_request_based(md))
2590 dm_start_queue(md->queue);
2597 int dm_resume(struct mapped_device *md)
2600 struct dm_table *map = NULL;
2604 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2606 if (!dm_suspended_md(md))
2609 if (dm_suspended_internally_md(md)) {
2610 /* already internally suspended, wait for internal resume */
2611 mutex_unlock(&md->suspend_lock);
2612 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2618 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2619 if (!map || !dm_table_get_size(map))
2622 r = __dm_resume(md, map);
2626 clear_bit(DMF_SUSPENDED, &md->flags);
2628 mutex_unlock(&md->suspend_lock);
2634 * Internal suspend/resume works like userspace-driven suspend. It waits
2635 * until all bios finish and prevents issuing new bios to the target drivers.
2636 * It may be used only from the kernel.
2639 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2641 struct dm_table *map = NULL;
2643 lockdep_assert_held(&md->suspend_lock);
2645 if (md->internal_suspend_count++)
2646 return; /* nested internal suspend */
2648 if (dm_suspended_md(md)) {
2649 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2650 return; /* nest suspend */
2653 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2656 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2657 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2658 * would require changing .presuspend to return an error -- avoid this
2659 * until there is a need for more elaborate variants of internal suspend.
2661 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2662 DMF_SUSPENDED_INTERNALLY);
2664 set_bit(DMF_POST_SUSPENDING, &md->flags);
2665 dm_table_postsuspend_targets(map);
2666 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2669 static void __dm_internal_resume(struct mapped_device *md)
2671 BUG_ON(!md->internal_suspend_count);
2673 if (--md->internal_suspend_count)
2674 return; /* resume from nested internal suspend */
2676 if (dm_suspended_md(md))
2677 goto done; /* resume from nested suspend */
2680 * NOTE: existing callers don't need to call dm_table_resume_targets
2681 * (which may fail -- so best to avoid it for now by passing NULL map)
2683 (void) __dm_resume(md, NULL);
2686 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2687 smp_mb__after_atomic();
2688 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2691 void dm_internal_suspend_noflush(struct mapped_device *md)
2693 mutex_lock(&md->suspend_lock);
2694 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2695 mutex_unlock(&md->suspend_lock);
2697 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2699 void dm_internal_resume(struct mapped_device *md)
2701 mutex_lock(&md->suspend_lock);
2702 __dm_internal_resume(md);
2703 mutex_unlock(&md->suspend_lock);
2705 EXPORT_SYMBOL_GPL(dm_internal_resume);
2708 * Fast variants of internal suspend/resume hold md->suspend_lock,
2709 * which prevents interaction with userspace-driven suspend.
2712 void dm_internal_suspend_fast(struct mapped_device *md)
2714 mutex_lock(&md->suspend_lock);
2715 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2718 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2719 synchronize_srcu(&md->io_barrier);
2720 flush_workqueue(md->wq);
2721 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2723 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2725 void dm_internal_resume_fast(struct mapped_device *md)
2727 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2733 mutex_unlock(&md->suspend_lock);
2735 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2737 /*-----------------------------------------------------------------
2738 * Event notification.
2739 *---------------------------------------------------------------*/
2740 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2745 char udev_cookie[DM_COOKIE_LENGTH];
2746 char *envp[] = { udev_cookie, NULL };
2748 noio_flag = memalloc_noio_save();
2751 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2753 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2754 DM_COOKIE_ENV_VAR_NAME, cookie);
2755 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2759 memalloc_noio_restore(noio_flag);
2764 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2766 return atomic_add_return(1, &md->uevent_seq);
2769 uint32_t dm_get_event_nr(struct mapped_device *md)
2771 return atomic_read(&md->event_nr);
2774 int dm_wait_event(struct mapped_device *md, int event_nr)
2776 return wait_event_interruptible(md->eventq,
2777 (event_nr != atomic_read(&md->event_nr)));
2780 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2782 unsigned long flags;
2784 spin_lock_irqsave(&md->uevent_lock, flags);
2785 list_add(elist, &md->uevent_list);
2786 spin_unlock_irqrestore(&md->uevent_lock, flags);
2790 * The gendisk is only valid as long as you have a reference
2793 struct gendisk *dm_disk(struct mapped_device *md)
2797 EXPORT_SYMBOL_GPL(dm_disk);
2799 struct kobject *dm_kobject(struct mapped_device *md)
2801 return &md->kobj_holder.kobj;
2804 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2806 struct mapped_device *md;
2808 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2810 spin_lock(&_minor_lock);
2811 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2817 spin_unlock(&_minor_lock);
2822 int dm_suspended_md(struct mapped_device *md)
2824 return test_bit(DMF_SUSPENDED, &md->flags);
2827 static int dm_post_suspending_md(struct mapped_device *md)
2829 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2832 int dm_suspended_internally_md(struct mapped_device *md)
2834 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2837 int dm_test_deferred_remove_flag(struct mapped_device *md)
2839 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2842 int dm_suspended(struct dm_target *ti)
2844 return dm_suspended_md(ti->table->md);
2846 EXPORT_SYMBOL_GPL(dm_suspended);
2848 int dm_post_suspending(struct dm_target *ti)
2850 return dm_post_suspending_md(ti->table->md);
2852 EXPORT_SYMBOL_GPL(dm_post_suspending);
2854 int dm_noflush_suspending(struct dm_target *ti)
2856 return __noflush_suspending(ti->table->md);
2858 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2860 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2861 unsigned integrity, unsigned per_io_data_size,
2862 unsigned min_pool_size)
2864 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2865 unsigned int pool_size = 0;
2866 unsigned int front_pad, io_front_pad;
2873 case DM_TYPE_BIO_BASED:
2874 case DM_TYPE_DAX_BIO_BASED:
2875 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2876 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2877 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2878 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2881 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2884 case DM_TYPE_REQUEST_BASED:
2885 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2886 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2887 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2893 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2897 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2903 dm_free_md_mempools(pools);
2908 void dm_free_md_mempools(struct dm_md_mempools *pools)
2913 bioset_exit(&pools->bs);
2914 bioset_exit(&pools->io_bs);
2926 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2929 struct mapped_device *md = bdev->bd_disk->private_data;
2930 struct dm_table *table;
2931 struct dm_target *ti;
2932 int ret = -ENOTTY, srcu_idx;
2934 table = dm_get_live_table(md, &srcu_idx);
2935 if (!table || !dm_table_get_size(table))
2938 /* We only support devices that have a single target */
2939 if (dm_table_get_num_targets(table) != 1)
2941 ti = dm_table_get_target(table, 0);
2944 if (!ti->type->iterate_devices)
2947 ret = ti->type->iterate_devices(ti, fn, data);
2949 dm_put_live_table(md, srcu_idx);
2954 * For register / unregister we need to manually call out to every path.
2956 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2957 sector_t start, sector_t len, void *data)
2959 struct dm_pr *pr = data;
2960 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2962 if (!ops || !ops->pr_register)
2964 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2967 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2978 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2979 if (ret && new_key) {
2980 /* unregister all paths if we failed to register any path */
2981 pr.old_key = new_key;
2984 pr.fail_early = false;
2985 dm_call_pr(bdev, __dm_pr_register, &pr);
2991 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2994 struct mapped_device *md = bdev->bd_disk->private_data;
2995 const struct pr_ops *ops;
2998 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3002 ops = bdev->bd_disk->fops->pr_ops;
3003 if (ops && ops->pr_reserve)
3004 r = ops->pr_reserve(bdev, key, type, flags);
3008 dm_unprepare_ioctl(md, srcu_idx);
3012 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3014 struct mapped_device *md = bdev->bd_disk->private_data;
3015 const struct pr_ops *ops;
3018 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3022 ops = bdev->bd_disk->fops->pr_ops;
3023 if (ops && ops->pr_release)
3024 r = ops->pr_release(bdev, key, type);
3028 dm_unprepare_ioctl(md, srcu_idx);
3032 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3033 enum pr_type type, bool abort)
3035 struct mapped_device *md = bdev->bd_disk->private_data;
3036 const struct pr_ops *ops;
3039 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3043 ops = bdev->bd_disk->fops->pr_ops;
3044 if (ops && ops->pr_preempt)
3045 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3049 dm_unprepare_ioctl(md, srcu_idx);
3053 static int dm_pr_clear(struct block_device *bdev, u64 key)
3055 struct mapped_device *md = bdev->bd_disk->private_data;
3056 const struct pr_ops *ops;
3059 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3063 ops = bdev->bd_disk->fops->pr_ops;
3064 if (ops && ops->pr_clear)
3065 r = ops->pr_clear(bdev, key);
3069 dm_unprepare_ioctl(md, srcu_idx);
3073 static const struct pr_ops dm_pr_ops = {
3074 .pr_register = dm_pr_register,
3075 .pr_reserve = dm_pr_reserve,
3076 .pr_release = dm_pr_release,
3077 .pr_preempt = dm_pr_preempt,
3078 .pr_clear = dm_pr_clear,
3081 static const struct block_device_operations dm_blk_dops = {
3082 .submit_bio = dm_submit_bio,
3083 .open = dm_blk_open,
3084 .release = dm_blk_close,
3085 .ioctl = dm_blk_ioctl,
3086 .getgeo = dm_blk_getgeo,
3087 .report_zones = dm_blk_report_zones,
3088 .pr_ops = &dm_pr_ops,
3089 .owner = THIS_MODULE
3092 static const struct block_device_operations dm_rq_blk_dops = {
3093 .open = dm_blk_open,
3094 .release = dm_blk_close,
3095 .ioctl = dm_blk_ioctl,
3096 .getgeo = dm_blk_getgeo,
3097 .pr_ops = &dm_pr_ops,
3098 .owner = THIS_MODULE
3101 static const struct dax_operations dm_dax_ops = {
3102 .direct_access = dm_dax_direct_access,
3103 .dax_supported = dm_dax_supported,
3104 .copy_from_iter = dm_dax_copy_from_iter,
3105 .copy_to_iter = dm_dax_copy_to_iter,
3106 .zero_page_range = dm_dax_zero_page_range,
3112 module_init(dm_init);
3113 module_exit(dm_exit);
3115 module_param(major, uint, 0);
3116 MODULE_PARM_DESC(major, "The major number of the device mapper");
3118 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3119 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3121 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3122 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3124 MODULE_DESCRIPTION(DM_NAME " driver");
3125 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3126 MODULE_LICENSE("GPL");