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 sector_t dm_get_size(struct mapped_device *md)
427 return get_capacity(md->disk);
430 struct request_queue *dm_get_md_queue(struct mapped_device *md)
435 struct dm_stats *dm_get_stats(struct mapped_device *md)
440 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
442 struct mapped_device *md = bdev->bd_disk->private_data;
444 return dm_get_geometry(md, geo);
447 #ifdef CONFIG_BLK_DEV_ZONED
448 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
450 struct dm_report_zones_args *args = data;
451 sector_t sector_diff = args->tgt->begin - args->start;
454 * Ignore zones beyond the target range.
456 if (zone->start >= args->start + args->tgt->len)
460 * Remap the start sector and write pointer position of the zone
461 * to match its position in the target range.
463 zone->start += sector_diff;
464 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
465 if (zone->cond == BLK_ZONE_COND_FULL)
466 zone->wp = zone->start + zone->len;
467 else if (zone->cond == BLK_ZONE_COND_EMPTY)
468 zone->wp = zone->start;
470 zone->wp += sector_diff;
473 args->next_sector = zone->start + zone->len;
474 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
476 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
478 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
479 unsigned int nr_zones, report_zones_cb cb, void *data)
481 struct mapped_device *md = disk->private_data;
482 struct dm_table *map;
484 struct dm_report_zones_args args = {
485 .next_sector = sector,
490 if (dm_suspended_md(md))
493 map = dm_get_live_table(md, &srcu_idx);
498 struct dm_target *tgt;
500 tgt = dm_table_find_target(map, args.next_sector);
501 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
507 ret = tgt->type->report_zones(tgt, &args,
508 nr_zones - args.zone_idx);
511 } while (args.zone_idx < nr_zones &&
512 args.next_sector < get_capacity(disk));
516 dm_put_live_table(md, srcu_idx);
520 #define dm_blk_report_zones NULL
521 #endif /* CONFIG_BLK_DEV_ZONED */
523 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
524 struct block_device **bdev)
525 __acquires(md->io_barrier)
527 struct dm_target *tgt;
528 struct dm_table *map;
533 map = dm_get_live_table(md, srcu_idx);
534 if (!map || !dm_table_get_size(map))
537 /* We only support devices that have a single target */
538 if (dm_table_get_num_targets(map) != 1)
541 tgt = dm_table_get_target(map, 0);
542 if (!tgt->type->prepare_ioctl)
545 if (dm_suspended_md(md))
548 r = tgt->type->prepare_ioctl(tgt, bdev);
549 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
550 dm_put_live_table(md, *srcu_idx);
558 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
559 __releases(md->io_barrier)
561 dm_put_live_table(md, srcu_idx);
564 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
565 unsigned int cmd, unsigned long arg)
567 struct mapped_device *md = bdev->bd_disk->private_data;
570 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
576 * Target determined this ioctl is being issued against a
577 * subset of the parent bdev; require extra privileges.
579 if (!capable(CAP_SYS_RAWIO)) {
581 "%s: sending ioctl %x to DM device without required privilege.",
588 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
590 dm_unprepare_ioctl(md, srcu_idx);
594 static void start_io_acct(struct dm_io *io);
596 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
599 struct dm_target_io *tio;
602 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
606 tio = container_of(clone, struct dm_target_io, clone);
607 tio->inside_dm_io = true;
610 io = container_of(tio, struct dm_io, tio);
611 io->magic = DM_IO_MAGIC;
613 atomic_set(&io->io_count, 1);
616 spin_lock_init(&io->endio_lock);
623 static void free_io(struct mapped_device *md, struct dm_io *io)
625 bio_put(&io->tio.clone);
628 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
629 unsigned target_bio_nr, gfp_t gfp_mask)
631 struct dm_target_io *tio;
633 if (!ci->io->tio.io) {
634 /* the dm_target_io embedded in ci->io is available */
637 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
641 tio = container_of(clone, struct dm_target_io, clone);
642 tio->inside_dm_io = false;
645 tio->magic = DM_TIO_MAGIC;
648 tio->target_bio_nr = target_bio_nr;
653 static void free_tio(struct dm_target_io *tio)
655 if (tio->inside_dm_io)
657 bio_put(&tio->clone);
660 u64 dm_start_time_ns_from_clone(struct bio *bio)
662 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
663 struct dm_io *io = tio->io;
665 return jiffies_to_nsecs(io->start_time);
667 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
669 static void start_io_acct(struct dm_io *io)
671 struct mapped_device *md = io->md;
672 struct bio *bio = io->orig_bio;
674 io->start_time = bio_start_io_acct(bio);
675 if (unlikely(dm_stats_used(&md->stats)))
676 dm_stats_account_io(&md->stats, bio_data_dir(bio),
677 bio->bi_iter.bi_sector, bio_sectors(bio),
678 false, 0, &io->stats_aux);
681 static void end_io_acct(struct dm_io *io)
683 struct mapped_device *md = io->md;
684 struct bio *bio = io->orig_bio;
685 unsigned long duration = jiffies - io->start_time;
687 bio_end_io_acct(bio, io->start_time);
689 if (unlikely(dm_stats_used(&md->stats)))
690 dm_stats_account_io(&md->stats, bio_data_dir(bio),
691 bio->bi_iter.bi_sector, bio_sectors(bio),
692 true, duration, &io->stats_aux);
694 /* nudge anyone waiting on suspend queue */
695 if (unlikely(wq_has_sleeper(&md->wait)))
700 * Add the bio to the list of deferred io.
702 static void queue_io(struct mapped_device *md, struct bio *bio)
706 spin_lock_irqsave(&md->deferred_lock, flags);
707 bio_list_add(&md->deferred, bio);
708 spin_unlock_irqrestore(&md->deferred_lock, flags);
709 queue_work(md->wq, &md->work);
713 * Everyone (including functions in this file), should use this
714 * function to access the md->map field, and make sure they call
715 * dm_put_live_table() when finished.
717 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
719 *srcu_idx = srcu_read_lock(&md->io_barrier);
721 return srcu_dereference(md->map, &md->io_barrier);
724 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
726 srcu_read_unlock(&md->io_barrier, srcu_idx);
729 void dm_sync_table(struct mapped_device *md)
731 synchronize_srcu(&md->io_barrier);
732 synchronize_rcu_expedited();
736 * A fast alternative to dm_get_live_table/dm_put_live_table.
737 * The caller must not block between these two functions.
739 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
742 return rcu_dereference(md->map);
745 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
750 static char *_dm_claim_ptr = "I belong to device-mapper";
753 * Open a table device so we can use it as a map destination.
755 static int open_table_device(struct table_device *td, dev_t dev,
756 struct mapped_device *md)
758 struct block_device *bdev;
762 BUG_ON(td->dm_dev.bdev);
764 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
766 return PTR_ERR(bdev);
768 r = bd_link_disk_holder(bdev, dm_disk(md));
770 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
774 td->dm_dev.bdev = bdev;
775 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
780 * Close a table device that we've been using.
782 static void close_table_device(struct table_device *td, struct mapped_device *md)
784 if (!td->dm_dev.bdev)
787 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
788 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
789 put_dax(td->dm_dev.dax_dev);
790 td->dm_dev.bdev = NULL;
791 td->dm_dev.dax_dev = NULL;
794 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
797 struct table_device *td;
799 list_for_each_entry(td, l, list)
800 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
806 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
807 struct dm_dev **result)
810 struct table_device *td;
812 mutex_lock(&md->table_devices_lock);
813 td = find_table_device(&md->table_devices, dev, mode);
815 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
817 mutex_unlock(&md->table_devices_lock);
821 td->dm_dev.mode = mode;
822 td->dm_dev.bdev = NULL;
824 if ((r = open_table_device(td, dev, md))) {
825 mutex_unlock(&md->table_devices_lock);
830 format_dev_t(td->dm_dev.name, dev);
832 refcount_set(&td->count, 1);
833 list_add(&td->list, &md->table_devices);
835 refcount_inc(&td->count);
837 mutex_unlock(&md->table_devices_lock);
839 *result = &td->dm_dev;
842 EXPORT_SYMBOL_GPL(dm_get_table_device);
844 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
846 struct table_device *td = container_of(d, struct table_device, dm_dev);
848 mutex_lock(&md->table_devices_lock);
849 if (refcount_dec_and_test(&td->count)) {
850 close_table_device(td, md);
854 mutex_unlock(&md->table_devices_lock);
856 EXPORT_SYMBOL(dm_put_table_device);
858 static void free_table_devices(struct list_head *devices)
860 struct list_head *tmp, *next;
862 list_for_each_safe(tmp, next, devices) {
863 struct table_device *td = list_entry(tmp, struct table_device, list);
865 DMWARN("dm_destroy: %s still exists with %d references",
866 td->dm_dev.name, refcount_read(&td->count));
872 * Get the geometry associated with a dm device
874 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
882 * Set the geometry of a device.
884 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
886 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
888 if (geo->start > sz) {
889 DMWARN("Start sector is beyond the geometry limits.");
898 static int __noflush_suspending(struct mapped_device *md)
900 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
904 * Decrements the number of outstanding ios that a bio has been
905 * cloned into, completing the original io if necc.
907 static void dec_pending(struct dm_io *io, blk_status_t error)
910 blk_status_t io_error;
912 struct mapped_device *md = io->md;
914 /* Push-back supersedes any I/O errors */
915 if (unlikely(error)) {
916 spin_lock_irqsave(&io->endio_lock, flags);
917 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
919 spin_unlock_irqrestore(&io->endio_lock, flags);
922 if (atomic_dec_and_test(&io->io_count)) {
923 if (io->status == BLK_STS_DM_REQUEUE) {
925 * Target requested pushing back the I/O.
927 spin_lock_irqsave(&md->deferred_lock, flags);
928 if (__noflush_suspending(md))
929 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
930 bio_list_add_head(&md->deferred, io->orig_bio);
932 /* noflush suspend was interrupted. */
933 io->status = BLK_STS_IOERR;
934 spin_unlock_irqrestore(&md->deferred_lock, flags);
937 io_error = io->status;
942 if (io_error == BLK_STS_DM_REQUEUE)
945 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
947 * Preflush done for flush with data, reissue
948 * without REQ_PREFLUSH.
950 bio->bi_opf &= ~REQ_PREFLUSH;
953 /* done with normal IO or empty flush */
955 bio->bi_status = io_error;
961 void disable_discard(struct mapped_device *md)
963 struct queue_limits *limits = dm_get_queue_limits(md);
965 /* device doesn't really support DISCARD, disable it */
966 limits->max_discard_sectors = 0;
967 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
970 void disable_write_same(struct mapped_device *md)
972 struct queue_limits *limits = dm_get_queue_limits(md);
974 /* device doesn't really support WRITE SAME, disable it */
975 limits->max_write_same_sectors = 0;
978 void disable_write_zeroes(struct mapped_device *md)
980 struct queue_limits *limits = dm_get_queue_limits(md);
982 /* device doesn't really support WRITE ZEROES, disable it */
983 limits->max_write_zeroes_sectors = 0;
986 static void clone_endio(struct bio *bio)
988 blk_status_t error = bio->bi_status;
989 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
990 struct dm_io *io = tio->io;
991 struct mapped_device *md = tio->io->md;
992 dm_endio_fn endio = tio->ti->type->end_io;
993 struct bio *orig_bio = io->orig_bio;
995 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
996 if (bio_op(bio) == REQ_OP_DISCARD &&
997 !bio->bi_disk->queue->limits.max_discard_sectors)
999 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1000 !bio->bi_disk->queue->limits.max_write_same_sectors)
1001 disable_write_same(md);
1002 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1003 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1004 disable_write_zeroes(md);
1008 * For zone-append bios get offset in zone of the written
1009 * sector and add that to the original bio sector pos.
1011 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1012 sector_t written_sector = bio->bi_iter.bi_sector;
1013 struct request_queue *q = orig_bio->bi_disk->queue;
1014 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1016 orig_bio->bi_iter.bi_sector += written_sector & mask;
1020 int r = endio(tio->ti, bio, &error);
1022 case DM_ENDIO_REQUEUE:
1023 error = BLK_STS_DM_REQUEUE;
1027 case DM_ENDIO_INCOMPLETE:
1028 /* The target will handle the io */
1031 DMWARN("unimplemented target endio return value: %d", r);
1037 dec_pending(io, error);
1041 * Return maximum size of I/O possible at the supplied sector up to the current
1044 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1046 sector_t target_offset = dm_target_offset(ti, sector);
1048 return ti->len - target_offset;
1051 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1053 sector_t len = max_io_len_target_boundary(sector, ti);
1057 * Does the target need to split even further?
1058 * - q->limits.chunk_sectors reflects ti->max_io_len so
1059 * blk_max_size_offset() provides required splitting.
1060 * - blk_max_size_offset() also respects q->limits.max_sectors
1062 max_len = blk_max_size_offset(dm_table_get_md(ti->table)->queue,
1063 dm_target_offset(ti, sector));
1070 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1072 if (len > UINT_MAX) {
1073 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1074 (unsigned long long)len, UINT_MAX);
1075 ti->error = "Maximum size of target IO is too large";
1079 ti->max_io_len = (uint32_t) len;
1083 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1085 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1086 sector_t sector, int *srcu_idx)
1087 __acquires(md->io_barrier)
1089 struct dm_table *map;
1090 struct dm_target *ti;
1092 map = dm_get_live_table(md, srcu_idx);
1096 ti = dm_table_find_target(map, sector);
1103 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1104 long nr_pages, void **kaddr, pfn_t *pfn)
1106 struct mapped_device *md = dax_get_private(dax_dev);
1107 sector_t sector = pgoff * PAGE_SECTORS;
1108 struct dm_target *ti;
1109 long len, ret = -EIO;
1112 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1116 if (!ti->type->direct_access)
1118 len = max_io_len(sector, ti) / PAGE_SECTORS;
1121 nr_pages = min(len, nr_pages);
1122 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1125 dm_put_live_table(md, srcu_idx);
1130 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1131 int blocksize, sector_t start, sector_t len)
1133 struct mapped_device *md = dax_get_private(dax_dev);
1134 struct dm_table *map;
1138 map = dm_get_live_table(md, &srcu_idx);
1142 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1145 dm_put_live_table(md, srcu_idx);
1150 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1151 void *addr, size_t bytes, struct iov_iter *i)
1153 struct mapped_device *md = dax_get_private(dax_dev);
1154 sector_t sector = pgoff * PAGE_SECTORS;
1155 struct dm_target *ti;
1159 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1163 if (!ti->type->dax_copy_from_iter) {
1164 ret = copy_from_iter(addr, bytes, i);
1167 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1169 dm_put_live_table(md, srcu_idx);
1174 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1175 void *addr, size_t bytes, struct iov_iter *i)
1177 struct mapped_device *md = dax_get_private(dax_dev);
1178 sector_t sector = pgoff * PAGE_SECTORS;
1179 struct dm_target *ti;
1183 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1187 if (!ti->type->dax_copy_to_iter) {
1188 ret = copy_to_iter(addr, bytes, i);
1191 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1193 dm_put_live_table(md, srcu_idx);
1198 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1201 struct mapped_device *md = dax_get_private(dax_dev);
1202 sector_t sector = pgoff * PAGE_SECTORS;
1203 struct dm_target *ti;
1207 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1211 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1213 * ->zero_page_range() is mandatory dax operation. If we are
1214 * here, something is wrong.
1216 dm_put_live_table(md, srcu_idx);
1219 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1222 dm_put_live_table(md, srcu_idx);
1228 * A target may call dm_accept_partial_bio only from the map routine. It is
1229 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1230 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1232 * dm_accept_partial_bio informs the dm that the target only wants to process
1233 * additional n_sectors sectors of the bio and the rest of the data should be
1234 * sent in a next bio.
1236 * A diagram that explains the arithmetics:
1237 * +--------------------+---------------+-------+
1239 * +--------------------+---------------+-------+
1241 * <-------------- *tio->len_ptr --------------->
1242 * <------- bi_size ------->
1245 * Region 1 was already iterated over with bio_advance or similar function.
1246 * (it may be empty if the target doesn't use bio_advance)
1247 * Region 2 is the remaining bio size that the target wants to process.
1248 * (it may be empty if region 1 is non-empty, although there is no reason
1250 * The target requires that region 3 is to be sent in the next bio.
1252 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1253 * the partially processed part (the sum of regions 1+2) must be the same for all
1254 * copies of the bio.
1256 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1258 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1259 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1260 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1261 BUG_ON(bi_size > *tio->len_ptr);
1262 BUG_ON(n_sectors > bi_size);
1263 *tio->len_ptr -= bi_size - n_sectors;
1264 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1266 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1268 static blk_qc_t __map_bio(struct dm_target_io *tio)
1272 struct bio *clone = &tio->clone;
1273 struct dm_io *io = tio->io;
1274 struct dm_target *ti = tio->ti;
1275 blk_qc_t ret = BLK_QC_T_NONE;
1277 clone->bi_end_io = clone_endio;
1280 * Map the clone. If r == 0 we don't need to do
1281 * anything, the target has assumed ownership of
1284 atomic_inc(&io->io_count);
1285 sector = clone->bi_iter.bi_sector;
1287 r = ti->type->map(ti, clone);
1289 case DM_MAPIO_SUBMITTED:
1291 case DM_MAPIO_REMAPPED:
1292 /* the bio has been remapped so dispatch it */
1293 trace_block_bio_remap(clone->bi_disk->queue, clone,
1294 bio_dev(io->orig_bio), sector);
1295 ret = submit_bio_noacct(clone);
1299 dec_pending(io, BLK_STS_IOERR);
1301 case DM_MAPIO_REQUEUE:
1303 dec_pending(io, BLK_STS_DM_REQUEUE);
1306 DMWARN("unimplemented target map return value: %d", r);
1313 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1315 bio->bi_iter.bi_sector = sector;
1316 bio->bi_iter.bi_size = to_bytes(len);
1320 * Creates a bio that consists of range of complete bvecs.
1322 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1323 sector_t sector, unsigned len)
1325 struct bio *clone = &tio->clone;
1327 __bio_clone_fast(clone, bio);
1329 bio_crypt_clone(clone, bio, GFP_NOIO);
1331 if (bio_integrity(bio)) {
1334 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1335 !dm_target_passes_integrity(tio->ti->type))) {
1336 DMWARN("%s: the target %s doesn't support integrity data.",
1337 dm_device_name(tio->io->md),
1338 tio->ti->type->name);
1342 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1347 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1348 clone->bi_iter.bi_size = to_bytes(len);
1350 if (bio_integrity(bio))
1351 bio_integrity_trim(clone);
1356 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1357 struct dm_target *ti, unsigned num_bios)
1359 struct dm_target_io *tio;
1365 if (num_bios == 1) {
1366 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1367 bio_list_add(blist, &tio->clone);
1371 for (try = 0; try < 2; try++) {
1376 mutex_lock(&ci->io->md->table_devices_lock);
1377 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1378 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1382 bio_list_add(blist, &tio->clone);
1385 mutex_unlock(&ci->io->md->table_devices_lock);
1386 if (bio_nr == num_bios)
1389 while ((bio = bio_list_pop(blist))) {
1390 tio = container_of(bio, struct dm_target_io, clone);
1396 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1397 struct dm_target_io *tio, unsigned *len)
1399 struct bio *clone = &tio->clone;
1403 __bio_clone_fast(clone, ci->bio);
1405 bio_setup_sector(clone, ci->sector, *len);
1407 return __map_bio(tio);
1410 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1411 unsigned num_bios, unsigned *len)
1413 struct bio_list blist = BIO_EMPTY_LIST;
1415 struct dm_target_io *tio;
1417 alloc_multiple_bios(&blist, ci, ti, num_bios);
1419 while ((bio = bio_list_pop(&blist))) {
1420 tio = container_of(bio, struct dm_target_io, clone);
1421 (void) __clone_and_map_simple_bio(ci, tio, len);
1425 static int __send_empty_flush(struct clone_info *ci)
1427 unsigned target_nr = 0;
1428 struct dm_target *ti;
1431 * Empty flush uses a statically initialized bio, as the base for
1432 * cloning. However, blkg association requires that a bdev is
1433 * associated with a gendisk, which doesn't happen until the bdev is
1434 * opened. So, blkg association is done at issue time of the flush
1435 * rather than when the device is created in alloc_dev().
1437 bio_set_dev(ci->bio, ci->io->md->bdev);
1439 BUG_ON(bio_has_data(ci->bio));
1440 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1441 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
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 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1466 static unsigned get_num_discard_bios(struct dm_target *ti)
1468 return ti->num_discard_bios;
1471 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1473 return ti->num_secure_erase_bios;
1476 static unsigned get_num_write_same_bios(struct dm_target *ti)
1478 return ti->num_write_same_bios;
1481 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1483 return ti->num_write_zeroes_bios;
1486 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1492 * Even though the device advertised support for this type of
1493 * request, that does not mean every target supports it, and
1494 * reconfiguration might also have changed that since the
1495 * check was performed.
1500 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1502 __send_duplicate_bios(ci, ti, num_bios, &len);
1505 ci->sector_count -= len;
1510 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1512 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1515 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1517 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1520 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1522 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1525 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1527 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1530 static bool is_abnormal_io(struct bio *bio)
1534 switch (bio_op(bio)) {
1535 case REQ_OP_DISCARD:
1536 case REQ_OP_SECURE_ERASE:
1537 case REQ_OP_WRITE_SAME:
1538 case REQ_OP_WRITE_ZEROES:
1546 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1549 struct bio *bio = ci->bio;
1551 if (bio_op(bio) == REQ_OP_DISCARD)
1552 *result = __send_discard(ci, ti);
1553 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1554 *result = __send_secure_erase(ci, ti);
1555 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1556 *result = __send_write_same(ci, ti);
1557 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1558 *result = __send_write_zeroes(ci, ti);
1566 * Select the correct strategy for processing a non-flush bio.
1568 static int __split_and_process_non_flush(struct clone_info *ci)
1570 struct dm_target *ti;
1574 ti = dm_table_find_target(ci->map, ci->sector);
1578 if (__process_abnormal_io(ci, ti, &r))
1581 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1583 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1588 ci->sector_count -= len;
1593 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1594 struct dm_table *map, struct bio *bio)
1597 ci->io = alloc_io(md, bio);
1598 ci->sector = bio->bi_iter.bi_sector;
1601 #define __dm_part_stat_sub(part, field, subnd) \
1602 (part_stat_get(part, field) -= (subnd))
1605 * Entry point to split a bio into clones and submit them to the targets.
1607 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1608 struct dm_table *map, struct bio *bio)
1610 struct clone_info ci;
1611 blk_qc_t ret = BLK_QC_T_NONE;
1614 init_clone_info(&ci, md, map, bio);
1616 if (bio->bi_opf & REQ_PREFLUSH) {
1617 struct bio flush_bio;
1620 * Use an on-stack bio for this, it's safe since we don't
1621 * need to reference it after submit. It's just used as
1622 * the basis for the clone(s).
1624 bio_init(&flush_bio, NULL, 0);
1625 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1626 ci.bio = &flush_bio;
1627 ci.sector_count = 0;
1628 error = __send_empty_flush(&ci);
1630 /* dec_pending submits any data associated with flush */
1631 } else if (op_is_zone_mgmt(bio_op(bio))) {
1633 ci.sector_count = 0;
1634 error = __split_and_process_non_flush(&ci);
1637 ci.sector_count = bio_sectors(bio);
1638 while (ci.sector_count && !error) {
1639 error = __split_and_process_non_flush(&ci);
1640 if (current->bio_list && ci.sector_count && !error) {
1642 * Remainder must be passed to submit_bio_noacct()
1643 * so that it gets handled *after* bios already submitted
1644 * have been completely processed.
1645 * We take a clone of the original to store in
1646 * ci.io->orig_bio to be used by end_io_acct() and
1647 * for dec_pending to use for completion handling.
1649 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1650 GFP_NOIO, &md->queue->bio_split);
1651 ci.io->orig_bio = b;
1654 * Adjust IO stats for each split, otherwise upon queue
1655 * reentry there will be redundant IO accounting.
1656 * NOTE: this is a stop-gap fix, a proper fix involves
1657 * significant refactoring of DM core's bio splitting
1658 * (by eliminating DM's splitting and just using bio_split)
1661 __dm_part_stat_sub(&dm_disk(md)->part0,
1662 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1666 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1667 ret = submit_bio_noacct(bio);
1673 /* drop the extra reference count */
1674 dec_pending(ci.io, errno_to_blk_status(error));
1679 * Optimized variant of __split_and_process_bio that leverages the
1680 * fact that targets that use it do _not_ have a need to split bios.
1682 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1683 struct bio *bio, struct dm_target *ti)
1685 struct clone_info ci;
1686 blk_qc_t ret = BLK_QC_T_NONE;
1689 init_clone_info(&ci, md, map, bio);
1691 if (bio->bi_opf & REQ_PREFLUSH) {
1692 struct bio flush_bio;
1695 * Use an on-stack bio for this, it's safe since we don't
1696 * need to reference it after submit. It's just used as
1697 * the basis for the clone(s).
1699 bio_init(&flush_bio, NULL, 0);
1700 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1701 ci.bio = &flush_bio;
1702 ci.sector_count = 0;
1703 error = __send_empty_flush(&ci);
1705 /* dec_pending submits any data associated with flush */
1707 struct dm_target_io *tio;
1710 ci.sector_count = bio_sectors(bio);
1711 if (__process_abnormal_io(&ci, ti, &error))
1714 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1715 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1718 /* drop the extra reference count */
1719 dec_pending(ci.io, errno_to_blk_status(error));
1723 static blk_qc_t dm_process_bio(struct mapped_device *md,
1724 struct dm_table *map, struct bio *bio)
1726 blk_qc_t ret = BLK_QC_T_NONE;
1727 struct dm_target *ti = md->immutable_target;
1729 if (unlikely(!map)) {
1735 ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1736 if (unlikely(!ti)) {
1743 * If in ->submit_bio we need to use blk_queue_split(), otherwise
1744 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1746 * If called from dm_wq_work() for deferred bio processing, bio
1747 * was already handled by following code with previous ->submit_bio.
1749 if (current->bio_list) {
1750 if (is_abnormal_io(bio))
1751 blk_queue_split(&bio);
1752 /* regular IO is split by __split_and_process_bio */
1755 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1756 return __process_bio(md, map, bio, ti);
1757 return __split_and_process_bio(md, map, bio);
1760 static blk_qc_t dm_submit_bio(struct bio *bio)
1762 struct mapped_device *md = bio->bi_disk->private_data;
1763 blk_qc_t ret = BLK_QC_T_NONE;
1765 struct dm_table *map;
1767 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
1769 * We are called with a live reference on q_usage_counter, but
1770 * that one will be released as soon as we return. Grab an
1771 * extra one as blk_mq_submit_bio expects to be able to consume
1772 * a reference (which lives until the request is freed in case a
1773 * request is allocated).
1775 percpu_ref_get(&bio->bi_disk->queue->q_usage_counter);
1776 return blk_mq_submit_bio(bio);
1779 map = dm_get_live_table(md, &srcu_idx);
1781 /* if we're suspended, we have to queue this io for later */
1782 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1783 dm_put_live_table(md, srcu_idx);
1785 if (bio->bi_opf & REQ_NOWAIT)
1786 bio_wouldblock_error(bio);
1787 else if (!(bio->bi_opf & REQ_RAHEAD))
1794 ret = dm_process_bio(md, map, bio);
1796 dm_put_live_table(md, srcu_idx);
1800 /*-----------------------------------------------------------------
1801 * An IDR is used to keep track of allocated minor numbers.
1802 *---------------------------------------------------------------*/
1803 static void free_minor(int minor)
1805 spin_lock(&_minor_lock);
1806 idr_remove(&_minor_idr, minor);
1807 spin_unlock(&_minor_lock);
1811 * See if the device with a specific minor # is free.
1813 static int specific_minor(int minor)
1817 if (minor >= (1 << MINORBITS))
1820 idr_preload(GFP_KERNEL);
1821 spin_lock(&_minor_lock);
1823 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1825 spin_unlock(&_minor_lock);
1828 return r == -ENOSPC ? -EBUSY : r;
1832 static int next_free_minor(int *minor)
1836 idr_preload(GFP_KERNEL);
1837 spin_lock(&_minor_lock);
1839 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1841 spin_unlock(&_minor_lock);
1849 static const struct block_device_operations dm_blk_dops;
1850 static const struct dax_operations dm_dax_ops;
1852 static void dm_wq_work(struct work_struct *work);
1854 static void cleanup_mapped_device(struct mapped_device *md)
1857 destroy_workqueue(md->wq);
1858 bioset_exit(&md->bs);
1859 bioset_exit(&md->io_bs);
1862 kill_dax(md->dax_dev);
1863 put_dax(md->dax_dev);
1868 spin_lock(&_minor_lock);
1869 md->disk->private_data = NULL;
1870 spin_unlock(&_minor_lock);
1871 del_gendisk(md->disk);
1876 blk_cleanup_queue(md->queue);
1878 cleanup_srcu_struct(&md->io_barrier);
1885 mutex_destroy(&md->suspend_lock);
1886 mutex_destroy(&md->type_lock);
1887 mutex_destroy(&md->table_devices_lock);
1889 dm_mq_cleanup_mapped_device(md);
1893 * Allocate and initialise a blank device with a given minor.
1895 static struct mapped_device *alloc_dev(int minor)
1897 int r, numa_node_id = dm_get_numa_node();
1898 struct mapped_device *md;
1901 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1903 DMWARN("unable to allocate device, out of memory.");
1907 if (!try_module_get(THIS_MODULE))
1908 goto bad_module_get;
1910 /* get a minor number for the dev */
1911 if (minor == DM_ANY_MINOR)
1912 r = next_free_minor(&minor);
1914 r = specific_minor(minor);
1918 r = init_srcu_struct(&md->io_barrier);
1920 goto bad_io_barrier;
1922 md->numa_node_id = numa_node_id;
1923 md->init_tio_pdu = false;
1924 md->type = DM_TYPE_NONE;
1925 mutex_init(&md->suspend_lock);
1926 mutex_init(&md->type_lock);
1927 mutex_init(&md->table_devices_lock);
1928 spin_lock_init(&md->deferred_lock);
1929 atomic_set(&md->holders, 1);
1930 atomic_set(&md->open_count, 0);
1931 atomic_set(&md->event_nr, 0);
1932 atomic_set(&md->uevent_seq, 0);
1933 INIT_LIST_HEAD(&md->uevent_list);
1934 INIT_LIST_HEAD(&md->table_devices);
1935 spin_lock_init(&md->uevent_lock);
1938 * default to bio-based until DM table is loaded and md->type
1939 * established. If request-based table is loaded: blk-mq will
1940 * override accordingly.
1942 md->queue = blk_alloc_queue(numa_node_id);
1946 md->disk = alloc_disk_node(1, md->numa_node_id);
1950 init_waitqueue_head(&md->wait);
1951 INIT_WORK(&md->work, dm_wq_work);
1952 init_waitqueue_head(&md->eventq);
1953 init_completion(&md->kobj_holder.completion);
1955 md->disk->major = _major;
1956 md->disk->first_minor = minor;
1957 md->disk->fops = &dm_blk_dops;
1958 md->disk->queue = md->queue;
1959 md->disk->private_data = md;
1960 sprintf(md->disk->disk_name, "dm-%d", minor);
1962 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1963 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1965 if (IS_ERR(md->dax_dev))
1969 add_disk_no_queue_reg(md->disk);
1970 format_dev_t(md->name, MKDEV(_major, minor));
1972 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1976 md->bdev = bdget_disk(md->disk, 0);
1980 dm_stats_init(&md->stats);
1982 /* Populate the mapping, nobody knows we exist yet */
1983 spin_lock(&_minor_lock);
1984 old_md = idr_replace(&_minor_idr, md, minor);
1985 spin_unlock(&_minor_lock);
1987 BUG_ON(old_md != MINOR_ALLOCED);
1992 cleanup_mapped_device(md);
1996 module_put(THIS_MODULE);
2002 static void unlock_fs(struct mapped_device *md);
2004 static void free_dev(struct mapped_device *md)
2006 int minor = MINOR(disk_devt(md->disk));
2010 cleanup_mapped_device(md);
2012 free_table_devices(&md->table_devices);
2013 dm_stats_cleanup(&md->stats);
2016 module_put(THIS_MODULE);
2020 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2022 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2025 if (dm_table_bio_based(t)) {
2027 * The md may already have mempools that need changing.
2028 * If so, reload bioset because front_pad may have changed
2029 * because a different table was loaded.
2031 bioset_exit(&md->bs);
2032 bioset_exit(&md->io_bs);
2034 } else if (bioset_initialized(&md->bs)) {
2036 * There's no need to reload with request-based dm
2037 * because the size of front_pad doesn't change.
2038 * Note for future: If you are to reload bioset,
2039 * prep-ed requests in the queue may refer
2040 * to bio from the old bioset, so you must walk
2041 * through the queue to unprep.
2047 bioset_initialized(&md->bs) ||
2048 bioset_initialized(&md->io_bs));
2050 ret = bioset_init_from_src(&md->bs, &p->bs);
2053 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2055 bioset_exit(&md->bs);
2057 /* mempool bind completed, no longer need any mempools in the table */
2058 dm_table_free_md_mempools(t);
2063 * Bind a table to the device.
2065 static void event_callback(void *context)
2067 unsigned long flags;
2069 struct mapped_device *md = (struct mapped_device *) context;
2071 spin_lock_irqsave(&md->uevent_lock, flags);
2072 list_splice_init(&md->uevent_list, &uevents);
2073 spin_unlock_irqrestore(&md->uevent_lock, flags);
2075 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2077 atomic_inc(&md->event_nr);
2078 wake_up(&md->eventq);
2079 dm_issue_global_event();
2083 * Returns old map, which caller must destroy.
2085 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2086 struct queue_limits *limits)
2088 struct dm_table *old_map;
2089 struct request_queue *q = md->queue;
2090 bool request_based = dm_table_request_based(t);
2094 lockdep_assert_held(&md->suspend_lock);
2096 size = dm_table_get_size(t);
2099 * Wipe any geometry if the size of the table changed.
2101 if (size != dm_get_size(md))
2102 memset(&md->geometry, 0, sizeof(md->geometry));
2104 set_capacity(md->disk, size);
2105 bd_set_nr_sectors(md->bdev, size);
2107 dm_table_event_callback(t, event_callback, md);
2110 * The queue hasn't been stopped yet, if the old table type wasn't
2111 * for request-based during suspension. So stop it to prevent
2112 * I/O mapping before resume.
2113 * This must be done before setting the queue restrictions,
2114 * because request-based dm may be run just after the setting.
2119 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2121 * Leverage the fact that request-based DM targets and
2122 * NVMe bio based targets are immutable singletons
2123 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2124 * and __process_bio.
2126 md->immutable_target = dm_table_get_immutable_target(t);
2129 ret = __bind_mempools(md, t);
2131 old_map = ERR_PTR(ret);
2135 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2136 rcu_assign_pointer(md->map, (void *)t);
2137 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2139 dm_table_set_restrictions(t, q, limits);
2148 * Returns unbound table for the caller to free.
2150 static struct dm_table *__unbind(struct mapped_device *md)
2152 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2157 dm_table_event_callback(map, NULL, NULL);
2158 RCU_INIT_POINTER(md->map, NULL);
2165 * Constructor for a new device.
2167 int dm_create(int minor, struct mapped_device **result)
2170 struct mapped_device *md;
2172 md = alloc_dev(minor);
2176 r = dm_sysfs_init(md);
2187 * Functions to manage md->type.
2188 * All are required to hold md->type_lock.
2190 void dm_lock_md_type(struct mapped_device *md)
2192 mutex_lock(&md->type_lock);
2195 void dm_unlock_md_type(struct mapped_device *md)
2197 mutex_unlock(&md->type_lock);
2200 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2202 BUG_ON(!mutex_is_locked(&md->type_lock));
2206 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2211 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2213 return md->immutable_target_type;
2217 * The queue_limits are only valid as long as you have a reference
2220 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2222 BUG_ON(!atomic_read(&md->holders));
2223 return &md->queue->limits;
2225 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2228 * Setup the DM device's queue based on md's type
2230 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2233 struct queue_limits limits;
2234 enum dm_queue_mode type = dm_get_md_type(md);
2237 case DM_TYPE_REQUEST_BASED:
2238 r = dm_mq_init_request_queue(md, t);
2240 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2244 case DM_TYPE_BIO_BASED:
2245 case DM_TYPE_DAX_BIO_BASED:
2246 case DM_TYPE_NVME_BIO_BASED:
2253 r = dm_calculate_queue_limits(t, &limits);
2255 DMERR("Cannot calculate initial queue limits");
2258 dm_table_set_restrictions(t, md->queue, &limits);
2259 blk_register_queue(md->disk);
2264 struct mapped_device *dm_get_md(dev_t dev)
2266 struct mapped_device *md;
2267 unsigned minor = MINOR(dev);
2269 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2272 spin_lock(&_minor_lock);
2274 md = idr_find(&_minor_idr, minor);
2275 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2276 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2282 spin_unlock(&_minor_lock);
2286 EXPORT_SYMBOL_GPL(dm_get_md);
2288 void *dm_get_mdptr(struct mapped_device *md)
2290 return md->interface_ptr;
2293 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2295 md->interface_ptr = ptr;
2298 void dm_get(struct mapped_device *md)
2300 atomic_inc(&md->holders);
2301 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2304 int dm_hold(struct mapped_device *md)
2306 spin_lock(&_minor_lock);
2307 if (test_bit(DMF_FREEING, &md->flags)) {
2308 spin_unlock(&_minor_lock);
2312 spin_unlock(&_minor_lock);
2315 EXPORT_SYMBOL_GPL(dm_hold);
2317 const char *dm_device_name(struct mapped_device *md)
2321 EXPORT_SYMBOL_GPL(dm_device_name);
2323 static void __dm_destroy(struct mapped_device *md, bool wait)
2325 struct dm_table *map;
2330 spin_lock(&_minor_lock);
2331 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2332 set_bit(DMF_FREEING, &md->flags);
2333 spin_unlock(&_minor_lock);
2335 blk_set_queue_dying(md->queue);
2338 * Take suspend_lock so that presuspend and postsuspend methods
2339 * do not race with internal suspend.
2341 mutex_lock(&md->suspend_lock);
2342 map = dm_get_live_table(md, &srcu_idx);
2343 if (!dm_suspended_md(md)) {
2344 dm_table_presuspend_targets(map);
2345 set_bit(DMF_SUSPENDED, &md->flags);
2346 set_bit(DMF_POST_SUSPENDING, &md->flags);
2347 dm_table_postsuspend_targets(map);
2349 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2350 dm_put_live_table(md, srcu_idx);
2351 mutex_unlock(&md->suspend_lock);
2354 * Rare, but there may be I/O requests still going to complete,
2355 * for example. Wait for all references to disappear.
2356 * No one should increment the reference count of the mapped_device,
2357 * after the mapped_device state becomes DMF_FREEING.
2360 while (atomic_read(&md->holders))
2362 else if (atomic_read(&md->holders))
2363 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2364 dm_device_name(md), atomic_read(&md->holders));
2367 dm_table_destroy(__unbind(md));
2371 void dm_destroy(struct mapped_device *md)
2373 __dm_destroy(md, true);
2376 void dm_destroy_immediate(struct mapped_device *md)
2378 __dm_destroy(md, false);
2381 void dm_put(struct mapped_device *md)
2383 atomic_dec(&md->holders);
2385 EXPORT_SYMBOL_GPL(dm_put);
2387 static bool md_in_flight_bios(struct mapped_device *md)
2390 struct hd_struct *part = &dm_disk(md)->part0;
2393 for_each_possible_cpu(cpu) {
2394 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2395 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2401 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2407 prepare_to_wait(&md->wait, &wait, task_state);
2409 if (!md_in_flight_bios(md))
2412 if (signal_pending_state(task_state, current)) {
2419 finish_wait(&md->wait, &wait);
2424 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2428 if (!queue_is_mq(md->queue))
2429 return dm_wait_for_bios_completion(md, task_state);
2432 if (!blk_mq_queue_inflight(md->queue))
2435 if (signal_pending_state(task_state, current)) {
2447 * Process the deferred bios
2449 static void dm_wq_work(struct work_struct *work)
2451 struct mapped_device *md = container_of(work, struct mapped_device,
2455 struct dm_table *map;
2457 map = dm_get_live_table(md, &srcu_idx);
2459 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2460 spin_lock_irq(&md->deferred_lock);
2461 c = bio_list_pop(&md->deferred);
2462 spin_unlock_irq(&md->deferred_lock);
2467 if (dm_request_based(md))
2468 (void) submit_bio_noacct(c);
2470 (void) dm_process_bio(md, map, c);
2473 dm_put_live_table(md, srcu_idx);
2476 static void dm_queue_flush(struct mapped_device *md)
2478 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2479 smp_mb__after_atomic();
2480 queue_work(md->wq, &md->work);
2484 * Swap in a new table, returning the old one for the caller to destroy.
2486 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2488 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2489 struct queue_limits limits;
2492 mutex_lock(&md->suspend_lock);
2494 /* device must be suspended */
2495 if (!dm_suspended_md(md))
2499 * If the new table has no data devices, retain the existing limits.
2500 * This helps multipath with queue_if_no_path if all paths disappear,
2501 * then new I/O is queued based on these limits, and then some paths
2504 if (dm_table_has_no_data_devices(table)) {
2505 live_map = dm_get_live_table_fast(md);
2507 limits = md->queue->limits;
2508 dm_put_live_table_fast(md);
2512 r = dm_calculate_queue_limits(table, &limits);
2519 map = __bind(md, table, &limits);
2520 dm_issue_global_event();
2523 mutex_unlock(&md->suspend_lock);
2528 * Functions to lock and unlock any filesystem running on the
2531 static int lock_fs(struct mapped_device *md)
2535 WARN_ON(md->frozen_sb);
2537 md->frozen_sb = freeze_bdev(md->bdev);
2538 if (IS_ERR(md->frozen_sb)) {
2539 r = PTR_ERR(md->frozen_sb);
2540 md->frozen_sb = NULL;
2544 set_bit(DMF_FROZEN, &md->flags);
2549 static void unlock_fs(struct mapped_device *md)
2551 if (!test_bit(DMF_FROZEN, &md->flags))
2554 thaw_bdev(md->bdev, md->frozen_sb);
2555 md->frozen_sb = NULL;
2556 clear_bit(DMF_FROZEN, &md->flags);
2560 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2561 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2562 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2564 * If __dm_suspend returns 0, the device is completely quiescent
2565 * now. There is no request-processing activity. All new requests
2566 * are being added to md->deferred list.
2568 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2569 unsigned suspend_flags, long task_state,
2570 int dmf_suspended_flag)
2572 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2573 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2576 lockdep_assert_held(&md->suspend_lock);
2579 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2580 * This flag is cleared before dm_suspend returns.
2583 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2585 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2588 * This gets reverted if there's an error later and the targets
2589 * provide the .presuspend_undo hook.
2591 dm_table_presuspend_targets(map);
2594 * Flush I/O to the device.
2595 * Any I/O submitted after lock_fs() may not be flushed.
2596 * noflush takes precedence over do_lockfs.
2597 * (lock_fs() flushes I/Os and waits for them to complete.)
2599 if (!noflush && do_lockfs) {
2602 dm_table_presuspend_undo_targets(map);
2608 * Here we must make sure that no processes are submitting requests
2609 * to target drivers i.e. no one may be executing
2610 * __split_and_process_bio. This is called from dm_request and
2613 * To get all processes out of __split_and_process_bio in dm_request,
2614 * we take the write lock. To prevent any process from reentering
2615 * __split_and_process_bio from dm_request and quiesce the thread
2616 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2617 * flush_workqueue(md->wq).
2619 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2621 synchronize_srcu(&md->io_barrier);
2624 * Stop md->queue before flushing md->wq in case request-based
2625 * dm defers requests to md->wq from md->queue.
2627 if (dm_request_based(md))
2628 dm_stop_queue(md->queue);
2630 flush_workqueue(md->wq);
2633 * At this point no more requests are entering target request routines.
2634 * We call dm_wait_for_completion to wait for all existing requests
2637 r = dm_wait_for_completion(md, task_state);
2639 set_bit(dmf_suspended_flag, &md->flags);
2642 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2644 synchronize_srcu(&md->io_barrier);
2646 /* were we interrupted ? */
2650 if (dm_request_based(md))
2651 dm_start_queue(md->queue);
2654 dm_table_presuspend_undo_targets(map);
2655 /* pushback list is already flushed, so skip flush */
2662 * We need to be able to change a mapping table under a mounted
2663 * filesystem. For example we might want to move some data in
2664 * the background. Before the table can be swapped with
2665 * dm_bind_table, dm_suspend must be called to flush any in
2666 * flight bios and ensure that any further io gets deferred.
2669 * Suspend mechanism in request-based dm.
2671 * 1. Flush all I/Os by lock_fs() if needed.
2672 * 2. Stop dispatching any I/O by stopping the request_queue.
2673 * 3. Wait for all in-flight I/Os to be completed or requeued.
2675 * To abort suspend, start the request_queue.
2677 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2679 struct dm_table *map = NULL;
2683 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2685 if (dm_suspended_md(md)) {
2690 if (dm_suspended_internally_md(md)) {
2691 /* already internally suspended, wait for internal resume */
2692 mutex_unlock(&md->suspend_lock);
2693 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2699 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2701 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2705 set_bit(DMF_POST_SUSPENDING, &md->flags);
2706 dm_table_postsuspend_targets(map);
2707 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2710 mutex_unlock(&md->suspend_lock);
2714 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2717 int r = dm_table_resume_targets(map);
2725 * Flushing deferred I/Os must be done after targets are resumed
2726 * so that mapping of targets can work correctly.
2727 * Request-based dm is queueing the deferred I/Os in its request_queue.
2729 if (dm_request_based(md))
2730 dm_start_queue(md->queue);
2737 int dm_resume(struct mapped_device *md)
2740 struct dm_table *map = NULL;
2744 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2746 if (!dm_suspended_md(md))
2749 if (dm_suspended_internally_md(md)) {
2750 /* already internally suspended, wait for internal resume */
2751 mutex_unlock(&md->suspend_lock);
2752 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2758 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2759 if (!map || !dm_table_get_size(map))
2762 r = __dm_resume(md, map);
2766 clear_bit(DMF_SUSPENDED, &md->flags);
2768 mutex_unlock(&md->suspend_lock);
2774 * Internal suspend/resume works like userspace-driven suspend. It waits
2775 * until all bios finish and prevents issuing new bios to the target drivers.
2776 * It may be used only from the kernel.
2779 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2781 struct dm_table *map = NULL;
2783 lockdep_assert_held(&md->suspend_lock);
2785 if (md->internal_suspend_count++)
2786 return; /* nested internal suspend */
2788 if (dm_suspended_md(md)) {
2789 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2790 return; /* nest suspend */
2793 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2796 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2797 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2798 * would require changing .presuspend to return an error -- avoid this
2799 * until there is a need for more elaborate variants of internal suspend.
2801 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2802 DMF_SUSPENDED_INTERNALLY);
2804 set_bit(DMF_POST_SUSPENDING, &md->flags);
2805 dm_table_postsuspend_targets(map);
2806 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2809 static void __dm_internal_resume(struct mapped_device *md)
2811 BUG_ON(!md->internal_suspend_count);
2813 if (--md->internal_suspend_count)
2814 return; /* resume from nested internal suspend */
2816 if (dm_suspended_md(md))
2817 goto done; /* resume from nested suspend */
2820 * NOTE: existing callers don't need to call dm_table_resume_targets
2821 * (which may fail -- so best to avoid it for now by passing NULL map)
2823 (void) __dm_resume(md, NULL);
2826 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2827 smp_mb__after_atomic();
2828 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2831 void dm_internal_suspend_noflush(struct mapped_device *md)
2833 mutex_lock(&md->suspend_lock);
2834 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2835 mutex_unlock(&md->suspend_lock);
2837 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2839 void dm_internal_resume(struct mapped_device *md)
2841 mutex_lock(&md->suspend_lock);
2842 __dm_internal_resume(md);
2843 mutex_unlock(&md->suspend_lock);
2845 EXPORT_SYMBOL_GPL(dm_internal_resume);
2848 * Fast variants of internal suspend/resume hold md->suspend_lock,
2849 * which prevents interaction with userspace-driven suspend.
2852 void dm_internal_suspend_fast(struct mapped_device *md)
2854 mutex_lock(&md->suspend_lock);
2855 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2858 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2859 synchronize_srcu(&md->io_barrier);
2860 flush_workqueue(md->wq);
2861 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2863 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2865 void dm_internal_resume_fast(struct mapped_device *md)
2867 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2873 mutex_unlock(&md->suspend_lock);
2875 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2877 /*-----------------------------------------------------------------
2878 * Event notification.
2879 *---------------------------------------------------------------*/
2880 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2885 char udev_cookie[DM_COOKIE_LENGTH];
2886 char *envp[] = { udev_cookie, NULL };
2888 noio_flag = memalloc_noio_save();
2891 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2893 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2894 DM_COOKIE_ENV_VAR_NAME, cookie);
2895 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2899 memalloc_noio_restore(noio_flag);
2904 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2906 return atomic_add_return(1, &md->uevent_seq);
2909 uint32_t dm_get_event_nr(struct mapped_device *md)
2911 return atomic_read(&md->event_nr);
2914 int dm_wait_event(struct mapped_device *md, int event_nr)
2916 return wait_event_interruptible(md->eventq,
2917 (event_nr != atomic_read(&md->event_nr)));
2920 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2922 unsigned long flags;
2924 spin_lock_irqsave(&md->uevent_lock, flags);
2925 list_add(elist, &md->uevent_list);
2926 spin_unlock_irqrestore(&md->uevent_lock, flags);
2930 * The gendisk is only valid as long as you have a reference
2933 struct gendisk *dm_disk(struct mapped_device *md)
2937 EXPORT_SYMBOL_GPL(dm_disk);
2939 struct kobject *dm_kobject(struct mapped_device *md)
2941 return &md->kobj_holder.kobj;
2944 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2946 struct mapped_device *md;
2948 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2950 spin_lock(&_minor_lock);
2951 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2957 spin_unlock(&_minor_lock);
2962 int dm_suspended_md(struct mapped_device *md)
2964 return test_bit(DMF_SUSPENDED, &md->flags);
2967 static int dm_post_suspending_md(struct mapped_device *md)
2969 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2972 int dm_suspended_internally_md(struct mapped_device *md)
2974 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2977 int dm_test_deferred_remove_flag(struct mapped_device *md)
2979 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2982 int dm_suspended(struct dm_target *ti)
2984 return dm_suspended_md(dm_table_get_md(ti->table));
2986 EXPORT_SYMBOL_GPL(dm_suspended);
2988 int dm_post_suspending(struct dm_target *ti)
2990 return dm_post_suspending_md(dm_table_get_md(ti->table));
2992 EXPORT_SYMBOL_GPL(dm_post_suspending);
2994 int dm_noflush_suspending(struct dm_target *ti)
2996 return __noflush_suspending(dm_table_get_md(ti->table));
2998 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3000 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3001 unsigned integrity, unsigned per_io_data_size,
3002 unsigned min_pool_size)
3004 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3005 unsigned int pool_size = 0;
3006 unsigned int front_pad, io_front_pad;
3013 case DM_TYPE_BIO_BASED:
3014 case DM_TYPE_DAX_BIO_BASED:
3015 case DM_TYPE_NVME_BIO_BASED:
3016 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3017 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3018 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3019 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3022 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3025 case DM_TYPE_REQUEST_BASED:
3026 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3027 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3028 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3034 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3038 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3044 dm_free_md_mempools(pools);
3049 void dm_free_md_mempools(struct dm_md_mempools *pools)
3054 bioset_exit(&pools->bs);
3055 bioset_exit(&pools->io_bs);
3067 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3070 struct mapped_device *md = bdev->bd_disk->private_data;
3071 struct dm_table *table;
3072 struct dm_target *ti;
3073 int ret = -ENOTTY, srcu_idx;
3075 table = dm_get_live_table(md, &srcu_idx);
3076 if (!table || !dm_table_get_size(table))
3079 /* We only support devices that have a single target */
3080 if (dm_table_get_num_targets(table) != 1)
3082 ti = dm_table_get_target(table, 0);
3085 if (!ti->type->iterate_devices)
3088 ret = ti->type->iterate_devices(ti, fn, data);
3090 dm_put_live_table(md, srcu_idx);
3095 * For register / unregister we need to manually call out to every path.
3097 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3098 sector_t start, sector_t len, void *data)
3100 struct dm_pr *pr = data;
3101 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3103 if (!ops || !ops->pr_register)
3105 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3108 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3119 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3120 if (ret && new_key) {
3121 /* unregister all paths if we failed to register any path */
3122 pr.old_key = new_key;
3125 pr.fail_early = false;
3126 dm_call_pr(bdev, __dm_pr_register, &pr);
3132 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3135 struct mapped_device *md = bdev->bd_disk->private_data;
3136 const struct pr_ops *ops;
3139 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3143 ops = bdev->bd_disk->fops->pr_ops;
3144 if (ops && ops->pr_reserve)
3145 r = ops->pr_reserve(bdev, key, type, flags);
3149 dm_unprepare_ioctl(md, srcu_idx);
3153 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3155 struct mapped_device *md = bdev->bd_disk->private_data;
3156 const struct pr_ops *ops;
3159 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3163 ops = bdev->bd_disk->fops->pr_ops;
3164 if (ops && ops->pr_release)
3165 r = ops->pr_release(bdev, key, type);
3169 dm_unprepare_ioctl(md, srcu_idx);
3173 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3174 enum pr_type type, bool abort)
3176 struct mapped_device *md = bdev->bd_disk->private_data;
3177 const struct pr_ops *ops;
3180 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3184 ops = bdev->bd_disk->fops->pr_ops;
3185 if (ops && ops->pr_preempt)
3186 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3190 dm_unprepare_ioctl(md, srcu_idx);
3194 static int dm_pr_clear(struct block_device *bdev, u64 key)
3196 struct mapped_device *md = bdev->bd_disk->private_data;
3197 const struct pr_ops *ops;
3200 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3204 ops = bdev->bd_disk->fops->pr_ops;
3205 if (ops && ops->pr_clear)
3206 r = ops->pr_clear(bdev, key);
3210 dm_unprepare_ioctl(md, srcu_idx);
3214 static const struct pr_ops dm_pr_ops = {
3215 .pr_register = dm_pr_register,
3216 .pr_reserve = dm_pr_reserve,
3217 .pr_release = dm_pr_release,
3218 .pr_preempt = dm_pr_preempt,
3219 .pr_clear = dm_pr_clear,
3222 static const struct block_device_operations dm_blk_dops = {
3223 .submit_bio = dm_submit_bio,
3224 .open = dm_blk_open,
3225 .release = dm_blk_close,
3226 .ioctl = dm_blk_ioctl,
3227 .getgeo = dm_blk_getgeo,
3228 .report_zones = dm_blk_report_zones,
3229 .pr_ops = &dm_pr_ops,
3230 .owner = THIS_MODULE
3233 static const struct dax_operations dm_dax_ops = {
3234 .direct_access = dm_dax_direct_access,
3235 .dax_supported = dm_dax_supported,
3236 .copy_from_iter = dm_dax_copy_from_iter,
3237 .copy_to_iter = dm_dax_copy_to_iter,
3238 .zero_page_range = dm_dax_zero_page_range,
3244 module_init(dm_init);
3245 module_exit(dm_exit);
3247 module_param(major, uint, 0);
3248 MODULE_PARM_DESC(major, "The major number of the device mapper");
3250 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3251 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3253 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3254 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3256 MODULE_DESCRIPTION(DM_NAME " driver");
3257 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3258 MODULE_LICENSE("GPL");