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>
30 #define DM_MSG_PREFIX "core"
33 * Cookies are numeric values sent with CHANGE and REMOVE
34 * uevents while resuming, removing or renaming the device.
36 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
37 #define DM_COOKIE_LENGTH 24
39 static const char *_name = DM_NAME;
41 static unsigned int major = 0;
42 static unsigned int _major = 0;
44 static DEFINE_IDR(_minor_idr);
46 static DEFINE_SPINLOCK(_minor_lock);
48 static void do_deferred_remove(struct work_struct *w);
50 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
52 static struct workqueue_struct *deferred_remove_workqueue;
54 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
55 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
57 void dm_issue_global_event(void)
59 atomic_inc(&dm_global_event_nr);
60 wake_up(&dm_global_eventq);
64 * One of these is allocated (on-stack) per original bio.
71 unsigned sector_count;
75 * One of these is allocated per clone bio.
77 #define DM_TIO_MAGIC 7282014
82 unsigned target_bio_nr;
89 * One of these is allocated per original bio.
90 * It contains the first clone used for that original.
92 #define DM_IO_MAGIC 5191977
95 struct mapped_device *md;
99 unsigned long start_time;
100 spinlock_t endio_lock;
101 struct dm_stats_aux stats_aux;
102 /* last member of dm_target_io is 'struct bio' */
103 struct dm_target_io tio;
106 void *dm_per_bio_data(struct bio *bio, size_t data_size)
108 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
109 if (!tio->inside_dm_io)
110 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
111 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
113 EXPORT_SYMBOL_GPL(dm_per_bio_data);
115 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
117 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
118 if (io->magic == DM_IO_MAGIC)
119 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
120 BUG_ON(io->magic != DM_TIO_MAGIC);
121 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
123 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
125 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
127 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
129 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
131 #define MINOR_ALLOCED ((void *)-1)
134 * Bits for the md->flags field.
136 #define DMF_BLOCK_IO_FOR_SUSPEND 0
137 #define DMF_SUSPENDED 1
139 #define DMF_FREEING 3
140 #define DMF_DELETING 4
141 #define DMF_NOFLUSH_SUSPENDING 5
142 #define DMF_DEFERRED_REMOVE 6
143 #define DMF_SUSPENDED_INTERNALLY 7
144 #define DMF_POST_SUSPENDING 8
146 #define DM_NUMA_NODE NUMA_NO_NODE
147 static int dm_numa_node = DM_NUMA_NODE;
149 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
150 static int swap_bios = DEFAULT_SWAP_BIOS;
151 static int get_swap_bios(void)
153 int latch = READ_ONCE(swap_bios);
154 if (unlikely(latch <= 0))
155 latch = DEFAULT_SWAP_BIOS;
160 * For mempools pre-allocation at the table loading time.
162 struct dm_md_mempools {
164 struct bio_set io_bs;
167 struct table_device {
168 struct list_head list;
170 struct dm_dev dm_dev;
173 static struct kmem_cache *_rq_tio_cache;
174 static struct kmem_cache *_rq_cache;
177 * Bio-based DM's mempools' reserved IOs set by the user.
179 #define RESERVED_BIO_BASED_IOS 16
180 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
182 static int __dm_get_module_param_int(int *module_param, int min, int max)
184 int param = READ_ONCE(*module_param);
185 int modified_param = 0;
186 bool modified = true;
189 modified_param = min;
190 else if (param > max)
191 modified_param = max;
196 (void)cmpxchg(module_param, param, modified_param);
197 param = modified_param;
203 unsigned __dm_get_module_param(unsigned *module_param,
204 unsigned def, unsigned max)
206 unsigned param = READ_ONCE(*module_param);
207 unsigned modified_param = 0;
210 modified_param = def;
211 else if (param > max)
212 modified_param = max;
214 if (modified_param) {
215 (void)cmpxchg(module_param, param, modified_param);
216 param = modified_param;
222 unsigned dm_get_reserved_bio_based_ios(void)
224 return __dm_get_module_param(&reserved_bio_based_ios,
225 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
227 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
229 static unsigned dm_get_numa_node(void)
231 return __dm_get_module_param_int(&dm_numa_node,
232 DM_NUMA_NODE, num_online_nodes() - 1);
235 static int __init local_init(void)
239 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
243 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
244 __alignof__(struct request), 0, NULL);
246 goto out_free_rq_tio_cache;
248 r = dm_uevent_init();
250 goto out_free_rq_cache;
252 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
253 if (!deferred_remove_workqueue) {
255 goto out_uevent_exit;
259 r = register_blkdev(_major, _name);
261 goto out_free_workqueue;
269 destroy_workqueue(deferred_remove_workqueue);
273 kmem_cache_destroy(_rq_cache);
274 out_free_rq_tio_cache:
275 kmem_cache_destroy(_rq_tio_cache);
280 static void local_exit(void)
282 destroy_workqueue(deferred_remove_workqueue);
284 kmem_cache_destroy(_rq_cache);
285 kmem_cache_destroy(_rq_tio_cache);
286 unregister_blkdev(_major, _name);
291 DMINFO("cleaned up");
294 static int (*_inits[])(void) __initdata = {
305 static void (*_exits[])(void) = {
316 static int __init dm_init(void)
318 const int count = ARRAY_SIZE(_inits);
322 for (i = 0; i < count; i++) {
337 static void __exit dm_exit(void)
339 int i = ARRAY_SIZE(_exits);
345 * Should be empty by this point.
347 idr_destroy(&_minor_idr);
351 * Block device functions
353 int dm_deleting_md(struct mapped_device *md)
355 return test_bit(DMF_DELETING, &md->flags);
358 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
360 struct mapped_device *md;
362 spin_lock(&_minor_lock);
364 md = bdev->bd_disk->private_data;
368 if (test_bit(DMF_FREEING, &md->flags) ||
369 dm_deleting_md(md)) {
375 atomic_inc(&md->open_count);
377 spin_unlock(&_minor_lock);
379 return md ? 0 : -ENXIO;
382 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
384 struct mapped_device *md;
386 spin_lock(&_minor_lock);
388 md = disk->private_data;
392 if (atomic_dec_and_test(&md->open_count) &&
393 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
394 queue_work(deferred_remove_workqueue, &deferred_remove_work);
398 spin_unlock(&_minor_lock);
401 int dm_open_count(struct mapped_device *md)
403 return atomic_read(&md->open_count);
407 * Guarantees nothing is using the device before it's deleted.
409 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
413 spin_lock(&_minor_lock);
415 if (dm_open_count(md)) {
418 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
419 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
422 set_bit(DMF_DELETING, &md->flags);
424 spin_unlock(&_minor_lock);
429 int dm_cancel_deferred_remove(struct mapped_device *md)
433 spin_lock(&_minor_lock);
435 if (test_bit(DMF_DELETING, &md->flags))
438 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
440 spin_unlock(&_minor_lock);
445 static void do_deferred_remove(struct work_struct *w)
447 dm_deferred_remove();
450 sector_t dm_get_size(struct mapped_device *md)
452 return get_capacity(md->disk);
455 struct request_queue *dm_get_md_queue(struct mapped_device *md)
460 struct dm_stats *dm_get_stats(struct mapped_device *md)
465 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
467 struct mapped_device *md = bdev->bd_disk->private_data;
469 return dm_get_geometry(md, geo);
472 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
473 struct block_device **bdev)
475 struct dm_target *tgt;
476 struct dm_table *map;
481 map = dm_get_live_table(md, srcu_idx);
482 if (!map || !dm_table_get_size(map))
485 /* We only support devices that have a single target */
486 if (dm_table_get_num_targets(map) != 1)
489 tgt = dm_table_get_target(map, 0);
490 if (!tgt->type->prepare_ioctl)
493 if (dm_suspended_md(md))
496 r = tgt->type->prepare_ioctl(tgt, bdev);
497 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
498 dm_put_live_table(md, *srcu_idx);
506 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
508 dm_put_live_table(md, srcu_idx);
511 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
512 unsigned int cmd, unsigned long arg)
514 struct mapped_device *md = bdev->bd_disk->private_data;
517 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
523 * Target determined this ioctl is being issued against a
524 * subset of the parent bdev; require extra privileges.
526 if (!capable(CAP_SYS_RAWIO)) {
528 "%s: sending ioctl %x to DM device without required privilege.",
535 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
537 dm_unprepare_ioctl(md, srcu_idx);
541 static void start_io_acct(struct dm_io *io);
543 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
546 struct dm_target_io *tio;
549 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
553 tio = container_of(clone, struct dm_target_io, clone);
554 tio->inside_dm_io = true;
557 io = container_of(tio, struct dm_io, tio);
558 io->magic = DM_IO_MAGIC;
560 atomic_set(&io->io_count, 1);
563 spin_lock_init(&io->endio_lock);
570 static void free_io(struct mapped_device *md, struct dm_io *io)
572 bio_put(&io->tio.clone);
575 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
576 unsigned target_bio_nr, gfp_t gfp_mask)
578 struct dm_target_io *tio;
580 if (!ci->io->tio.io) {
581 /* the dm_target_io embedded in ci->io is available */
584 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
588 tio = container_of(clone, struct dm_target_io, clone);
589 tio->inside_dm_io = false;
592 tio->magic = DM_TIO_MAGIC;
595 tio->target_bio_nr = target_bio_nr;
600 static void free_tio(struct dm_target_io *tio)
602 if (tio->inside_dm_io)
604 bio_put(&tio->clone);
607 int md_in_flight(struct mapped_device *md)
609 return atomic_read(&md->pending[READ]) +
610 atomic_read(&md->pending[WRITE]);
613 static void start_io_acct(struct dm_io *io)
615 struct mapped_device *md = io->md;
616 struct bio *bio = io->orig_bio;
617 int rw = bio_data_dir(bio);
619 io->start_time = jiffies;
621 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
622 &dm_disk(md)->part0);
624 atomic_set(&dm_disk(md)->part0.in_flight[rw],
625 atomic_inc_return(&md->pending[rw]));
627 if (unlikely(dm_stats_used(&md->stats)))
628 dm_stats_account_io(&md->stats, bio_data_dir(bio),
629 bio->bi_iter.bi_sector, bio_sectors(bio),
630 false, 0, &io->stats_aux);
633 static void end_io_acct(struct mapped_device *md, struct bio *bio,
634 unsigned long start_time, struct dm_stats_aux *stats_aux)
636 unsigned long duration = jiffies - start_time;
638 int rw = bio_data_dir(bio);
640 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
643 if (unlikely(dm_stats_used(&md->stats)))
644 dm_stats_account_io(&md->stats, bio_data_dir(bio),
645 bio->bi_iter.bi_sector, bio_sectors(bio),
646 true, duration, stats_aux);
649 * After this is decremented the bio must not be touched if it is
652 pending = atomic_dec_return(&md->pending[rw]);
653 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
654 pending += atomic_read(&md->pending[rw^0x1]);
656 /* nudge anyone waiting on suspend queue */
662 * Add the bio to the list of deferred io.
664 static void queue_io(struct mapped_device *md, struct bio *bio)
668 spin_lock_irqsave(&md->deferred_lock, flags);
669 bio_list_add(&md->deferred, bio);
670 spin_unlock_irqrestore(&md->deferred_lock, flags);
671 queue_work(md->wq, &md->work);
675 * Everyone (including functions in this file), should use this
676 * function to access the md->map field, and make sure they call
677 * dm_put_live_table() when finished.
679 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
681 *srcu_idx = srcu_read_lock(&md->io_barrier);
683 return srcu_dereference(md->map, &md->io_barrier);
686 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
688 srcu_read_unlock(&md->io_barrier, srcu_idx);
691 void dm_sync_table(struct mapped_device *md)
693 synchronize_srcu(&md->io_barrier);
694 synchronize_rcu_expedited();
698 * A fast alternative to dm_get_live_table/dm_put_live_table.
699 * The caller must not block between these two functions.
701 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
704 return rcu_dereference(md->map);
707 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
712 static char *_dm_claim_ptr = "I belong to device-mapper";
715 * Open a table device so we can use it as a map destination.
717 static int open_table_device(struct table_device *td, dev_t dev,
718 struct mapped_device *md)
720 struct block_device *bdev;
724 BUG_ON(td->dm_dev.bdev);
726 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
728 return PTR_ERR(bdev);
730 r = bd_link_disk_holder(bdev, dm_disk(md));
732 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
736 td->dm_dev.bdev = bdev;
737 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
742 * Close a table device that we've been using.
744 static void close_table_device(struct table_device *td, struct mapped_device *md)
746 if (!td->dm_dev.bdev)
749 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
750 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
751 put_dax(td->dm_dev.dax_dev);
752 td->dm_dev.bdev = NULL;
753 td->dm_dev.dax_dev = NULL;
756 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
758 struct table_device *td;
760 list_for_each_entry(td, l, list)
761 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
767 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
768 struct dm_dev **result) {
770 struct table_device *td;
772 mutex_lock(&md->table_devices_lock);
773 td = find_table_device(&md->table_devices, dev, mode);
775 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
777 mutex_unlock(&md->table_devices_lock);
781 td->dm_dev.mode = mode;
782 td->dm_dev.bdev = NULL;
784 if ((r = open_table_device(td, dev, md))) {
785 mutex_unlock(&md->table_devices_lock);
790 format_dev_t(td->dm_dev.name, dev);
792 refcount_set(&td->count, 1);
793 list_add(&td->list, &md->table_devices);
795 refcount_inc(&td->count);
797 mutex_unlock(&md->table_devices_lock);
799 *result = &td->dm_dev;
802 EXPORT_SYMBOL_GPL(dm_get_table_device);
804 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
806 struct table_device *td = container_of(d, struct table_device, dm_dev);
808 mutex_lock(&md->table_devices_lock);
809 if (refcount_dec_and_test(&td->count)) {
810 close_table_device(td, md);
814 mutex_unlock(&md->table_devices_lock);
816 EXPORT_SYMBOL(dm_put_table_device);
818 static void free_table_devices(struct list_head *devices)
820 struct list_head *tmp, *next;
822 list_for_each_safe(tmp, next, devices) {
823 struct table_device *td = list_entry(tmp, struct table_device, list);
825 DMWARN("dm_destroy: %s still exists with %d references",
826 td->dm_dev.name, refcount_read(&td->count));
832 * Get the geometry associated with a dm device
834 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
842 * Set the geometry of a device.
844 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
846 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
848 if (geo->start > sz) {
849 DMWARN("Start sector is beyond the geometry limits.");
858 static int __noflush_suspending(struct mapped_device *md)
860 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
864 * Decrements the number of outstanding ios that a bio has been
865 * cloned into, completing the original io if necc.
867 static void dec_pending(struct dm_io *io, blk_status_t error)
870 blk_status_t io_error;
872 struct mapped_device *md = io->md;
873 unsigned long start_time = 0;
874 struct dm_stats_aux stats_aux;
876 /* Push-back supersedes any I/O errors */
877 if (unlikely(error)) {
878 spin_lock_irqsave(&io->endio_lock, flags);
879 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
881 spin_unlock_irqrestore(&io->endio_lock, flags);
884 if (atomic_dec_and_test(&io->io_count)) {
885 if (io->status == BLK_STS_DM_REQUEUE) {
887 * Target requested pushing back the I/O.
889 spin_lock_irqsave(&md->deferred_lock, flags);
890 if (__noflush_suspending(md))
891 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
892 bio_list_add_head(&md->deferred, io->orig_bio);
894 /* noflush suspend was interrupted. */
895 io->status = BLK_STS_IOERR;
896 spin_unlock_irqrestore(&md->deferred_lock, flags);
899 io_error = io->status;
901 start_time = io->start_time;
902 stats_aux = io->stats_aux;
904 end_io_acct(md, bio, start_time, &stats_aux);
906 if (io_error == BLK_STS_DM_REQUEUE)
909 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
911 * Preflush done for flush with data, reissue
912 * without REQ_PREFLUSH.
914 bio->bi_opf &= ~REQ_PREFLUSH;
917 /* done with normal IO or empty flush */
919 bio->bi_status = io_error;
925 void disable_discard(struct mapped_device *md)
927 struct queue_limits *limits = dm_get_queue_limits(md);
929 /* device doesn't really support DISCARD, disable it */
930 limits->max_discard_sectors = 0;
931 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
934 void disable_write_same(struct mapped_device *md)
936 struct queue_limits *limits = dm_get_queue_limits(md);
938 /* device doesn't really support WRITE SAME, disable it */
939 limits->max_write_same_sectors = 0;
942 void disable_write_zeroes(struct mapped_device *md)
944 struct queue_limits *limits = dm_get_queue_limits(md);
946 /* device doesn't really support WRITE ZEROES, disable it */
947 limits->max_write_zeroes_sectors = 0;
950 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
952 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
955 static void clone_endio(struct bio *bio)
957 blk_status_t error = bio->bi_status;
958 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
959 struct dm_io *io = tio->io;
960 struct mapped_device *md = tio->io->md;
961 dm_endio_fn endio = tio->ti->type->end_io;
963 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
964 if (bio_op(bio) == REQ_OP_DISCARD &&
965 !bio->bi_disk->queue->limits.max_discard_sectors)
967 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
968 !bio->bi_disk->queue->limits.max_write_same_sectors)
969 disable_write_same(md);
970 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
971 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
972 disable_write_zeroes(md);
976 int r = endio(tio->ti, bio, &error);
978 case DM_ENDIO_REQUEUE:
979 error = BLK_STS_DM_REQUEUE;
983 case DM_ENDIO_INCOMPLETE:
984 /* The target will handle the io */
987 DMWARN("unimplemented target endio return value: %d", r);
992 if (unlikely(swap_bios_limit(tio->ti, bio))) {
993 struct mapped_device *md = io->md;
994 up(&md->swap_bios_semaphore);
998 dec_pending(io, error);
1002 * Return maximum size of I/O possible at the supplied sector up to the current
1005 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1007 sector_t target_offset = dm_target_offset(ti, sector);
1009 return ti->len - target_offset;
1012 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1014 sector_t len = max_io_len_target_boundary(sector, ti);
1015 sector_t offset, max_len;
1018 * Does the target need to split even further?
1020 if (ti->max_io_len) {
1021 offset = dm_target_offset(ti, sector);
1022 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1023 max_len = sector_div(offset, ti->max_io_len);
1025 max_len = offset & (ti->max_io_len - 1);
1026 max_len = ti->max_io_len - max_len;
1035 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1037 if (len > UINT_MAX) {
1038 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1039 (unsigned long long)len, UINT_MAX);
1040 ti->error = "Maximum size of target IO is too large";
1044 ti->max_io_len = (uint32_t) len;
1048 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1050 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1051 sector_t sector, int *srcu_idx)
1052 __acquires(md->io_barrier)
1054 struct dm_table *map;
1055 struct dm_target *ti;
1057 map = dm_get_live_table(md, srcu_idx);
1061 ti = dm_table_find_target(map, sector);
1062 if (!dm_target_is_valid(ti))
1068 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1069 long nr_pages, void **kaddr, pfn_t *pfn)
1071 struct mapped_device *md = dax_get_private(dax_dev);
1072 sector_t sector = pgoff * PAGE_SECTORS;
1073 struct dm_target *ti;
1074 long len, ret = -EIO;
1077 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1081 if (!ti->type->direct_access)
1083 len = max_io_len(sector, ti) / PAGE_SECTORS;
1086 nr_pages = min(len, nr_pages);
1087 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1090 dm_put_live_table(md, srcu_idx);
1095 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1096 void *addr, size_t bytes, struct iov_iter *i)
1098 struct mapped_device *md = dax_get_private(dax_dev);
1099 sector_t sector = pgoff * PAGE_SECTORS;
1100 struct dm_target *ti;
1104 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1108 if (!ti->type->dax_copy_from_iter) {
1109 ret = copy_from_iter(addr, bytes, i);
1112 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1114 dm_put_live_table(md, srcu_idx);
1119 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1120 void *addr, size_t bytes, struct iov_iter *i)
1122 struct mapped_device *md = dax_get_private(dax_dev);
1123 sector_t sector = pgoff * PAGE_SECTORS;
1124 struct dm_target *ti;
1128 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1132 if (!ti->type->dax_copy_to_iter) {
1133 ret = copy_to_iter(addr, bytes, i);
1136 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1138 dm_put_live_table(md, srcu_idx);
1144 * A target may call dm_accept_partial_bio only from the map routine. It is
1145 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1147 * dm_accept_partial_bio informs the dm that the target only wants to process
1148 * additional n_sectors sectors of the bio and the rest of the data should be
1149 * sent in a next bio.
1151 * A diagram that explains the arithmetics:
1152 * +--------------------+---------------+-------+
1154 * +--------------------+---------------+-------+
1156 * <-------------- *tio->len_ptr --------------->
1157 * <------- bi_size ------->
1160 * Region 1 was already iterated over with bio_advance or similar function.
1161 * (it may be empty if the target doesn't use bio_advance)
1162 * Region 2 is the remaining bio size that the target wants to process.
1163 * (it may be empty if region 1 is non-empty, although there is no reason
1165 * The target requires that region 3 is to be sent in the next bio.
1167 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1168 * the partially processed part (the sum of regions 1+2) must be the same for all
1169 * copies of the bio.
1171 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1173 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1174 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1175 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1176 BUG_ON(bi_size > *tio->len_ptr);
1177 BUG_ON(n_sectors > bi_size);
1178 *tio->len_ptr -= bi_size - n_sectors;
1179 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1181 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1184 * The zone descriptors obtained with a zone report indicate zone positions
1185 * within the target backing device, regardless of that device is a partition
1186 * and regardless of the target mapping start sector on the device or partition.
1187 * The zone descriptors start sector and write pointer position must be adjusted
1188 * to match their relative position within the dm device.
1189 * A target may call dm_remap_zone_report() after completion of a
1190 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1193 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1195 #ifdef CONFIG_BLK_DEV_ZONED
1196 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1197 struct bio *report_bio = tio->io->orig_bio;
1198 struct blk_zone_report_hdr *hdr = NULL;
1199 struct blk_zone *zone;
1200 unsigned int nr_rep = 0;
1202 sector_t part_offset;
1203 struct bio_vec bvec;
1204 struct bvec_iter iter;
1211 * bio sector was incremented by the request size on completion. Taking
1212 * into account the original request sector, the target start offset on
1213 * the backing device and the target mapping offset (ti->begin), the
1214 * start sector of the backing device. The partition offset is always 0
1215 * if the target uses a whole device.
1217 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1220 * Remap the start sector of the reported zones. For sequential zones,
1221 * also remap the write pointer position.
1223 bio_for_each_segment(bvec, report_bio, iter) {
1224 addr = kmap_atomic(bvec.bv_page);
1226 /* Remember the report header in the first page */
1229 ofst = sizeof(struct blk_zone_report_hdr);
1233 /* Set zones start sector */
1234 while (hdr->nr_zones && ofst < bvec.bv_len) {
1236 zone->start -= part_offset;
1237 if (zone->start >= start + ti->len) {
1241 zone->start = zone->start + ti->begin - start;
1242 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1243 if (zone->cond == BLK_ZONE_COND_FULL)
1244 zone->wp = zone->start + zone->len;
1245 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1246 zone->wp = zone->start;
1248 zone->wp = zone->wp + ti->begin - start - part_offset;
1250 ofst += sizeof(struct blk_zone);
1256 kunmap_atomic(addr);
1263 hdr->nr_zones = nr_rep;
1267 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1269 #else /* !CONFIG_BLK_DEV_ZONED */
1270 bio->bi_status = BLK_STS_NOTSUPP;
1273 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1275 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1277 mutex_lock(&md->swap_bios_lock);
1278 while (latch < md->swap_bios) {
1280 down(&md->swap_bios_semaphore);
1283 while (latch > md->swap_bios) {
1285 up(&md->swap_bios_semaphore);
1288 mutex_unlock(&md->swap_bios_lock);
1291 static blk_qc_t __map_bio(struct dm_target_io *tio)
1295 struct bio *clone = &tio->clone;
1296 struct dm_io *io = tio->io;
1297 struct mapped_device *md = io->md;
1298 struct dm_target *ti = tio->ti;
1299 blk_qc_t ret = BLK_QC_T_NONE;
1301 clone->bi_end_io = clone_endio;
1304 * Map the clone. If r == 0 we don't need to do
1305 * anything, the target has assumed ownership of
1308 atomic_inc(&io->io_count);
1309 sector = clone->bi_iter.bi_sector;
1311 if (unlikely(swap_bios_limit(ti, clone))) {
1312 struct mapped_device *md = io->md;
1313 int latch = get_swap_bios();
1314 if (unlikely(latch != md->swap_bios))
1315 __set_swap_bios_limit(md, latch);
1316 down(&md->swap_bios_semaphore);
1319 r = ti->type->map(ti, clone);
1321 case DM_MAPIO_SUBMITTED:
1323 case DM_MAPIO_REMAPPED:
1324 /* the bio has been remapped so dispatch it */
1325 trace_block_bio_remap(clone->bi_disk->queue, clone,
1326 bio_dev(io->orig_bio), sector);
1327 if (md->type == DM_TYPE_NVME_BIO_BASED)
1328 ret = direct_make_request(clone);
1330 ret = generic_make_request(clone);
1333 if (unlikely(swap_bios_limit(ti, clone))) {
1334 struct mapped_device *md = io->md;
1335 up(&md->swap_bios_semaphore);
1338 dec_pending(io, BLK_STS_IOERR);
1340 case DM_MAPIO_REQUEUE:
1341 if (unlikely(swap_bios_limit(ti, clone))) {
1342 struct mapped_device *md = io->md;
1343 up(&md->swap_bios_semaphore);
1346 dec_pending(io, BLK_STS_DM_REQUEUE);
1349 DMWARN("unimplemented target map return value: %d", r);
1356 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1358 bio->bi_iter.bi_sector = sector;
1359 bio->bi_iter.bi_size = to_bytes(len);
1363 * Creates a bio that consists of range of complete bvecs.
1365 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1366 sector_t sector, unsigned len)
1368 struct bio *clone = &tio->clone;
1370 __bio_clone_fast(clone, bio);
1372 if (unlikely(bio_integrity(bio) != NULL)) {
1375 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1376 !dm_target_passes_integrity(tio->ti->type))) {
1377 DMWARN("%s: the target %s doesn't support integrity data.",
1378 dm_device_name(tio->io->md),
1379 tio->ti->type->name);
1383 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1388 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1389 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1390 clone->bi_iter.bi_size = to_bytes(len);
1392 if (unlikely(bio_integrity(bio) != NULL))
1393 bio_integrity_trim(clone);
1398 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1399 struct dm_target *ti, unsigned num_bios)
1401 struct dm_target_io *tio;
1407 if (num_bios == 1) {
1408 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1409 bio_list_add(blist, &tio->clone);
1413 for (try = 0; try < 2; try++) {
1418 mutex_lock(&ci->io->md->table_devices_lock);
1419 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1420 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1424 bio_list_add(blist, &tio->clone);
1427 mutex_unlock(&ci->io->md->table_devices_lock);
1428 if (bio_nr == num_bios)
1431 while ((bio = bio_list_pop(blist))) {
1432 tio = container_of(bio, struct dm_target_io, clone);
1438 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1439 struct dm_target_io *tio, unsigned *len)
1441 struct bio *clone = &tio->clone;
1445 __bio_clone_fast(clone, ci->bio);
1447 bio_setup_sector(clone, ci->sector, *len);
1449 return __map_bio(tio);
1452 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1453 unsigned num_bios, unsigned *len)
1455 struct bio_list blist = BIO_EMPTY_LIST;
1457 struct dm_target_io *tio;
1459 alloc_multiple_bios(&blist, ci, ti, num_bios);
1461 while ((bio = bio_list_pop(&blist))) {
1462 tio = container_of(bio, struct dm_target_io, clone);
1463 (void) __clone_and_map_simple_bio(ci, tio, len);
1467 static int __send_empty_flush(struct clone_info *ci)
1469 unsigned target_nr = 0;
1470 struct dm_target *ti;
1472 BUG_ON(bio_has_data(ci->bio));
1473 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1474 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1479 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1480 sector_t sector, unsigned *len)
1482 struct bio *bio = ci->bio;
1483 struct dm_target_io *tio;
1486 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1488 r = clone_bio(tio, bio, sector, *len);
1493 (void) __map_bio(tio);
1498 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1500 static unsigned get_num_discard_bios(struct dm_target *ti)
1502 return ti->num_discard_bios;
1505 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1507 return ti->num_secure_erase_bios;
1510 static unsigned get_num_write_same_bios(struct dm_target *ti)
1512 return ti->num_write_same_bios;
1515 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1517 return ti->num_write_zeroes_bios;
1520 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1522 static bool is_split_required_for_discard(struct dm_target *ti)
1524 return ti->split_discard_bios;
1527 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1528 get_num_bios_fn get_num_bios,
1529 is_split_required_fn is_split_required)
1535 * Even though the device advertised support for this type of
1536 * request, that does not mean every target supports it, and
1537 * reconfiguration might also have changed that since the
1538 * check was performed.
1540 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1544 if (is_split_required && !is_split_required(ti))
1545 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1547 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1549 __send_duplicate_bios(ci, ti, num_bios, &len);
1552 ci->sector_count -= len;
1557 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1559 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1560 is_split_required_for_discard);
1563 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1565 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1568 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1570 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1573 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1575 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1578 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1581 struct bio *bio = ci->bio;
1583 if (bio_op(bio) == REQ_OP_DISCARD)
1584 *result = __send_discard(ci, ti);
1585 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1586 *result = __send_secure_erase(ci, ti);
1587 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1588 *result = __send_write_same(ci, ti);
1589 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1590 *result = __send_write_zeroes(ci, ti);
1598 * Select the correct strategy for processing a non-flush bio.
1600 static int __split_and_process_non_flush(struct clone_info *ci)
1602 struct bio *bio = ci->bio;
1603 struct dm_target *ti;
1607 ti = dm_table_find_target(ci->map, ci->sector);
1608 if (!dm_target_is_valid(ti))
1611 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1614 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1615 len = ci->sector_count;
1617 len = min_t(sector_t, max_io_len(ci->sector, ti),
1620 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1625 ci->sector_count -= len;
1630 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1631 struct dm_table *map, struct bio *bio)
1634 ci->io = alloc_io(md, bio);
1635 ci->sector = bio->bi_iter.bi_sector;
1639 * Entry point to split a bio into clones and submit them to the targets.
1641 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1642 struct dm_table *map, struct bio *bio)
1644 struct clone_info ci;
1645 blk_qc_t ret = BLK_QC_T_NONE;
1648 if (unlikely(!map)) {
1653 blk_queue_split(md->queue, &bio);
1655 init_clone_info(&ci, md, map, bio);
1657 if (bio->bi_opf & REQ_PREFLUSH) {
1658 ci.bio = &ci.io->md->flush_bio;
1659 ci.sector_count = 0;
1660 error = __send_empty_flush(&ci);
1661 /* dec_pending submits any data associated with flush */
1662 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1664 ci.sector_count = 0;
1665 error = __split_and_process_non_flush(&ci);
1668 ci.sector_count = bio_sectors(bio);
1669 while (ci.sector_count && !error) {
1670 error = __split_and_process_non_flush(&ci);
1671 if (current->bio_list && ci.sector_count && !error) {
1673 * Remainder must be passed to generic_make_request()
1674 * so that it gets handled *after* bios already submitted
1675 * have been completely processed.
1676 * We take a clone of the original to store in
1677 * ci.io->orig_bio to be used by end_io_acct() and
1678 * for dec_pending to use for completion handling.
1679 * As this path is not used for REQ_OP_ZONE_REPORT,
1680 * the usage of io->orig_bio in dm_remap_zone_report()
1681 * won't be affected by this reassignment.
1683 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1684 GFP_NOIO, &md->queue->bio_split);
1685 ci.io->orig_bio = b;
1687 ret = generic_make_request(bio);
1693 /* drop the extra reference count */
1694 dec_pending(ci.io, errno_to_blk_status(error));
1699 * Optimized variant of __split_and_process_bio that leverages the
1700 * fact that targets that use it do _not_ have a need to split bios.
1702 static blk_qc_t __process_bio(struct mapped_device *md,
1703 struct dm_table *map, struct bio *bio)
1705 struct clone_info ci;
1706 blk_qc_t ret = BLK_QC_T_NONE;
1709 if (unlikely(!map)) {
1714 init_clone_info(&ci, md, map, bio);
1716 if (bio->bi_opf & REQ_PREFLUSH) {
1717 ci.bio = &ci.io->md->flush_bio;
1718 ci.sector_count = 0;
1719 error = __send_empty_flush(&ci);
1720 /* dec_pending submits any data associated with flush */
1722 struct dm_target *ti = md->immutable_target;
1723 struct dm_target_io *tio;
1726 * Defend against IO still getting in during teardown
1727 * - as was seen for a time with nvme-fcloop
1729 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1735 ci.sector_count = bio_sectors(bio);
1736 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1739 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1740 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1743 /* drop the extra reference count */
1744 dec_pending(ci.io, errno_to_blk_status(error));
1748 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1750 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1751 process_bio_fn process_bio)
1753 struct mapped_device *md = q->queuedata;
1754 blk_qc_t ret = BLK_QC_T_NONE;
1756 struct dm_table *map;
1758 map = dm_get_live_table(md, &srcu_idx);
1760 /* if we're suspended, we have to queue this io for later */
1761 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1762 dm_put_live_table(md, srcu_idx);
1764 if (!(bio->bi_opf & REQ_RAHEAD))
1771 ret = process_bio(md, map, bio);
1773 dm_put_live_table(md, srcu_idx);
1778 * The request function that remaps the bio to one target and
1779 * splits off any remainder.
1781 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1783 return __dm_make_request(q, bio, __split_and_process_bio);
1786 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1788 return __dm_make_request(q, bio, __process_bio);
1791 static int dm_any_congested(void *congested_data, int bdi_bits)
1794 struct mapped_device *md = congested_data;
1795 struct dm_table *map;
1797 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1798 if (dm_request_based(md)) {
1800 * With request-based DM we only need to check the
1801 * top-level queue for congestion.
1803 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1805 map = dm_get_live_table_fast(md);
1807 r = dm_table_any_congested(map, bdi_bits);
1808 dm_put_live_table_fast(md);
1815 /*-----------------------------------------------------------------
1816 * An IDR is used to keep track of allocated minor numbers.
1817 *---------------------------------------------------------------*/
1818 static void free_minor(int minor)
1820 spin_lock(&_minor_lock);
1821 idr_remove(&_minor_idr, minor);
1822 spin_unlock(&_minor_lock);
1826 * See if the device with a specific minor # is free.
1828 static int specific_minor(int minor)
1832 if (minor >= (1 << MINORBITS))
1835 idr_preload(GFP_KERNEL);
1836 spin_lock(&_minor_lock);
1838 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1840 spin_unlock(&_minor_lock);
1843 return r == -ENOSPC ? -EBUSY : r;
1847 static int next_free_minor(int *minor)
1851 idr_preload(GFP_KERNEL);
1852 spin_lock(&_minor_lock);
1854 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1856 spin_unlock(&_minor_lock);
1864 static const struct block_device_operations dm_blk_dops;
1865 static const struct dax_operations dm_dax_ops;
1867 static void dm_wq_work(struct work_struct *work);
1869 static void dm_init_normal_md_queue(struct mapped_device *md)
1871 md->use_blk_mq = false;
1874 * Initialize aspects of queue that aren't relevant for blk-mq
1876 md->queue->backing_dev_info->congested_data = md;
1877 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1880 static void cleanup_mapped_device(struct mapped_device *md)
1883 destroy_workqueue(md->wq);
1884 if (md->kworker_task)
1885 kthread_stop(md->kworker_task);
1886 bioset_exit(&md->bs);
1887 bioset_exit(&md->io_bs);
1890 kill_dax(md->dax_dev);
1891 put_dax(md->dax_dev);
1896 spin_lock(&_minor_lock);
1897 md->disk->private_data = NULL;
1898 spin_unlock(&_minor_lock);
1899 del_gendisk(md->disk);
1904 blk_cleanup_queue(md->queue);
1906 cleanup_srcu_struct(&md->io_barrier);
1913 mutex_destroy(&md->suspend_lock);
1914 mutex_destroy(&md->type_lock);
1915 mutex_destroy(&md->table_devices_lock);
1916 mutex_destroy(&md->swap_bios_lock);
1918 dm_mq_cleanup_mapped_device(md);
1922 * Allocate and initialise a blank device with a given minor.
1924 static struct mapped_device *alloc_dev(int minor)
1926 int r, numa_node_id = dm_get_numa_node();
1927 struct dax_device *dax_dev = NULL;
1928 struct mapped_device *md;
1931 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1933 DMWARN("unable to allocate device, out of memory.");
1937 if (!try_module_get(THIS_MODULE))
1938 goto bad_module_get;
1940 /* get a minor number for the dev */
1941 if (minor == DM_ANY_MINOR)
1942 r = next_free_minor(&minor);
1944 r = specific_minor(minor);
1948 r = init_srcu_struct(&md->io_barrier);
1950 goto bad_io_barrier;
1952 md->numa_node_id = numa_node_id;
1953 md->use_blk_mq = dm_use_blk_mq_default();
1954 md->init_tio_pdu = false;
1955 md->type = DM_TYPE_NONE;
1956 mutex_init(&md->suspend_lock);
1957 mutex_init(&md->type_lock);
1958 mutex_init(&md->table_devices_lock);
1959 spin_lock_init(&md->deferred_lock);
1960 atomic_set(&md->holders, 1);
1961 atomic_set(&md->open_count, 0);
1962 atomic_set(&md->event_nr, 0);
1963 atomic_set(&md->uevent_seq, 0);
1964 INIT_LIST_HEAD(&md->uevent_list);
1965 INIT_LIST_HEAD(&md->table_devices);
1966 spin_lock_init(&md->uevent_lock);
1968 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1971 md->queue->queuedata = md;
1973 * default to bio-based required ->make_request_fn until DM
1974 * table is loaded and md->type established. If request-based
1975 * table is loaded: blk-mq will override accordingly.
1977 blk_queue_make_request(md->queue, dm_make_request);
1979 md->disk = alloc_disk_node(1, md->numa_node_id);
1983 atomic_set(&md->pending[0], 0);
1984 atomic_set(&md->pending[1], 0);
1985 init_waitqueue_head(&md->wait);
1986 INIT_WORK(&md->work, dm_wq_work);
1987 init_waitqueue_head(&md->eventq);
1988 init_completion(&md->kobj_holder.completion);
1989 md->kworker_task = NULL;
1991 md->swap_bios = get_swap_bios();
1992 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1993 mutex_init(&md->swap_bios_lock);
1995 md->disk->major = _major;
1996 md->disk->first_minor = minor;
1997 md->disk->fops = &dm_blk_dops;
1998 md->disk->queue = md->queue;
1999 md->disk->private_data = md;
2000 sprintf(md->disk->disk_name, "dm-%d", minor);
2002 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
2003 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
2007 md->dax_dev = dax_dev;
2009 add_disk_no_queue_reg(md->disk);
2010 format_dev_t(md->name, MKDEV(_major, minor));
2012 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2016 md->bdev = bdget_disk(md->disk, 0);
2020 bio_init(&md->flush_bio, NULL, 0);
2021 bio_set_dev(&md->flush_bio, md->bdev);
2022 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
2024 r = dm_stats_init(&md->stats);
2028 /* Populate the mapping, nobody knows we exist yet */
2029 spin_lock(&_minor_lock);
2030 old_md = idr_replace(&_minor_idr, md, minor);
2031 spin_unlock(&_minor_lock);
2033 BUG_ON(old_md != MINOR_ALLOCED);
2038 cleanup_mapped_device(md);
2042 module_put(THIS_MODULE);
2048 static void unlock_fs(struct mapped_device *md);
2050 static void free_dev(struct mapped_device *md)
2052 int minor = MINOR(disk_devt(md->disk));
2056 cleanup_mapped_device(md);
2058 free_table_devices(&md->table_devices);
2059 dm_stats_cleanup(&md->stats);
2062 module_put(THIS_MODULE);
2066 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2068 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2071 if (dm_table_bio_based(t)) {
2073 * The md may already have mempools that need changing.
2074 * If so, reload bioset because front_pad may have changed
2075 * because a different table was loaded.
2077 bioset_exit(&md->bs);
2078 bioset_exit(&md->io_bs);
2080 } else if (bioset_initialized(&md->bs)) {
2082 * There's no need to reload with request-based dm
2083 * because the size of front_pad doesn't change.
2084 * Note for future: If you are to reload bioset,
2085 * prep-ed requests in the queue may refer
2086 * to bio from the old bioset, so you must walk
2087 * through the queue to unprep.
2093 bioset_initialized(&md->bs) ||
2094 bioset_initialized(&md->io_bs));
2096 ret = bioset_init_from_src(&md->bs, &p->bs);
2099 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2101 bioset_exit(&md->bs);
2103 /* mempool bind completed, no longer need any mempools in the table */
2104 dm_table_free_md_mempools(t);
2109 * Bind a table to the device.
2111 static void event_callback(void *context)
2113 unsigned long flags;
2115 struct mapped_device *md = (struct mapped_device *) context;
2117 spin_lock_irqsave(&md->uevent_lock, flags);
2118 list_splice_init(&md->uevent_list, &uevents);
2119 spin_unlock_irqrestore(&md->uevent_lock, flags);
2121 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2123 atomic_inc(&md->event_nr);
2124 wake_up(&md->eventq);
2125 dm_issue_global_event();
2129 * Protected by md->suspend_lock obtained by dm_swap_table().
2131 static void __set_size(struct mapped_device *md, sector_t size)
2133 lockdep_assert_held(&md->suspend_lock);
2135 set_capacity(md->disk, size);
2137 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2141 * Returns old map, which caller must destroy.
2143 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2144 struct queue_limits *limits)
2146 struct dm_table *old_map;
2147 struct request_queue *q = md->queue;
2148 bool request_based = dm_table_request_based(t);
2152 lockdep_assert_held(&md->suspend_lock);
2154 size = dm_table_get_size(t);
2157 * Wipe any geometry if the size of the table changed.
2159 if (size != dm_get_size(md))
2160 memset(&md->geometry, 0, sizeof(md->geometry));
2162 __set_size(md, size);
2164 dm_table_event_callback(t, event_callback, md);
2167 * The queue hasn't been stopped yet, if the old table type wasn't
2168 * for request-based during suspension. So stop it to prevent
2169 * I/O mapping before resume.
2170 * This must be done before setting the queue restrictions,
2171 * because request-based dm may be run just after the setting.
2176 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2178 * Leverage the fact that request-based DM targets and
2179 * NVMe bio based targets are immutable singletons
2180 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2181 * and __process_bio.
2183 md->immutable_target = dm_table_get_immutable_target(t);
2186 ret = __bind_mempools(md, t);
2188 old_map = ERR_PTR(ret);
2192 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2193 rcu_assign_pointer(md->map, (void *)t);
2194 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2196 dm_table_set_restrictions(t, q, limits);
2205 * Returns unbound table for the caller to free.
2207 static struct dm_table *__unbind(struct mapped_device *md)
2209 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2214 dm_table_event_callback(map, NULL, NULL);
2215 RCU_INIT_POINTER(md->map, NULL);
2222 * Constructor for a new device.
2224 int dm_create(int minor, struct mapped_device **result)
2227 struct mapped_device *md;
2229 md = alloc_dev(minor);
2233 r = dm_sysfs_init(md);
2244 * Functions to manage md->type.
2245 * All are required to hold md->type_lock.
2247 void dm_lock_md_type(struct mapped_device *md)
2249 mutex_lock(&md->type_lock);
2252 void dm_unlock_md_type(struct mapped_device *md)
2254 mutex_unlock(&md->type_lock);
2257 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2259 BUG_ON(!mutex_is_locked(&md->type_lock));
2263 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2268 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2270 return md->immutable_target_type;
2274 * The queue_limits are only valid as long as you have a reference
2277 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2279 BUG_ON(!atomic_read(&md->holders));
2280 return &md->queue->limits;
2282 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2285 * Setup the DM device's queue based on md's type
2287 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2290 struct queue_limits limits;
2291 enum dm_queue_mode type = dm_get_md_type(md);
2294 case DM_TYPE_REQUEST_BASED:
2295 dm_init_normal_md_queue(md);
2296 r = dm_old_init_request_queue(md, t);
2298 DMERR("Cannot initialize queue for request-based mapped device");
2302 case DM_TYPE_MQ_REQUEST_BASED:
2303 r = dm_mq_init_request_queue(md, t);
2305 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2309 case DM_TYPE_BIO_BASED:
2310 case DM_TYPE_DAX_BIO_BASED:
2311 dm_init_normal_md_queue(md);
2313 case DM_TYPE_NVME_BIO_BASED:
2314 dm_init_normal_md_queue(md);
2315 blk_queue_make_request(md->queue, dm_make_request_nvme);
2322 r = dm_calculate_queue_limits(t, &limits);
2324 DMERR("Cannot calculate initial queue limits");
2327 dm_table_set_restrictions(t, md->queue, &limits);
2328 blk_register_queue(md->disk);
2333 struct mapped_device *dm_get_md(dev_t dev)
2335 struct mapped_device *md;
2336 unsigned minor = MINOR(dev);
2338 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2341 spin_lock(&_minor_lock);
2343 md = idr_find(&_minor_idr, minor);
2344 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2345 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2351 spin_unlock(&_minor_lock);
2355 EXPORT_SYMBOL_GPL(dm_get_md);
2357 void *dm_get_mdptr(struct mapped_device *md)
2359 return md->interface_ptr;
2362 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2364 md->interface_ptr = ptr;
2367 void dm_get(struct mapped_device *md)
2369 atomic_inc(&md->holders);
2370 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2373 int dm_hold(struct mapped_device *md)
2375 spin_lock(&_minor_lock);
2376 if (test_bit(DMF_FREEING, &md->flags)) {
2377 spin_unlock(&_minor_lock);
2381 spin_unlock(&_minor_lock);
2384 EXPORT_SYMBOL_GPL(dm_hold);
2386 const char *dm_device_name(struct mapped_device *md)
2390 EXPORT_SYMBOL_GPL(dm_device_name);
2392 static void __dm_destroy(struct mapped_device *md, bool wait)
2394 struct dm_table *map;
2399 spin_lock(&_minor_lock);
2400 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2401 set_bit(DMF_FREEING, &md->flags);
2402 spin_unlock(&_minor_lock);
2404 blk_set_queue_dying(md->queue);
2406 if (dm_request_based(md) && md->kworker_task)
2407 kthread_flush_worker(&md->kworker);
2410 * Take suspend_lock so that presuspend and postsuspend methods
2411 * do not race with internal suspend.
2413 mutex_lock(&md->suspend_lock);
2414 map = dm_get_live_table(md, &srcu_idx);
2415 if (!dm_suspended_md(md)) {
2416 dm_table_presuspend_targets(map);
2417 set_bit(DMF_SUSPENDED, &md->flags);
2418 set_bit(DMF_POST_SUSPENDING, &md->flags);
2419 dm_table_postsuspend_targets(map);
2421 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2422 dm_put_live_table(md, srcu_idx);
2423 mutex_unlock(&md->suspend_lock);
2426 * Rare, but there may be I/O requests still going to complete,
2427 * for example. Wait for all references to disappear.
2428 * No one should increment the reference count of the mapped_device,
2429 * after the mapped_device state becomes DMF_FREEING.
2432 while (atomic_read(&md->holders))
2434 else if (atomic_read(&md->holders))
2435 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2436 dm_device_name(md), atomic_read(&md->holders));
2439 dm_table_destroy(__unbind(md));
2443 void dm_destroy(struct mapped_device *md)
2445 __dm_destroy(md, true);
2448 void dm_destroy_immediate(struct mapped_device *md)
2450 __dm_destroy(md, false);
2453 void dm_put(struct mapped_device *md)
2455 atomic_dec(&md->holders);
2457 EXPORT_SYMBOL_GPL(dm_put);
2459 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2465 prepare_to_wait(&md->wait, &wait, task_state);
2467 if (!md_in_flight(md))
2470 if (signal_pending_state(task_state, current)) {
2477 finish_wait(&md->wait, &wait);
2479 smp_rmb(); /* paired with atomic_dec_return in end_io_acct */
2485 * Process the deferred bios
2487 static void dm_wq_work(struct work_struct *work)
2489 struct mapped_device *md = container_of(work, struct mapped_device,
2493 struct dm_table *map;
2495 map = dm_get_live_table(md, &srcu_idx);
2497 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2498 spin_lock_irq(&md->deferred_lock);
2499 c = bio_list_pop(&md->deferred);
2500 spin_unlock_irq(&md->deferred_lock);
2505 if (dm_request_based(md))
2506 generic_make_request(c);
2508 __split_and_process_bio(md, map, c);
2511 dm_put_live_table(md, srcu_idx);
2514 static void dm_queue_flush(struct mapped_device *md)
2516 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2517 smp_mb__after_atomic();
2518 queue_work(md->wq, &md->work);
2522 * Swap in a new table, returning the old one for the caller to destroy.
2524 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2526 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2527 struct queue_limits limits;
2530 mutex_lock(&md->suspend_lock);
2532 /* device must be suspended */
2533 if (!dm_suspended_md(md))
2537 * If the new table has no data devices, retain the existing limits.
2538 * This helps multipath with queue_if_no_path if all paths disappear,
2539 * then new I/O is queued based on these limits, and then some paths
2542 if (dm_table_has_no_data_devices(table)) {
2543 live_map = dm_get_live_table_fast(md);
2545 limits = md->queue->limits;
2546 dm_put_live_table_fast(md);
2550 r = dm_calculate_queue_limits(table, &limits);
2557 map = __bind(md, table, &limits);
2558 dm_issue_global_event();
2561 mutex_unlock(&md->suspend_lock);
2566 * Functions to lock and unlock any filesystem running on the
2569 static int lock_fs(struct mapped_device *md)
2573 WARN_ON(md->frozen_sb);
2575 md->frozen_sb = freeze_bdev(md->bdev);
2576 if (IS_ERR(md->frozen_sb)) {
2577 r = PTR_ERR(md->frozen_sb);
2578 md->frozen_sb = NULL;
2582 set_bit(DMF_FROZEN, &md->flags);
2587 static void unlock_fs(struct mapped_device *md)
2589 if (!test_bit(DMF_FROZEN, &md->flags))
2592 thaw_bdev(md->bdev, md->frozen_sb);
2593 md->frozen_sb = NULL;
2594 clear_bit(DMF_FROZEN, &md->flags);
2598 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2599 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2600 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2602 * If __dm_suspend returns 0, the device is completely quiescent
2603 * now. There is no request-processing activity. All new requests
2604 * are being added to md->deferred list.
2606 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2607 unsigned suspend_flags, long task_state,
2608 int dmf_suspended_flag)
2610 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2611 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2614 lockdep_assert_held(&md->suspend_lock);
2617 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2618 * This flag is cleared before dm_suspend returns.
2621 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2623 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2626 * This gets reverted if there's an error later and the targets
2627 * provide the .presuspend_undo hook.
2629 dm_table_presuspend_targets(map);
2632 * Flush I/O to the device.
2633 * Any I/O submitted after lock_fs() may not be flushed.
2634 * noflush takes precedence over do_lockfs.
2635 * (lock_fs() flushes I/Os and waits for them to complete.)
2637 if (!noflush && do_lockfs) {
2640 dm_table_presuspend_undo_targets(map);
2646 * Here we must make sure that no processes are submitting requests
2647 * to target drivers i.e. no one may be executing
2648 * __split_and_process_bio. This is called from dm_request and
2651 * To get all processes out of __split_and_process_bio in dm_request,
2652 * we take the write lock. To prevent any process from reentering
2653 * __split_and_process_bio from dm_request and quiesce the thread
2654 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2655 * flush_workqueue(md->wq).
2657 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2659 synchronize_srcu(&md->io_barrier);
2662 * Stop md->queue before flushing md->wq in case request-based
2663 * dm defers requests to md->wq from md->queue.
2665 if (dm_request_based(md)) {
2666 dm_stop_queue(md->queue);
2667 if (md->kworker_task)
2668 kthread_flush_worker(&md->kworker);
2671 flush_workqueue(md->wq);
2674 * At this point no more requests are entering target request routines.
2675 * We call dm_wait_for_completion to wait for all existing requests
2678 r = dm_wait_for_completion(md, task_state);
2680 set_bit(dmf_suspended_flag, &md->flags);
2683 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2685 synchronize_srcu(&md->io_barrier);
2687 /* were we interrupted ? */
2691 if (dm_request_based(md))
2692 dm_start_queue(md->queue);
2695 dm_table_presuspend_undo_targets(map);
2696 /* pushback list is already flushed, so skip flush */
2703 * We need to be able to change a mapping table under a mounted
2704 * filesystem. For example we might want to move some data in
2705 * the background. Before the table can be swapped with
2706 * dm_bind_table, dm_suspend must be called to flush any in
2707 * flight bios and ensure that any further io gets deferred.
2710 * Suspend mechanism in request-based dm.
2712 * 1. Flush all I/Os by lock_fs() if needed.
2713 * 2. Stop dispatching any I/O by stopping the request_queue.
2714 * 3. Wait for all in-flight I/Os to be completed or requeued.
2716 * To abort suspend, start the request_queue.
2718 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2720 struct dm_table *map = NULL;
2724 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2726 if (dm_suspended_md(md)) {
2731 if (dm_suspended_internally_md(md)) {
2732 /* already internally suspended, wait for internal resume */
2733 mutex_unlock(&md->suspend_lock);
2734 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2740 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2742 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2746 set_bit(DMF_POST_SUSPENDING, &md->flags);
2747 dm_table_postsuspend_targets(map);
2748 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2751 mutex_unlock(&md->suspend_lock);
2755 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2758 int r = dm_table_resume_targets(map);
2766 * Flushing deferred I/Os must be done after targets are resumed
2767 * so that mapping of targets can work correctly.
2768 * Request-based dm is queueing the deferred I/Os in its request_queue.
2770 if (dm_request_based(md))
2771 dm_start_queue(md->queue);
2778 int dm_resume(struct mapped_device *md)
2781 struct dm_table *map = NULL;
2785 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2787 if (!dm_suspended_md(md))
2790 if (dm_suspended_internally_md(md)) {
2791 /* already internally suspended, wait for internal resume */
2792 mutex_unlock(&md->suspend_lock);
2793 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2799 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2800 if (!map || !dm_table_get_size(map))
2803 r = __dm_resume(md, map);
2807 clear_bit(DMF_SUSPENDED, &md->flags);
2809 mutex_unlock(&md->suspend_lock);
2815 * Internal suspend/resume works like userspace-driven suspend. It waits
2816 * until all bios finish and prevents issuing new bios to the target drivers.
2817 * It may be used only from the kernel.
2820 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2822 struct dm_table *map = NULL;
2824 lockdep_assert_held(&md->suspend_lock);
2826 if (md->internal_suspend_count++)
2827 return; /* nested internal suspend */
2829 if (dm_suspended_md(md)) {
2830 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2831 return; /* nest suspend */
2834 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2837 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2838 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2839 * would require changing .presuspend to return an error -- avoid this
2840 * until there is a need for more elaborate variants of internal suspend.
2842 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2843 DMF_SUSPENDED_INTERNALLY);
2845 set_bit(DMF_POST_SUSPENDING, &md->flags);
2846 dm_table_postsuspend_targets(map);
2847 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2850 static void __dm_internal_resume(struct mapped_device *md)
2853 struct dm_table *map;
2855 BUG_ON(!md->internal_suspend_count);
2857 if (--md->internal_suspend_count)
2858 return; /* resume from nested internal suspend */
2860 if (dm_suspended_md(md))
2861 goto done; /* resume from nested suspend */
2863 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2864 r = __dm_resume(md, map);
2867 * If a preresume method of some target failed, we are in a
2868 * tricky situation. We can't return an error to the caller. We
2869 * can't fake success because then the "resume" and
2870 * "postsuspend" methods would not be paired correctly, and it
2871 * would break various targets, for example it would cause list
2872 * corruption in the "origin" target.
2874 * So, we fake normal suspend here, to make sure that the
2875 * "resume" and "postsuspend" methods will be paired correctly.
2877 DMERR("Preresume method failed: %d", r);
2878 set_bit(DMF_SUSPENDED, &md->flags);
2881 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2882 smp_mb__after_atomic();
2883 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2886 void dm_internal_suspend_noflush(struct mapped_device *md)
2888 mutex_lock(&md->suspend_lock);
2889 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2890 mutex_unlock(&md->suspend_lock);
2892 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2894 void dm_internal_resume(struct mapped_device *md)
2896 mutex_lock(&md->suspend_lock);
2897 __dm_internal_resume(md);
2898 mutex_unlock(&md->suspend_lock);
2900 EXPORT_SYMBOL_GPL(dm_internal_resume);
2903 * Fast variants of internal suspend/resume hold md->suspend_lock,
2904 * which prevents interaction with userspace-driven suspend.
2907 void dm_internal_suspend_fast(struct mapped_device *md)
2909 mutex_lock(&md->suspend_lock);
2910 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2913 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2914 synchronize_srcu(&md->io_barrier);
2915 flush_workqueue(md->wq);
2916 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2918 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2920 void dm_internal_resume_fast(struct mapped_device *md)
2922 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2928 mutex_unlock(&md->suspend_lock);
2930 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2932 /*-----------------------------------------------------------------
2933 * Event notification.
2934 *---------------------------------------------------------------*/
2935 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2940 char udev_cookie[DM_COOKIE_LENGTH];
2941 char *envp[] = { udev_cookie, NULL };
2943 noio_flag = memalloc_noio_save();
2946 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2948 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2949 DM_COOKIE_ENV_VAR_NAME, cookie);
2950 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2954 memalloc_noio_restore(noio_flag);
2959 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2961 return atomic_add_return(1, &md->uevent_seq);
2964 uint32_t dm_get_event_nr(struct mapped_device *md)
2966 return atomic_read(&md->event_nr);
2969 int dm_wait_event(struct mapped_device *md, int event_nr)
2971 return wait_event_interruptible(md->eventq,
2972 (event_nr != atomic_read(&md->event_nr)));
2975 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2977 unsigned long flags;
2979 spin_lock_irqsave(&md->uevent_lock, flags);
2980 list_add(elist, &md->uevent_list);
2981 spin_unlock_irqrestore(&md->uevent_lock, flags);
2985 * The gendisk is only valid as long as you have a reference
2988 struct gendisk *dm_disk(struct mapped_device *md)
2992 EXPORT_SYMBOL_GPL(dm_disk);
2994 struct kobject *dm_kobject(struct mapped_device *md)
2996 return &md->kobj_holder.kobj;
2999 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3001 struct mapped_device *md;
3003 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3005 spin_lock(&_minor_lock);
3006 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3012 spin_unlock(&_minor_lock);
3017 int dm_suspended_md(struct mapped_device *md)
3019 return test_bit(DMF_SUSPENDED, &md->flags);
3022 static int dm_post_suspending_md(struct mapped_device *md)
3024 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3027 int dm_suspended_internally_md(struct mapped_device *md)
3029 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3032 int dm_test_deferred_remove_flag(struct mapped_device *md)
3034 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3037 int dm_suspended(struct dm_target *ti)
3039 return dm_suspended_md(dm_table_get_md(ti->table));
3041 EXPORT_SYMBOL_GPL(dm_suspended);
3043 int dm_post_suspending(struct dm_target *ti)
3045 return dm_post_suspending_md(dm_table_get_md(ti->table));
3047 EXPORT_SYMBOL_GPL(dm_post_suspending);
3049 int dm_noflush_suspending(struct dm_target *ti)
3051 return __noflush_suspending(dm_table_get_md(ti->table));
3053 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3055 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3056 unsigned integrity, unsigned per_io_data_size,
3057 unsigned min_pool_size)
3059 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3060 unsigned int pool_size = 0;
3061 unsigned int front_pad, io_front_pad;
3068 case DM_TYPE_BIO_BASED:
3069 case DM_TYPE_DAX_BIO_BASED:
3070 case DM_TYPE_NVME_BIO_BASED:
3071 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3072 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3073 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3074 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3077 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3080 case DM_TYPE_REQUEST_BASED:
3081 case DM_TYPE_MQ_REQUEST_BASED:
3082 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3083 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3084 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3090 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3094 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3100 dm_free_md_mempools(pools);
3105 void dm_free_md_mempools(struct dm_md_mempools *pools)
3110 bioset_exit(&pools->bs);
3111 bioset_exit(&pools->io_bs);
3123 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3126 struct mapped_device *md = bdev->bd_disk->private_data;
3127 struct dm_table *table;
3128 struct dm_target *ti;
3129 int ret = -ENOTTY, srcu_idx;
3131 table = dm_get_live_table(md, &srcu_idx);
3132 if (!table || !dm_table_get_size(table))
3135 /* We only support devices that have a single target */
3136 if (dm_table_get_num_targets(table) != 1)
3138 ti = dm_table_get_target(table, 0);
3140 if (dm_suspended_md(md)) {
3146 if (!ti->type->iterate_devices)
3149 ret = ti->type->iterate_devices(ti, fn, data);
3151 dm_put_live_table(md, srcu_idx);
3156 * For register / unregister we need to manually call out to every path.
3158 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3159 sector_t start, sector_t len, void *data)
3161 struct dm_pr *pr = data;
3162 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3164 if (!ops || !ops->pr_register)
3166 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3169 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3180 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3181 if (ret && new_key) {
3182 /* unregister all paths if we failed to register any path */
3183 pr.old_key = new_key;
3186 pr.fail_early = false;
3187 dm_call_pr(bdev, __dm_pr_register, &pr);
3193 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
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_reserve)
3206 r = ops->pr_reserve(bdev, key, type, flags);
3210 dm_unprepare_ioctl(md, srcu_idx);
3214 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3216 struct mapped_device *md = bdev->bd_disk->private_data;
3217 const struct pr_ops *ops;
3220 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3224 ops = bdev->bd_disk->fops->pr_ops;
3225 if (ops && ops->pr_release)
3226 r = ops->pr_release(bdev, key, type);
3230 dm_unprepare_ioctl(md, srcu_idx);
3234 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3235 enum pr_type type, bool abort)
3237 struct mapped_device *md = bdev->bd_disk->private_data;
3238 const struct pr_ops *ops;
3241 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3245 ops = bdev->bd_disk->fops->pr_ops;
3246 if (ops && ops->pr_preempt)
3247 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3251 dm_unprepare_ioctl(md, srcu_idx);
3255 static int dm_pr_clear(struct block_device *bdev, u64 key)
3257 struct mapped_device *md = bdev->bd_disk->private_data;
3258 const struct pr_ops *ops;
3261 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3265 ops = bdev->bd_disk->fops->pr_ops;
3266 if (ops && ops->pr_clear)
3267 r = ops->pr_clear(bdev, key);
3271 dm_unprepare_ioctl(md, srcu_idx);
3275 static const struct pr_ops dm_pr_ops = {
3276 .pr_register = dm_pr_register,
3277 .pr_reserve = dm_pr_reserve,
3278 .pr_release = dm_pr_release,
3279 .pr_preempt = dm_pr_preempt,
3280 .pr_clear = dm_pr_clear,
3283 static const struct block_device_operations dm_blk_dops = {
3284 .open = dm_blk_open,
3285 .release = dm_blk_close,
3286 .ioctl = dm_blk_ioctl,
3287 .getgeo = dm_blk_getgeo,
3288 .pr_ops = &dm_pr_ops,
3289 .owner = THIS_MODULE
3292 static const struct dax_operations dm_dax_ops = {
3293 .direct_access = dm_dax_direct_access,
3294 .copy_from_iter = dm_dax_copy_from_iter,
3295 .copy_to_iter = dm_dax_copy_to_iter,
3301 module_init(dm_init);
3302 module_exit(dm_exit);
3304 module_param(major, uint, 0);
3305 MODULE_PARM_DESC(major, "The major number of the device mapper");
3307 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3308 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3310 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3311 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3313 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3314 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3316 MODULE_DESCRIPTION(DM_NAME " driver");
3317 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3318 MODULE_LICENSE("GPL");