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/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
27 #include <linux/refcount.h>
29 #define DM_MSG_PREFIX "core"
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name = DM_NAME;
40 static unsigned int major = 0;
41 static unsigned int _major = 0;
43 static DEFINE_IDR(_minor_idr);
45 static DEFINE_SPINLOCK(_minor_lock);
47 static void do_deferred_remove(struct work_struct *w);
49 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
51 static struct workqueue_struct *deferred_remove_workqueue;
53 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
56 void dm_issue_global_event(void)
58 atomic_inc(&dm_global_event_nr);
59 wake_up(&dm_global_eventq);
63 * One of these is allocated (on-stack) per original bio.
70 unsigned sector_count;
74 * One of these is allocated per clone bio.
76 #define DM_TIO_MAGIC 7282014
81 unsigned target_bio_nr;
88 * One of these is allocated per original bio.
89 * It contains the first clone used for that original.
91 #define DM_IO_MAGIC 5191977
94 struct mapped_device *md;
98 unsigned long start_time;
99 spinlock_t endio_lock;
100 struct dm_stats_aux stats_aux;
101 /* last member of dm_target_io is 'struct bio' */
102 struct dm_target_io tio;
105 void *dm_per_bio_data(struct bio *bio, size_t data_size)
107 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
108 if (!tio->inside_dm_io)
109 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
110 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
112 EXPORT_SYMBOL_GPL(dm_per_bio_data);
114 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
116 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
117 if (io->magic == DM_IO_MAGIC)
118 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
119 BUG_ON(io->magic != DM_TIO_MAGIC);
120 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
122 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
124 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
126 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
128 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
130 #define MINOR_ALLOCED ((void *)-1)
133 * Bits for the md->flags field.
135 #define DMF_BLOCK_IO_FOR_SUSPEND 0
136 #define DMF_SUSPENDED 1
138 #define DMF_FREEING 3
139 #define DMF_DELETING 4
140 #define DMF_NOFLUSH_SUSPENDING 5
141 #define DMF_DEFERRED_REMOVE 6
142 #define DMF_SUSPENDED_INTERNALLY 7
144 #define DM_NUMA_NODE NUMA_NO_NODE
145 static int dm_numa_node = DM_NUMA_NODE;
148 * For mempools pre-allocation at the table loading time.
150 struct dm_md_mempools {
152 struct bio_set io_bs;
155 struct table_device {
156 struct list_head list;
158 struct dm_dev dm_dev;
161 static struct kmem_cache *_rq_tio_cache;
162 static struct kmem_cache *_rq_cache;
165 * Bio-based DM's mempools' reserved IOs set by the user.
167 #define RESERVED_BIO_BASED_IOS 16
168 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
170 static int __dm_get_module_param_int(int *module_param, int min, int max)
172 int param = READ_ONCE(*module_param);
173 int modified_param = 0;
174 bool modified = true;
177 modified_param = min;
178 else if (param > max)
179 modified_param = max;
184 (void)cmpxchg(module_param, param, modified_param);
185 param = modified_param;
191 unsigned __dm_get_module_param(unsigned *module_param,
192 unsigned def, unsigned max)
194 unsigned param = READ_ONCE(*module_param);
195 unsigned modified_param = 0;
198 modified_param = def;
199 else if (param > max)
200 modified_param = max;
202 if (modified_param) {
203 (void)cmpxchg(module_param, param, modified_param);
204 param = modified_param;
210 unsigned dm_get_reserved_bio_based_ios(void)
212 return __dm_get_module_param(&reserved_bio_based_ios,
213 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
215 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
217 static unsigned dm_get_numa_node(void)
219 return __dm_get_module_param_int(&dm_numa_node,
220 DM_NUMA_NODE, num_online_nodes() - 1);
223 static int __init local_init(void)
227 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
231 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
232 __alignof__(struct request), 0, NULL);
234 goto out_free_rq_tio_cache;
236 r = dm_uevent_init();
238 goto out_free_rq_cache;
240 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
241 if (!deferred_remove_workqueue) {
243 goto out_uevent_exit;
247 r = register_blkdev(_major, _name);
249 goto out_free_workqueue;
257 destroy_workqueue(deferred_remove_workqueue);
261 kmem_cache_destroy(_rq_cache);
262 out_free_rq_tio_cache:
263 kmem_cache_destroy(_rq_tio_cache);
268 static void local_exit(void)
270 flush_scheduled_work();
271 destroy_workqueue(deferred_remove_workqueue);
273 kmem_cache_destroy(_rq_cache);
274 kmem_cache_destroy(_rq_tio_cache);
275 unregister_blkdev(_major, _name);
280 DMINFO("cleaned up");
283 static int (*_inits[])(void) __initdata = {
294 static void (*_exits[])(void) = {
305 static int __init dm_init(void)
307 const int count = ARRAY_SIZE(_inits);
311 for (i = 0; i < count; i++) {
326 static void __exit dm_exit(void)
328 int i = ARRAY_SIZE(_exits);
334 * Should be empty by this point.
336 idr_destroy(&_minor_idr);
340 * Block device functions
342 int dm_deleting_md(struct mapped_device *md)
344 return test_bit(DMF_DELETING, &md->flags);
347 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
349 struct mapped_device *md;
351 spin_lock(&_minor_lock);
353 md = bdev->bd_disk->private_data;
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
364 atomic_inc(&md->open_count);
366 spin_unlock(&_minor_lock);
368 return md ? 0 : -ENXIO;
371 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
373 struct mapped_device *md;
375 spin_lock(&_minor_lock);
377 md = disk->private_data;
381 if (atomic_dec_and_test(&md->open_count) &&
382 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
383 queue_work(deferred_remove_workqueue, &deferred_remove_work);
387 spin_unlock(&_minor_lock);
390 int dm_open_count(struct mapped_device *md)
392 return atomic_read(&md->open_count);
396 * Guarantees nothing is using the device before it's deleted.
398 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
402 spin_lock(&_minor_lock);
404 if (dm_open_count(md)) {
407 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
408 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
411 set_bit(DMF_DELETING, &md->flags);
413 spin_unlock(&_minor_lock);
418 int dm_cancel_deferred_remove(struct mapped_device *md)
422 spin_lock(&_minor_lock);
424 if (test_bit(DMF_DELETING, &md->flags))
427 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
429 spin_unlock(&_minor_lock);
434 static void do_deferred_remove(struct work_struct *w)
436 dm_deferred_remove();
439 sector_t dm_get_size(struct mapped_device *md)
441 return get_capacity(md->disk);
444 struct request_queue *dm_get_md_queue(struct mapped_device *md)
449 struct dm_stats *dm_get_stats(struct mapped_device *md)
454 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
456 struct mapped_device *md = bdev->bd_disk->private_data;
458 return dm_get_geometry(md, geo);
461 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
462 struct blk_zone *zones, unsigned int *nr_zones,
465 #ifdef CONFIG_BLK_DEV_ZONED
466 struct mapped_device *md = disk->private_data;
467 struct dm_target *tgt;
468 struct dm_table *map;
471 if (dm_suspended_md(md))
474 map = dm_get_live_table(md, &srcu_idx);
478 tgt = dm_table_find_target(map, sector);
479 if (!dm_target_is_valid(tgt)) {
485 * If we are executing this, we already know that the block device
486 * is a zoned device and so each target should have support for that
487 * type of drive. A missing report_zones method means that the target
488 * driver has a problem.
490 if (WARN_ON(!tgt->type->report_zones)) {
496 * blkdev_report_zones() will loop and call this again to cover all the
497 * zones of the target, eventually moving on to the next target.
498 * So there is no need to loop here trying to fill the entire array
501 ret = tgt->type->report_zones(tgt, sector, zones,
505 dm_put_live_table(md, srcu_idx);
512 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
513 struct block_device **bdev)
514 __acquires(md->io_barrier)
516 struct dm_target *tgt;
517 struct dm_table *map;
522 map = dm_get_live_table(md, srcu_idx);
523 if (!map || !dm_table_get_size(map))
526 /* We only support devices that have a single target */
527 if (dm_table_get_num_targets(map) != 1)
530 tgt = dm_table_get_target(map, 0);
531 if (!tgt->type->prepare_ioctl)
534 if (dm_suspended_md(md))
537 r = tgt->type->prepare_ioctl(tgt, bdev);
538 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
539 dm_put_live_table(md, *srcu_idx);
547 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
548 __releases(md->io_barrier)
550 dm_put_live_table(md, srcu_idx);
553 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
554 unsigned int cmd, unsigned long arg)
556 struct mapped_device *md = bdev->bd_disk->private_data;
559 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
565 * Target determined this ioctl is being issued against a
566 * subset of the parent bdev; require extra privileges.
568 if (!capable(CAP_SYS_RAWIO)) {
570 "%s: sending ioctl %x to DM device without required privilege.",
577 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
579 dm_unprepare_ioctl(md, srcu_idx);
583 static void start_io_acct(struct dm_io *io);
585 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
588 struct dm_target_io *tio;
591 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
595 tio = container_of(clone, struct dm_target_io, clone);
596 tio->inside_dm_io = true;
599 io = container_of(tio, struct dm_io, tio);
600 io->magic = DM_IO_MAGIC;
602 atomic_set(&io->io_count, 1);
605 spin_lock_init(&io->endio_lock);
612 static void free_io(struct mapped_device *md, struct dm_io *io)
614 bio_put(&io->tio.clone);
617 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
618 unsigned target_bio_nr, gfp_t gfp_mask)
620 struct dm_target_io *tio;
622 if (!ci->io->tio.io) {
623 /* the dm_target_io embedded in ci->io is available */
626 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
630 tio = container_of(clone, struct dm_target_io, clone);
631 tio->inside_dm_io = false;
634 tio->magic = DM_TIO_MAGIC;
637 tio->target_bio_nr = target_bio_nr;
642 static void free_tio(struct dm_target_io *tio)
644 if (tio->inside_dm_io)
646 bio_put(&tio->clone);
649 static bool md_in_flight_bios(struct mapped_device *md)
652 struct hd_struct *part = &dm_disk(md)->part0;
655 for_each_possible_cpu(cpu) {
656 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
657 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
663 static bool md_in_flight(struct mapped_device *md)
665 if (queue_is_mq(md->queue))
666 return blk_mq_queue_inflight(md->queue);
668 return md_in_flight_bios(md);
671 static void start_io_acct(struct dm_io *io)
673 struct mapped_device *md = io->md;
674 struct bio *bio = io->orig_bio;
676 io->start_time = jiffies;
678 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
679 &dm_disk(md)->part0);
681 if (unlikely(dm_stats_used(&md->stats)))
682 dm_stats_account_io(&md->stats, bio_data_dir(bio),
683 bio->bi_iter.bi_sector, bio_sectors(bio),
684 false, 0, &io->stats_aux);
687 static void end_io_acct(struct dm_io *io)
689 struct mapped_device *md = io->md;
690 struct bio *bio = io->orig_bio;
691 unsigned long duration = jiffies - io->start_time;
693 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
696 if (unlikely(dm_stats_used(&md->stats)))
697 dm_stats_account_io(&md->stats, bio_data_dir(bio),
698 bio->bi_iter.bi_sector, bio_sectors(bio),
699 true, duration, &io->stats_aux);
701 /* nudge anyone waiting on suspend queue */
702 if (unlikely(wq_has_sleeper(&md->wait)))
707 * Add the bio to the list of deferred io.
709 static void queue_io(struct mapped_device *md, struct bio *bio)
713 spin_lock_irqsave(&md->deferred_lock, flags);
714 bio_list_add(&md->deferred, bio);
715 spin_unlock_irqrestore(&md->deferred_lock, flags);
716 queue_work(md->wq, &md->work);
720 * Everyone (including functions in this file), should use this
721 * function to access the md->map field, and make sure they call
722 * dm_put_live_table() when finished.
724 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
726 *srcu_idx = srcu_read_lock(&md->io_barrier);
728 return srcu_dereference(md->map, &md->io_barrier);
731 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
733 srcu_read_unlock(&md->io_barrier, srcu_idx);
736 void dm_sync_table(struct mapped_device *md)
738 synchronize_srcu(&md->io_barrier);
739 synchronize_rcu_expedited();
743 * A fast alternative to dm_get_live_table/dm_put_live_table.
744 * The caller must not block between these two functions.
746 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
749 return rcu_dereference(md->map);
752 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
757 static char *_dm_claim_ptr = "I belong to device-mapper";
760 * Open a table device so we can use it as a map destination.
762 static int open_table_device(struct table_device *td, dev_t dev,
763 struct mapped_device *md)
765 struct block_device *bdev;
769 BUG_ON(td->dm_dev.bdev);
771 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
773 return PTR_ERR(bdev);
775 r = bd_link_disk_holder(bdev, dm_disk(md));
777 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
781 td->dm_dev.bdev = bdev;
782 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
787 * Close a table device that we've been using.
789 static void close_table_device(struct table_device *td, struct mapped_device *md)
791 if (!td->dm_dev.bdev)
794 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
795 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
796 put_dax(td->dm_dev.dax_dev);
797 td->dm_dev.bdev = NULL;
798 td->dm_dev.dax_dev = NULL;
801 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
803 struct table_device *td;
805 list_for_each_entry(td, l, list)
806 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
812 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
813 struct dm_dev **result) {
815 struct table_device *td;
817 mutex_lock(&md->table_devices_lock);
818 td = find_table_device(&md->table_devices, dev, mode);
820 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
822 mutex_unlock(&md->table_devices_lock);
826 td->dm_dev.mode = mode;
827 td->dm_dev.bdev = NULL;
829 if ((r = open_table_device(td, dev, md))) {
830 mutex_unlock(&md->table_devices_lock);
835 format_dev_t(td->dm_dev.name, dev);
837 refcount_set(&td->count, 1);
838 list_add(&td->list, &md->table_devices);
840 refcount_inc(&td->count);
842 mutex_unlock(&md->table_devices_lock);
844 *result = &td->dm_dev;
847 EXPORT_SYMBOL_GPL(dm_get_table_device);
849 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
851 struct table_device *td = container_of(d, struct table_device, dm_dev);
853 mutex_lock(&md->table_devices_lock);
854 if (refcount_dec_and_test(&td->count)) {
855 close_table_device(td, md);
859 mutex_unlock(&md->table_devices_lock);
861 EXPORT_SYMBOL(dm_put_table_device);
863 static void free_table_devices(struct list_head *devices)
865 struct list_head *tmp, *next;
867 list_for_each_safe(tmp, next, devices) {
868 struct table_device *td = list_entry(tmp, struct table_device, list);
870 DMWARN("dm_destroy: %s still exists with %d references",
871 td->dm_dev.name, refcount_read(&td->count));
877 * Get the geometry associated with a dm device
879 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
887 * Set the geometry of a device.
889 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
891 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
893 if (geo->start > sz) {
894 DMWARN("Start sector is beyond the geometry limits.");
903 static int __noflush_suspending(struct mapped_device *md)
905 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
909 * Decrements the number of outstanding ios that a bio has been
910 * cloned into, completing the original io if necc.
912 static void dec_pending(struct dm_io *io, blk_status_t error)
915 blk_status_t io_error;
917 struct mapped_device *md = io->md;
919 /* Push-back supersedes any I/O errors */
920 if (unlikely(error)) {
921 spin_lock_irqsave(&io->endio_lock, flags);
922 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
924 spin_unlock_irqrestore(&io->endio_lock, flags);
927 if (atomic_dec_and_test(&io->io_count)) {
928 if (io->status == BLK_STS_DM_REQUEUE) {
930 * Target requested pushing back the I/O.
932 spin_lock_irqsave(&md->deferred_lock, flags);
933 if (__noflush_suspending(md))
934 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
935 bio_list_add_head(&md->deferred, io->orig_bio);
937 /* noflush suspend was interrupted. */
938 io->status = BLK_STS_IOERR;
939 spin_unlock_irqrestore(&md->deferred_lock, flags);
942 io_error = io->status;
947 if (io_error == BLK_STS_DM_REQUEUE)
950 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
952 * Preflush done for flush with data, reissue
953 * without REQ_PREFLUSH.
955 bio->bi_opf &= ~REQ_PREFLUSH;
958 /* done with normal IO or empty flush */
960 bio->bi_status = io_error;
966 void disable_write_same(struct mapped_device *md)
968 struct queue_limits *limits = dm_get_queue_limits(md);
970 /* device doesn't really support WRITE SAME, disable it */
971 limits->max_write_same_sectors = 0;
974 void disable_write_zeroes(struct mapped_device *md)
976 struct queue_limits *limits = dm_get_queue_limits(md);
978 /* device doesn't really support WRITE ZEROES, disable it */
979 limits->max_write_zeroes_sectors = 0;
982 static void clone_endio(struct bio *bio)
984 blk_status_t error = bio->bi_status;
985 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
986 struct dm_io *io = tio->io;
987 struct mapped_device *md = tio->io->md;
988 dm_endio_fn endio = tio->ti->type->end_io;
990 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
991 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
992 !bio->bi_disk->queue->limits.max_write_same_sectors)
993 disable_write_same(md);
994 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
995 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
996 disable_write_zeroes(md);
1000 int r = endio(tio->ti, bio, &error);
1002 case DM_ENDIO_REQUEUE:
1003 error = BLK_STS_DM_REQUEUE;
1007 case DM_ENDIO_INCOMPLETE:
1008 /* The target will handle the io */
1011 DMWARN("unimplemented target endio return value: %d", r);
1017 dec_pending(io, error);
1021 * Return maximum size of I/O possible at the supplied sector up to the current
1024 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1026 sector_t target_offset = dm_target_offset(ti, sector);
1028 return ti->len - target_offset;
1031 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1033 sector_t len = max_io_len_target_boundary(sector, ti);
1034 sector_t offset, max_len;
1037 * Does the target need to split even further?
1039 if (ti->max_io_len) {
1040 offset = dm_target_offset(ti, sector);
1041 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1042 max_len = sector_div(offset, ti->max_io_len);
1044 max_len = offset & (ti->max_io_len - 1);
1045 max_len = ti->max_io_len - max_len;
1054 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1056 if (len > UINT_MAX) {
1057 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1058 (unsigned long long)len, UINT_MAX);
1059 ti->error = "Maximum size of target IO is too large";
1064 * BIO based queue uses its own splitting. When multipage bvecs
1065 * is switched on, size of the incoming bio may be too big to
1066 * be handled in some targets, such as crypt.
1068 * When these targets are ready for the big bio, we can remove
1071 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1075 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1077 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1078 sector_t sector, int *srcu_idx)
1079 __acquires(md->io_barrier)
1081 struct dm_table *map;
1082 struct dm_target *ti;
1084 map = dm_get_live_table(md, srcu_idx);
1088 ti = dm_table_find_target(map, sector);
1089 if (!dm_target_is_valid(ti))
1095 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1096 long nr_pages, void **kaddr, pfn_t *pfn)
1098 struct mapped_device *md = dax_get_private(dax_dev);
1099 sector_t sector = pgoff * PAGE_SECTORS;
1100 struct dm_target *ti;
1101 long len, ret = -EIO;
1104 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1108 if (!ti->type->direct_access)
1110 len = max_io_len(sector, ti) / PAGE_SECTORS;
1113 nr_pages = min(len, nr_pages);
1114 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1117 dm_put_live_table(md, srcu_idx);
1122 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1123 void *addr, size_t bytes, struct iov_iter *i)
1125 struct mapped_device *md = dax_get_private(dax_dev);
1126 sector_t sector = pgoff * PAGE_SECTORS;
1127 struct dm_target *ti;
1131 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1135 if (!ti->type->dax_copy_from_iter) {
1136 ret = copy_from_iter(addr, bytes, i);
1139 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1141 dm_put_live_table(md, srcu_idx);
1146 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1147 void *addr, size_t bytes, struct iov_iter *i)
1149 struct mapped_device *md = dax_get_private(dax_dev);
1150 sector_t sector = pgoff * PAGE_SECTORS;
1151 struct dm_target *ti;
1155 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1159 if (!ti->type->dax_copy_to_iter) {
1160 ret = copy_to_iter(addr, bytes, i);
1163 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1165 dm_put_live_table(md, srcu_idx);
1171 * A target may call dm_accept_partial_bio only from the map routine. It is
1172 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1174 * dm_accept_partial_bio informs the dm that the target only wants to process
1175 * additional n_sectors sectors of the bio and the rest of the data should be
1176 * sent in a next bio.
1178 * A diagram that explains the arithmetics:
1179 * +--------------------+---------------+-------+
1181 * +--------------------+---------------+-------+
1183 * <-------------- *tio->len_ptr --------------->
1184 * <------- bi_size ------->
1187 * Region 1 was already iterated over with bio_advance or similar function.
1188 * (it may be empty if the target doesn't use bio_advance)
1189 * Region 2 is the remaining bio size that the target wants to process.
1190 * (it may be empty if region 1 is non-empty, although there is no reason
1192 * The target requires that region 3 is to be sent in the next bio.
1194 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1195 * the partially processed part (the sum of regions 1+2) must be the same for all
1196 * copies of the bio.
1198 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1200 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1201 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1202 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1203 BUG_ON(bi_size > *tio->len_ptr);
1204 BUG_ON(n_sectors > bi_size);
1205 *tio->len_ptr -= bi_size - n_sectors;
1206 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1208 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1211 * The zone descriptors obtained with a zone report indicate
1212 * zone positions within the underlying device of the target. The zone
1213 * descriptors must be remapped to match their position within the dm device.
1214 * The caller target should obtain the zones information using
1215 * blkdev_report_zones() to ensure that remapping for partition offset is
1218 void dm_remap_zone_report(struct dm_target *ti, sector_t start,
1219 struct blk_zone *zones, unsigned int *nr_zones)
1221 #ifdef CONFIG_BLK_DEV_ZONED
1222 struct blk_zone *zone;
1223 unsigned int nrz = *nr_zones;
1227 * Remap the start sector and write pointer position of the zones in
1228 * the array. Since we may have obtained from the target underlying
1229 * device more zones that the target size, also adjust the number
1232 for (i = 0; i < nrz; i++) {
1234 if (zone->start >= start + ti->len) {
1235 memset(zone, 0, sizeof(struct blk_zone) * (nrz - i));
1239 zone->start = zone->start + ti->begin - start;
1240 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;
1252 #else /* !CONFIG_BLK_DEV_ZONED */
1256 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1258 static blk_qc_t __map_bio(struct dm_target_io *tio)
1262 struct bio *clone = &tio->clone;
1263 struct dm_io *io = tio->io;
1264 struct mapped_device *md = io->md;
1265 struct dm_target *ti = tio->ti;
1266 blk_qc_t ret = BLK_QC_T_NONE;
1268 clone->bi_end_io = clone_endio;
1271 * Map the clone. If r == 0 we don't need to do
1272 * anything, the target has assumed ownership of
1275 atomic_inc(&io->io_count);
1276 sector = clone->bi_iter.bi_sector;
1278 r = ti->type->map(ti, clone);
1280 case DM_MAPIO_SUBMITTED:
1282 case DM_MAPIO_REMAPPED:
1283 /* the bio has been remapped so dispatch it */
1284 trace_block_bio_remap(clone->bi_disk->queue, clone,
1285 bio_dev(io->orig_bio), sector);
1286 if (md->type == DM_TYPE_NVME_BIO_BASED)
1287 ret = direct_make_request(clone);
1289 ret = generic_make_request(clone);
1293 dec_pending(io, BLK_STS_IOERR);
1295 case DM_MAPIO_REQUEUE:
1297 dec_pending(io, BLK_STS_DM_REQUEUE);
1300 DMWARN("unimplemented target map return value: %d", r);
1307 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1309 bio->bi_iter.bi_sector = sector;
1310 bio->bi_iter.bi_size = to_bytes(len);
1314 * Creates a bio that consists of range of complete bvecs.
1316 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1317 sector_t sector, unsigned len)
1319 struct bio *clone = &tio->clone;
1321 __bio_clone_fast(clone, bio);
1323 if (bio_integrity(bio)) {
1326 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1327 !dm_target_passes_integrity(tio->ti->type))) {
1328 DMWARN("%s: the target %s doesn't support integrity data.",
1329 dm_device_name(tio->io->md),
1330 tio->ti->type->name);
1334 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1339 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1340 clone->bi_iter.bi_size = to_bytes(len);
1342 if (bio_integrity(bio))
1343 bio_integrity_trim(clone);
1348 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1349 struct dm_target *ti, unsigned num_bios)
1351 struct dm_target_io *tio;
1357 if (num_bios == 1) {
1358 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1359 bio_list_add(blist, &tio->clone);
1363 for (try = 0; try < 2; try++) {
1368 mutex_lock(&ci->io->md->table_devices_lock);
1369 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1370 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1374 bio_list_add(blist, &tio->clone);
1377 mutex_unlock(&ci->io->md->table_devices_lock);
1378 if (bio_nr == num_bios)
1381 while ((bio = bio_list_pop(blist))) {
1382 tio = container_of(bio, struct dm_target_io, clone);
1388 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1389 struct dm_target_io *tio, unsigned *len)
1391 struct bio *clone = &tio->clone;
1395 __bio_clone_fast(clone, ci->bio);
1397 bio_setup_sector(clone, ci->sector, *len);
1399 return __map_bio(tio);
1402 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1403 unsigned num_bios, unsigned *len)
1405 struct bio_list blist = BIO_EMPTY_LIST;
1407 struct dm_target_io *tio;
1409 alloc_multiple_bios(&blist, ci, ti, num_bios);
1411 while ((bio = bio_list_pop(&blist))) {
1412 tio = container_of(bio, struct dm_target_io, clone);
1413 (void) __clone_and_map_simple_bio(ci, tio, len);
1417 static int __send_empty_flush(struct clone_info *ci)
1419 unsigned target_nr = 0;
1420 struct dm_target *ti;
1423 * Empty flush uses a statically initialized bio, as the base for
1424 * cloning. However, blkg association requires that a bdev is
1425 * associated with a gendisk, which doesn't happen until the bdev is
1426 * opened. So, blkg association is done at issue time of the flush
1427 * rather than when the device is created in alloc_dev().
1429 bio_set_dev(ci->bio, ci->io->md->bdev);
1431 BUG_ON(bio_has_data(ci->bio));
1432 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1433 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1435 bio_disassociate_blkg(ci->bio);
1440 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1441 sector_t sector, unsigned *len)
1443 struct bio *bio = ci->bio;
1444 struct dm_target_io *tio;
1447 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1449 r = clone_bio(tio, bio, sector, *len);
1454 (void) __map_bio(tio);
1459 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1461 static unsigned get_num_discard_bios(struct dm_target *ti)
1463 return ti->num_discard_bios;
1466 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1468 return ti->num_secure_erase_bios;
1471 static unsigned get_num_write_same_bios(struct dm_target *ti)
1473 return ti->num_write_same_bios;
1476 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1478 return ti->num_write_zeroes_bios;
1481 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1483 static bool is_split_required_for_discard(struct dm_target *ti)
1485 return ti->split_discard_bios;
1488 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1489 unsigned num_bios, bool is_split_required)
1494 * Even though the device advertised support for this type of
1495 * request, that does not mean every target supports it, and
1496 * reconfiguration might also have changed that since the
1497 * check was performed.
1502 if (!is_split_required)
1503 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1505 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1507 __send_duplicate_bios(ci, ti, num_bios, &len);
1510 ci->sector_count -= len;
1515 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1517 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti),
1518 is_split_required_for_discard(ti));
1521 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1523 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti), false);
1526 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1528 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti), false);
1531 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1533 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti), false);
1536 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1539 struct bio *bio = ci->bio;
1541 if (bio_op(bio) == REQ_OP_DISCARD)
1542 *result = __send_discard(ci, ti);
1543 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1544 *result = __send_secure_erase(ci, ti);
1545 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1546 *result = __send_write_same(ci, ti);
1547 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1548 *result = __send_write_zeroes(ci, ti);
1556 * Select the correct strategy for processing a non-flush bio.
1558 static int __split_and_process_non_flush(struct clone_info *ci)
1560 struct dm_target *ti;
1564 ti = dm_table_find_target(ci->map, ci->sector);
1565 if (!dm_target_is_valid(ti))
1568 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1571 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1573 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1578 ci->sector_count -= len;
1583 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1584 struct dm_table *map, struct bio *bio)
1587 ci->io = alloc_io(md, bio);
1588 ci->sector = bio->bi_iter.bi_sector;
1591 #define __dm_part_stat_sub(part, field, subnd) \
1592 (part_stat_get(part, field) -= (subnd))
1595 * Entry point to split a bio into clones and submit them to the targets.
1597 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1598 struct dm_table *map, struct bio *bio)
1600 struct clone_info ci;
1601 blk_qc_t ret = BLK_QC_T_NONE;
1604 if (unlikely(!map)) {
1609 blk_queue_split(md->queue, &bio);
1611 init_clone_info(&ci, md, map, bio);
1613 if (bio->bi_opf & REQ_PREFLUSH) {
1614 struct bio flush_bio;
1617 * Use an on-stack bio for this, it's safe since we don't
1618 * need to reference it after submit. It's just used as
1619 * the basis for the clone(s).
1621 bio_init(&flush_bio, NULL, 0);
1622 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1623 ci.bio = &flush_bio;
1624 ci.sector_count = 0;
1625 error = __send_empty_flush(&ci);
1626 /* dec_pending submits any data associated with flush */
1627 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1629 ci.sector_count = 0;
1630 error = __split_and_process_non_flush(&ci);
1633 ci.sector_count = bio_sectors(bio);
1634 while (ci.sector_count && !error) {
1635 error = __split_and_process_non_flush(&ci);
1636 if (current->bio_list && ci.sector_count && !error) {
1638 * Remainder must be passed to generic_make_request()
1639 * so that it gets handled *after* bios already submitted
1640 * have been completely processed.
1641 * We take a clone of the original to store in
1642 * ci.io->orig_bio to be used by end_io_acct() and
1643 * for dec_pending to use for completion handling.
1645 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1646 GFP_NOIO, &md->queue->bio_split);
1647 ci.io->orig_bio = b;
1650 * Adjust IO stats for each split, otherwise upon queue
1651 * reentry there will be redundant IO accounting.
1652 * NOTE: this is a stop-gap fix, a proper fix involves
1653 * significant refactoring of DM core's bio splitting
1654 * (by eliminating DM's splitting and just using bio_split)
1657 __dm_part_stat_sub(&dm_disk(md)->part0,
1658 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1662 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1663 ret = generic_make_request(bio);
1669 /* drop the extra reference count */
1670 dec_pending(ci.io, errno_to_blk_status(error));
1675 * Optimized variant of __split_and_process_bio that leverages the
1676 * fact that targets that use it do _not_ have a need to split bios.
1678 static blk_qc_t __process_bio(struct mapped_device *md,
1679 struct dm_table *map, struct bio *bio)
1681 struct clone_info ci;
1682 blk_qc_t ret = BLK_QC_T_NONE;
1685 if (unlikely(!map)) {
1690 init_clone_info(&ci, md, map, bio);
1692 if (bio->bi_opf & REQ_PREFLUSH) {
1693 struct bio flush_bio;
1696 * Use an on-stack bio for this, it's safe since we don't
1697 * need to reference it after submit. It's just used as
1698 * the basis for the clone(s).
1700 bio_init(&flush_bio, NULL, 0);
1701 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1702 ci.bio = &flush_bio;
1703 ci.sector_count = 0;
1704 error = __send_empty_flush(&ci);
1705 /* dec_pending submits any data associated with flush */
1707 struct dm_target *ti = md->immutable_target;
1708 struct dm_target_io *tio;
1711 * Defend against IO still getting in during teardown
1712 * - as was seen for a time with nvme-fcloop
1714 if (WARN_ON_ONCE(!ti || !dm_target_is_valid(ti))) {
1720 ci.sector_count = bio_sectors(bio);
1721 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1724 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1725 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1728 /* drop the extra reference count */
1729 dec_pending(ci.io, errno_to_blk_status(error));
1733 static blk_qc_t dm_process_bio(struct mapped_device *md,
1734 struct dm_table *map, struct bio *bio)
1736 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1737 return __process_bio(md, map, bio);
1739 return __split_and_process_bio(md, map, bio);
1742 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1744 struct mapped_device *md = q->queuedata;
1745 blk_qc_t ret = BLK_QC_T_NONE;
1747 struct dm_table *map;
1749 map = dm_get_live_table(md, &srcu_idx);
1751 /* if we're suspended, we have to queue this io for later */
1752 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1753 dm_put_live_table(md, srcu_idx);
1755 if (!(bio->bi_opf & REQ_RAHEAD))
1762 ret = dm_process_bio(md, map, bio);
1764 dm_put_live_table(md, srcu_idx);
1768 static int dm_any_congested(void *congested_data, int bdi_bits)
1771 struct mapped_device *md = congested_data;
1772 struct dm_table *map;
1774 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1775 if (dm_request_based(md)) {
1777 * With request-based DM we only need to check the
1778 * top-level queue for congestion.
1780 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1782 map = dm_get_live_table_fast(md);
1784 r = dm_table_any_congested(map, bdi_bits);
1785 dm_put_live_table_fast(md);
1792 /*-----------------------------------------------------------------
1793 * An IDR is used to keep track of allocated minor numbers.
1794 *---------------------------------------------------------------*/
1795 static void free_minor(int minor)
1797 spin_lock(&_minor_lock);
1798 idr_remove(&_minor_idr, minor);
1799 spin_unlock(&_minor_lock);
1803 * See if the device with a specific minor # is free.
1805 static int specific_minor(int minor)
1809 if (minor >= (1 << MINORBITS))
1812 idr_preload(GFP_KERNEL);
1813 spin_lock(&_minor_lock);
1815 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1817 spin_unlock(&_minor_lock);
1820 return r == -ENOSPC ? -EBUSY : r;
1824 static int next_free_minor(int *minor)
1828 idr_preload(GFP_KERNEL);
1829 spin_lock(&_minor_lock);
1831 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1833 spin_unlock(&_minor_lock);
1841 static const struct block_device_operations dm_blk_dops;
1842 static const struct dax_operations dm_dax_ops;
1844 static void dm_wq_work(struct work_struct *work);
1846 static void dm_init_normal_md_queue(struct mapped_device *md)
1849 * Initialize aspects of queue that aren't relevant for blk-mq
1851 md->queue->backing_dev_info->congested_fn = dm_any_congested;
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 dax_device *dax_dev = NULL;
1899 struct mapped_device *md;
1902 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1904 DMWARN("unable to allocate device, out of memory.");
1908 if (!try_module_get(THIS_MODULE))
1909 goto bad_module_get;
1911 /* get a minor number for the dev */
1912 if (minor == DM_ANY_MINOR)
1913 r = next_free_minor(&minor);
1915 r = specific_minor(minor);
1919 r = init_srcu_struct(&md->io_barrier);
1921 goto bad_io_barrier;
1923 md->numa_node_id = numa_node_id;
1924 md->init_tio_pdu = false;
1925 md->type = DM_TYPE_NONE;
1926 mutex_init(&md->suspend_lock);
1927 mutex_init(&md->type_lock);
1928 mutex_init(&md->table_devices_lock);
1929 spin_lock_init(&md->deferred_lock);
1930 atomic_set(&md->holders, 1);
1931 atomic_set(&md->open_count, 0);
1932 atomic_set(&md->event_nr, 0);
1933 atomic_set(&md->uevent_seq, 0);
1934 INIT_LIST_HEAD(&md->uevent_list);
1935 INIT_LIST_HEAD(&md->table_devices);
1936 spin_lock_init(&md->uevent_lock);
1938 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1941 md->queue->queuedata = md;
1942 md->queue->backing_dev_info->congested_data = md;
1944 md->disk = alloc_disk_node(1, md->numa_node_id);
1948 init_waitqueue_head(&md->wait);
1949 INIT_WORK(&md->work, dm_wq_work);
1950 init_waitqueue_head(&md->eventq);
1951 init_completion(&md->kobj_holder.completion);
1953 md->disk->major = _major;
1954 md->disk->first_minor = minor;
1955 md->disk->fops = &dm_blk_dops;
1956 md->disk->queue = md->queue;
1957 md->disk->private_data = md;
1958 sprintf(md->disk->disk_name, "dm-%d", minor);
1960 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1961 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1965 md->dax_dev = dax_dev;
1967 add_disk_no_queue_reg(md->disk);
1968 format_dev_t(md->name, MKDEV(_major, minor));
1970 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1974 md->bdev = bdget_disk(md->disk, 0);
1978 dm_stats_init(&md->stats);
1980 /* Populate the mapping, nobody knows we exist yet */
1981 spin_lock(&_minor_lock);
1982 old_md = idr_replace(&_minor_idr, md, minor);
1983 spin_unlock(&_minor_lock);
1985 BUG_ON(old_md != MINOR_ALLOCED);
1990 cleanup_mapped_device(md);
1994 module_put(THIS_MODULE);
2000 static void unlock_fs(struct mapped_device *md);
2002 static void free_dev(struct mapped_device *md)
2004 int minor = MINOR(disk_devt(md->disk));
2008 cleanup_mapped_device(md);
2010 free_table_devices(&md->table_devices);
2011 dm_stats_cleanup(&md->stats);
2014 module_put(THIS_MODULE);
2018 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2020 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2023 if (dm_table_bio_based(t)) {
2025 * The md may already have mempools that need changing.
2026 * If so, reload bioset because front_pad may have changed
2027 * because a different table was loaded.
2029 bioset_exit(&md->bs);
2030 bioset_exit(&md->io_bs);
2032 } else if (bioset_initialized(&md->bs)) {
2034 * There's no need to reload with request-based dm
2035 * because the size of front_pad doesn't change.
2036 * Note for future: If you are to reload bioset,
2037 * prep-ed requests in the queue may refer
2038 * to bio from the old bioset, so you must walk
2039 * through the queue to unprep.
2045 bioset_initialized(&md->bs) ||
2046 bioset_initialized(&md->io_bs));
2048 ret = bioset_init_from_src(&md->bs, &p->bs);
2051 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2053 bioset_exit(&md->bs);
2055 /* mempool bind completed, no longer need any mempools in the table */
2056 dm_table_free_md_mempools(t);
2061 * Bind a table to the device.
2063 static void event_callback(void *context)
2065 unsigned long flags;
2067 struct mapped_device *md = (struct mapped_device *) context;
2069 spin_lock_irqsave(&md->uevent_lock, flags);
2070 list_splice_init(&md->uevent_list, &uevents);
2071 spin_unlock_irqrestore(&md->uevent_lock, flags);
2073 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2075 atomic_inc(&md->event_nr);
2076 wake_up(&md->eventq);
2077 dm_issue_global_event();
2081 * Protected by md->suspend_lock obtained by dm_swap_table().
2083 static void __set_size(struct mapped_device *md, sector_t size)
2085 lockdep_assert_held(&md->suspend_lock);
2087 set_capacity(md->disk, size);
2089 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2093 * Returns old map, which caller must destroy.
2095 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2096 struct queue_limits *limits)
2098 struct dm_table *old_map;
2099 struct request_queue *q = md->queue;
2100 bool request_based = dm_table_request_based(t);
2104 lockdep_assert_held(&md->suspend_lock);
2106 size = dm_table_get_size(t);
2109 * Wipe any geometry if the size of the table changed.
2111 if (size != dm_get_size(md))
2112 memset(&md->geometry, 0, sizeof(md->geometry));
2114 __set_size(md, size);
2116 dm_table_event_callback(t, event_callback, md);
2119 * The queue hasn't been stopped yet, if the old table type wasn't
2120 * for request-based during suspension. So stop it to prevent
2121 * I/O mapping before resume.
2122 * This must be done before setting the queue restrictions,
2123 * because request-based dm may be run just after the setting.
2128 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2130 * Leverage the fact that request-based DM targets and
2131 * NVMe bio based targets are immutable singletons
2132 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2133 * and __process_bio.
2135 md->immutable_target = dm_table_get_immutable_target(t);
2138 ret = __bind_mempools(md, t);
2140 old_map = ERR_PTR(ret);
2144 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2145 rcu_assign_pointer(md->map, (void *)t);
2146 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2148 dm_table_set_restrictions(t, q, limits);
2157 * Returns unbound table for the caller to free.
2159 static struct dm_table *__unbind(struct mapped_device *md)
2161 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2166 dm_table_event_callback(map, NULL, NULL);
2167 RCU_INIT_POINTER(md->map, NULL);
2174 * Constructor for a new device.
2176 int dm_create(int minor, struct mapped_device **result)
2179 struct mapped_device *md;
2181 md = alloc_dev(minor);
2185 r = dm_sysfs_init(md);
2196 * Functions to manage md->type.
2197 * All are required to hold md->type_lock.
2199 void dm_lock_md_type(struct mapped_device *md)
2201 mutex_lock(&md->type_lock);
2204 void dm_unlock_md_type(struct mapped_device *md)
2206 mutex_unlock(&md->type_lock);
2209 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2211 BUG_ON(!mutex_is_locked(&md->type_lock));
2215 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2220 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2222 return md->immutable_target_type;
2226 * The queue_limits are only valid as long as you have a reference
2229 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2231 BUG_ON(!atomic_read(&md->holders));
2232 return &md->queue->limits;
2234 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2237 * Setup the DM device's queue based on md's type
2239 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2242 struct queue_limits limits;
2243 enum dm_queue_mode type = dm_get_md_type(md);
2246 case DM_TYPE_REQUEST_BASED:
2247 r = dm_mq_init_request_queue(md, t);
2249 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2253 case DM_TYPE_BIO_BASED:
2254 case DM_TYPE_DAX_BIO_BASED:
2255 case DM_TYPE_NVME_BIO_BASED:
2256 dm_init_normal_md_queue(md);
2257 blk_queue_make_request(md->queue, dm_make_request);
2264 r = dm_calculate_queue_limits(t, &limits);
2266 DMERR("Cannot calculate initial queue limits");
2269 dm_table_set_restrictions(t, md->queue, &limits);
2270 blk_register_queue(md->disk);
2275 struct mapped_device *dm_get_md(dev_t dev)
2277 struct mapped_device *md;
2278 unsigned minor = MINOR(dev);
2280 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2283 spin_lock(&_minor_lock);
2285 md = idr_find(&_minor_idr, minor);
2286 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2287 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2293 spin_unlock(&_minor_lock);
2297 EXPORT_SYMBOL_GPL(dm_get_md);
2299 void *dm_get_mdptr(struct mapped_device *md)
2301 return md->interface_ptr;
2304 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2306 md->interface_ptr = ptr;
2309 void dm_get(struct mapped_device *md)
2311 atomic_inc(&md->holders);
2312 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2315 int dm_hold(struct mapped_device *md)
2317 spin_lock(&_minor_lock);
2318 if (test_bit(DMF_FREEING, &md->flags)) {
2319 spin_unlock(&_minor_lock);
2323 spin_unlock(&_minor_lock);
2326 EXPORT_SYMBOL_GPL(dm_hold);
2328 const char *dm_device_name(struct mapped_device *md)
2332 EXPORT_SYMBOL_GPL(dm_device_name);
2334 static void __dm_destroy(struct mapped_device *md, bool wait)
2336 struct dm_table *map;
2341 spin_lock(&_minor_lock);
2342 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2343 set_bit(DMF_FREEING, &md->flags);
2344 spin_unlock(&_minor_lock);
2346 blk_set_queue_dying(md->queue);
2349 * Take suspend_lock so that presuspend and postsuspend methods
2350 * do not race with internal suspend.
2352 mutex_lock(&md->suspend_lock);
2353 map = dm_get_live_table(md, &srcu_idx);
2354 if (!dm_suspended_md(md)) {
2355 dm_table_presuspend_targets(map);
2356 dm_table_postsuspend_targets(map);
2358 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2359 dm_put_live_table(md, srcu_idx);
2360 mutex_unlock(&md->suspend_lock);
2363 * Rare, but there may be I/O requests still going to complete,
2364 * for example. Wait for all references to disappear.
2365 * No one should increment the reference count of the mapped_device,
2366 * after the mapped_device state becomes DMF_FREEING.
2369 while (atomic_read(&md->holders))
2371 else if (atomic_read(&md->holders))
2372 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2373 dm_device_name(md), atomic_read(&md->holders));
2376 dm_table_destroy(__unbind(md));
2380 void dm_destroy(struct mapped_device *md)
2382 __dm_destroy(md, true);
2385 void dm_destroy_immediate(struct mapped_device *md)
2387 __dm_destroy(md, false);
2390 void dm_put(struct mapped_device *md)
2392 atomic_dec(&md->holders);
2394 EXPORT_SYMBOL_GPL(dm_put);
2396 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2402 prepare_to_wait(&md->wait, &wait, task_state);
2404 if (!md_in_flight(md))
2407 if (signal_pending_state(task_state, current)) {
2414 finish_wait(&md->wait, &wait);
2420 * Process the deferred bios
2422 static void dm_wq_work(struct work_struct *work)
2424 struct mapped_device *md = container_of(work, struct mapped_device,
2428 struct dm_table *map;
2430 map = dm_get_live_table(md, &srcu_idx);
2432 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2433 spin_lock_irq(&md->deferred_lock);
2434 c = bio_list_pop(&md->deferred);
2435 spin_unlock_irq(&md->deferred_lock);
2440 if (dm_request_based(md))
2441 (void) generic_make_request(c);
2443 (void) dm_process_bio(md, map, c);
2446 dm_put_live_table(md, srcu_idx);
2449 static void dm_queue_flush(struct mapped_device *md)
2451 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2452 smp_mb__after_atomic();
2453 queue_work(md->wq, &md->work);
2457 * Swap in a new table, returning the old one for the caller to destroy.
2459 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2461 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2462 struct queue_limits limits;
2465 mutex_lock(&md->suspend_lock);
2467 /* device must be suspended */
2468 if (!dm_suspended_md(md))
2472 * If the new table has no data devices, retain the existing limits.
2473 * This helps multipath with queue_if_no_path if all paths disappear,
2474 * then new I/O is queued based on these limits, and then some paths
2477 if (dm_table_has_no_data_devices(table)) {
2478 live_map = dm_get_live_table_fast(md);
2480 limits = md->queue->limits;
2481 dm_put_live_table_fast(md);
2485 r = dm_calculate_queue_limits(table, &limits);
2492 map = __bind(md, table, &limits);
2493 dm_issue_global_event();
2496 mutex_unlock(&md->suspend_lock);
2501 * Functions to lock and unlock any filesystem running on the
2504 static int lock_fs(struct mapped_device *md)
2508 WARN_ON(md->frozen_sb);
2510 md->frozen_sb = freeze_bdev(md->bdev);
2511 if (IS_ERR(md->frozen_sb)) {
2512 r = PTR_ERR(md->frozen_sb);
2513 md->frozen_sb = NULL;
2517 set_bit(DMF_FROZEN, &md->flags);
2522 static void unlock_fs(struct mapped_device *md)
2524 if (!test_bit(DMF_FROZEN, &md->flags))
2527 thaw_bdev(md->bdev, md->frozen_sb);
2528 md->frozen_sb = NULL;
2529 clear_bit(DMF_FROZEN, &md->flags);
2533 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2534 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2535 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2537 * If __dm_suspend returns 0, the device is completely quiescent
2538 * now. There is no request-processing activity. All new requests
2539 * are being added to md->deferred list.
2541 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2542 unsigned suspend_flags, long task_state,
2543 int dmf_suspended_flag)
2545 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2546 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2549 lockdep_assert_held(&md->suspend_lock);
2552 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2553 * This flag is cleared before dm_suspend returns.
2556 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2558 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2561 * This gets reverted if there's an error later and the targets
2562 * provide the .presuspend_undo hook.
2564 dm_table_presuspend_targets(map);
2567 * Flush I/O to the device.
2568 * Any I/O submitted after lock_fs() may not be flushed.
2569 * noflush takes precedence over do_lockfs.
2570 * (lock_fs() flushes I/Os and waits for them to complete.)
2572 if (!noflush && do_lockfs) {
2575 dm_table_presuspend_undo_targets(map);
2581 * Here we must make sure that no processes are submitting requests
2582 * to target drivers i.e. no one may be executing
2583 * __split_and_process_bio. This is called from dm_request and
2586 * To get all processes out of __split_and_process_bio in dm_request,
2587 * we take the write lock. To prevent any process from reentering
2588 * __split_and_process_bio from dm_request and quiesce the thread
2589 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2590 * flush_workqueue(md->wq).
2592 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2594 synchronize_srcu(&md->io_barrier);
2597 * Stop md->queue before flushing md->wq in case request-based
2598 * dm defers requests to md->wq from md->queue.
2600 if (dm_request_based(md))
2601 dm_stop_queue(md->queue);
2603 flush_workqueue(md->wq);
2606 * At this point no more requests are entering target request routines.
2607 * We call dm_wait_for_completion to wait for all existing requests
2610 r = dm_wait_for_completion(md, task_state);
2612 set_bit(dmf_suspended_flag, &md->flags);
2615 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2617 synchronize_srcu(&md->io_barrier);
2619 /* were we interrupted ? */
2623 if (dm_request_based(md))
2624 dm_start_queue(md->queue);
2627 dm_table_presuspend_undo_targets(map);
2628 /* pushback list is already flushed, so skip flush */
2635 * We need to be able to change a mapping table under a mounted
2636 * filesystem. For example we might want to move some data in
2637 * the background. Before the table can be swapped with
2638 * dm_bind_table, dm_suspend must be called to flush any in
2639 * flight bios and ensure that any further io gets deferred.
2642 * Suspend mechanism in request-based dm.
2644 * 1. Flush all I/Os by lock_fs() if needed.
2645 * 2. Stop dispatching any I/O by stopping the request_queue.
2646 * 3. Wait for all in-flight I/Os to be completed or requeued.
2648 * To abort suspend, start the request_queue.
2650 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2652 struct dm_table *map = NULL;
2656 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2658 if (dm_suspended_md(md)) {
2663 if (dm_suspended_internally_md(md)) {
2664 /* already internally suspended, wait for internal resume */
2665 mutex_unlock(&md->suspend_lock);
2666 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2672 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2674 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2678 dm_table_postsuspend_targets(map);
2681 mutex_unlock(&md->suspend_lock);
2685 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2688 int r = dm_table_resume_targets(map);
2696 * Flushing deferred I/Os must be done after targets are resumed
2697 * so that mapping of targets can work correctly.
2698 * Request-based dm is queueing the deferred I/Os in its request_queue.
2700 if (dm_request_based(md))
2701 dm_start_queue(md->queue);
2708 int dm_resume(struct mapped_device *md)
2711 struct dm_table *map = NULL;
2715 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2717 if (!dm_suspended_md(md))
2720 if (dm_suspended_internally_md(md)) {
2721 /* already internally suspended, wait for internal resume */
2722 mutex_unlock(&md->suspend_lock);
2723 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2729 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2730 if (!map || !dm_table_get_size(map))
2733 r = __dm_resume(md, map);
2737 clear_bit(DMF_SUSPENDED, &md->flags);
2739 mutex_unlock(&md->suspend_lock);
2745 * Internal suspend/resume works like userspace-driven suspend. It waits
2746 * until all bios finish and prevents issuing new bios to the target drivers.
2747 * It may be used only from the kernel.
2750 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2752 struct dm_table *map = NULL;
2754 lockdep_assert_held(&md->suspend_lock);
2756 if (md->internal_suspend_count++)
2757 return; /* nested internal suspend */
2759 if (dm_suspended_md(md)) {
2760 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2761 return; /* nest suspend */
2764 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2767 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2768 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2769 * would require changing .presuspend to return an error -- avoid this
2770 * until there is a need for more elaborate variants of internal suspend.
2772 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2773 DMF_SUSPENDED_INTERNALLY);
2775 dm_table_postsuspend_targets(map);
2778 static void __dm_internal_resume(struct mapped_device *md)
2780 BUG_ON(!md->internal_suspend_count);
2782 if (--md->internal_suspend_count)
2783 return; /* resume from nested internal suspend */
2785 if (dm_suspended_md(md))
2786 goto done; /* resume from nested suspend */
2789 * NOTE: existing callers don't need to call dm_table_resume_targets
2790 * (which may fail -- so best to avoid it for now by passing NULL map)
2792 (void) __dm_resume(md, NULL);
2795 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2796 smp_mb__after_atomic();
2797 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2800 void dm_internal_suspend_noflush(struct mapped_device *md)
2802 mutex_lock(&md->suspend_lock);
2803 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2804 mutex_unlock(&md->suspend_lock);
2806 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2808 void dm_internal_resume(struct mapped_device *md)
2810 mutex_lock(&md->suspend_lock);
2811 __dm_internal_resume(md);
2812 mutex_unlock(&md->suspend_lock);
2814 EXPORT_SYMBOL_GPL(dm_internal_resume);
2817 * Fast variants of internal suspend/resume hold md->suspend_lock,
2818 * which prevents interaction with userspace-driven suspend.
2821 void dm_internal_suspend_fast(struct mapped_device *md)
2823 mutex_lock(&md->suspend_lock);
2824 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2827 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2828 synchronize_srcu(&md->io_barrier);
2829 flush_workqueue(md->wq);
2830 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2832 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2834 void dm_internal_resume_fast(struct mapped_device *md)
2836 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2842 mutex_unlock(&md->suspend_lock);
2844 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2846 /*-----------------------------------------------------------------
2847 * Event notification.
2848 *---------------------------------------------------------------*/
2849 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2852 char udev_cookie[DM_COOKIE_LENGTH];
2853 char *envp[] = { udev_cookie, NULL };
2856 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2858 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2859 DM_COOKIE_ENV_VAR_NAME, cookie);
2860 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2865 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2867 return atomic_add_return(1, &md->uevent_seq);
2870 uint32_t dm_get_event_nr(struct mapped_device *md)
2872 return atomic_read(&md->event_nr);
2875 int dm_wait_event(struct mapped_device *md, int event_nr)
2877 return wait_event_interruptible(md->eventq,
2878 (event_nr != atomic_read(&md->event_nr)));
2881 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2883 unsigned long flags;
2885 spin_lock_irqsave(&md->uevent_lock, flags);
2886 list_add(elist, &md->uevent_list);
2887 spin_unlock_irqrestore(&md->uevent_lock, flags);
2891 * The gendisk is only valid as long as you have a reference
2894 struct gendisk *dm_disk(struct mapped_device *md)
2898 EXPORT_SYMBOL_GPL(dm_disk);
2900 struct kobject *dm_kobject(struct mapped_device *md)
2902 return &md->kobj_holder.kobj;
2905 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2907 struct mapped_device *md;
2909 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2911 spin_lock(&_minor_lock);
2912 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2918 spin_unlock(&_minor_lock);
2923 int dm_suspended_md(struct mapped_device *md)
2925 return test_bit(DMF_SUSPENDED, &md->flags);
2928 int dm_suspended_internally_md(struct mapped_device *md)
2930 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2933 int dm_test_deferred_remove_flag(struct mapped_device *md)
2935 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2938 int dm_suspended(struct dm_target *ti)
2940 return dm_suspended_md(dm_table_get_md(ti->table));
2942 EXPORT_SYMBOL_GPL(dm_suspended);
2944 int dm_noflush_suspending(struct dm_target *ti)
2946 return __noflush_suspending(dm_table_get_md(ti->table));
2948 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2950 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2951 unsigned integrity, unsigned per_io_data_size,
2952 unsigned min_pool_size)
2954 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2955 unsigned int pool_size = 0;
2956 unsigned int front_pad, io_front_pad;
2963 case DM_TYPE_BIO_BASED:
2964 case DM_TYPE_DAX_BIO_BASED:
2965 case DM_TYPE_NVME_BIO_BASED:
2966 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2967 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2968 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2969 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2972 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2975 case DM_TYPE_REQUEST_BASED:
2976 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2977 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2978 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2984 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2988 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2994 dm_free_md_mempools(pools);
2999 void dm_free_md_mempools(struct dm_md_mempools *pools)
3004 bioset_exit(&pools->bs);
3005 bioset_exit(&pools->io_bs);
3017 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3020 struct mapped_device *md = bdev->bd_disk->private_data;
3021 struct dm_table *table;
3022 struct dm_target *ti;
3023 int ret = -ENOTTY, srcu_idx;
3025 table = dm_get_live_table(md, &srcu_idx);
3026 if (!table || !dm_table_get_size(table))
3029 /* We only support devices that have a single target */
3030 if (dm_table_get_num_targets(table) != 1)
3032 ti = dm_table_get_target(table, 0);
3035 if (!ti->type->iterate_devices)
3038 ret = ti->type->iterate_devices(ti, fn, data);
3040 dm_put_live_table(md, srcu_idx);
3045 * For register / unregister we need to manually call out to every path.
3047 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3048 sector_t start, sector_t len, void *data)
3050 struct dm_pr *pr = data;
3051 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3053 if (!ops || !ops->pr_register)
3055 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3058 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3069 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3070 if (ret && new_key) {
3071 /* unregister all paths if we failed to register any path */
3072 pr.old_key = new_key;
3075 pr.fail_early = false;
3076 dm_call_pr(bdev, __dm_pr_register, &pr);
3082 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3085 struct mapped_device *md = bdev->bd_disk->private_data;
3086 const struct pr_ops *ops;
3089 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3093 ops = bdev->bd_disk->fops->pr_ops;
3094 if (ops && ops->pr_reserve)
3095 r = ops->pr_reserve(bdev, key, type, flags);
3099 dm_unprepare_ioctl(md, srcu_idx);
3103 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3105 struct mapped_device *md = bdev->bd_disk->private_data;
3106 const struct pr_ops *ops;
3109 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3113 ops = bdev->bd_disk->fops->pr_ops;
3114 if (ops && ops->pr_release)
3115 r = ops->pr_release(bdev, key, type);
3119 dm_unprepare_ioctl(md, srcu_idx);
3123 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3124 enum pr_type type, bool abort)
3126 struct mapped_device *md = bdev->bd_disk->private_data;
3127 const struct pr_ops *ops;
3130 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3134 ops = bdev->bd_disk->fops->pr_ops;
3135 if (ops && ops->pr_preempt)
3136 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3140 dm_unprepare_ioctl(md, srcu_idx);
3144 static int dm_pr_clear(struct block_device *bdev, u64 key)
3146 struct mapped_device *md = bdev->bd_disk->private_data;
3147 const struct pr_ops *ops;
3150 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3154 ops = bdev->bd_disk->fops->pr_ops;
3155 if (ops && ops->pr_clear)
3156 r = ops->pr_clear(bdev, key);
3160 dm_unprepare_ioctl(md, srcu_idx);
3164 static const struct pr_ops dm_pr_ops = {
3165 .pr_register = dm_pr_register,
3166 .pr_reserve = dm_pr_reserve,
3167 .pr_release = dm_pr_release,
3168 .pr_preempt = dm_pr_preempt,
3169 .pr_clear = dm_pr_clear,
3172 static const struct block_device_operations dm_blk_dops = {
3173 .open = dm_blk_open,
3174 .release = dm_blk_close,
3175 .ioctl = dm_blk_ioctl,
3176 .getgeo = dm_blk_getgeo,
3177 .report_zones = dm_blk_report_zones,
3178 .pr_ops = &dm_pr_ops,
3179 .owner = THIS_MODULE
3182 static const struct dax_operations dm_dax_ops = {
3183 .direct_access = dm_dax_direct_access,
3184 .copy_from_iter = dm_dax_copy_from_iter,
3185 .copy_to_iter = dm_dax_copy_to_iter,
3191 module_init(dm_init);
3192 module_exit(dm_exit);
3194 module_param(major, uint, 0);
3195 MODULE_PARM_DESC(major, "The major number of the device mapper");
3197 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3198 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3200 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3201 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3203 MODULE_DESCRIPTION(DM_NAME " driver");
3204 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3205 MODULE_LICENSE("GPL");