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 int md_in_flight(struct mapped_device *md)
651 return atomic_read(&md->pending[READ]) +
652 atomic_read(&md->pending[WRITE]);
655 static void start_io_acct(struct dm_io *io)
657 struct mapped_device *md = io->md;
658 struct bio *bio = io->orig_bio;
659 int rw = bio_data_dir(bio);
661 io->start_time = jiffies;
663 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
664 &dm_disk(md)->part0);
666 atomic_set(&dm_disk(md)->part0.in_flight[rw],
667 atomic_inc_return(&md->pending[rw]));
669 if (unlikely(dm_stats_used(&md->stats)))
670 dm_stats_account_io(&md->stats, bio_data_dir(bio),
671 bio->bi_iter.bi_sector, bio_sectors(bio),
672 false, 0, &io->stats_aux);
675 static void end_io_acct(struct dm_io *io)
677 struct mapped_device *md = io->md;
678 struct bio *bio = io->orig_bio;
679 unsigned long duration = jiffies - io->start_time;
681 int rw = bio_data_dir(bio);
683 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
686 if (unlikely(dm_stats_used(&md->stats)))
687 dm_stats_account_io(&md->stats, bio_data_dir(bio),
688 bio->bi_iter.bi_sector, bio_sectors(bio),
689 true, duration, &io->stats_aux);
692 * After this is decremented the bio must not be touched if it is
695 pending = atomic_dec_return(&md->pending[rw]);
696 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
697 pending += atomic_read(&md->pending[rw^0x1]);
699 /* nudge anyone waiting on suspend queue */
705 * Add the bio to the list of deferred io.
707 static void queue_io(struct mapped_device *md, struct bio *bio)
711 spin_lock_irqsave(&md->deferred_lock, flags);
712 bio_list_add(&md->deferred, bio);
713 spin_unlock_irqrestore(&md->deferred_lock, flags);
714 queue_work(md->wq, &md->work);
718 * Everyone (including functions in this file), should use this
719 * function to access the md->map field, and make sure they call
720 * dm_put_live_table() when finished.
722 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
724 *srcu_idx = srcu_read_lock(&md->io_barrier);
726 return srcu_dereference(md->map, &md->io_barrier);
729 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
731 srcu_read_unlock(&md->io_barrier, srcu_idx);
734 void dm_sync_table(struct mapped_device *md)
736 synchronize_srcu(&md->io_barrier);
737 synchronize_rcu_expedited();
741 * A fast alternative to dm_get_live_table/dm_put_live_table.
742 * The caller must not block between these two functions.
744 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
747 return rcu_dereference(md->map);
750 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
755 static char *_dm_claim_ptr = "I belong to device-mapper";
758 * Open a table device so we can use it as a map destination.
760 static int open_table_device(struct table_device *td, dev_t dev,
761 struct mapped_device *md)
763 struct block_device *bdev;
767 BUG_ON(td->dm_dev.bdev);
769 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
771 return PTR_ERR(bdev);
773 r = bd_link_disk_holder(bdev, dm_disk(md));
775 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
779 td->dm_dev.bdev = bdev;
780 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
785 * Close a table device that we've been using.
787 static void close_table_device(struct table_device *td, struct mapped_device *md)
789 if (!td->dm_dev.bdev)
792 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
793 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
794 put_dax(td->dm_dev.dax_dev);
795 td->dm_dev.bdev = NULL;
796 td->dm_dev.dax_dev = NULL;
799 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
801 struct table_device *td;
803 list_for_each_entry(td, l, list)
804 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
810 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
811 struct dm_dev **result) {
813 struct table_device *td;
815 mutex_lock(&md->table_devices_lock);
816 td = find_table_device(&md->table_devices, dev, mode);
818 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
820 mutex_unlock(&md->table_devices_lock);
824 td->dm_dev.mode = mode;
825 td->dm_dev.bdev = NULL;
827 if ((r = open_table_device(td, dev, md))) {
828 mutex_unlock(&md->table_devices_lock);
833 format_dev_t(td->dm_dev.name, dev);
835 refcount_set(&td->count, 1);
836 list_add(&td->list, &md->table_devices);
838 refcount_inc(&td->count);
840 mutex_unlock(&md->table_devices_lock);
842 *result = &td->dm_dev;
845 EXPORT_SYMBOL_GPL(dm_get_table_device);
847 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
849 struct table_device *td = container_of(d, struct table_device, dm_dev);
851 mutex_lock(&md->table_devices_lock);
852 if (refcount_dec_and_test(&td->count)) {
853 close_table_device(td, md);
857 mutex_unlock(&md->table_devices_lock);
859 EXPORT_SYMBOL(dm_put_table_device);
861 static void free_table_devices(struct list_head *devices)
863 struct list_head *tmp, *next;
865 list_for_each_safe(tmp, next, devices) {
866 struct table_device *td = list_entry(tmp, struct table_device, list);
868 DMWARN("dm_destroy: %s still exists with %d references",
869 td->dm_dev.name, refcount_read(&td->count));
875 * Get the geometry associated with a dm device
877 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
885 * Set the geometry of a device.
887 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
889 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
891 if (geo->start > sz) {
892 DMWARN("Start sector is beyond the geometry limits.");
901 static int __noflush_suspending(struct mapped_device *md)
903 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
907 * Decrements the number of outstanding ios that a bio has been
908 * cloned into, completing the original io if necc.
910 static void dec_pending(struct dm_io *io, blk_status_t error)
913 blk_status_t io_error;
915 struct mapped_device *md = io->md;
917 /* Push-back supersedes any I/O errors */
918 if (unlikely(error)) {
919 spin_lock_irqsave(&io->endio_lock, flags);
920 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
922 spin_unlock_irqrestore(&io->endio_lock, flags);
925 if (atomic_dec_and_test(&io->io_count)) {
926 if (io->status == BLK_STS_DM_REQUEUE) {
928 * Target requested pushing back the I/O.
930 spin_lock_irqsave(&md->deferred_lock, flags);
931 if (__noflush_suspending(md))
932 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
933 bio_list_add_head(&md->deferred, io->orig_bio);
935 /* noflush suspend was interrupted. */
936 io->status = BLK_STS_IOERR;
937 spin_unlock_irqrestore(&md->deferred_lock, flags);
940 io_error = io->status;
945 if (io_error == BLK_STS_DM_REQUEUE)
948 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
950 * Preflush done for flush with data, reissue
951 * without REQ_PREFLUSH.
953 bio->bi_opf &= ~REQ_PREFLUSH;
956 /* done with normal IO or empty flush */
958 bio->bi_status = io_error;
964 void disable_write_same(struct mapped_device *md)
966 struct queue_limits *limits = dm_get_queue_limits(md);
968 /* device doesn't really support WRITE SAME, disable it */
969 limits->max_write_same_sectors = 0;
972 void disable_write_zeroes(struct mapped_device *md)
974 struct queue_limits *limits = dm_get_queue_limits(md);
976 /* device doesn't really support WRITE ZEROES, disable it */
977 limits->max_write_zeroes_sectors = 0;
980 static void clone_endio(struct bio *bio)
982 blk_status_t error = bio->bi_status;
983 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
984 struct dm_io *io = tio->io;
985 struct mapped_device *md = tio->io->md;
986 dm_endio_fn endio = tio->ti->type->end_io;
988 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
989 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
990 !bio->bi_disk->queue->limits.max_write_same_sectors)
991 disable_write_same(md);
992 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
993 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
994 disable_write_zeroes(md);
998 int r = endio(tio->ti, bio, &error);
1000 case DM_ENDIO_REQUEUE:
1001 error = BLK_STS_DM_REQUEUE;
1005 case DM_ENDIO_INCOMPLETE:
1006 /* The target will handle the io */
1009 DMWARN("unimplemented target endio return value: %d", r);
1015 dec_pending(io, error);
1019 * Return maximum size of I/O possible at the supplied sector up to the current
1022 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1024 sector_t target_offset = dm_target_offset(ti, sector);
1026 return ti->len - target_offset;
1029 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1031 sector_t len = max_io_len_target_boundary(sector, ti);
1032 sector_t offset, max_len;
1035 * Does the target need to split even further?
1037 if (ti->max_io_len) {
1038 offset = dm_target_offset(ti, sector);
1039 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1040 max_len = sector_div(offset, ti->max_io_len);
1042 max_len = offset & (ti->max_io_len - 1);
1043 max_len = ti->max_io_len - max_len;
1052 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1054 if (len > UINT_MAX) {
1055 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1056 (unsigned long long)len, UINT_MAX);
1057 ti->error = "Maximum size of target IO is too large";
1062 * BIO based queue uses its own splitting. When multipage bvecs
1063 * is switched on, size of the incoming bio may be too big to
1064 * be handled in some targets, such as crypt.
1066 * When these targets are ready for the big bio, we can remove
1069 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1073 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1075 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1076 sector_t sector, int *srcu_idx)
1077 __acquires(md->io_barrier)
1079 struct dm_table *map;
1080 struct dm_target *ti;
1082 map = dm_get_live_table(md, srcu_idx);
1086 ti = dm_table_find_target(map, sector);
1087 if (!dm_target_is_valid(ti))
1093 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1094 long nr_pages, void **kaddr, pfn_t *pfn)
1096 struct mapped_device *md = dax_get_private(dax_dev);
1097 sector_t sector = pgoff * PAGE_SECTORS;
1098 struct dm_target *ti;
1099 long len, ret = -EIO;
1102 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1106 if (!ti->type->direct_access)
1108 len = max_io_len(sector, ti) / PAGE_SECTORS;
1111 nr_pages = min(len, nr_pages);
1112 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1115 dm_put_live_table(md, srcu_idx);
1120 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1121 void *addr, size_t bytes, struct iov_iter *i)
1123 struct mapped_device *md = dax_get_private(dax_dev);
1124 sector_t sector = pgoff * PAGE_SECTORS;
1125 struct dm_target *ti;
1129 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1133 if (!ti->type->dax_copy_from_iter) {
1134 ret = copy_from_iter(addr, bytes, i);
1137 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1139 dm_put_live_table(md, srcu_idx);
1144 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1145 void *addr, size_t bytes, struct iov_iter *i)
1147 struct mapped_device *md = dax_get_private(dax_dev);
1148 sector_t sector = pgoff * PAGE_SECTORS;
1149 struct dm_target *ti;
1153 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1157 if (!ti->type->dax_copy_to_iter) {
1158 ret = copy_to_iter(addr, bytes, i);
1161 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1163 dm_put_live_table(md, srcu_idx);
1169 * A target may call dm_accept_partial_bio only from the map routine. It is
1170 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1172 * dm_accept_partial_bio informs the dm that the target only wants to process
1173 * additional n_sectors sectors of the bio and the rest of the data should be
1174 * sent in a next bio.
1176 * A diagram that explains the arithmetics:
1177 * +--------------------+---------------+-------+
1179 * +--------------------+---------------+-------+
1181 * <-------------- *tio->len_ptr --------------->
1182 * <------- bi_size ------->
1185 * Region 1 was already iterated over with bio_advance or similar function.
1186 * (it may be empty if the target doesn't use bio_advance)
1187 * Region 2 is the remaining bio size that the target wants to process.
1188 * (it may be empty if region 1 is non-empty, although there is no reason
1190 * The target requires that region 3 is to be sent in the next bio.
1192 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1193 * the partially processed part (the sum of regions 1+2) must be the same for all
1194 * copies of the bio.
1196 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1198 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1199 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1200 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1201 BUG_ON(bi_size > *tio->len_ptr);
1202 BUG_ON(n_sectors > bi_size);
1203 *tio->len_ptr -= bi_size - n_sectors;
1204 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1206 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1209 * The zone descriptors obtained with a zone report indicate
1210 * zone positions within the underlying device of the target. The zone
1211 * descriptors must be remapped to match their position within the dm device.
1212 * The caller target should obtain the zones information using
1213 * blkdev_report_zones() to ensure that remapping for partition offset is
1216 void dm_remap_zone_report(struct dm_target *ti, sector_t start,
1217 struct blk_zone *zones, unsigned int *nr_zones)
1219 #ifdef CONFIG_BLK_DEV_ZONED
1220 struct blk_zone *zone;
1221 unsigned int nrz = *nr_zones;
1225 * Remap the start sector and write pointer position of the zones in
1226 * the array. Since we may have obtained from the target underlying
1227 * device more zones that the target size, also adjust the number
1230 for (i = 0; i < nrz; i++) {
1232 if (zone->start >= start + ti->len) {
1233 memset(zone, 0, sizeof(struct blk_zone) * (nrz - i));
1237 zone->start = zone->start + ti->begin - start;
1238 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
1241 if (zone->cond == BLK_ZONE_COND_FULL)
1242 zone->wp = zone->start + zone->len;
1243 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1244 zone->wp = zone->start;
1246 zone->wp = zone->wp + ti->begin - start;
1250 #else /* !CONFIG_BLK_DEV_ZONED */
1254 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1256 static blk_qc_t __map_bio(struct dm_target_io *tio)
1260 struct bio *clone = &tio->clone;
1261 struct dm_io *io = tio->io;
1262 struct mapped_device *md = io->md;
1263 struct dm_target *ti = tio->ti;
1264 blk_qc_t ret = BLK_QC_T_NONE;
1266 clone->bi_end_io = clone_endio;
1269 * Map the clone. If r == 0 we don't need to do
1270 * anything, the target has assumed ownership of
1273 atomic_inc(&io->io_count);
1274 sector = clone->bi_iter.bi_sector;
1276 r = ti->type->map(ti, clone);
1278 case DM_MAPIO_SUBMITTED:
1280 case DM_MAPIO_REMAPPED:
1281 /* the bio has been remapped so dispatch it */
1282 trace_block_bio_remap(clone->bi_disk->queue, clone,
1283 bio_dev(io->orig_bio), sector);
1284 if (md->type == DM_TYPE_NVME_BIO_BASED)
1285 ret = direct_make_request(clone);
1287 ret = generic_make_request(clone);
1291 dec_pending(io, BLK_STS_IOERR);
1293 case DM_MAPIO_REQUEUE:
1295 dec_pending(io, BLK_STS_DM_REQUEUE);
1298 DMWARN("unimplemented target map return value: %d", r);
1305 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1307 bio->bi_iter.bi_sector = sector;
1308 bio->bi_iter.bi_size = to_bytes(len);
1312 * Creates a bio that consists of range of complete bvecs.
1314 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1315 sector_t sector, unsigned len)
1317 struct bio *clone = &tio->clone;
1319 __bio_clone_fast(clone, bio);
1321 if (unlikely(bio_integrity(bio) != NULL)) {
1324 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1325 !dm_target_passes_integrity(tio->ti->type))) {
1326 DMWARN("%s: the target %s doesn't support integrity data.",
1327 dm_device_name(tio->io->md),
1328 tio->ti->type->name);
1332 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1337 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1338 clone->bi_iter.bi_size = to_bytes(len);
1340 if (unlikely(bio_integrity(bio) != NULL))
1341 bio_integrity_trim(clone);
1346 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1347 struct dm_target *ti, unsigned num_bios)
1349 struct dm_target_io *tio;
1355 if (num_bios == 1) {
1356 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1357 bio_list_add(blist, &tio->clone);
1361 for (try = 0; try < 2; try++) {
1366 mutex_lock(&ci->io->md->table_devices_lock);
1367 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1368 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1372 bio_list_add(blist, &tio->clone);
1375 mutex_unlock(&ci->io->md->table_devices_lock);
1376 if (bio_nr == num_bios)
1379 while ((bio = bio_list_pop(blist))) {
1380 tio = container_of(bio, struct dm_target_io, clone);
1386 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1387 struct dm_target_io *tio, unsigned *len)
1389 struct bio *clone = &tio->clone;
1393 __bio_clone_fast(clone, ci->bio);
1395 bio_setup_sector(clone, ci->sector, *len);
1397 return __map_bio(tio);
1400 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1401 unsigned num_bios, unsigned *len)
1403 struct bio_list blist = BIO_EMPTY_LIST;
1405 struct dm_target_io *tio;
1407 alloc_multiple_bios(&blist, ci, ti, num_bios);
1409 while ((bio = bio_list_pop(&blist))) {
1410 tio = container_of(bio, struct dm_target_io, clone);
1411 (void) __clone_and_map_simple_bio(ci, tio, len);
1415 static int __send_empty_flush(struct clone_info *ci)
1417 unsigned target_nr = 0;
1418 struct dm_target *ti;
1420 BUG_ON(bio_has_data(ci->bio));
1421 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1422 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1427 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1428 sector_t sector, unsigned *len)
1430 struct bio *bio = ci->bio;
1431 struct dm_target_io *tio;
1434 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1436 r = clone_bio(tio, bio, sector, *len);
1441 (void) __map_bio(tio);
1446 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1448 static unsigned get_num_discard_bios(struct dm_target *ti)
1450 return ti->num_discard_bios;
1453 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1455 return ti->num_secure_erase_bios;
1458 static unsigned get_num_write_same_bios(struct dm_target *ti)
1460 return ti->num_write_same_bios;
1463 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1465 return ti->num_write_zeroes_bios;
1468 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1470 static bool is_split_required_for_discard(struct dm_target *ti)
1472 return ti->split_discard_bios;
1475 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1476 get_num_bios_fn get_num_bios,
1477 is_split_required_fn is_split_required)
1483 * Even though the device advertised support for this type of
1484 * request, that does not mean every target supports it, and
1485 * reconfiguration might also have changed that since the
1486 * check was performed.
1488 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1492 if (is_split_required && !is_split_required(ti))
1493 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1495 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1497 __send_duplicate_bios(ci, ti, num_bios, &len);
1500 ci->sector_count -= len;
1505 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1507 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1508 is_split_required_for_discard);
1511 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1513 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1516 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1518 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1521 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1523 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1526 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1529 struct bio *bio = ci->bio;
1531 if (bio_op(bio) == REQ_OP_DISCARD)
1532 *result = __send_discard(ci, ti);
1533 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1534 *result = __send_secure_erase(ci, ti);
1535 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1536 *result = __send_write_same(ci, ti);
1537 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1538 *result = __send_write_zeroes(ci, ti);
1546 * Select the correct strategy for processing a non-flush bio.
1548 static int __split_and_process_non_flush(struct clone_info *ci)
1550 struct dm_target *ti;
1554 ti = dm_table_find_target(ci->map, ci->sector);
1555 if (!dm_target_is_valid(ti))
1558 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1561 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1563 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1568 ci->sector_count -= len;
1573 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1574 struct dm_table *map, struct bio *bio)
1577 ci->io = alloc_io(md, bio);
1578 ci->sector = bio->bi_iter.bi_sector;
1582 * Entry point to split a bio into clones and submit them to the targets.
1584 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1585 struct dm_table *map, struct bio *bio)
1587 struct clone_info ci;
1588 blk_qc_t ret = BLK_QC_T_NONE;
1591 if (unlikely(!map)) {
1596 init_clone_info(&ci, md, map, bio);
1598 if (bio->bi_opf & REQ_PREFLUSH) {
1599 ci.bio = &ci.io->md->flush_bio;
1600 ci.sector_count = 0;
1601 error = __send_empty_flush(&ci);
1602 /* dec_pending submits any data associated with flush */
1603 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1605 ci.sector_count = 0;
1606 error = __split_and_process_non_flush(&ci);
1609 ci.sector_count = bio_sectors(bio);
1610 while (ci.sector_count && !error) {
1611 error = __split_and_process_non_flush(&ci);
1612 if (current->bio_list && ci.sector_count && !error) {
1614 * Remainder must be passed to generic_make_request()
1615 * so that it gets handled *after* bios already submitted
1616 * have been completely processed.
1617 * We take a clone of the original to store in
1618 * ci.io->orig_bio to be used by end_io_acct() and
1619 * for dec_pending to use for completion handling.
1621 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1622 GFP_NOIO, &md->queue->bio_split);
1623 ci.io->orig_bio = b;
1625 ret = generic_make_request(bio);
1631 /* drop the extra reference count */
1632 dec_pending(ci.io, errno_to_blk_status(error));
1637 * Optimized variant of __split_and_process_bio that leverages the
1638 * fact that targets that use it do _not_ have a need to split bios.
1640 static blk_qc_t __process_bio(struct mapped_device *md,
1641 struct dm_table *map, struct bio *bio)
1643 struct clone_info ci;
1644 blk_qc_t ret = BLK_QC_T_NONE;
1647 if (unlikely(!map)) {
1652 init_clone_info(&ci, md, map, bio);
1654 if (bio->bi_opf & REQ_PREFLUSH) {
1655 ci.bio = &ci.io->md->flush_bio;
1656 ci.sector_count = 0;
1657 error = __send_empty_flush(&ci);
1658 /* dec_pending submits any data associated with flush */
1660 struct dm_target *ti = md->immutable_target;
1661 struct dm_target_io *tio;
1664 * Defend against IO still getting in during teardown
1665 * - as was seen for a time with nvme-fcloop
1667 if (WARN_ON_ONCE(!ti || !dm_target_is_valid(ti))) {
1673 ci.sector_count = bio_sectors(bio);
1674 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1677 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1678 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1681 /* drop the extra reference count */
1682 dec_pending(ci.io, errno_to_blk_status(error));
1686 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1688 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1689 process_bio_fn process_bio)
1691 struct mapped_device *md = q->queuedata;
1692 blk_qc_t ret = BLK_QC_T_NONE;
1694 struct dm_table *map;
1696 map = dm_get_live_table(md, &srcu_idx);
1698 /* if we're suspended, we have to queue this io for later */
1699 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1700 dm_put_live_table(md, srcu_idx);
1702 if (!(bio->bi_opf & REQ_RAHEAD))
1709 ret = process_bio(md, map, bio);
1711 dm_put_live_table(md, srcu_idx);
1716 * The request function that remaps the bio to one target and
1717 * splits off any remainder.
1719 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1721 return __dm_make_request(q, bio, __split_and_process_bio);
1724 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1726 return __dm_make_request(q, bio, __process_bio);
1729 static int dm_any_congested(void *congested_data, int bdi_bits)
1732 struct mapped_device *md = congested_data;
1733 struct dm_table *map;
1735 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1736 if (dm_request_based(md)) {
1738 * With request-based DM we only need to check the
1739 * top-level queue for congestion.
1741 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1743 map = dm_get_live_table_fast(md);
1745 r = dm_table_any_congested(map, bdi_bits);
1746 dm_put_live_table_fast(md);
1753 /*-----------------------------------------------------------------
1754 * An IDR is used to keep track of allocated minor numbers.
1755 *---------------------------------------------------------------*/
1756 static void free_minor(int minor)
1758 spin_lock(&_minor_lock);
1759 idr_remove(&_minor_idr, minor);
1760 spin_unlock(&_minor_lock);
1764 * See if the device with a specific minor # is free.
1766 static int specific_minor(int minor)
1770 if (minor >= (1 << MINORBITS))
1773 idr_preload(GFP_KERNEL);
1774 spin_lock(&_minor_lock);
1776 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1778 spin_unlock(&_minor_lock);
1781 return r == -ENOSPC ? -EBUSY : r;
1785 static int next_free_minor(int *minor)
1789 idr_preload(GFP_KERNEL);
1790 spin_lock(&_minor_lock);
1792 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1794 spin_unlock(&_minor_lock);
1802 static const struct block_device_operations dm_blk_dops;
1803 static const struct dax_operations dm_dax_ops;
1805 static void dm_wq_work(struct work_struct *work);
1807 static void dm_init_normal_md_queue(struct mapped_device *md)
1810 * Initialize aspects of queue that aren't relevant for blk-mq
1812 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1815 static void cleanup_mapped_device(struct mapped_device *md)
1818 destroy_workqueue(md->wq);
1819 bioset_exit(&md->bs);
1820 bioset_exit(&md->io_bs);
1823 kill_dax(md->dax_dev);
1824 put_dax(md->dax_dev);
1829 spin_lock(&_minor_lock);
1830 md->disk->private_data = NULL;
1831 spin_unlock(&_minor_lock);
1832 del_gendisk(md->disk);
1837 blk_cleanup_queue(md->queue);
1839 cleanup_srcu_struct(&md->io_barrier);
1846 mutex_destroy(&md->suspend_lock);
1847 mutex_destroy(&md->type_lock);
1848 mutex_destroy(&md->table_devices_lock);
1850 dm_mq_cleanup_mapped_device(md);
1854 * Allocate and initialise a blank device with a given minor.
1856 static struct mapped_device *alloc_dev(int minor)
1858 int r, numa_node_id = dm_get_numa_node();
1859 struct dax_device *dax_dev = NULL;
1860 struct mapped_device *md;
1863 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1865 DMWARN("unable to allocate device, out of memory.");
1869 if (!try_module_get(THIS_MODULE))
1870 goto bad_module_get;
1872 /* get a minor number for the dev */
1873 if (minor == DM_ANY_MINOR)
1874 r = next_free_minor(&minor);
1876 r = specific_minor(minor);
1880 r = init_srcu_struct(&md->io_barrier);
1882 goto bad_io_barrier;
1884 md->numa_node_id = numa_node_id;
1885 md->init_tio_pdu = false;
1886 md->type = DM_TYPE_NONE;
1887 mutex_init(&md->suspend_lock);
1888 mutex_init(&md->type_lock);
1889 mutex_init(&md->table_devices_lock);
1890 spin_lock_init(&md->deferred_lock);
1891 atomic_set(&md->holders, 1);
1892 atomic_set(&md->open_count, 0);
1893 atomic_set(&md->event_nr, 0);
1894 atomic_set(&md->uevent_seq, 0);
1895 INIT_LIST_HEAD(&md->uevent_list);
1896 INIT_LIST_HEAD(&md->table_devices);
1897 spin_lock_init(&md->uevent_lock);
1899 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1902 md->queue->queuedata = md;
1903 md->queue->backing_dev_info->congested_data = md;
1905 md->disk = alloc_disk_node(1, md->numa_node_id);
1909 atomic_set(&md->pending[0], 0);
1910 atomic_set(&md->pending[1], 0);
1911 init_waitqueue_head(&md->wait);
1912 INIT_WORK(&md->work, dm_wq_work);
1913 init_waitqueue_head(&md->eventq);
1914 init_completion(&md->kobj_holder.completion);
1916 md->disk->major = _major;
1917 md->disk->first_minor = minor;
1918 md->disk->fops = &dm_blk_dops;
1919 md->disk->queue = md->queue;
1920 md->disk->private_data = md;
1921 sprintf(md->disk->disk_name, "dm-%d", minor);
1923 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1924 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1928 md->dax_dev = dax_dev;
1930 add_disk_no_queue_reg(md->disk);
1931 format_dev_t(md->name, MKDEV(_major, minor));
1933 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1937 md->bdev = bdget_disk(md->disk, 0);
1941 bio_init(&md->flush_bio, NULL, 0);
1942 bio_set_dev(&md->flush_bio, md->bdev);
1943 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1945 dm_stats_init(&md->stats);
1947 /* Populate the mapping, nobody knows we exist yet */
1948 spin_lock(&_minor_lock);
1949 old_md = idr_replace(&_minor_idr, md, minor);
1950 spin_unlock(&_minor_lock);
1952 BUG_ON(old_md != MINOR_ALLOCED);
1957 cleanup_mapped_device(md);
1961 module_put(THIS_MODULE);
1967 static void unlock_fs(struct mapped_device *md);
1969 static void free_dev(struct mapped_device *md)
1971 int minor = MINOR(disk_devt(md->disk));
1975 cleanup_mapped_device(md);
1977 free_table_devices(&md->table_devices);
1978 dm_stats_cleanup(&md->stats);
1981 module_put(THIS_MODULE);
1985 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1987 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1990 if (dm_table_bio_based(t)) {
1992 * The md may already have mempools that need changing.
1993 * If so, reload bioset because front_pad may have changed
1994 * because a different table was loaded.
1996 bioset_exit(&md->bs);
1997 bioset_exit(&md->io_bs);
1999 } else if (bioset_initialized(&md->bs)) {
2001 * There's no need to reload with request-based dm
2002 * because the size of front_pad doesn't change.
2003 * Note for future: If you are to reload bioset,
2004 * prep-ed requests in the queue may refer
2005 * to bio from the old bioset, so you must walk
2006 * through the queue to unprep.
2012 bioset_initialized(&md->bs) ||
2013 bioset_initialized(&md->io_bs));
2015 ret = bioset_init_from_src(&md->bs, &p->bs);
2018 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2020 bioset_exit(&md->bs);
2022 /* mempool bind completed, no longer need any mempools in the table */
2023 dm_table_free_md_mempools(t);
2028 * Bind a table to the device.
2030 static void event_callback(void *context)
2032 unsigned long flags;
2034 struct mapped_device *md = (struct mapped_device *) context;
2036 spin_lock_irqsave(&md->uevent_lock, flags);
2037 list_splice_init(&md->uevent_list, &uevents);
2038 spin_unlock_irqrestore(&md->uevent_lock, flags);
2040 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2042 atomic_inc(&md->event_nr);
2043 wake_up(&md->eventq);
2044 dm_issue_global_event();
2048 * Protected by md->suspend_lock obtained by dm_swap_table().
2050 static void __set_size(struct mapped_device *md, sector_t size)
2052 lockdep_assert_held(&md->suspend_lock);
2054 set_capacity(md->disk, size);
2056 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2060 * Returns old map, which caller must destroy.
2062 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2063 struct queue_limits *limits)
2065 struct dm_table *old_map;
2066 struct request_queue *q = md->queue;
2067 bool request_based = dm_table_request_based(t);
2071 lockdep_assert_held(&md->suspend_lock);
2073 size = dm_table_get_size(t);
2076 * Wipe any geometry if the size of the table changed.
2078 if (size != dm_get_size(md))
2079 memset(&md->geometry, 0, sizeof(md->geometry));
2081 __set_size(md, size);
2083 dm_table_event_callback(t, event_callback, md);
2086 * The queue hasn't been stopped yet, if the old table type wasn't
2087 * for request-based during suspension. So stop it to prevent
2088 * I/O mapping before resume.
2089 * This must be done before setting the queue restrictions,
2090 * because request-based dm may be run just after the setting.
2095 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2097 * Leverage the fact that request-based DM targets and
2098 * NVMe bio based targets are immutable singletons
2099 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2100 * and __process_bio.
2102 md->immutable_target = dm_table_get_immutable_target(t);
2105 ret = __bind_mempools(md, t);
2107 old_map = ERR_PTR(ret);
2111 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2112 rcu_assign_pointer(md->map, (void *)t);
2113 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2115 dm_table_set_restrictions(t, q, limits);
2124 * Returns unbound table for the caller to free.
2126 static struct dm_table *__unbind(struct mapped_device *md)
2128 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2133 dm_table_event_callback(map, NULL, NULL);
2134 RCU_INIT_POINTER(md->map, NULL);
2141 * Constructor for a new device.
2143 int dm_create(int minor, struct mapped_device **result)
2146 struct mapped_device *md;
2148 md = alloc_dev(minor);
2152 r = dm_sysfs_init(md);
2163 * Functions to manage md->type.
2164 * All are required to hold md->type_lock.
2166 void dm_lock_md_type(struct mapped_device *md)
2168 mutex_lock(&md->type_lock);
2171 void dm_unlock_md_type(struct mapped_device *md)
2173 mutex_unlock(&md->type_lock);
2176 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2178 BUG_ON(!mutex_is_locked(&md->type_lock));
2182 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2187 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2189 return md->immutable_target_type;
2193 * The queue_limits are only valid as long as you have a reference
2196 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2198 BUG_ON(!atomic_read(&md->holders));
2199 return &md->queue->limits;
2201 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2204 * Setup the DM device's queue based on md's type
2206 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2209 struct queue_limits limits;
2210 enum dm_queue_mode type = dm_get_md_type(md);
2213 case DM_TYPE_REQUEST_BASED:
2214 r = dm_mq_init_request_queue(md, t);
2216 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2220 case DM_TYPE_BIO_BASED:
2221 case DM_TYPE_DAX_BIO_BASED:
2222 dm_init_normal_md_queue(md);
2223 blk_queue_make_request(md->queue, dm_make_request);
2225 case DM_TYPE_NVME_BIO_BASED:
2226 dm_init_normal_md_queue(md);
2227 blk_queue_make_request(md->queue, dm_make_request_nvme);
2234 r = dm_calculate_queue_limits(t, &limits);
2236 DMERR("Cannot calculate initial queue limits");
2239 dm_table_set_restrictions(t, md->queue, &limits);
2240 blk_register_queue(md->disk);
2245 struct mapped_device *dm_get_md(dev_t dev)
2247 struct mapped_device *md;
2248 unsigned minor = MINOR(dev);
2250 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2253 spin_lock(&_minor_lock);
2255 md = idr_find(&_minor_idr, minor);
2256 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2257 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2263 spin_unlock(&_minor_lock);
2267 EXPORT_SYMBOL_GPL(dm_get_md);
2269 void *dm_get_mdptr(struct mapped_device *md)
2271 return md->interface_ptr;
2274 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2276 md->interface_ptr = ptr;
2279 void dm_get(struct mapped_device *md)
2281 atomic_inc(&md->holders);
2282 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2285 int dm_hold(struct mapped_device *md)
2287 spin_lock(&_minor_lock);
2288 if (test_bit(DMF_FREEING, &md->flags)) {
2289 spin_unlock(&_minor_lock);
2293 spin_unlock(&_minor_lock);
2296 EXPORT_SYMBOL_GPL(dm_hold);
2298 const char *dm_device_name(struct mapped_device *md)
2302 EXPORT_SYMBOL_GPL(dm_device_name);
2304 static void __dm_destroy(struct mapped_device *md, bool wait)
2306 struct dm_table *map;
2311 spin_lock(&_minor_lock);
2312 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2313 set_bit(DMF_FREEING, &md->flags);
2314 spin_unlock(&_minor_lock);
2316 blk_set_queue_dying(md->queue);
2319 * Take suspend_lock so that presuspend and postsuspend methods
2320 * do not race with internal suspend.
2322 mutex_lock(&md->suspend_lock);
2323 map = dm_get_live_table(md, &srcu_idx);
2324 if (!dm_suspended_md(md)) {
2325 dm_table_presuspend_targets(map);
2326 dm_table_postsuspend_targets(map);
2328 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2329 dm_put_live_table(md, srcu_idx);
2330 mutex_unlock(&md->suspend_lock);
2333 * Rare, but there may be I/O requests still going to complete,
2334 * for example. Wait for all references to disappear.
2335 * No one should increment the reference count of the mapped_device,
2336 * after the mapped_device state becomes DMF_FREEING.
2339 while (atomic_read(&md->holders))
2341 else if (atomic_read(&md->holders))
2342 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2343 dm_device_name(md), atomic_read(&md->holders));
2346 dm_table_destroy(__unbind(md));
2350 void dm_destroy(struct mapped_device *md)
2352 __dm_destroy(md, true);
2355 void dm_destroy_immediate(struct mapped_device *md)
2357 __dm_destroy(md, false);
2360 void dm_put(struct mapped_device *md)
2362 atomic_dec(&md->holders);
2364 EXPORT_SYMBOL_GPL(dm_put);
2366 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2372 prepare_to_wait(&md->wait, &wait, task_state);
2374 if (!md_in_flight(md))
2377 if (signal_pending_state(task_state, current)) {
2384 finish_wait(&md->wait, &wait);
2390 * Process the deferred bios
2392 static void dm_wq_work(struct work_struct *work)
2394 struct mapped_device *md = container_of(work, struct mapped_device,
2398 struct dm_table *map;
2400 map = dm_get_live_table(md, &srcu_idx);
2402 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2403 spin_lock_irq(&md->deferred_lock);
2404 c = bio_list_pop(&md->deferred);
2405 spin_unlock_irq(&md->deferred_lock);
2410 if (dm_request_based(md))
2411 generic_make_request(c);
2413 __split_and_process_bio(md, map, c);
2416 dm_put_live_table(md, srcu_idx);
2419 static void dm_queue_flush(struct mapped_device *md)
2421 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2422 smp_mb__after_atomic();
2423 queue_work(md->wq, &md->work);
2427 * Swap in a new table, returning the old one for the caller to destroy.
2429 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2431 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2432 struct queue_limits limits;
2435 mutex_lock(&md->suspend_lock);
2437 /* device must be suspended */
2438 if (!dm_suspended_md(md))
2442 * If the new table has no data devices, retain the existing limits.
2443 * This helps multipath with queue_if_no_path if all paths disappear,
2444 * then new I/O is queued based on these limits, and then some paths
2447 if (dm_table_has_no_data_devices(table)) {
2448 live_map = dm_get_live_table_fast(md);
2450 limits = md->queue->limits;
2451 dm_put_live_table_fast(md);
2455 r = dm_calculate_queue_limits(table, &limits);
2462 map = __bind(md, table, &limits);
2463 dm_issue_global_event();
2466 mutex_unlock(&md->suspend_lock);
2471 * Functions to lock and unlock any filesystem running on the
2474 static int lock_fs(struct mapped_device *md)
2478 WARN_ON(md->frozen_sb);
2480 md->frozen_sb = freeze_bdev(md->bdev);
2481 if (IS_ERR(md->frozen_sb)) {
2482 r = PTR_ERR(md->frozen_sb);
2483 md->frozen_sb = NULL;
2487 set_bit(DMF_FROZEN, &md->flags);
2492 static void unlock_fs(struct mapped_device *md)
2494 if (!test_bit(DMF_FROZEN, &md->flags))
2497 thaw_bdev(md->bdev, md->frozen_sb);
2498 md->frozen_sb = NULL;
2499 clear_bit(DMF_FROZEN, &md->flags);
2503 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2504 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2505 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2507 * If __dm_suspend returns 0, the device is completely quiescent
2508 * now. There is no request-processing activity. All new requests
2509 * are being added to md->deferred list.
2511 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2512 unsigned suspend_flags, long task_state,
2513 int dmf_suspended_flag)
2515 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2516 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2519 lockdep_assert_held(&md->suspend_lock);
2522 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2523 * This flag is cleared before dm_suspend returns.
2526 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2528 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2531 * This gets reverted if there's an error later and the targets
2532 * provide the .presuspend_undo hook.
2534 dm_table_presuspend_targets(map);
2537 * Flush I/O to the device.
2538 * Any I/O submitted after lock_fs() may not be flushed.
2539 * noflush takes precedence over do_lockfs.
2540 * (lock_fs() flushes I/Os and waits for them to complete.)
2542 if (!noflush && do_lockfs) {
2545 dm_table_presuspend_undo_targets(map);
2551 * Here we must make sure that no processes are submitting requests
2552 * to target drivers i.e. no one may be executing
2553 * __split_and_process_bio. This is called from dm_request and
2556 * To get all processes out of __split_and_process_bio in dm_request,
2557 * we take the write lock. To prevent any process from reentering
2558 * __split_and_process_bio from dm_request and quiesce the thread
2559 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2560 * flush_workqueue(md->wq).
2562 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2564 synchronize_srcu(&md->io_barrier);
2567 * Stop md->queue before flushing md->wq in case request-based
2568 * dm defers requests to md->wq from md->queue.
2570 if (dm_request_based(md))
2571 dm_stop_queue(md->queue);
2573 flush_workqueue(md->wq);
2576 * At this point no more requests are entering target request routines.
2577 * We call dm_wait_for_completion to wait for all existing requests
2580 r = dm_wait_for_completion(md, task_state);
2582 set_bit(dmf_suspended_flag, &md->flags);
2585 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2587 synchronize_srcu(&md->io_barrier);
2589 /* were we interrupted ? */
2593 if (dm_request_based(md))
2594 dm_start_queue(md->queue);
2597 dm_table_presuspend_undo_targets(map);
2598 /* pushback list is already flushed, so skip flush */
2605 * We need to be able to change a mapping table under a mounted
2606 * filesystem. For example we might want to move some data in
2607 * the background. Before the table can be swapped with
2608 * dm_bind_table, dm_suspend must be called to flush any in
2609 * flight bios and ensure that any further io gets deferred.
2612 * Suspend mechanism in request-based dm.
2614 * 1. Flush all I/Os by lock_fs() if needed.
2615 * 2. Stop dispatching any I/O by stopping the request_queue.
2616 * 3. Wait for all in-flight I/Os to be completed or requeued.
2618 * To abort suspend, start the request_queue.
2620 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2622 struct dm_table *map = NULL;
2626 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2628 if (dm_suspended_md(md)) {
2633 if (dm_suspended_internally_md(md)) {
2634 /* already internally suspended, wait for internal resume */
2635 mutex_unlock(&md->suspend_lock);
2636 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2642 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2644 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2648 dm_table_postsuspend_targets(map);
2651 mutex_unlock(&md->suspend_lock);
2655 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2658 int r = dm_table_resume_targets(map);
2666 * Flushing deferred I/Os must be done after targets are resumed
2667 * so that mapping of targets can work correctly.
2668 * Request-based dm is queueing the deferred I/Os in its request_queue.
2670 if (dm_request_based(md))
2671 dm_start_queue(md->queue);
2678 int dm_resume(struct mapped_device *md)
2681 struct dm_table *map = NULL;
2685 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2687 if (!dm_suspended_md(md))
2690 if (dm_suspended_internally_md(md)) {
2691 /* already internally suspended, wait for internal resume */
2692 mutex_unlock(&md->suspend_lock);
2693 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2699 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2700 if (!map || !dm_table_get_size(map))
2703 r = __dm_resume(md, map);
2707 clear_bit(DMF_SUSPENDED, &md->flags);
2709 mutex_unlock(&md->suspend_lock);
2715 * Internal suspend/resume works like userspace-driven suspend. It waits
2716 * until all bios finish and prevents issuing new bios to the target drivers.
2717 * It may be used only from the kernel.
2720 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2722 struct dm_table *map = NULL;
2724 lockdep_assert_held(&md->suspend_lock);
2726 if (md->internal_suspend_count++)
2727 return; /* nested internal suspend */
2729 if (dm_suspended_md(md)) {
2730 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2731 return; /* nest suspend */
2734 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2737 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2738 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2739 * would require changing .presuspend to return an error -- avoid this
2740 * until there is a need for more elaborate variants of internal suspend.
2742 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2743 DMF_SUSPENDED_INTERNALLY);
2745 dm_table_postsuspend_targets(map);
2748 static void __dm_internal_resume(struct mapped_device *md)
2750 BUG_ON(!md->internal_suspend_count);
2752 if (--md->internal_suspend_count)
2753 return; /* resume from nested internal suspend */
2755 if (dm_suspended_md(md))
2756 goto done; /* resume from nested suspend */
2759 * NOTE: existing callers don't need to call dm_table_resume_targets
2760 * (which may fail -- so best to avoid it for now by passing NULL map)
2762 (void) __dm_resume(md, NULL);
2765 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2766 smp_mb__after_atomic();
2767 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2770 void dm_internal_suspend_noflush(struct mapped_device *md)
2772 mutex_lock(&md->suspend_lock);
2773 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2774 mutex_unlock(&md->suspend_lock);
2776 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2778 void dm_internal_resume(struct mapped_device *md)
2780 mutex_lock(&md->suspend_lock);
2781 __dm_internal_resume(md);
2782 mutex_unlock(&md->suspend_lock);
2784 EXPORT_SYMBOL_GPL(dm_internal_resume);
2787 * Fast variants of internal suspend/resume hold md->suspend_lock,
2788 * which prevents interaction with userspace-driven suspend.
2791 void dm_internal_suspend_fast(struct mapped_device *md)
2793 mutex_lock(&md->suspend_lock);
2794 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2797 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2798 synchronize_srcu(&md->io_barrier);
2799 flush_workqueue(md->wq);
2800 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2802 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2804 void dm_internal_resume_fast(struct mapped_device *md)
2806 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2812 mutex_unlock(&md->suspend_lock);
2814 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2816 /*-----------------------------------------------------------------
2817 * Event notification.
2818 *---------------------------------------------------------------*/
2819 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2822 char udev_cookie[DM_COOKIE_LENGTH];
2823 char *envp[] = { udev_cookie, NULL };
2826 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2828 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2829 DM_COOKIE_ENV_VAR_NAME, cookie);
2830 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2835 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2837 return atomic_add_return(1, &md->uevent_seq);
2840 uint32_t dm_get_event_nr(struct mapped_device *md)
2842 return atomic_read(&md->event_nr);
2845 int dm_wait_event(struct mapped_device *md, int event_nr)
2847 return wait_event_interruptible(md->eventq,
2848 (event_nr != atomic_read(&md->event_nr)));
2851 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2853 unsigned long flags;
2855 spin_lock_irqsave(&md->uevent_lock, flags);
2856 list_add(elist, &md->uevent_list);
2857 spin_unlock_irqrestore(&md->uevent_lock, flags);
2861 * The gendisk is only valid as long as you have a reference
2864 struct gendisk *dm_disk(struct mapped_device *md)
2868 EXPORT_SYMBOL_GPL(dm_disk);
2870 struct kobject *dm_kobject(struct mapped_device *md)
2872 return &md->kobj_holder.kobj;
2875 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2877 struct mapped_device *md;
2879 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2881 spin_lock(&_minor_lock);
2882 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2888 spin_unlock(&_minor_lock);
2893 int dm_suspended_md(struct mapped_device *md)
2895 return test_bit(DMF_SUSPENDED, &md->flags);
2898 int dm_suspended_internally_md(struct mapped_device *md)
2900 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2903 int dm_test_deferred_remove_flag(struct mapped_device *md)
2905 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2908 int dm_suspended(struct dm_target *ti)
2910 return dm_suspended_md(dm_table_get_md(ti->table));
2912 EXPORT_SYMBOL_GPL(dm_suspended);
2914 int dm_noflush_suspending(struct dm_target *ti)
2916 return __noflush_suspending(dm_table_get_md(ti->table));
2918 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2920 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2921 unsigned integrity, unsigned per_io_data_size,
2922 unsigned min_pool_size)
2924 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2925 unsigned int pool_size = 0;
2926 unsigned int front_pad, io_front_pad;
2933 case DM_TYPE_BIO_BASED:
2934 case DM_TYPE_DAX_BIO_BASED:
2935 case DM_TYPE_NVME_BIO_BASED:
2936 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2937 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2938 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2939 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2942 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2945 case DM_TYPE_REQUEST_BASED:
2946 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2947 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2948 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2954 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2958 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2964 dm_free_md_mempools(pools);
2969 void dm_free_md_mempools(struct dm_md_mempools *pools)
2974 bioset_exit(&pools->bs);
2975 bioset_exit(&pools->io_bs);
2987 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2990 struct mapped_device *md = bdev->bd_disk->private_data;
2991 struct dm_table *table;
2992 struct dm_target *ti;
2993 int ret = -ENOTTY, srcu_idx;
2995 table = dm_get_live_table(md, &srcu_idx);
2996 if (!table || !dm_table_get_size(table))
2999 /* We only support devices that have a single target */
3000 if (dm_table_get_num_targets(table) != 1)
3002 ti = dm_table_get_target(table, 0);
3005 if (!ti->type->iterate_devices)
3008 ret = ti->type->iterate_devices(ti, fn, data);
3010 dm_put_live_table(md, srcu_idx);
3015 * For register / unregister we need to manually call out to every path.
3017 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3018 sector_t start, sector_t len, void *data)
3020 struct dm_pr *pr = data;
3021 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3023 if (!ops || !ops->pr_register)
3025 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3028 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3039 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3040 if (ret && new_key) {
3041 /* unregister all paths if we failed to register any path */
3042 pr.old_key = new_key;
3045 pr.fail_early = false;
3046 dm_call_pr(bdev, __dm_pr_register, &pr);
3052 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3055 struct mapped_device *md = bdev->bd_disk->private_data;
3056 const struct pr_ops *ops;
3059 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3063 ops = bdev->bd_disk->fops->pr_ops;
3064 if (ops && ops->pr_reserve)
3065 r = ops->pr_reserve(bdev, key, type, flags);
3069 dm_unprepare_ioctl(md, srcu_idx);
3073 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3075 struct mapped_device *md = bdev->bd_disk->private_data;
3076 const struct pr_ops *ops;
3079 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3083 ops = bdev->bd_disk->fops->pr_ops;
3084 if (ops && ops->pr_release)
3085 r = ops->pr_release(bdev, key, type);
3089 dm_unprepare_ioctl(md, srcu_idx);
3093 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3094 enum pr_type type, bool abort)
3096 struct mapped_device *md = bdev->bd_disk->private_data;
3097 const struct pr_ops *ops;
3100 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3104 ops = bdev->bd_disk->fops->pr_ops;
3105 if (ops && ops->pr_preempt)
3106 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3110 dm_unprepare_ioctl(md, srcu_idx);
3114 static int dm_pr_clear(struct block_device *bdev, u64 key)
3116 struct mapped_device *md = bdev->bd_disk->private_data;
3117 const struct pr_ops *ops;
3120 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3124 ops = bdev->bd_disk->fops->pr_ops;
3125 if (ops && ops->pr_clear)
3126 r = ops->pr_clear(bdev, key);
3130 dm_unprepare_ioctl(md, srcu_idx);
3134 static const struct pr_ops dm_pr_ops = {
3135 .pr_register = dm_pr_register,
3136 .pr_reserve = dm_pr_reserve,
3137 .pr_release = dm_pr_release,
3138 .pr_preempt = dm_pr_preempt,
3139 .pr_clear = dm_pr_clear,
3142 static const struct block_device_operations dm_blk_dops = {
3143 .open = dm_blk_open,
3144 .release = dm_blk_close,
3145 .ioctl = dm_blk_ioctl,
3146 .getgeo = dm_blk_getgeo,
3147 .report_zones = dm_blk_report_zones,
3148 .pr_ops = &dm_pr_ops,
3149 .owner = THIS_MODULE
3152 static const struct dax_operations dm_dax_ops = {
3153 .direct_access = dm_dax_direct_access,
3154 .copy_from_iter = dm_dax_copy_from_iter,
3155 .copy_to_iter = dm_dax_copy_to_iter,
3161 module_init(dm_init);
3162 module_exit(dm_exit);
3164 module_param(major, uint, 0);
3165 MODULE_PARM_DESC(major, "The major number of the device mapper");
3167 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3168 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3170 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3171 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3173 MODULE_DESCRIPTION(DM_NAME " driver");
3174 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3175 MODULE_LICENSE("GPL");