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;
162 * Bio-based DM's mempools' reserved IOs set by the user.
164 #define RESERVED_BIO_BASED_IOS 16
165 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
167 static int __dm_get_module_param_int(int *module_param, int min, int max)
169 int param = READ_ONCE(*module_param);
170 int modified_param = 0;
171 bool modified = true;
174 modified_param = min;
175 else if (param > max)
176 modified_param = max;
181 (void)cmpxchg(module_param, param, modified_param);
182 param = modified_param;
188 unsigned __dm_get_module_param(unsigned *module_param,
189 unsigned def, unsigned max)
191 unsigned param = READ_ONCE(*module_param);
192 unsigned modified_param = 0;
195 modified_param = def;
196 else if (param > max)
197 modified_param = max;
199 if (modified_param) {
200 (void)cmpxchg(module_param, param, modified_param);
201 param = modified_param;
207 unsigned dm_get_reserved_bio_based_ios(void)
209 return __dm_get_module_param(&reserved_bio_based_ios,
210 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
212 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
214 static unsigned dm_get_numa_node(void)
216 return __dm_get_module_param_int(&dm_numa_node,
217 DM_NUMA_NODE, num_online_nodes() - 1);
220 static int __init local_init(void)
224 r = dm_uevent_init();
228 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
229 if (!deferred_remove_workqueue) {
231 goto out_uevent_exit;
235 r = register_blkdev(_major, _name);
237 goto out_free_workqueue;
245 destroy_workqueue(deferred_remove_workqueue);
252 static void local_exit(void)
254 flush_scheduled_work();
255 destroy_workqueue(deferred_remove_workqueue);
257 unregister_blkdev(_major, _name);
262 DMINFO("cleaned up");
265 static int (*_inits[])(void) __initdata = {
276 static void (*_exits[])(void) = {
287 static int __init dm_init(void)
289 const int count = ARRAY_SIZE(_inits);
293 for (i = 0; i < count; i++) {
308 static void __exit dm_exit(void)
310 int i = ARRAY_SIZE(_exits);
316 * Should be empty by this point.
318 idr_destroy(&_minor_idr);
322 * Block device functions
324 int dm_deleting_md(struct mapped_device *md)
326 return test_bit(DMF_DELETING, &md->flags);
329 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
331 struct mapped_device *md;
333 spin_lock(&_minor_lock);
335 md = bdev->bd_disk->private_data;
339 if (test_bit(DMF_FREEING, &md->flags) ||
340 dm_deleting_md(md)) {
346 atomic_inc(&md->open_count);
348 spin_unlock(&_minor_lock);
350 return md ? 0 : -ENXIO;
353 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
355 struct mapped_device *md;
357 spin_lock(&_minor_lock);
359 md = disk->private_data;
363 if (atomic_dec_and_test(&md->open_count) &&
364 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
365 queue_work(deferred_remove_workqueue, &deferred_remove_work);
369 spin_unlock(&_minor_lock);
372 int dm_open_count(struct mapped_device *md)
374 return atomic_read(&md->open_count);
378 * Guarantees nothing is using the device before it's deleted.
380 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
384 spin_lock(&_minor_lock);
386 if (dm_open_count(md)) {
389 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
390 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
393 set_bit(DMF_DELETING, &md->flags);
395 spin_unlock(&_minor_lock);
400 int dm_cancel_deferred_remove(struct mapped_device *md)
404 spin_lock(&_minor_lock);
406 if (test_bit(DMF_DELETING, &md->flags))
409 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
411 spin_unlock(&_minor_lock);
416 static void do_deferred_remove(struct work_struct *w)
418 dm_deferred_remove();
421 sector_t dm_get_size(struct mapped_device *md)
423 return get_capacity(md->disk);
426 struct request_queue *dm_get_md_queue(struct mapped_device *md)
431 struct dm_stats *dm_get_stats(struct mapped_device *md)
436 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
438 struct mapped_device *md = bdev->bd_disk->private_data;
440 return dm_get_geometry(md, geo);
443 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
444 struct blk_zone *zones, unsigned int *nr_zones,
447 #ifdef CONFIG_BLK_DEV_ZONED
448 struct mapped_device *md = disk->private_data;
449 struct dm_target *tgt;
450 struct dm_table *map;
453 if (dm_suspended_md(md))
456 map = dm_get_live_table(md, &srcu_idx);
460 tgt = dm_table_find_target(map, sector);
461 if (!dm_target_is_valid(tgt)) {
467 * If we are executing this, we already know that the block device
468 * is a zoned device and so each target should have support for that
469 * type of drive. A missing report_zones method means that the target
470 * driver has a problem.
472 if (WARN_ON(!tgt->type->report_zones)) {
478 * blkdev_report_zones() will loop and call this again to cover all the
479 * zones of the target, eventually moving on to the next target.
480 * So there is no need to loop here trying to fill the entire array
483 ret = tgt->type->report_zones(tgt, sector, zones,
487 dm_put_live_table(md, srcu_idx);
494 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
495 struct block_device **bdev)
496 __acquires(md->io_barrier)
498 struct dm_target *tgt;
499 struct dm_table *map;
504 map = dm_get_live_table(md, srcu_idx);
505 if (!map || !dm_table_get_size(map))
508 /* We only support devices that have a single target */
509 if (dm_table_get_num_targets(map) != 1)
512 tgt = dm_table_get_target(map, 0);
513 if (!tgt->type->prepare_ioctl)
516 if (dm_suspended_md(md))
519 r = tgt->type->prepare_ioctl(tgt, bdev);
520 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
521 dm_put_live_table(md, *srcu_idx);
529 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
530 __releases(md->io_barrier)
532 dm_put_live_table(md, srcu_idx);
535 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
536 unsigned int cmd, unsigned long arg)
538 struct mapped_device *md = bdev->bd_disk->private_data;
541 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
547 * Target determined this ioctl is being issued against a
548 * subset of the parent bdev; require extra privileges.
550 if (!capable(CAP_SYS_RAWIO)) {
552 "%s: sending ioctl %x to DM device without required privilege.",
559 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
561 dm_unprepare_ioctl(md, srcu_idx);
565 static void start_io_acct(struct dm_io *io);
567 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
570 struct dm_target_io *tio;
573 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
577 tio = container_of(clone, struct dm_target_io, clone);
578 tio->inside_dm_io = true;
581 io = container_of(tio, struct dm_io, tio);
582 io->magic = DM_IO_MAGIC;
584 atomic_set(&io->io_count, 1);
587 spin_lock_init(&io->endio_lock);
594 static void free_io(struct mapped_device *md, struct dm_io *io)
596 bio_put(&io->tio.clone);
599 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
600 unsigned target_bio_nr, gfp_t gfp_mask)
602 struct dm_target_io *tio;
604 if (!ci->io->tio.io) {
605 /* the dm_target_io embedded in ci->io is available */
608 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
612 tio = container_of(clone, struct dm_target_io, clone);
613 tio->inside_dm_io = false;
616 tio->magic = DM_TIO_MAGIC;
619 tio->target_bio_nr = target_bio_nr;
624 static void free_tio(struct dm_target_io *tio)
626 if (tio->inside_dm_io)
628 bio_put(&tio->clone);
631 static bool md_in_flight_bios(struct mapped_device *md)
634 struct hd_struct *part = &dm_disk(md)->part0;
637 for_each_possible_cpu(cpu) {
638 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
639 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
645 static bool md_in_flight(struct mapped_device *md)
647 if (queue_is_mq(md->queue))
648 return blk_mq_queue_inflight(md->queue);
650 return md_in_flight_bios(md);
653 static void start_io_acct(struct dm_io *io)
655 struct mapped_device *md = io->md;
656 struct bio *bio = io->orig_bio;
658 io->start_time = jiffies;
660 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
661 &dm_disk(md)->part0);
663 if (unlikely(dm_stats_used(&md->stats)))
664 dm_stats_account_io(&md->stats, bio_data_dir(bio),
665 bio->bi_iter.bi_sector, bio_sectors(bio),
666 false, 0, &io->stats_aux);
669 static void end_io_acct(struct dm_io *io)
671 struct mapped_device *md = io->md;
672 struct bio *bio = io->orig_bio;
673 unsigned long duration = jiffies - io->start_time;
675 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
678 if (unlikely(dm_stats_used(&md->stats)))
679 dm_stats_account_io(&md->stats, bio_data_dir(bio),
680 bio->bi_iter.bi_sector, bio_sectors(bio),
681 true, duration, &io->stats_aux);
683 /* nudge anyone waiting on suspend queue */
684 if (unlikely(wq_has_sleeper(&md->wait)))
689 * Add the bio to the list of deferred io.
691 static void queue_io(struct mapped_device *md, struct bio *bio)
695 spin_lock_irqsave(&md->deferred_lock, flags);
696 bio_list_add(&md->deferred, bio);
697 spin_unlock_irqrestore(&md->deferred_lock, flags);
698 queue_work(md->wq, &md->work);
702 * Everyone (including functions in this file), should use this
703 * function to access the md->map field, and make sure they call
704 * dm_put_live_table() when finished.
706 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
708 *srcu_idx = srcu_read_lock(&md->io_barrier);
710 return srcu_dereference(md->map, &md->io_barrier);
713 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
715 srcu_read_unlock(&md->io_barrier, srcu_idx);
718 void dm_sync_table(struct mapped_device *md)
720 synchronize_srcu(&md->io_barrier);
721 synchronize_rcu_expedited();
725 * A fast alternative to dm_get_live_table/dm_put_live_table.
726 * The caller must not block between these two functions.
728 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
731 return rcu_dereference(md->map);
734 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
739 static char *_dm_claim_ptr = "I belong to device-mapper";
742 * Open a table device so we can use it as a map destination.
744 static int open_table_device(struct table_device *td, dev_t dev,
745 struct mapped_device *md)
747 struct block_device *bdev;
751 BUG_ON(td->dm_dev.bdev);
753 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
755 return PTR_ERR(bdev);
757 r = bd_link_disk_holder(bdev, dm_disk(md));
759 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
763 td->dm_dev.bdev = bdev;
764 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
769 * Close a table device that we've been using.
771 static void close_table_device(struct table_device *td, struct mapped_device *md)
773 if (!td->dm_dev.bdev)
776 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
777 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
778 put_dax(td->dm_dev.dax_dev);
779 td->dm_dev.bdev = NULL;
780 td->dm_dev.dax_dev = NULL;
783 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
785 struct table_device *td;
787 list_for_each_entry(td, l, list)
788 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
794 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
795 struct dm_dev **result) {
797 struct table_device *td;
799 mutex_lock(&md->table_devices_lock);
800 td = find_table_device(&md->table_devices, dev, mode);
802 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
804 mutex_unlock(&md->table_devices_lock);
808 td->dm_dev.mode = mode;
809 td->dm_dev.bdev = NULL;
811 if ((r = open_table_device(td, dev, md))) {
812 mutex_unlock(&md->table_devices_lock);
817 format_dev_t(td->dm_dev.name, dev);
819 refcount_set(&td->count, 1);
820 list_add(&td->list, &md->table_devices);
822 refcount_inc(&td->count);
824 mutex_unlock(&md->table_devices_lock);
826 *result = &td->dm_dev;
829 EXPORT_SYMBOL_GPL(dm_get_table_device);
831 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
833 struct table_device *td = container_of(d, struct table_device, dm_dev);
835 mutex_lock(&md->table_devices_lock);
836 if (refcount_dec_and_test(&td->count)) {
837 close_table_device(td, md);
841 mutex_unlock(&md->table_devices_lock);
843 EXPORT_SYMBOL(dm_put_table_device);
845 static void free_table_devices(struct list_head *devices)
847 struct list_head *tmp, *next;
849 list_for_each_safe(tmp, next, devices) {
850 struct table_device *td = list_entry(tmp, struct table_device, list);
852 DMWARN("dm_destroy: %s still exists with %d references",
853 td->dm_dev.name, refcount_read(&td->count));
859 * Get the geometry associated with a dm device
861 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
869 * Set the geometry of a device.
871 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
873 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
875 if (geo->start > sz) {
876 DMWARN("Start sector is beyond the geometry limits.");
885 static int __noflush_suspending(struct mapped_device *md)
887 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
891 * Decrements the number of outstanding ios that a bio has been
892 * cloned into, completing the original io if necc.
894 static void dec_pending(struct dm_io *io, blk_status_t error)
897 blk_status_t io_error;
899 struct mapped_device *md = io->md;
901 /* Push-back supersedes any I/O errors */
902 if (unlikely(error)) {
903 spin_lock_irqsave(&io->endio_lock, flags);
904 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
906 spin_unlock_irqrestore(&io->endio_lock, flags);
909 if (atomic_dec_and_test(&io->io_count)) {
910 if (io->status == BLK_STS_DM_REQUEUE) {
912 * Target requested pushing back the I/O.
914 spin_lock_irqsave(&md->deferred_lock, flags);
915 if (__noflush_suspending(md))
916 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
917 bio_list_add_head(&md->deferred, io->orig_bio);
919 /* noflush suspend was interrupted. */
920 io->status = BLK_STS_IOERR;
921 spin_unlock_irqrestore(&md->deferred_lock, flags);
924 io_error = io->status;
929 if (io_error == BLK_STS_DM_REQUEUE)
932 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
934 * Preflush done for flush with data, reissue
935 * without REQ_PREFLUSH.
937 bio->bi_opf &= ~REQ_PREFLUSH;
940 /* done with normal IO or empty flush */
942 bio->bi_status = io_error;
948 void disable_write_same(struct mapped_device *md)
950 struct queue_limits *limits = dm_get_queue_limits(md);
952 /* device doesn't really support WRITE SAME, disable it */
953 limits->max_write_same_sectors = 0;
956 void disable_write_zeroes(struct mapped_device *md)
958 struct queue_limits *limits = dm_get_queue_limits(md);
960 /* device doesn't really support WRITE ZEROES, disable it */
961 limits->max_write_zeroes_sectors = 0;
964 static void clone_endio(struct bio *bio)
966 blk_status_t error = bio->bi_status;
967 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
968 struct dm_io *io = tio->io;
969 struct mapped_device *md = tio->io->md;
970 dm_endio_fn endio = tio->ti->type->end_io;
972 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
973 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
974 !bio->bi_disk->queue->limits.max_write_same_sectors)
975 disable_write_same(md);
976 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
977 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
978 disable_write_zeroes(md);
982 int r = endio(tio->ti, bio, &error);
984 case DM_ENDIO_REQUEUE:
985 error = BLK_STS_DM_REQUEUE;
989 case DM_ENDIO_INCOMPLETE:
990 /* The target will handle the io */
993 DMWARN("unimplemented target endio return value: %d", r);
999 dec_pending(io, error);
1003 * Return maximum size of I/O possible at the supplied sector up to the current
1006 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1008 sector_t target_offset = dm_target_offset(ti, sector);
1010 return ti->len - target_offset;
1013 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1015 sector_t len = max_io_len_target_boundary(sector, ti);
1016 sector_t offset, max_len;
1019 * Does the target need to split even further?
1021 if (ti->max_io_len) {
1022 offset = dm_target_offset(ti, sector);
1023 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1024 max_len = sector_div(offset, ti->max_io_len);
1026 max_len = offset & (ti->max_io_len - 1);
1027 max_len = ti->max_io_len - max_len;
1036 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1038 if (len > UINT_MAX) {
1039 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1040 (unsigned long long)len, UINT_MAX);
1041 ti->error = "Maximum size of target IO is too large";
1046 * BIO based queue uses its own splitting. When multipage bvecs
1047 * is switched on, size of the incoming bio may be too big to
1048 * be handled in some targets, such as crypt.
1050 * When these targets are ready for the big bio, we can remove
1053 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1057 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1059 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1060 sector_t sector, int *srcu_idx)
1061 __acquires(md->io_barrier)
1063 struct dm_table *map;
1064 struct dm_target *ti;
1066 map = dm_get_live_table(md, srcu_idx);
1070 ti = dm_table_find_target(map, sector);
1071 if (!dm_target_is_valid(ti))
1077 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1078 long nr_pages, void **kaddr, pfn_t *pfn)
1080 struct mapped_device *md = dax_get_private(dax_dev);
1081 sector_t sector = pgoff * PAGE_SECTORS;
1082 struct dm_target *ti;
1083 long len, ret = -EIO;
1086 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1090 if (!ti->type->direct_access)
1092 len = max_io_len(sector, ti) / PAGE_SECTORS;
1095 nr_pages = min(len, nr_pages);
1096 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1099 dm_put_live_table(md, srcu_idx);
1104 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1105 void *addr, size_t bytes, struct iov_iter *i)
1107 struct mapped_device *md = dax_get_private(dax_dev);
1108 sector_t sector = pgoff * PAGE_SECTORS;
1109 struct dm_target *ti;
1113 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1117 if (!ti->type->dax_copy_from_iter) {
1118 ret = copy_from_iter(addr, bytes, i);
1121 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1123 dm_put_live_table(md, srcu_idx);
1128 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1129 void *addr, size_t bytes, struct iov_iter *i)
1131 struct mapped_device *md = dax_get_private(dax_dev);
1132 sector_t sector = pgoff * PAGE_SECTORS;
1133 struct dm_target *ti;
1137 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1141 if (!ti->type->dax_copy_to_iter) {
1142 ret = copy_to_iter(addr, bytes, i);
1145 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1147 dm_put_live_table(md, srcu_idx);
1153 * A target may call dm_accept_partial_bio only from the map routine. It is
1154 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1156 * dm_accept_partial_bio informs the dm that the target only wants to process
1157 * additional n_sectors sectors of the bio and the rest of the data should be
1158 * sent in a next bio.
1160 * A diagram that explains the arithmetics:
1161 * +--------------------+---------------+-------+
1163 * +--------------------+---------------+-------+
1165 * <-------------- *tio->len_ptr --------------->
1166 * <------- bi_size ------->
1169 * Region 1 was already iterated over with bio_advance or similar function.
1170 * (it may be empty if the target doesn't use bio_advance)
1171 * Region 2 is the remaining bio size that the target wants to process.
1172 * (it may be empty if region 1 is non-empty, although there is no reason
1174 * The target requires that region 3 is to be sent in the next bio.
1176 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1177 * the partially processed part (the sum of regions 1+2) must be the same for all
1178 * copies of the bio.
1180 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1182 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1183 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1184 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1185 BUG_ON(bi_size > *tio->len_ptr);
1186 BUG_ON(n_sectors > bi_size);
1187 *tio->len_ptr -= bi_size - n_sectors;
1188 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1190 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1193 * The zone descriptors obtained with a zone report indicate
1194 * zone positions within the underlying device of the target. The zone
1195 * descriptors must be remapped to match their position within the dm device.
1196 * The caller target should obtain the zones information using
1197 * blkdev_report_zones() to ensure that remapping for partition offset is
1200 void dm_remap_zone_report(struct dm_target *ti, sector_t start,
1201 struct blk_zone *zones, unsigned int *nr_zones)
1203 #ifdef CONFIG_BLK_DEV_ZONED
1204 struct blk_zone *zone;
1205 unsigned int nrz = *nr_zones;
1209 * Remap the start sector and write pointer position of the zones in
1210 * the array. Since we may have obtained from the target underlying
1211 * device more zones that the target size, also adjust the number
1214 for (i = 0; i < nrz; i++) {
1216 if (zone->start >= start + ti->len) {
1217 memset(zone, 0, sizeof(struct blk_zone) * (nrz - i));
1221 zone->start = zone->start + ti->begin - start;
1222 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
1225 if (zone->cond == BLK_ZONE_COND_FULL)
1226 zone->wp = zone->start + zone->len;
1227 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1228 zone->wp = zone->start;
1230 zone->wp = zone->wp + ti->begin - start;
1234 #else /* !CONFIG_BLK_DEV_ZONED */
1238 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1240 static blk_qc_t __map_bio(struct dm_target_io *tio)
1244 struct bio *clone = &tio->clone;
1245 struct dm_io *io = tio->io;
1246 struct mapped_device *md = io->md;
1247 struct dm_target *ti = tio->ti;
1248 blk_qc_t ret = BLK_QC_T_NONE;
1250 clone->bi_end_io = clone_endio;
1253 * Map the clone. If r == 0 we don't need to do
1254 * anything, the target has assumed ownership of
1257 atomic_inc(&io->io_count);
1258 sector = clone->bi_iter.bi_sector;
1260 r = ti->type->map(ti, clone);
1262 case DM_MAPIO_SUBMITTED:
1264 case DM_MAPIO_REMAPPED:
1265 /* the bio has been remapped so dispatch it */
1266 trace_block_bio_remap(clone->bi_disk->queue, clone,
1267 bio_dev(io->orig_bio), sector);
1268 if (md->type == DM_TYPE_NVME_BIO_BASED)
1269 ret = direct_make_request(clone);
1271 ret = generic_make_request(clone);
1275 dec_pending(io, BLK_STS_IOERR);
1277 case DM_MAPIO_REQUEUE:
1279 dec_pending(io, BLK_STS_DM_REQUEUE);
1282 DMWARN("unimplemented target map return value: %d", r);
1289 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1291 bio->bi_iter.bi_sector = sector;
1292 bio->bi_iter.bi_size = to_bytes(len);
1296 * Creates a bio that consists of range of complete bvecs.
1298 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1299 sector_t sector, unsigned len)
1301 struct bio *clone = &tio->clone;
1303 __bio_clone_fast(clone, bio);
1305 if (bio_integrity(bio)) {
1308 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1309 !dm_target_passes_integrity(tio->ti->type))) {
1310 DMWARN("%s: the target %s doesn't support integrity data.",
1311 dm_device_name(tio->io->md),
1312 tio->ti->type->name);
1316 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1321 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1322 clone->bi_iter.bi_size = to_bytes(len);
1324 if (bio_integrity(bio))
1325 bio_integrity_trim(clone);
1330 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1331 struct dm_target *ti, unsigned num_bios)
1333 struct dm_target_io *tio;
1339 if (num_bios == 1) {
1340 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1341 bio_list_add(blist, &tio->clone);
1345 for (try = 0; try < 2; try++) {
1350 mutex_lock(&ci->io->md->table_devices_lock);
1351 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1352 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1356 bio_list_add(blist, &tio->clone);
1359 mutex_unlock(&ci->io->md->table_devices_lock);
1360 if (bio_nr == num_bios)
1363 while ((bio = bio_list_pop(blist))) {
1364 tio = container_of(bio, struct dm_target_io, clone);
1370 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1371 struct dm_target_io *tio, unsigned *len)
1373 struct bio *clone = &tio->clone;
1377 __bio_clone_fast(clone, ci->bio);
1379 bio_setup_sector(clone, ci->sector, *len);
1381 return __map_bio(tio);
1384 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1385 unsigned num_bios, unsigned *len)
1387 struct bio_list blist = BIO_EMPTY_LIST;
1389 struct dm_target_io *tio;
1391 alloc_multiple_bios(&blist, ci, ti, num_bios);
1393 while ((bio = bio_list_pop(&blist))) {
1394 tio = container_of(bio, struct dm_target_io, clone);
1395 (void) __clone_and_map_simple_bio(ci, tio, len);
1399 static int __send_empty_flush(struct clone_info *ci)
1401 unsigned target_nr = 0;
1402 struct dm_target *ti;
1405 * Empty flush uses a statically initialized bio, as the base for
1406 * cloning. However, blkg association requires that a bdev is
1407 * associated with a gendisk, which doesn't happen until the bdev is
1408 * opened. So, blkg association is done at issue time of the flush
1409 * rather than when the device is created in alloc_dev().
1411 bio_set_dev(ci->bio, ci->io->md->bdev);
1413 BUG_ON(bio_has_data(ci->bio));
1414 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1415 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1417 bio_disassociate_blkg(ci->bio);
1422 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1423 sector_t sector, unsigned *len)
1425 struct bio *bio = ci->bio;
1426 struct dm_target_io *tio;
1429 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1431 r = clone_bio(tio, bio, sector, *len);
1436 (void) __map_bio(tio);
1441 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1443 static unsigned get_num_discard_bios(struct dm_target *ti)
1445 return ti->num_discard_bios;
1448 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1450 return ti->num_secure_erase_bios;
1453 static unsigned get_num_write_same_bios(struct dm_target *ti)
1455 return ti->num_write_same_bios;
1458 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1460 return ti->num_write_zeroes_bios;
1463 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1466 unsigned len = ci->sector_count;
1469 * Even though the device advertised support for this type of
1470 * request, that does not mean every target supports it, and
1471 * reconfiguration might also have changed that since the
1472 * check was performed.
1477 __send_duplicate_bios(ci, ti, num_bios, &len);
1480 ci->sector_count -= len;
1485 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1487 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1490 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1492 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1495 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1497 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1500 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1502 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1505 static bool is_abnormal_io(struct bio *bio)
1509 switch (bio_op(bio)) {
1510 case REQ_OP_DISCARD:
1511 case REQ_OP_SECURE_ERASE:
1512 case REQ_OP_WRITE_SAME:
1513 case REQ_OP_WRITE_ZEROES:
1521 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1524 struct bio *bio = ci->bio;
1526 if (bio_op(bio) == REQ_OP_DISCARD)
1527 *result = __send_discard(ci, ti);
1528 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1529 *result = __send_secure_erase(ci, ti);
1530 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1531 *result = __send_write_same(ci, ti);
1532 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1533 *result = __send_write_zeroes(ci, ti);
1541 * Select the correct strategy for processing a non-flush bio.
1543 static int __split_and_process_non_flush(struct clone_info *ci)
1545 struct dm_target *ti;
1549 ti = dm_table_find_target(ci->map, ci->sector);
1550 if (!dm_target_is_valid(ti))
1553 if (__process_abnormal_io(ci, ti, &r))
1556 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1558 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1563 ci->sector_count -= len;
1568 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1569 struct dm_table *map, struct bio *bio)
1572 ci->io = alloc_io(md, bio);
1573 ci->sector = bio->bi_iter.bi_sector;
1576 #define __dm_part_stat_sub(part, field, subnd) \
1577 (part_stat_get(part, field) -= (subnd))
1580 * Entry point to split a bio into clones and submit them to the targets.
1582 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1583 struct dm_table *map, struct bio *bio)
1585 struct clone_info ci;
1586 blk_qc_t ret = BLK_QC_T_NONE;
1589 init_clone_info(&ci, md, map, bio);
1591 if (bio->bi_opf & REQ_PREFLUSH) {
1592 struct bio flush_bio;
1595 * Use an on-stack bio for this, it's safe since we don't
1596 * need to reference it after submit. It's just used as
1597 * the basis for the clone(s).
1599 bio_init(&flush_bio, NULL, 0);
1600 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1601 ci.bio = &flush_bio;
1602 ci.sector_count = 0;
1603 error = __send_empty_flush(&ci);
1604 /* dec_pending submits any data associated with flush */
1605 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1607 ci.sector_count = 0;
1608 error = __split_and_process_non_flush(&ci);
1611 ci.sector_count = bio_sectors(bio);
1612 while (ci.sector_count && !error) {
1613 error = __split_and_process_non_flush(&ci);
1614 if (current->bio_list && ci.sector_count && !error) {
1616 * Remainder must be passed to generic_make_request()
1617 * so that it gets handled *after* bios already submitted
1618 * have been completely processed.
1619 * We take a clone of the original to store in
1620 * ci.io->orig_bio to be used by end_io_acct() and
1621 * for dec_pending to use for completion handling.
1623 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1624 GFP_NOIO, &md->queue->bio_split);
1625 ci.io->orig_bio = b;
1628 * Adjust IO stats for each split, otherwise upon queue
1629 * reentry there will be redundant IO accounting.
1630 * NOTE: this is a stop-gap fix, a proper fix involves
1631 * significant refactoring of DM core's bio splitting
1632 * (by eliminating DM's splitting and just using bio_split)
1635 __dm_part_stat_sub(&dm_disk(md)->part0,
1636 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1640 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1641 ret = generic_make_request(bio);
1647 /* drop the extra reference count */
1648 dec_pending(ci.io, errno_to_blk_status(error));
1653 * Optimized variant of __split_and_process_bio that leverages the
1654 * fact that targets that use it do _not_ have a need to split bios.
1656 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1657 struct bio *bio, struct dm_target *ti)
1659 struct clone_info ci;
1660 blk_qc_t ret = BLK_QC_T_NONE;
1663 init_clone_info(&ci, md, map, bio);
1665 if (bio->bi_opf & REQ_PREFLUSH) {
1666 struct bio flush_bio;
1669 * Use an on-stack bio for this, it's safe since we don't
1670 * need to reference it after submit. It's just used as
1671 * the basis for the clone(s).
1673 bio_init(&flush_bio, NULL, 0);
1674 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1675 ci.bio = &flush_bio;
1676 ci.sector_count = 0;
1677 error = __send_empty_flush(&ci);
1678 /* dec_pending submits any data associated with flush */
1680 struct dm_target_io *tio;
1683 ci.sector_count = bio_sectors(bio);
1684 if (__process_abnormal_io(&ci, ti, &error))
1687 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1688 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1691 /* drop the extra reference count */
1692 dec_pending(ci.io, errno_to_blk_status(error));
1696 static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio)
1698 unsigned len, sector_count;
1700 sector_count = bio_sectors(*bio);
1701 len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count);
1703 if (sector_count > len) {
1704 struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split);
1706 bio_chain(split, *bio);
1707 trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector);
1708 generic_make_request(*bio);
1713 static blk_qc_t dm_process_bio(struct mapped_device *md,
1714 struct dm_table *map, struct bio *bio)
1716 blk_qc_t ret = BLK_QC_T_NONE;
1717 struct dm_target *ti = md->immutable_target;
1719 if (unlikely(!map)) {
1725 ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1726 if (unlikely(!ti || !dm_target_is_valid(ti))) {
1733 * If in ->make_request_fn we need to use blk_queue_split(), otherwise
1734 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1737 if (current->bio_list) {
1738 blk_queue_split(md->queue, &bio);
1739 if (!is_abnormal_io(bio))
1740 dm_queue_split(md, ti, &bio);
1743 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1744 return __process_bio(md, map, bio, ti);
1746 return __split_and_process_bio(md, map, bio);
1749 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1751 struct mapped_device *md = q->queuedata;
1752 blk_qc_t ret = BLK_QC_T_NONE;
1754 struct dm_table *map;
1756 map = dm_get_live_table(md, &srcu_idx);
1758 /* if we're suspended, we have to queue this io for later */
1759 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1760 dm_put_live_table(md, srcu_idx);
1762 if (!(bio->bi_opf & REQ_RAHEAD))
1769 ret = dm_process_bio(md, map, bio);
1771 dm_put_live_table(md, srcu_idx);
1775 static int dm_any_congested(void *congested_data, int bdi_bits)
1778 struct mapped_device *md = congested_data;
1779 struct dm_table *map;
1781 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1782 if (dm_request_based(md)) {
1784 * With request-based DM we only need to check the
1785 * top-level queue for congestion.
1787 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1789 map = dm_get_live_table_fast(md);
1791 r = dm_table_any_congested(map, bdi_bits);
1792 dm_put_live_table_fast(md);
1799 /*-----------------------------------------------------------------
1800 * An IDR is used to keep track of allocated minor numbers.
1801 *---------------------------------------------------------------*/
1802 static void free_minor(int minor)
1804 spin_lock(&_minor_lock);
1805 idr_remove(&_minor_idr, minor);
1806 spin_unlock(&_minor_lock);
1810 * See if the device with a specific minor # is free.
1812 static int specific_minor(int minor)
1816 if (minor >= (1 << MINORBITS))
1819 idr_preload(GFP_KERNEL);
1820 spin_lock(&_minor_lock);
1822 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1824 spin_unlock(&_minor_lock);
1827 return r == -ENOSPC ? -EBUSY : r;
1831 static int next_free_minor(int *minor)
1835 idr_preload(GFP_KERNEL);
1836 spin_lock(&_minor_lock);
1838 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1840 spin_unlock(&_minor_lock);
1848 static const struct block_device_operations dm_blk_dops;
1849 static const struct dax_operations dm_dax_ops;
1851 static void dm_wq_work(struct work_struct *work);
1853 static void dm_init_normal_md_queue(struct mapped_device *md)
1856 * Initialize aspects of queue that aren't relevant for blk-mq
1858 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1861 static void cleanup_mapped_device(struct mapped_device *md)
1864 destroy_workqueue(md->wq);
1865 bioset_exit(&md->bs);
1866 bioset_exit(&md->io_bs);
1869 kill_dax(md->dax_dev);
1870 put_dax(md->dax_dev);
1875 spin_lock(&_minor_lock);
1876 md->disk->private_data = NULL;
1877 spin_unlock(&_minor_lock);
1878 del_gendisk(md->disk);
1883 blk_cleanup_queue(md->queue);
1885 cleanup_srcu_struct(&md->io_barrier);
1892 mutex_destroy(&md->suspend_lock);
1893 mutex_destroy(&md->type_lock);
1894 mutex_destroy(&md->table_devices_lock);
1896 dm_mq_cleanup_mapped_device(md);
1900 * Allocate and initialise a blank device with a given minor.
1902 static struct mapped_device *alloc_dev(int minor)
1904 int r, numa_node_id = dm_get_numa_node();
1905 struct dax_device *dax_dev = NULL;
1906 struct mapped_device *md;
1909 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1911 DMWARN("unable to allocate device, out of memory.");
1915 if (!try_module_get(THIS_MODULE))
1916 goto bad_module_get;
1918 /* get a minor number for the dev */
1919 if (minor == DM_ANY_MINOR)
1920 r = next_free_minor(&minor);
1922 r = specific_minor(minor);
1926 r = init_srcu_struct(&md->io_barrier);
1928 goto bad_io_barrier;
1930 md->numa_node_id = numa_node_id;
1931 md->init_tio_pdu = false;
1932 md->type = DM_TYPE_NONE;
1933 mutex_init(&md->suspend_lock);
1934 mutex_init(&md->type_lock);
1935 mutex_init(&md->table_devices_lock);
1936 spin_lock_init(&md->deferred_lock);
1937 atomic_set(&md->holders, 1);
1938 atomic_set(&md->open_count, 0);
1939 atomic_set(&md->event_nr, 0);
1940 atomic_set(&md->uevent_seq, 0);
1941 INIT_LIST_HEAD(&md->uevent_list);
1942 INIT_LIST_HEAD(&md->table_devices);
1943 spin_lock_init(&md->uevent_lock);
1945 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1948 md->queue->queuedata = md;
1949 md->queue->backing_dev_info->congested_data = md;
1951 md->disk = alloc_disk_node(1, md->numa_node_id);
1955 init_waitqueue_head(&md->wait);
1956 INIT_WORK(&md->work, dm_wq_work);
1957 init_waitqueue_head(&md->eventq);
1958 init_completion(&md->kobj_holder.completion);
1960 md->disk->major = _major;
1961 md->disk->first_minor = minor;
1962 md->disk->fops = &dm_blk_dops;
1963 md->disk->queue = md->queue;
1964 md->disk->private_data = md;
1965 sprintf(md->disk->disk_name, "dm-%d", minor);
1967 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1968 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1972 md->dax_dev = dax_dev;
1974 add_disk_no_queue_reg(md->disk);
1975 format_dev_t(md->name, MKDEV(_major, minor));
1977 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1981 md->bdev = bdget_disk(md->disk, 0);
1985 dm_stats_init(&md->stats);
1987 /* Populate the mapping, nobody knows we exist yet */
1988 spin_lock(&_minor_lock);
1989 old_md = idr_replace(&_minor_idr, md, minor);
1990 spin_unlock(&_minor_lock);
1992 BUG_ON(old_md != MINOR_ALLOCED);
1997 cleanup_mapped_device(md);
2001 module_put(THIS_MODULE);
2007 static void unlock_fs(struct mapped_device *md);
2009 static void free_dev(struct mapped_device *md)
2011 int minor = MINOR(disk_devt(md->disk));
2015 cleanup_mapped_device(md);
2017 free_table_devices(&md->table_devices);
2018 dm_stats_cleanup(&md->stats);
2021 module_put(THIS_MODULE);
2025 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2027 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2030 if (dm_table_bio_based(t)) {
2032 * The md may already have mempools that need changing.
2033 * If so, reload bioset because front_pad may have changed
2034 * because a different table was loaded.
2036 bioset_exit(&md->bs);
2037 bioset_exit(&md->io_bs);
2039 } else if (bioset_initialized(&md->bs)) {
2041 * There's no need to reload with request-based dm
2042 * because the size of front_pad doesn't change.
2043 * Note for future: If you are to reload bioset,
2044 * prep-ed requests in the queue may refer
2045 * to bio from the old bioset, so you must walk
2046 * through the queue to unprep.
2052 bioset_initialized(&md->bs) ||
2053 bioset_initialized(&md->io_bs));
2055 ret = bioset_init_from_src(&md->bs, &p->bs);
2058 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2060 bioset_exit(&md->bs);
2062 /* mempool bind completed, no longer need any mempools in the table */
2063 dm_table_free_md_mempools(t);
2068 * Bind a table to the device.
2070 static void event_callback(void *context)
2072 unsigned long flags;
2074 struct mapped_device *md = (struct mapped_device *) context;
2076 spin_lock_irqsave(&md->uevent_lock, flags);
2077 list_splice_init(&md->uevent_list, &uevents);
2078 spin_unlock_irqrestore(&md->uevent_lock, flags);
2080 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2082 atomic_inc(&md->event_nr);
2083 wake_up(&md->eventq);
2084 dm_issue_global_event();
2088 * Protected by md->suspend_lock obtained by dm_swap_table().
2090 static void __set_size(struct mapped_device *md, sector_t size)
2092 lockdep_assert_held(&md->suspend_lock);
2094 set_capacity(md->disk, size);
2096 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2100 * Returns old map, which caller must destroy.
2102 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2103 struct queue_limits *limits)
2105 struct dm_table *old_map;
2106 struct request_queue *q = md->queue;
2107 bool request_based = dm_table_request_based(t);
2111 lockdep_assert_held(&md->suspend_lock);
2113 size = dm_table_get_size(t);
2116 * Wipe any geometry if the size of the table changed.
2118 if (size != dm_get_size(md))
2119 memset(&md->geometry, 0, sizeof(md->geometry));
2121 __set_size(md, size);
2123 dm_table_event_callback(t, event_callback, md);
2126 * The queue hasn't been stopped yet, if the old table type wasn't
2127 * for request-based during suspension. So stop it to prevent
2128 * I/O mapping before resume.
2129 * This must be done before setting the queue restrictions,
2130 * because request-based dm may be run just after the setting.
2135 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2137 * Leverage the fact that request-based DM targets and
2138 * NVMe bio based targets are immutable singletons
2139 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2140 * and __process_bio.
2142 md->immutable_target = dm_table_get_immutable_target(t);
2145 ret = __bind_mempools(md, t);
2147 old_map = ERR_PTR(ret);
2151 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2152 rcu_assign_pointer(md->map, (void *)t);
2153 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2155 dm_table_set_restrictions(t, q, limits);
2164 * Returns unbound table for the caller to free.
2166 static struct dm_table *__unbind(struct mapped_device *md)
2168 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2173 dm_table_event_callback(map, NULL, NULL);
2174 RCU_INIT_POINTER(md->map, NULL);
2181 * Constructor for a new device.
2183 int dm_create(int minor, struct mapped_device **result)
2186 struct mapped_device *md;
2188 md = alloc_dev(minor);
2192 r = dm_sysfs_init(md);
2203 * Functions to manage md->type.
2204 * All are required to hold md->type_lock.
2206 void dm_lock_md_type(struct mapped_device *md)
2208 mutex_lock(&md->type_lock);
2211 void dm_unlock_md_type(struct mapped_device *md)
2213 mutex_unlock(&md->type_lock);
2216 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2218 BUG_ON(!mutex_is_locked(&md->type_lock));
2222 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2227 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2229 return md->immutable_target_type;
2233 * The queue_limits are only valid as long as you have a reference
2236 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2238 BUG_ON(!atomic_read(&md->holders));
2239 return &md->queue->limits;
2241 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2244 * Setup the DM device's queue based on md's type
2246 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2249 struct queue_limits limits;
2250 enum dm_queue_mode type = dm_get_md_type(md);
2253 case DM_TYPE_REQUEST_BASED:
2254 r = dm_mq_init_request_queue(md, t);
2256 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2260 case DM_TYPE_BIO_BASED:
2261 case DM_TYPE_DAX_BIO_BASED:
2262 case DM_TYPE_NVME_BIO_BASED:
2263 dm_init_normal_md_queue(md);
2264 blk_queue_make_request(md->queue, dm_make_request);
2271 r = dm_calculate_queue_limits(t, &limits);
2273 DMERR("Cannot calculate initial queue limits");
2276 dm_table_set_restrictions(t, md->queue, &limits);
2277 blk_register_queue(md->disk);
2282 struct mapped_device *dm_get_md(dev_t dev)
2284 struct mapped_device *md;
2285 unsigned minor = MINOR(dev);
2287 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2290 spin_lock(&_minor_lock);
2292 md = idr_find(&_minor_idr, minor);
2293 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2294 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2300 spin_unlock(&_minor_lock);
2304 EXPORT_SYMBOL_GPL(dm_get_md);
2306 void *dm_get_mdptr(struct mapped_device *md)
2308 return md->interface_ptr;
2311 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2313 md->interface_ptr = ptr;
2316 void dm_get(struct mapped_device *md)
2318 atomic_inc(&md->holders);
2319 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2322 int dm_hold(struct mapped_device *md)
2324 spin_lock(&_minor_lock);
2325 if (test_bit(DMF_FREEING, &md->flags)) {
2326 spin_unlock(&_minor_lock);
2330 spin_unlock(&_minor_lock);
2333 EXPORT_SYMBOL_GPL(dm_hold);
2335 const char *dm_device_name(struct mapped_device *md)
2339 EXPORT_SYMBOL_GPL(dm_device_name);
2341 static void __dm_destroy(struct mapped_device *md, bool wait)
2343 struct dm_table *map;
2348 spin_lock(&_minor_lock);
2349 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2350 set_bit(DMF_FREEING, &md->flags);
2351 spin_unlock(&_minor_lock);
2353 blk_set_queue_dying(md->queue);
2356 * Take suspend_lock so that presuspend and postsuspend methods
2357 * do not race with internal suspend.
2359 mutex_lock(&md->suspend_lock);
2360 map = dm_get_live_table(md, &srcu_idx);
2361 if (!dm_suspended_md(md)) {
2362 dm_table_presuspend_targets(map);
2363 dm_table_postsuspend_targets(map);
2365 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2366 dm_put_live_table(md, srcu_idx);
2367 mutex_unlock(&md->suspend_lock);
2370 * Rare, but there may be I/O requests still going to complete,
2371 * for example. Wait for all references to disappear.
2372 * No one should increment the reference count of the mapped_device,
2373 * after the mapped_device state becomes DMF_FREEING.
2376 while (atomic_read(&md->holders))
2378 else if (atomic_read(&md->holders))
2379 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2380 dm_device_name(md), atomic_read(&md->holders));
2383 dm_table_destroy(__unbind(md));
2387 void dm_destroy(struct mapped_device *md)
2389 __dm_destroy(md, true);
2392 void dm_destroy_immediate(struct mapped_device *md)
2394 __dm_destroy(md, false);
2397 void dm_put(struct mapped_device *md)
2399 atomic_dec(&md->holders);
2401 EXPORT_SYMBOL_GPL(dm_put);
2403 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2409 prepare_to_wait(&md->wait, &wait, task_state);
2411 if (!md_in_flight(md))
2414 if (signal_pending_state(task_state, current)) {
2421 finish_wait(&md->wait, &wait);
2427 * Process the deferred bios
2429 static void dm_wq_work(struct work_struct *work)
2431 struct mapped_device *md = container_of(work, struct mapped_device,
2435 struct dm_table *map;
2437 map = dm_get_live_table(md, &srcu_idx);
2439 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2440 spin_lock_irq(&md->deferred_lock);
2441 c = bio_list_pop(&md->deferred);
2442 spin_unlock_irq(&md->deferred_lock);
2447 if (dm_request_based(md))
2448 (void) generic_make_request(c);
2450 (void) dm_process_bio(md, map, c);
2453 dm_put_live_table(md, srcu_idx);
2456 static void dm_queue_flush(struct mapped_device *md)
2458 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2459 smp_mb__after_atomic();
2460 queue_work(md->wq, &md->work);
2464 * Swap in a new table, returning the old one for the caller to destroy.
2466 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2468 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2469 struct queue_limits limits;
2472 mutex_lock(&md->suspend_lock);
2474 /* device must be suspended */
2475 if (!dm_suspended_md(md))
2479 * If the new table has no data devices, retain the existing limits.
2480 * This helps multipath with queue_if_no_path if all paths disappear,
2481 * then new I/O is queued based on these limits, and then some paths
2484 if (dm_table_has_no_data_devices(table)) {
2485 live_map = dm_get_live_table_fast(md);
2487 limits = md->queue->limits;
2488 dm_put_live_table_fast(md);
2492 r = dm_calculate_queue_limits(table, &limits);
2499 map = __bind(md, table, &limits);
2500 dm_issue_global_event();
2503 mutex_unlock(&md->suspend_lock);
2508 * Functions to lock and unlock any filesystem running on the
2511 static int lock_fs(struct mapped_device *md)
2515 WARN_ON(md->frozen_sb);
2517 md->frozen_sb = freeze_bdev(md->bdev);
2518 if (IS_ERR(md->frozen_sb)) {
2519 r = PTR_ERR(md->frozen_sb);
2520 md->frozen_sb = NULL;
2524 set_bit(DMF_FROZEN, &md->flags);
2529 static void unlock_fs(struct mapped_device *md)
2531 if (!test_bit(DMF_FROZEN, &md->flags))
2534 thaw_bdev(md->bdev, md->frozen_sb);
2535 md->frozen_sb = NULL;
2536 clear_bit(DMF_FROZEN, &md->flags);
2540 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2541 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2542 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2544 * If __dm_suspend returns 0, the device is completely quiescent
2545 * now. There is no request-processing activity. All new requests
2546 * are being added to md->deferred list.
2548 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2549 unsigned suspend_flags, long task_state,
2550 int dmf_suspended_flag)
2552 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2553 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2556 lockdep_assert_held(&md->suspend_lock);
2559 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2560 * This flag is cleared before dm_suspend returns.
2563 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2565 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2568 * This gets reverted if there's an error later and the targets
2569 * provide the .presuspend_undo hook.
2571 dm_table_presuspend_targets(map);
2574 * Flush I/O to the device.
2575 * Any I/O submitted after lock_fs() may not be flushed.
2576 * noflush takes precedence over do_lockfs.
2577 * (lock_fs() flushes I/Os and waits for them to complete.)
2579 if (!noflush && do_lockfs) {
2582 dm_table_presuspend_undo_targets(map);
2588 * Here we must make sure that no processes are submitting requests
2589 * to target drivers i.e. no one may be executing
2590 * __split_and_process_bio. This is called from dm_request and
2593 * To get all processes out of __split_and_process_bio in dm_request,
2594 * we take the write lock. To prevent any process from reentering
2595 * __split_and_process_bio from dm_request and quiesce the thread
2596 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2597 * flush_workqueue(md->wq).
2599 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2601 synchronize_srcu(&md->io_barrier);
2604 * Stop md->queue before flushing md->wq in case request-based
2605 * dm defers requests to md->wq from md->queue.
2607 if (dm_request_based(md))
2608 dm_stop_queue(md->queue);
2610 flush_workqueue(md->wq);
2613 * At this point no more requests are entering target request routines.
2614 * We call dm_wait_for_completion to wait for all existing requests
2617 r = dm_wait_for_completion(md, task_state);
2619 set_bit(dmf_suspended_flag, &md->flags);
2622 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2624 synchronize_srcu(&md->io_barrier);
2626 /* were we interrupted ? */
2630 if (dm_request_based(md))
2631 dm_start_queue(md->queue);
2634 dm_table_presuspend_undo_targets(map);
2635 /* pushback list is already flushed, so skip flush */
2642 * We need to be able to change a mapping table under a mounted
2643 * filesystem. For example we might want to move some data in
2644 * the background. Before the table can be swapped with
2645 * dm_bind_table, dm_suspend must be called to flush any in
2646 * flight bios and ensure that any further io gets deferred.
2649 * Suspend mechanism in request-based dm.
2651 * 1. Flush all I/Os by lock_fs() if needed.
2652 * 2. Stop dispatching any I/O by stopping the request_queue.
2653 * 3. Wait for all in-flight I/Os to be completed or requeued.
2655 * To abort suspend, start the request_queue.
2657 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2659 struct dm_table *map = NULL;
2663 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2665 if (dm_suspended_md(md)) {
2670 if (dm_suspended_internally_md(md)) {
2671 /* already internally suspended, wait for internal resume */
2672 mutex_unlock(&md->suspend_lock);
2673 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2679 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2681 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2685 dm_table_postsuspend_targets(map);
2688 mutex_unlock(&md->suspend_lock);
2692 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2695 int r = dm_table_resume_targets(map);
2703 * Flushing deferred I/Os must be done after targets are resumed
2704 * so that mapping of targets can work correctly.
2705 * Request-based dm is queueing the deferred I/Os in its request_queue.
2707 if (dm_request_based(md))
2708 dm_start_queue(md->queue);
2715 int dm_resume(struct mapped_device *md)
2718 struct dm_table *map = NULL;
2722 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2724 if (!dm_suspended_md(md))
2727 if (dm_suspended_internally_md(md)) {
2728 /* already internally suspended, wait for internal resume */
2729 mutex_unlock(&md->suspend_lock);
2730 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2736 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2737 if (!map || !dm_table_get_size(map))
2740 r = __dm_resume(md, map);
2744 clear_bit(DMF_SUSPENDED, &md->flags);
2746 mutex_unlock(&md->suspend_lock);
2752 * Internal suspend/resume works like userspace-driven suspend. It waits
2753 * until all bios finish and prevents issuing new bios to the target drivers.
2754 * It may be used only from the kernel.
2757 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2759 struct dm_table *map = NULL;
2761 lockdep_assert_held(&md->suspend_lock);
2763 if (md->internal_suspend_count++)
2764 return; /* nested internal suspend */
2766 if (dm_suspended_md(md)) {
2767 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2768 return; /* nest suspend */
2771 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2774 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2775 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2776 * would require changing .presuspend to return an error -- avoid this
2777 * until there is a need for more elaborate variants of internal suspend.
2779 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2780 DMF_SUSPENDED_INTERNALLY);
2782 dm_table_postsuspend_targets(map);
2785 static void __dm_internal_resume(struct mapped_device *md)
2787 BUG_ON(!md->internal_suspend_count);
2789 if (--md->internal_suspend_count)
2790 return; /* resume from nested internal suspend */
2792 if (dm_suspended_md(md))
2793 goto done; /* resume from nested suspend */
2796 * NOTE: existing callers don't need to call dm_table_resume_targets
2797 * (which may fail -- so best to avoid it for now by passing NULL map)
2799 (void) __dm_resume(md, NULL);
2802 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2803 smp_mb__after_atomic();
2804 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2807 void dm_internal_suspend_noflush(struct mapped_device *md)
2809 mutex_lock(&md->suspend_lock);
2810 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2811 mutex_unlock(&md->suspend_lock);
2813 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2815 void dm_internal_resume(struct mapped_device *md)
2817 mutex_lock(&md->suspend_lock);
2818 __dm_internal_resume(md);
2819 mutex_unlock(&md->suspend_lock);
2821 EXPORT_SYMBOL_GPL(dm_internal_resume);
2824 * Fast variants of internal suspend/resume hold md->suspend_lock,
2825 * which prevents interaction with userspace-driven suspend.
2828 void dm_internal_suspend_fast(struct mapped_device *md)
2830 mutex_lock(&md->suspend_lock);
2831 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2834 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2835 synchronize_srcu(&md->io_barrier);
2836 flush_workqueue(md->wq);
2837 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2839 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2841 void dm_internal_resume_fast(struct mapped_device *md)
2843 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2849 mutex_unlock(&md->suspend_lock);
2851 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2853 /*-----------------------------------------------------------------
2854 * Event notification.
2855 *---------------------------------------------------------------*/
2856 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2859 char udev_cookie[DM_COOKIE_LENGTH];
2860 char *envp[] = { udev_cookie, NULL };
2863 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2865 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2866 DM_COOKIE_ENV_VAR_NAME, cookie);
2867 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2872 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2874 return atomic_add_return(1, &md->uevent_seq);
2877 uint32_t dm_get_event_nr(struct mapped_device *md)
2879 return atomic_read(&md->event_nr);
2882 int dm_wait_event(struct mapped_device *md, int event_nr)
2884 return wait_event_interruptible(md->eventq,
2885 (event_nr != atomic_read(&md->event_nr)));
2888 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2890 unsigned long flags;
2892 spin_lock_irqsave(&md->uevent_lock, flags);
2893 list_add(elist, &md->uevent_list);
2894 spin_unlock_irqrestore(&md->uevent_lock, flags);
2898 * The gendisk is only valid as long as you have a reference
2901 struct gendisk *dm_disk(struct mapped_device *md)
2905 EXPORT_SYMBOL_GPL(dm_disk);
2907 struct kobject *dm_kobject(struct mapped_device *md)
2909 return &md->kobj_holder.kobj;
2912 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2914 struct mapped_device *md;
2916 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2918 spin_lock(&_minor_lock);
2919 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2925 spin_unlock(&_minor_lock);
2930 int dm_suspended_md(struct mapped_device *md)
2932 return test_bit(DMF_SUSPENDED, &md->flags);
2935 int dm_suspended_internally_md(struct mapped_device *md)
2937 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2940 int dm_test_deferred_remove_flag(struct mapped_device *md)
2942 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2945 int dm_suspended(struct dm_target *ti)
2947 return dm_suspended_md(dm_table_get_md(ti->table));
2949 EXPORT_SYMBOL_GPL(dm_suspended);
2951 int dm_noflush_suspending(struct dm_target *ti)
2953 return __noflush_suspending(dm_table_get_md(ti->table));
2955 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2957 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2958 unsigned integrity, unsigned per_io_data_size,
2959 unsigned min_pool_size)
2961 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2962 unsigned int pool_size = 0;
2963 unsigned int front_pad, io_front_pad;
2970 case DM_TYPE_BIO_BASED:
2971 case DM_TYPE_DAX_BIO_BASED:
2972 case DM_TYPE_NVME_BIO_BASED:
2973 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2974 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2975 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2976 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2979 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2982 case DM_TYPE_REQUEST_BASED:
2983 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2984 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2985 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2991 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2995 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3001 dm_free_md_mempools(pools);
3006 void dm_free_md_mempools(struct dm_md_mempools *pools)
3011 bioset_exit(&pools->bs);
3012 bioset_exit(&pools->io_bs);
3024 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3027 struct mapped_device *md = bdev->bd_disk->private_data;
3028 struct dm_table *table;
3029 struct dm_target *ti;
3030 int ret = -ENOTTY, srcu_idx;
3032 table = dm_get_live_table(md, &srcu_idx);
3033 if (!table || !dm_table_get_size(table))
3036 /* We only support devices that have a single target */
3037 if (dm_table_get_num_targets(table) != 1)
3039 ti = dm_table_get_target(table, 0);
3042 if (!ti->type->iterate_devices)
3045 ret = ti->type->iterate_devices(ti, fn, data);
3047 dm_put_live_table(md, srcu_idx);
3052 * For register / unregister we need to manually call out to every path.
3054 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3055 sector_t start, sector_t len, void *data)
3057 struct dm_pr *pr = data;
3058 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3060 if (!ops || !ops->pr_register)
3062 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3065 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3076 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3077 if (ret && new_key) {
3078 /* unregister all paths if we failed to register any path */
3079 pr.old_key = new_key;
3082 pr.fail_early = false;
3083 dm_call_pr(bdev, __dm_pr_register, &pr);
3089 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3092 struct mapped_device *md = bdev->bd_disk->private_data;
3093 const struct pr_ops *ops;
3096 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3100 ops = bdev->bd_disk->fops->pr_ops;
3101 if (ops && ops->pr_reserve)
3102 r = ops->pr_reserve(bdev, key, type, flags);
3106 dm_unprepare_ioctl(md, srcu_idx);
3110 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3112 struct mapped_device *md = bdev->bd_disk->private_data;
3113 const struct pr_ops *ops;
3116 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3120 ops = bdev->bd_disk->fops->pr_ops;
3121 if (ops && ops->pr_release)
3122 r = ops->pr_release(bdev, key, type);
3126 dm_unprepare_ioctl(md, srcu_idx);
3130 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3131 enum pr_type type, bool abort)
3133 struct mapped_device *md = bdev->bd_disk->private_data;
3134 const struct pr_ops *ops;
3137 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3141 ops = bdev->bd_disk->fops->pr_ops;
3142 if (ops && ops->pr_preempt)
3143 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3147 dm_unprepare_ioctl(md, srcu_idx);
3151 static int dm_pr_clear(struct block_device *bdev, u64 key)
3153 struct mapped_device *md = bdev->bd_disk->private_data;
3154 const struct pr_ops *ops;
3157 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3161 ops = bdev->bd_disk->fops->pr_ops;
3162 if (ops && ops->pr_clear)
3163 r = ops->pr_clear(bdev, key);
3167 dm_unprepare_ioctl(md, srcu_idx);
3171 static const struct pr_ops dm_pr_ops = {
3172 .pr_register = dm_pr_register,
3173 .pr_reserve = dm_pr_reserve,
3174 .pr_release = dm_pr_release,
3175 .pr_preempt = dm_pr_preempt,
3176 .pr_clear = dm_pr_clear,
3179 static const struct block_device_operations dm_blk_dops = {
3180 .open = dm_blk_open,
3181 .release = dm_blk_close,
3182 .ioctl = dm_blk_ioctl,
3183 .getgeo = dm_blk_getgeo,
3184 .report_zones = dm_blk_report_zones,
3185 .pr_ops = &dm_pr_ops,
3186 .owner = THIS_MODULE
3189 static const struct dax_operations dm_dax_ops = {
3190 .direct_access = dm_dax_direct_access,
3191 .copy_from_iter = dm_dax_copy_from_iter,
3192 .copy_to_iter = dm_dax_copy_to_iter,
3198 module_init(dm_init);
3199 module_exit(dm_exit);
3201 module_param(major, uint, 0);
3202 MODULE_PARM_DESC(major, "The major number of the device mapper");
3204 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3205 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3207 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3208 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3210 MODULE_DESCRIPTION(DM_NAME " driver");
3211 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3212 MODULE_LICENSE("GPL");