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>
28 #include <linux/part_stat.h>
30 #define DM_MSG_PREFIX "core"
33 * Cookies are numeric values sent with CHANGE and REMOVE
34 * uevents while resuming, removing or renaming the device.
36 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
37 #define DM_COOKIE_LENGTH 24
39 static const char *_name = DM_NAME;
41 static unsigned int major = 0;
42 static unsigned int _major = 0;
44 static DEFINE_IDR(_minor_idr);
46 static DEFINE_SPINLOCK(_minor_lock);
48 static void do_deferred_remove(struct work_struct *w);
50 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
52 static struct workqueue_struct *deferred_remove_workqueue;
54 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
55 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
57 void dm_issue_global_event(void)
59 atomic_inc(&dm_global_event_nr);
60 wake_up(&dm_global_eventq);
64 * One of these is allocated (on-stack) per original bio.
71 unsigned sector_count;
75 * One of these is allocated per clone bio.
77 #define DM_TIO_MAGIC 7282014
82 unsigned target_bio_nr;
89 * One of these is allocated per original bio.
90 * It contains the first clone used for that original.
92 #define DM_IO_MAGIC 5191977
95 struct mapped_device *md;
99 unsigned long start_time;
100 spinlock_t endio_lock;
101 struct dm_stats_aux stats_aux;
102 /* last member of dm_target_io is 'struct bio' */
103 struct dm_target_io tio;
106 void *dm_per_bio_data(struct bio *bio, size_t data_size)
108 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
109 if (!tio->inside_dm_io)
110 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
111 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
113 EXPORT_SYMBOL_GPL(dm_per_bio_data);
115 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
117 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
118 if (io->magic == DM_IO_MAGIC)
119 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
120 BUG_ON(io->magic != DM_TIO_MAGIC);
121 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
123 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
125 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
127 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
129 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
131 #define MINOR_ALLOCED ((void *)-1)
134 * Bits for the md->flags field.
136 #define DMF_BLOCK_IO_FOR_SUSPEND 0
137 #define DMF_SUSPENDED 1
139 #define DMF_FREEING 3
140 #define DMF_DELETING 4
141 #define DMF_NOFLUSH_SUSPENDING 5
142 #define DMF_DEFERRED_REMOVE 6
143 #define DMF_SUSPENDED_INTERNALLY 7
145 #define DM_NUMA_NODE NUMA_NO_NODE
146 static int dm_numa_node = DM_NUMA_NODE;
149 * For mempools pre-allocation at the table loading time.
151 struct dm_md_mempools {
153 struct bio_set io_bs;
156 struct table_device {
157 struct list_head list;
159 struct dm_dev dm_dev;
163 * Bio-based DM's mempools' reserved IOs set by the user.
165 #define RESERVED_BIO_BASED_IOS 16
166 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
168 static int __dm_get_module_param_int(int *module_param, int min, int max)
170 int param = READ_ONCE(*module_param);
171 int modified_param = 0;
172 bool modified = true;
175 modified_param = min;
176 else if (param > max)
177 modified_param = max;
182 (void)cmpxchg(module_param, param, modified_param);
183 param = modified_param;
189 unsigned __dm_get_module_param(unsigned *module_param,
190 unsigned def, unsigned max)
192 unsigned param = READ_ONCE(*module_param);
193 unsigned modified_param = 0;
196 modified_param = def;
197 else if (param > max)
198 modified_param = max;
200 if (modified_param) {
201 (void)cmpxchg(module_param, param, modified_param);
202 param = modified_param;
208 unsigned dm_get_reserved_bio_based_ios(void)
210 return __dm_get_module_param(&reserved_bio_based_ios,
211 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
213 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
215 static unsigned dm_get_numa_node(void)
217 return __dm_get_module_param_int(&dm_numa_node,
218 DM_NUMA_NODE, num_online_nodes() - 1);
221 static int __init local_init(void)
225 r = dm_uevent_init();
229 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
230 if (!deferred_remove_workqueue) {
232 goto out_uevent_exit;
236 r = register_blkdev(_major, _name);
238 goto out_free_workqueue;
246 destroy_workqueue(deferred_remove_workqueue);
253 static void local_exit(void)
255 flush_scheduled_work();
256 destroy_workqueue(deferred_remove_workqueue);
258 unregister_blkdev(_major, _name);
263 DMINFO("cleaned up");
266 static int (*_inits[])(void) __initdata = {
277 static void (*_exits[])(void) = {
288 static int __init dm_init(void)
290 const int count = ARRAY_SIZE(_inits);
294 for (i = 0; i < count; i++) {
309 static void __exit dm_exit(void)
311 int i = ARRAY_SIZE(_exits);
317 * Should be empty by this point.
319 idr_destroy(&_minor_idr);
323 * Block device functions
325 int dm_deleting_md(struct mapped_device *md)
327 return test_bit(DMF_DELETING, &md->flags);
330 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
332 struct mapped_device *md;
334 spin_lock(&_minor_lock);
336 md = bdev->bd_disk->private_data;
340 if (test_bit(DMF_FREEING, &md->flags) ||
341 dm_deleting_md(md)) {
347 atomic_inc(&md->open_count);
349 spin_unlock(&_minor_lock);
351 return md ? 0 : -ENXIO;
354 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
356 struct mapped_device *md;
358 spin_lock(&_minor_lock);
360 md = disk->private_data;
364 if (atomic_dec_and_test(&md->open_count) &&
365 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
366 queue_work(deferred_remove_workqueue, &deferred_remove_work);
370 spin_unlock(&_minor_lock);
373 int dm_open_count(struct mapped_device *md)
375 return atomic_read(&md->open_count);
379 * Guarantees nothing is using the device before it's deleted.
381 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
385 spin_lock(&_minor_lock);
387 if (dm_open_count(md)) {
390 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
391 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
394 set_bit(DMF_DELETING, &md->flags);
396 spin_unlock(&_minor_lock);
401 int dm_cancel_deferred_remove(struct mapped_device *md)
405 spin_lock(&_minor_lock);
407 if (test_bit(DMF_DELETING, &md->flags))
410 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
412 spin_unlock(&_minor_lock);
417 static void do_deferred_remove(struct work_struct *w)
419 dm_deferred_remove();
422 sector_t dm_get_size(struct mapped_device *md)
424 return get_capacity(md->disk);
427 struct request_queue *dm_get_md_queue(struct mapped_device *md)
432 struct dm_stats *dm_get_stats(struct mapped_device *md)
437 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
439 struct mapped_device *md = bdev->bd_disk->private_data;
441 return dm_get_geometry(md, geo);
444 #ifdef CONFIG_BLK_DEV_ZONED
445 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
447 struct dm_report_zones_args *args = data;
448 sector_t sector_diff = args->tgt->begin - args->start;
451 * Ignore zones beyond the target range.
453 if (zone->start >= args->start + args->tgt->len)
457 * Remap the start sector and write pointer position of the zone
458 * to match its position in the target range.
460 zone->start += sector_diff;
461 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
462 if (zone->cond == BLK_ZONE_COND_FULL)
463 zone->wp = zone->start + zone->len;
464 else if (zone->cond == BLK_ZONE_COND_EMPTY)
465 zone->wp = zone->start;
467 zone->wp += sector_diff;
470 args->next_sector = zone->start + zone->len;
471 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
473 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
475 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
476 unsigned int nr_zones, report_zones_cb cb, void *data)
478 struct mapped_device *md = disk->private_data;
479 struct dm_table *map;
481 struct dm_report_zones_args args = {
482 .next_sector = sector,
487 if (dm_suspended_md(md))
490 map = dm_get_live_table(md, &srcu_idx);
495 struct dm_target *tgt;
497 tgt = dm_table_find_target(map, args.next_sector);
498 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
504 ret = tgt->type->report_zones(tgt, &args, nr_zones);
507 } while (args.zone_idx < nr_zones &&
508 args.next_sector < get_capacity(disk));
512 dm_put_live_table(md, srcu_idx);
516 #define dm_blk_report_zones NULL
517 #endif /* CONFIG_BLK_DEV_ZONED */
519 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
520 struct block_device **bdev)
521 __acquires(md->io_barrier)
523 struct dm_target *tgt;
524 struct dm_table *map;
529 map = dm_get_live_table(md, srcu_idx);
530 if (!map || !dm_table_get_size(map))
533 /* We only support devices that have a single target */
534 if (dm_table_get_num_targets(map) != 1)
537 tgt = dm_table_get_target(map, 0);
538 if (!tgt->type->prepare_ioctl)
541 if (dm_suspended_md(md))
544 r = tgt->type->prepare_ioctl(tgt, bdev);
545 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
546 dm_put_live_table(md, *srcu_idx);
554 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
555 __releases(md->io_barrier)
557 dm_put_live_table(md, srcu_idx);
560 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
561 unsigned int cmd, unsigned long arg)
563 struct mapped_device *md = bdev->bd_disk->private_data;
566 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
572 * Target determined this ioctl is being issued against a
573 * subset of the parent bdev; require extra privileges.
575 if (!capable(CAP_SYS_RAWIO)) {
577 "%s: sending ioctl %x to DM device without required privilege.",
584 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
586 dm_unprepare_ioctl(md, srcu_idx);
590 static void start_io_acct(struct dm_io *io);
592 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
595 struct dm_target_io *tio;
598 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
602 tio = container_of(clone, struct dm_target_io, clone);
603 tio->inside_dm_io = true;
606 io = container_of(tio, struct dm_io, tio);
607 io->magic = DM_IO_MAGIC;
609 atomic_set(&io->io_count, 1);
612 spin_lock_init(&io->endio_lock);
619 static void free_io(struct mapped_device *md, struct dm_io *io)
621 bio_put(&io->tio.clone);
624 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
625 unsigned target_bio_nr, gfp_t gfp_mask)
627 struct dm_target_io *tio;
629 if (!ci->io->tio.io) {
630 /* the dm_target_io embedded in ci->io is available */
633 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
637 tio = container_of(clone, struct dm_target_io, clone);
638 tio->inside_dm_io = false;
641 tio->magic = DM_TIO_MAGIC;
644 tio->target_bio_nr = target_bio_nr;
649 static void free_tio(struct dm_target_io *tio)
651 if (tio->inside_dm_io)
653 bio_put(&tio->clone);
656 static bool md_in_flight_bios(struct mapped_device *md)
659 struct hd_struct *part = &dm_disk(md)->part0;
662 for_each_possible_cpu(cpu) {
663 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
664 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
670 static bool md_in_flight(struct mapped_device *md)
672 if (queue_is_mq(md->queue))
673 return blk_mq_queue_inflight(md->queue);
675 return md_in_flight_bios(md);
678 static void start_io_acct(struct dm_io *io)
680 struct mapped_device *md = io->md;
681 struct bio *bio = io->orig_bio;
683 io->start_time = jiffies;
685 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
686 &dm_disk(md)->part0);
688 if (unlikely(dm_stats_used(&md->stats)))
689 dm_stats_account_io(&md->stats, bio_data_dir(bio),
690 bio->bi_iter.bi_sector, bio_sectors(bio),
691 false, 0, &io->stats_aux);
694 static void end_io_acct(struct dm_io *io)
696 struct mapped_device *md = io->md;
697 struct bio *bio = io->orig_bio;
698 unsigned long duration = jiffies - io->start_time;
700 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
703 if (unlikely(dm_stats_used(&md->stats)))
704 dm_stats_account_io(&md->stats, bio_data_dir(bio),
705 bio->bi_iter.bi_sector, bio_sectors(bio),
706 true, duration, &io->stats_aux);
708 /* nudge anyone waiting on suspend queue */
709 if (unlikely(wq_has_sleeper(&md->wait)))
714 * Add the bio to the list of deferred io.
716 static void queue_io(struct mapped_device *md, struct bio *bio)
720 spin_lock_irqsave(&md->deferred_lock, flags);
721 bio_list_add(&md->deferred, bio);
722 spin_unlock_irqrestore(&md->deferred_lock, flags);
723 queue_work(md->wq, &md->work);
727 * Everyone (including functions in this file), should use this
728 * function to access the md->map field, and make sure they call
729 * dm_put_live_table() when finished.
731 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
733 *srcu_idx = srcu_read_lock(&md->io_barrier);
735 return srcu_dereference(md->map, &md->io_barrier);
738 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
740 srcu_read_unlock(&md->io_barrier, srcu_idx);
743 void dm_sync_table(struct mapped_device *md)
745 synchronize_srcu(&md->io_barrier);
746 synchronize_rcu_expedited();
750 * A fast alternative to dm_get_live_table/dm_put_live_table.
751 * The caller must not block between these two functions.
753 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
756 return rcu_dereference(md->map);
759 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
764 static char *_dm_claim_ptr = "I belong to device-mapper";
767 * Open a table device so we can use it as a map destination.
769 static int open_table_device(struct table_device *td, dev_t dev,
770 struct mapped_device *md)
772 struct block_device *bdev;
776 BUG_ON(td->dm_dev.bdev);
778 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
780 return PTR_ERR(bdev);
782 r = bd_link_disk_holder(bdev, dm_disk(md));
784 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
788 td->dm_dev.bdev = bdev;
789 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
794 * Close a table device that we've been using.
796 static void close_table_device(struct table_device *td, struct mapped_device *md)
798 if (!td->dm_dev.bdev)
801 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
802 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
803 put_dax(td->dm_dev.dax_dev);
804 td->dm_dev.bdev = NULL;
805 td->dm_dev.dax_dev = NULL;
808 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
811 struct table_device *td;
813 list_for_each_entry(td, l, list)
814 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
820 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
821 struct dm_dev **result)
824 struct table_device *td;
826 mutex_lock(&md->table_devices_lock);
827 td = find_table_device(&md->table_devices, dev, mode);
829 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
831 mutex_unlock(&md->table_devices_lock);
835 td->dm_dev.mode = mode;
836 td->dm_dev.bdev = NULL;
838 if ((r = open_table_device(td, dev, md))) {
839 mutex_unlock(&md->table_devices_lock);
844 format_dev_t(td->dm_dev.name, dev);
846 refcount_set(&td->count, 1);
847 list_add(&td->list, &md->table_devices);
849 refcount_inc(&td->count);
851 mutex_unlock(&md->table_devices_lock);
853 *result = &td->dm_dev;
856 EXPORT_SYMBOL_GPL(dm_get_table_device);
858 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
860 struct table_device *td = container_of(d, struct table_device, dm_dev);
862 mutex_lock(&md->table_devices_lock);
863 if (refcount_dec_and_test(&td->count)) {
864 close_table_device(td, md);
868 mutex_unlock(&md->table_devices_lock);
870 EXPORT_SYMBOL(dm_put_table_device);
872 static void free_table_devices(struct list_head *devices)
874 struct list_head *tmp, *next;
876 list_for_each_safe(tmp, next, devices) {
877 struct table_device *td = list_entry(tmp, struct table_device, list);
879 DMWARN("dm_destroy: %s still exists with %d references",
880 td->dm_dev.name, refcount_read(&td->count));
886 * Get the geometry associated with a dm device
888 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
896 * Set the geometry of a device.
898 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
900 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
902 if (geo->start > sz) {
903 DMWARN("Start sector is beyond the geometry limits.");
912 static int __noflush_suspending(struct mapped_device *md)
914 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
918 * Decrements the number of outstanding ios that a bio has been
919 * cloned into, completing the original io if necc.
921 static void dec_pending(struct dm_io *io, blk_status_t error)
924 blk_status_t io_error;
926 struct mapped_device *md = io->md;
928 /* Push-back supersedes any I/O errors */
929 if (unlikely(error)) {
930 spin_lock_irqsave(&io->endio_lock, flags);
931 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
933 spin_unlock_irqrestore(&io->endio_lock, flags);
936 if (atomic_dec_and_test(&io->io_count)) {
937 if (io->status == BLK_STS_DM_REQUEUE) {
939 * Target requested pushing back the I/O.
941 spin_lock_irqsave(&md->deferred_lock, flags);
942 if (__noflush_suspending(md))
943 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
944 bio_list_add_head(&md->deferred, io->orig_bio);
946 /* noflush suspend was interrupted. */
947 io->status = BLK_STS_IOERR;
948 spin_unlock_irqrestore(&md->deferred_lock, flags);
951 io_error = io->status;
956 if (io_error == BLK_STS_DM_REQUEUE)
959 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
961 * Preflush done for flush with data, reissue
962 * without REQ_PREFLUSH.
964 bio->bi_opf &= ~REQ_PREFLUSH;
967 /* done with normal IO or empty flush */
969 bio->bi_status = io_error;
975 void disable_discard(struct mapped_device *md)
977 struct queue_limits *limits = dm_get_queue_limits(md);
979 /* device doesn't really support DISCARD, disable it */
980 limits->max_discard_sectors = 0;
981 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
984 void disable_write_same(struct mapped_device *md)
986 struct queue_limits *limits = dm_get_queue_limits(md);
988 /* device doesn't really support WRITE SAME, disable it */
989 limits->max_write_same_sectors = 0;
992 void disable_write_zeroes(struct mapped_device *md)
994 struct queue_limits *limits = dm_get_queue_limits(md);
996 /* device doesn't really support WRITE ZEROES, disable it */
997 limits->max_write_zeroes_sectors = 0;
1000 static void clone_endio(struct bio *bio)
1002 blk_status_t error = bio->bi_status;
1003 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1004 struct dm_io *io = tio->io;
1005 struct mapped_device *md = tio->io->md;
1006 dm_endio_fn endio = tio->ti->type->end_io;
1008 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
1009 if (bio_op(bio) == REQ_OP_DISCARD &&
1010 !bio->bi_disk->queue->limits.max_discard_sectors)
1011 disable_discard(md);
1012 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1013 !bio->bi_disk->queue->limits.max_write_same_sectors)
1014 disable_write_same(md);
1015 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1016 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1017 disable_write_zeroes(md);
1021 int r = endio(tio->ti, bio, &error);
1023 case DM_ENDIO_REQUEUE:
1024 error = BLK_STS_DM_REQUEUE;
1028 case DM_ENDIO_INCOMPLETE:
1029 /* The target will handle the io */
1032 DMWARN("unimplemented target endio return value: %d", r);
1038 dec_pending(io, error);
1042 * Return maximum size of I/O possible at the supplied sector up to the current
1045 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1047 sector_t target_offset = dm_target_offset(ti, sector);
1049 return ti->len - target_offset;
1052 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1054 sector_t len = max_io_len_target_boundary(sector, ti);
1055 sector_t offset, max_len;
1058 * Does the target need to split even further?
1060 if (ti->max_io_len) {
1061 offset = dm_target_offset(ti, sector);
1062 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1063 max_len = sector_div(offset, ti->max_io_len);
1065 max_len = offset & (ti->max_io_len - 1);
1066 max_len = ti->max_io_len - max_len;
1075 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1077 if (len > UINT_MAX) {
1078 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1079 (unsigned long long)len, UINT_MAX);
1080 ti->error = "Maximum size of target IO is too large";
1084 ti->max_io_len = (uint32_t) len;
1088 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1090 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1091 sector_t sector, int *srcu_idx)
1092 __acquires(md->io_barrier)
1094 struct dm_table *map;
1095 struct dm_target *ti;
1097 map = dm_get_live_table(md, srcu_idx);
1101 ti = dm_table_find_target(map, sector);
1108 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1109 long nr_pages, void **kaddr, pfn_t *pfn)
1111 struct mapped_device *md = dax_get_private(dax_dev);
1112 sector_t sector = pgoff * PAGE_SECTORS;
1113 struct dm_target *ti;
1114 long len, ret = -EIO;
1117 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1121 if (!ti->type->direct_access)
1123 len = max_io_len(sector, ti) / PAGE_SECTORS;
1126 nr_pages = min(len, nr_pages);
1127 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1130 dm_put_live_table(md, srcu_idx);
1135 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1136 int blocksize, sector_t start, sector_t len)
1138 struct mapped_device *md = dax_get_private(dax_dev);
1139 struct dm_table *map;
1143 map = dm_get_live_table(md, &srcu_idx);
1147 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1149 dm_put_live_table(md, srcu_idx);
1154 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1155 void *addr, size_t bytes, struct iov_iter *i)
1157 struct mapped_device *md = dax_get_private(dax_dev);
1158 sector_t sector = pgoff * PAGE_SECTORS;
1159 struct dm_target *ti;
1163 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1167 if (!ti->type->dax_copy_from_iter) {
1168 ret = copy_from_iter(addr, bytes, i);
1171 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1173 dm_put_live_table(md, srcu_idx);
1178 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1179 void *addr, size_t bytes, struct iov_iter *i)
1181 struct mapped_device *md = dax_get_private(dax_dev);
1182 sector_t sector = pgoff * PAGE_SECTORS;
1183 struct dm_target *ti;
1187 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1191 if (!ti->type->dax_copy_to_iter) {
1192 ret = copy_to_iter(addr, bytes, i);
1195 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1197 dm_put_live_table(md, srcu_idx);
1202 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1205 struct mapped_device *md = dax_get_private(dax_dev);
1206 sector_t sector = pgoff * PAGE_SECTORS;
1207 struct dm_target *ti;
1211 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1215 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1217 * ->zero_page_range() is mandatory dax operation. If we are
1218 * here, something is wrong.
1220 dm_put_live_table(md, srcu_idx);
1223 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1226 dm_put_live_table(md, srcu_idx);
1232 * A target may call dm_accept_partial_bio only from the map routine. It is
1233 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1234 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1236 * dm_accept_partial_bio informs the dm that the target only wants to process
1237 * additional n_sectors sectors of the bio and the rest of the data should be
1238 * sent in a next bio.
1240 * A diagram that explains the arithmetics:
1241 * +--------------------+---------------+-------+
1243 * +--------------------+---------------+-------+
1245 * <-------------- *tio->len_ptr --------------->
1246 * <------- bi_size ------->
1249 * Region 1 was already iterated over with bio_advance or similar function.
1250 * (it may be empty if the target doesn't use bio_advance)
1251 * Region 2 is the remaining bio size that the target wants to process.
1252 * (it may be empty if region 1 is non-empty, although there is no reason
1254 * The target requires that region 3 is to be sent in the next bio.
1256 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1257 * the partially processed part (the sum of regions 1+2) must be the same for all
1258 * copies of the bio.
1260 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1262 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1263 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1264 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1265 BUG_ON(bi_size > *tio->len_ptr);
1266 BUG_ON(n_sectors > bi_size);
1267 *tio->len_ptr -= bi_size - n_sectors;
1268 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1270 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1272 static blk_qc_t __map_bio(struct dm_target_io *tio)
1276 struct bio *clone = &tio->clone;
1277 struct dm_io *io = tio->io;
1278 struct mapped_device *md = io->md;
1279 struct dm_target *ti = tio->ti;
1280 blk_qc_t ret = BLK_QC_T_NONE;
1282 clone->bi_end_io = clone_endio;
1285 * Map the clone. If r == 0 we don't need to do
1286 * anything, the target has assumed ownership of
1289 atomic_inc(&io->io_count);
1290 sector = clone->bi_iter.bi_sector;
1292 r = ti->type->map(ti, clone);
1294 case DM_MAPIO_SUBMITTED:
1296 case DM_MAPIO_REMAPPED:
1297 /* the bio has been remapped so dispatch it */
1298 trace_block_bio_remap(clone->bi_disk->queue, clone,
1299 bio_dev(io->orig_bio), sector);
1300 if (md->type == DM_TYPE_NVME_BIO_BASED)
1301 ret = direct_make_request(clone);
1303 ret = generic_make_request(clone);
1307 dec_pending(io, BLK_STS_IOERR);
1309 case DM_MAPIO_REQUEUE:
1311 dec_pending(io, BLK_STS_DM_REQUEUE);
1314 DMWARN("unimplemented target map return value: %d", r);
1321 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1323 bio->bi_iter.bi_sector = sector;
1324 bio->bi_iter.bi_size = to_bytes(len);
1328 * Creates a bio that consists of range of complete bvecs.
1330 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1331 sector_t sector, unsigned len)
1333 struct bio *clone = &tio->clone;
1335 __bio_clone_fast(clone, bio);
1337 if (bio_integrity(bio)) {
1340 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1341 !dm_target_passes_integrity(tio->ti->type))) {
1342 DMWARN("%s: the target %s doesn't support integrity data.",
1343 dm_device_name(tio->io->md),
1344 tio->ti->type->name);
1348 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1353 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1354 clone->bi_iter.bi_size = to_bytes(len);
1356 if (bio_integrity(bio))
1357 bio_integrity_trim(clone);
1362 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1363 struct dm_target *ti, unsigned num_bios)
1365 struct dm_target_io *tio;
1371 if (num_bios == 1) {
1372 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1373 bio_list_add(blist, &tio->clone);
1377 for (try = 0; try < 2; try++) {
1382 mutex_lock(&ci->io->md->table_devices_lock);
1383 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1384 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1388 bio_list_add(blist, &tio->clone);
1391 mutex_unlock(&ci->io->md->table_devices_lock);
1392 if (bio_nr == num_bios)
1395 while ((bio = bio_list_pop(blist))) {
1396 tio = container_of(bio, struct dm_target_io, clone);
1402 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1403 struct dm_target_io *tio, unsigned *len)
1405 struct bio *clone = &tio->clone;
1409 __bio_clone_fast(clone, ci->bio);
1411 bio_setup_sector(clone, ci->sector, *len);
1413 return __map_bio(tio);
1416 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1417 unsigned num_bios, unsigned *len)
1419 struct bio_list blist = BIO_EMPTY_LIST;
1421 struct dm_target_io *tio;
1423 alloc_multiple_bios(&blist, ci, ti, num_bios);
1425 while ((bio = bio_list_pop(&blist))) {
1426 tio = container_of(bio, struct dm_target_io, clone);
1427 (void) __clone_and_map_simple_bio(ci, tio, len);
1431 static int __send_empty_flush(struct clone_info *ci)
1433 unsigned target_nr = 0;
1434 struct dm_target *ti;
1437 * Empty flush uses a statically initialized bio, as the base for
1438 * cloning. However, blkg association requires that a bdev is
1439 * associated with a gendisk, which doesn't happen until the bdev is
1440 * opened. So, blkg association is done at issue time of the flush
1441 * rather than when the device is created in alloc_dev().
1443 bio_set_dev(ci->bio, ci->io->md->bdev);
1445 BUG_ON(bio_has_data(ci->bio));
1446 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1447 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1449 bio_disassociate_blkg(ci->bio);
1454 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1455 sector_t sector, unsigned *len)
1457 struct bio *bio = ci->bio;
1458 struct dm_target_io *tio;
1461 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1463 r = clone_bio(tio, bio, sector, *len);
1468 (void) __map_bio(tio);
1473 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1475 static unsigned get_num_discard_bios(struct dm_target *ti)
1477 return ti->num_discard_bios;
1480 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1482 return ti->num_secure_erase_bios;
1485 static unsigned get_num_write_same_bios(struct dm_target *ti)
1487 return ti->num_write_same_bios;
1490 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1492 return ti->num_write_zeroes_bios;
1495 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1501 * Even though the device advertised support for this type of
1502 * request, that does not mean every target supports it, and
1503 * reconfiguration might also have changed that since the
1504 * check was performed.
1509 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1511 __send_duplicate_bios(ci, ti, num_bios, &len);
1514 ci->sector_count -= len;
1519 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1521 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1524 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1526 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1529 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1531 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1534 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1536 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1539 static bool is_abnormal_io(struct bio *bio)
1543 switch (bio_op(bio)) {
1544 case REQ_OP_DISCARD:
1545 case REQ_OP_SECURE_ERASE:
1546 case REQ_OP_WRITE_SAME:
1547 case REQ_OP_WRITE_ZEROES:
1555 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1558 struct bio *bio = ci->bio;
1560 if (bio_op(bio) == REQ_OP_DISCARD)
1561 *result = __send_discard(ci, ti);
1562 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1563 *result = __send_secure_erase(ci, ti);
1564 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1565 *result = __send_write_same(ci, ti);
1566 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1567 *result = __send_write_zeroes(ci, ti);
1575 * Select the correct strategy for processing a non-flush bio.
1577 static int __split_and_process_non_flush(struct clone_info *ci)
1579 struct dm_target *ti;
1583 ti = dm_table_find_target(ci->map, ci->sector);
1587 if (__process_abnormal_io(ci, ti, &r))
1590 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1592 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1597 ci->sector_count -= len;
1602 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1603 struct dm_table *map, struct bio *bio)
1606 ci->io = alloc_io(md, bio);
1607 ci->sector = bio->bi_iter.bi_sector;
1610 #define __dm_part_stat_sub(part, field, subnd) \
1611 (part_stat_get(part, field) -= (subnd))
1614 * Entry point to split a bio into clones and submit them to the targets.
1616 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1617 struct dm_table *map, struct bio *bio)
1619 struct clone_info ci;
1620 blk_qc_t ret = BLK_QC_T_NONE;
1623 init_clone_info(&ci, md, map, bio);
1625 if (bio->bi_opf & REQ_PREFLUSH) {
1626 struct bio flush_bio;
1629 * Use an on-stack bio for this, it's safe since we don't
1630 * need to reference it after submit. It's just used as
1631 * the basis for the clone(s).
1633 bio_init(&flush_bio, NULL, 0);
1634 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1635 ci.bio = &flush_bio;
1636 ci.sector_count = 0;
1637 error = __send_empty_flush(&ci);
1638 /* dec_pending submits any data associated with flush */
1639 } else if (op_is_zone_mgmt(bio_op(bio))) {
1641 ci.sector_count = 0;
1642 error = __split_and_process_non_flush(&ci);
1645 ci.sector_count = bio_sectors(bio);
1646 while (ci.sector_count && !error) {
1647 error = __split_and_process_non_flush(&ci);
1648 if (current->bio_list && ci.sector_count && !error) {
1650 * Remainder must be passed to generic_make_request()
1651 * so that it gets handled *after* bios already submitted
1652 * have been completely processed.
1653 * We take a clone of the original to store in
1654 * ci.io->orig_bio to be used by end_io_acct() and
1655 * for dec_pending to use for completion handling.
1657 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1658 GFP_NOIO, &md->queue->bio_split);
1659 ci.io->orig_bio = b;
1662 * Adjust IO stats for each split, otherwise upon queue
1663 * reentry there will be redundant IO accounting.
1664 * NOTE: this is a stop-gap fix, a proper fix involves
1665 * significant refactoring of DM core's bio splitting
1666 * (by eliminating DM's splitting and just using bio_split)
1669 __dm_part_stat_sub(&dm_disk(md)->part0,
1670 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1674 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1675 ret = generic_make_request(bio);
1681 /* drop the extra reference count */
1682 dec_pending(ci.io, errno_to_blk_status(error));
1687 * Optimized variant of __split_and_process_bio that leverages the
1688 * fact that targets that use it do _not_ have a need to split bios.
1690 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1691 struct bio *bio, struct dm_target *ti)
1693 struct clone_info ci;
1694 blk_qc_t ret = BLK_QC_T_NONE;
1697 init_clone_info(&ci, md, map, bio);
1699 if (bio->bi_opf & REQ_PREFLUSH) {
1700 struct bio flush_bio;
1703 * Use an on-stack bio for this, it's safe since we don't
1704 * need to reference it after submit. It's just used as
1705 * the basis for the clone(s).
1707 bio_init(&flush_bio, NULL, 0);
1708 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1709 ci.bio = &flush_bio;
1710 ci.sector_count = 0;
1711 error = __send_empty_flush(&ci);
1712 /* dec_pending submits any data associated with flush */
1714 struct dm_target_io *tio;
1717 ci.sector_count = bio_sectors(bio);
1718 if (__process_abnormal_io(&ci, ti, &error))
1721 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1722 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1725 /* drop the extra reference count */
1726 dec_pending(ci.io, errno_to_blk_status(error));
1730 static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio)
1732 unsigned len, sector_count;
1734 sector_count = bio_sectors(*bio);
1735 len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count);
1737 if (sector_count > len) {
1738 struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split);
1740 bio_chain(split, *bio);
1741 trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector);
1742 generic_make_request(*bio);
1747 static blk_qc_t dm_process_bio(struct mapped_device *md,
1748 struct dm_table *map, struct bio *bio)
1750 blk_qc_t ret = BLK_QC_T_NONE;
1751 struct dm_target *ti = md->immutable_target;
1753 if (unlikely(!map)) {
1759 ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1760 if (unlikely(!ti)) {
1767 * If in ->make_request_fn we need to use blk_queue_split(), otherwise
1768 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1771 if (current->bio_list) {
1772 if (is_abnormal_io(bio))
1773 blk_queue_split(md->queue, &bio);
1775 dm_queue_split(md, ti, &bio);
1778 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1779 return __process_bio(md, map, bio, ti);
1781 return __split_and_process_bio(md, map, bio);
1784 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1786 struct mapped_device *md = q->queuedata;
1787 blk_qc_t ret = BLK_QC_T_NONE;
1789 struct dm_table *map;
1791 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED)
1792 return blk_mq_make_request(q, bio);
1794 map = dm_get_live_table(md, &srcu_idx);
1796 /* if we're suspended, we have to queue this io for later */
1797 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1798 dm_put_live_table(md, srcu_idx);
1800 if (!(bio->bi_opf & REQ_RAHEAD))
1807 ret = dm_process_bio(md, map, bio);
1809 dm_put_live_table(md, srcu_idx);
1813 static int dm_any_congested(void *congested_data, int bdi_bits)
1816 struct mapped_device *md = congested_data;
1817 struct dm_table *map;
1819 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1820 if (dm_request_based(md)) {
1822 * With request-based DM we only need to check the
1823 * top-level queue for congestion.
1825 struct backing_dev_info *bdi = md->queue->backing_dev_info;
1826 r = bdi->wb.congested->state & bdi_bits;
1828 map = dm_get_live_table_fast(md);
1830 r = dm_table_any_congested(map, bdi_bits);
1831 dm_put_live_table_fast(md);
1838 /*-----------------------------------------------------------------
1839 * An IDR is used to keep track of allocated minor numbers.
1840 *---------------------------------------------------------------*/
1841 static void free_minor(int minor)
1843 spin_lock(&_minor_lock);
1844 idr_remove(&_minor_idr, minor);
1845 spin_unlock(&_minor_lock);
1849 * See if the device with a specific minor # is free.
1851 static int specific_minor(int minor)
1855 if (minor >= (1 << MINORBITS))
1858 idr_preload(GFP_KERNEL);
1859 spin_lock(&_minor_lock);
1861 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1863 spin_unlock(&_minor_lock);
1866 return r == -ENOSPC ? -EBUSY : r;
1870 static int next_free_minor(int *minor)
1874 idr_preload(GFP_KERNEL);
1875 spin_lock(&_minor_lock);
1877 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1879 spin_unlock(&_minor_lock);
1887 static const struct block_device_operations dm_blk_dops;
1888 static const struct dax_operations dm_dax_ops;
1890 static void dm_wq_work(struct work_struct *work);
1892 static void cleanup_mapped_device(struct mapped_device *md)
1895 destroy_workqueue(md->wq);
1896 bioset_exit(&md->bs);
1897 bioset_exit(&md->io_bs);
1900 kill_dax(md->dax_dev);
1901 put_dax(md->dax_dev);
1906 spin_lock(&_minor_lock);
1907 md->disk->private_data = NULL;
1908 spin_unlock(&_minor_lock);
1909 del_gendisk(md->disk);
1914 blk_cleanup_queue(md->queue);
1916 cleanup_srcu_struct(&md->io_barrier);
1923 mutex_destroy(&md->suspend_lock);
1924 mutex_destroy(&md->type_lock);
1925 mutex_destroy(&md->table_devices_lock);
1927 dm_mq_cleanup_mapped_device(md);
1931 * Allocate and initialise a blank device with a given minor.
1933 static struct mapped_device *alloc_dev(int minor)
1935 int r, numa_node_id = dm_get_numa_node();
1936 struct mapped_device *md;
1939 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1941 DMWARN("unable to allocate device, out of memory.");
1945 if (!try_module_get(THIS_MODULE))
1946 goto bad_module_get;
1948 /* get a minor number for the dev */
1949 if (minor == DM_ANY_MINOR)
1950 r = next_free_minor(&minor);
1952 r = specific_minor(minor);
1956 r = init_srcu_struct(&md->io_barrier);
1958 goto bad_io_barrier;
1960 md->numa_node_id = numa_node_id;
1961 md->init_tio_pdu = false;
1962 md->type = DM_TYPE_NONE;
1963 mutex_init(&md->suspend_lock);
1964 mutex_init(&md->type_lock);
1965 mutex_init(&md->table_devices_lock);
1966 spin_lock_init(&md->deferred_lock);
1967 atomic_set(&md->holders, 1);
1968 atomic_set(&md->open_count, 0);
1969 atomic_set(&md->event_nr, 0);
1970 atomic_set(&md->uevent_seq, 0);
1971 INIT_LIST_HEAD(&md->uevent_list);
1972 INIT_LIST_HEAD(&md->table_devices);
1973 spin_lock_init(&md->uevent_lock);
1976 * default to bio-based required ->make_request_fn until DM
1977 * table is loaded and md->type established. If request-based
1978 * table is loaded: blk-mq will override accordingly.
1980 md->queue = blk_alloc_queue(dm_make_request, numa_node_id);
1983 md->queue->queuedata = md;
1985 md->disk = alloc_disk_node(1, md->numa_node_id);
1989 init_waitqueue_head(&md->wait);
1990 INIT_WORK(&md->work, dm_wq_work);
1991 init_waitqueue_head(&md->eventq);
1992 init_completion(&md->kobj_holder.completion);
1994 md->disk->major = _major;
1995 md->disk->first_minor = minor;
1996 md->disk->fops = &dm_blk_dops;
1997 md->disk->queue = md->queue;
1998 md->disk->private_data = md;
1999 sprintf(md->disk->disk_name, "dm-%d", minor);
2001 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
2002 md->dax_dev = alloc_dax(md, md->disk->disk_name,
2004 if (IS_ERR(md->dax_dev))
2008 add_disk_no_queue_reg(md->disk);
2009 format_dev_t(md->name, MKDEV(_major, minor));
2011 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2015 md->bdev = bdget_disk(md->disk, 0);
2019 dm_stats_init(&md->stats);
2021 /* Populate the mapping, nobody knows we exist yet */
2022 spin_lock(&_minor_lock);
2023 old_md = idr_replace(&_minor_idr, md, minor);
2024 spin_unlock(&_minor_lock);
2026 BUG_ON(old_md != MINOR_ALLOCED);
2031 cleanup_mapped_device(md);
2035 module_put(THIS_MODULE);
2041 static void unlock_fs(struct mapped_device *md);
2043 static void free_dev(struct mapped_device *md)
2045 int minor = MINOR(disk_devt(md->disk));
2049 cleanup_mapped_device(md);
2051 free_table_devices(&md->table_devices);
2052 dm_stats_cleanup(&md->stats);
2055 module_put(THIS_MODULE);
2059 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2061 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2064 if (dm_table_bio_based(t)) {
2066 * The md may already have mempools that need changing.
2067 * If so, reload bioset because front_pad may have changed
2068 * because a different table was loaded.
2070 bioset_exit(&md->bs);
2071 bioset_exit(&md->io_bs);
2073 } else if (bioset_initialized(&md->bs)) {
2075 * There's no need to reload with request-based dm
2076 * because the size of front_pad doesn't change.
2077 * Note for future: If you are to reload bioset,
2078 * prep-ed requests in the queue may refer
2079 * to bio from the old bioset, so you must walk
2080 * through the queue to unprep.
2086 bioset_initialized(&md->bs) ||
2087 bioset_initialized(&md->io_bs));
2089 ret = bioset_init_from_src(&md->bs, &p->bs);
2092 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2094 bioset_exit(&md->bs);
2096 /* mempool bind completed, no longer need any mempools in the table */
2097 dm_table_free_md_mempools(t);
2102 * Bind a table to the device.
2104 static void event_callback(void *context)
2106 unsigned long flags;
2108 struct mapped_device *md = (struct mapped_device *) context;
2110 spin_lock_irqsave(&md->uevent_lock, flags);
2111 list_splice_init(&md->uevent_list, &uevents);
2112 spin_unlock_irqrestore(&md->uevent_lock, flags);
2114 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2116 atomic_inc(&md->event_nr);
2117 wake_up(&md->eventq);
2118 dm_issue_global_event();
2122 * Protected by md->suspend_lock obtained by dm_swap_table().
2124 static void __set_size(struct mapped_device *md, sector_t size)
2126 lockdep_assert_held(&md->suspend_lock);
2128 set_capacity(md->disk, size);
2130 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2134 * Returns old map, which caller must destroy.
2136 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2137 struct queue_limits *limits)
2139 struct dm_table *old_map;
2140 struct request_queue *q = md->queue;
2141 bool request_based = dm_table_request_based(t);
2145 lockdep_assert_held(&md->suspend_lock);
2147 size = dm_table_get_size(t);
2150 * Wipe any geometry if the size of the table changed.
2152 if (size != dm_get_size(md))
2153 memset(&md->geometry, 0, sizeof(md->geometry));
2155 __set_size(md, size);
2157 dm_table_event_callback(t, event_callback, md);
2160 * The queue hasn't been stopped yet, if the old table type wasn't
2161 * for request-based during suspension. So stop it to prevent
2162 * I/O mapping before resume.
2163 * This must be done before setting the queue restrictions,
2164 * because request-based dm may be run just after the setting.
2169 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2171 * Leverage the fact that request-based DM targets and
2172 * NVMe bio based targets are immutable singletons
2173 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2174 * and __process_bio.
2176 md->immutable_target = dm_table_get_immutable_target(t);
2179 ret = __bind_mempools(md, t);
2181 old_map = ERR_PTR(ret);
2185 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2186 rcu_assign_pointer(md->map, (void *)t);
2187 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2189 dm_table_set_restrictions(t, q, limits);
2198 * Returns unbound table for the caller to free.
2200 static struct dm_table *__unbind(struct mapped_device *md)
2202 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2207 dm_table_event_callback(map, NULL, NULL);
2208 RCU_INIT_POINTER(md->map, NULL);
2215 * Constructor for a new device.
2217 int dm_create(int minor, struct mapped_device **result)
2220 struct mapped_device *md;
2222 md = alloc_dev(minor);
2226 r = dm_sysfs_init(md);
2237 * Functions to manage md->type.
2238 * All are required to hold md->type_lock.
2240 void dm_lock_md_type(struct mapped_device *md)
2242 mutex_lock(&md->type_lock);
2245 void dm_unlock_md_type(struct mapped_device *md)
2247 mutex_unlock(&md->type_lock);
2250 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2252 BUG_ON(!mutex_is_locked(&md->type_lock));
2256 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2261 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2263 return md->immutable_target_type;
2267 * The queue_limits are only valid as long as you have a reference
2270 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2272 BUG_ON(!atomic_read(&md->holders));
2273 return &md->queue->limits;
2275 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2277 static void dm_init_congested_fn(struct mapped_device *md)
2279 md->queue->backing_dev_info->congested_data = md;
2280 md->queue->backing_dev_info->congested_fn = dm_any_congested;
2284 * Setup the DM device's queue based on md's type
2286 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2289 struct queue_limits limits;
2290 enum dm_queue_mode type = dm_get_md_type(md);
2293 case DM_TYPE_REQUEST_BASED:
2294 r = dm_mq_init_request_queue(md, t);
2296 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2299 dm_init_congested_fn(md);
2301 case DM_TYPE_BIO_BASED:
2302 case DM_TYPE_DAX_BIO_BASED:
2303 case DM_TYPE_NVME_BIO_BASED:
2304 dm_init_congested_fn(md);
2311 r = dm_calculate_queue_limits(t, &limits);
2313 DMERR("Cannot calculate initial queue limits");
2316 dm_table_set_restrictions(t, md->queue, &limits);
2317 blk_register_queue(md->disk);
2322 struct mapped_device *dm_get_md(dev_t dev)
2324 struct mapped_device *md;
2325 unsigned minor = MINOR(dev);
2327 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2330 spin_lock(&_minor_lock);
2332 md = idr_find(&_minor_idr, minor);
2333 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2334 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2340 spin_unlock(&_minor_lock);
2344 EXPORT_SYMBOL_GPL(dm_get_md);
2346 void *dm_get_mdptr(struct mapped_device *md)
2348 return md->interface_ptr;
2351 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2353 md->interface_ptr = ptr;
2356 void dm_get(struct mapped_device *md)
2358 atomic_inc(&md->holders);
2359 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2362 int dm_hold(struct mapped_device *md)
2364 spin_lock(&_minor_lock);
2365 if (test_bit(DMF_FREEING, &md->flags)) {
2366 spin_unlock(&_minor_lock);
2370 spin_unlock(&_minor_lock);
2373 EXPORT_SYMBOL_GPL(dm_hold);
2375 const char *dm_device_name(struct mapped_device *md)
2379 EXPORT_SYMBOL_GPL(dm_device_name);
2381 static void __dm_destroy(struct mapped_device *md, bool wait)
2383 struct dm_table *map;
2388 spin_lock(&_minor_lock);
2389 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2390 set_bit(DMF_FREEING, &md->flags);
2391 spin_unlock(&_minor_lock);
2393 blk_set_queue_dying(md->queue);
2396 * Take suspend_lock so that presuspend and postsuspend methods
2397 * do not race with internal suspend.
2399 mutex_lock(&md->suspend_lock);
2400 map = dm_get_live_table(md, &srcu_idx);
2401 if (!dm_suspended_md(md)) {
2402 dm_table_presuspend_targets(map);
2403 set_bit(DMF_SUSPENDED, &md->flags);
2404 dm_table_postsuspend_targets(map);
2406 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2407 dm_put_live_table(md, srcu_idx);
2408 mutex_unlock(&md->suspend_lock);
2411 * Rare, but there may be I/O requests still going to complete,
2412 * for example. Wait for all references to disappear.
2413 * No one should increment the reference count of the mapped_device,
2414 * after the mapped_device state becomes DMF_FREEING.
2417 while (atomic_read(&md->holders))
2419 else if (atomic_read(&md->holders))
2420 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2421 dm_device_name(md), atomic_read(&md->holders));
2424 dm_table_destroy(__unbind(md));
2428 void dm_destroy(struct mapped_device *md)
2430 __dm_destroy(md, true);
2433 void dm_destroy_immediate(struct mapped_device *md)
2435 __dm_destroy(md, false);
2438 void dm_put(struct mapped_device *md)
2440 atomic_dec(&md->holders);
2442 EXPORT_SYMBOL_GPL(dm_put);
2444 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2450 prepare_to_wait(&md->wait, &wait, task_state);
2452 if (!md_in_flight(md))
2455 if (signal_pending_state(task_state, current)) {
2462 finish_wait(&md->wait, &wait);
2468 * Process the deferred bios
2470 static void dm_wq_work(struct work_struct *work)
2472 struct mapped_device *md = container_of(work, struct mapped_device,
2476 struct dm_table *map;
2478 map = dm_get_live_table(md, &srcu_idx);
2480 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2481 spin_lock_irq(&md->deferred_lock);
2482 c = bio_list_pop(&md->deferred);
2483 spin_unlock_irq(&md->deferred_lock);
2488 if (dm_request_based(md))
2489 (void) generic_make_request(c);
2491 (void) dm_process_bio(md, map, c);
2494 dm_put_live_table(md, srcu_idx);
2497 static void dm_queue_flush(struct mapped_device *md)
2499 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2500 smp_mb__after_atomic();
2501 queue_work(md->wq, &md->work);
2505 * Swap in a new table, returning the old one for the caller to destroy.
2507 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2509 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2510 struct queue_limits limits;
2513 mutex_lock(&md->suspend_lock);
2515 /* device must be suspended */
2516 if (!dm_suspended_md(md))
2520 * If the new table has no data devices, retain the existing limits.
2521 * This helps multipath with queue_if_no_path if all paths disappear,
2522 * then new I/O is queued based on these limits, and then some paths
2525 if (dm_table_has_no_data_devices(table)) {
2526 live_map = dm_get_live_table_fast(md);
2528 limits = md->queue->limits;
2529 dm_put_live_table_fast(md);
2533 r = dm_calculate_queue_limits(table, &limits);
2540 map = __bind(md, table, &limits);
2541 dm_issue_global_event();
2544 mutex_unlock(&md->suspend_lock);
2549 * Functions to lock and unlock any filesystem running on the
2552 static int lock_fs(struct mapped_device *md)
2556 WARN_ON(md->frozen_sb);
2558 md->frozen_sb = freeze_bdev(md->bdev);
2559 if (IS_ERR(md->frozen_sb)) {
2560 r = PTR_ERR(md->frozen_sb);
2561 md->frozen_sb = NULL;
2565 set_bit(DMF_FROZEN, &md->flags);
2570 static void unlock_fs(struct mapped_device *md)
2572 if (!test_bit(DMF_FROZEN, &md->flags))
2575 thaw_bdev(md->bdev, md->frozen_sb);
2576 md->frozen_sb = NULL;
2577 clear_bit(DMF_FROZEN, &md->flags);
2581 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2582 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2583 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2585 * If __dm_suspend returns 0, the device is completely quiescent
2586 * now. There is no request-processing activity. All new requests
2587 * are being added to md->deferred list.
2589 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2590 unsigned suspend_flags, long task_state,
2591 int dmf_suspended_flag)
2593 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2594 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2597 lockdep_assert_held(&md->suspend_lock);
2600 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2601 * This flag is cleared before dm_suspend returns.
2604 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2606 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2609 * This gets reverted if there's an error later and the targets
2610 * provide the .presuspend_undo hook.
2612 dm_table_presuspend_targets(map);
2615 * Flush I/O to the device.
2616 * Any I/O submitted after lock_fs() may not be flushed.
2617 * noflush takes precedence over do_lockfs.
2618 * (lock_fs() flushes I/Os and waits for them to complete.)
2620 if (!noflush && do_lockfs) {
2623 dm_table_presuspend_undo_targets(map);
2629 * Here we must make sure that no processes are submitting requests
2630 * to target drivers i.e. no one may be executing
2631 * __split_and_process_bio. This is called from dm_request and
2634 * To get all processes out of __split_and_process_bio in dm_request,
2635 * we take the write lock. To prevent any process from reentering
2636 * __split_and_process_bio from dm_request and quiesce the thread
2637 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2638 * flush_workqueue(md->wq).
2640 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2642 synchronize_srcu(&md->io_barrier);
2645 * Stop md->queue before flushing md->wq in case request-based
2646 * dm defers requests to md->wq from md->queue.
2648 if (dm_request_based(md))
2649 dm_stop_queue(md->queue);
2651 flush_workqueue(md->wq);
2654 * At this point no more requests are entering target request routines.
2655 * We call dm_wait_for_completion to wait for all existing requests
2658 r = dm_wait_for_completion(md, task_state);
2660 set_bit(dmf_suspended_flag, &md->flags);
2663 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2665 synchronize_srcu(&md->io_barrier);
2667 /* were we interrupted ? */
2671 if (dm_request_based(md))
2672 dm_start_queue(md->queue);
2675 dm_table_presuspend_undo_targets(map);
2676 /* pushback list is already flushed, so skip flush */
2683 * We need to be able to change a mapping table under a mounted
2684 * filesystem. For example we might want to move some data in
2685 * the background. Before the table can be swapped with
2686 * dm_bind_table, dm_suspend must be called to flush any in
2687 * flight bios and ensure that any further io gets deferred.
2690 * Suspend mechanism in request-based dm.
2692 * 1. Flush all I/Os by lock_fs() if needed.
2693 * 2. Stop dispatching any I/O by stopping the request_queue.
2694 * 3. Wait for all in-flight I/Os to be completed or requeued.
2696 * To abort suspend, start the request_queue.
2698 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2700 struct dm_table *map = NULL;
2704 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2706 if (dm_suspended_md(md)) {
2711 if (dm_suspended_internally_md(md)) {
2712 /* already internally suspended, wait for internal resume */
2713 mutex_unlock(&md->suspend_lock);
2714 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2720 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2722 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2726 dm_table_postsuspend_targets(map);
2729 mutex_unlock(&md->suspend_lock);
2733 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2736 int r = dm_table_resume_targets(map);
2744 * Flushing deferred I/Os must be done after targets are resumed
2745 * so that mapping of targets can work correctly.
2746 * Request-based dm is queueing the deferred I/Os in its request_queue.
2748 if (dm_request_based(md))
2749 dm_start_queue(md->queue);
2756 int dm_resume(struct mapped_device *md)
2759 struct dm_table *map = NULL;
2763 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2765 if (!dm_suspended_md(md))
2768 if (dm_suspended_internally_md(md)) {
2769 /* already internally suspended, wait for internal resume */
2770 mutex_unlock(&md->suspend_lock);
2771 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2777 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2778 if (!map || !dm_table_get_size(map))
2781 r = __dm_resume(md, map);
2785 clear_bit(DMF_SUSPENDED, &md->flags);
2787 mutex_unlock(&md->suspend_lock);
2793 * Internal suspend/resume works like userspace-driven suspend. It waits
2794 * until all bios finish and prevents issuing new bios to the target drivers.
2795 * It may be used only from the kernel.
2798 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2800 struct dm_table *map = NULL;
2802 lockdep_assert_held(&md->suspend_lock);
2804 if (md->internal_suspend_count++)
2805 return; /* nested internal suspend */
2807 if (dm_suspended_md(md)) {
2808 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2809 return; /* nest suspend */
2812 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2815 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2816 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2817 * would require changing .presuspend to return an error -- avoid this
2818 * until there is a need for more elaborate variants of internal suspend.
2820 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2821 DMF_SUSPENDED_INTERNALLY);
2823 dm_table_postsuspend_targets(map);
2826 static void __dm_internal_resume(struct mapped_device *md)
2828 BUG_ON(!md->internal_suspend_count);
2830 if (--md->internal_suspend_count)
2831 return; /* resume from nested internal suspend */
2833 if (dm_suspended_md(md))
2834 goto done; /* resume from nested suspend */
2837 * NOTE: existing callers don't need to call dm_table_resume_targets
2838 * (which may fail -- so best to avoid it for now by passing NULL map)
2840 (void) __dm_resume(md, NULL);
2843 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2844 smp_mb__after_atomic();
2845 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2848 void dm_internal_suspend_noflush(struct mapped_device *md)
2850 mutex_lock(&md->suspend_lock);
2851 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2852 mutex_unlock(&md->suspend_lock);
2854 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2856 void dm_internal_resume(struct mapped_device *md)
2858 mutex_lock(&md->suspend_lock);
2859 __dm_internal_resume(md);
2860 mutex_unlock(&md->suspend_lock);
2862 EXPORT_SYMBOL_GPL(dm_internal_resume);
2865 * Fast variants of internal suspend/resume hold md->suspend_lock,
2866 * which prevents interaction with userspace-driven suspend.
2869 void dm_internal_suspend_fast(struct mapped_device *md)
2871 mutex_lock(&md->suspend_lock);
2872 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2875 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2876 synchronize_srcu(&md->io_barrier);
2877 flush_workqueue(md->wq);
2878 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2880 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2882 void dm_internal_resume_fast(struct mapped_device *md)
2884 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2890 mutex_unlock(&md->suspend_lock);
2892 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2894 /*-----------------------------------------------------------------
2895 * Event notification.
2896 *---------------------------------------------------------------*/
2897 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2900 char udev_cookie[DM_COOKIE_LENGTH];
2901 char *envp[] = { udev_cookie, NULL };
2904 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2906 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2907 DM_COOKIE_ENV_VAR_NAME, cookie);
2908 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2913 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2915 return atomic_add_return(1, &md->uevent_seq);
2918 uint32_t dm_get_event_nr(struct mapped_device *md)
2920 return atomic_read(&md->event_nr);
2923 int dm_wait_event(struct mapped_device *md, int event_nr)
2925 return wait_event_interruptible(md->eventq,
2926 (event_nr != atomic_read(&md->event_nr)));
2929 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2931 unsigned long flags;
2933 spin_lock_irqsave(&md->uevent_lock, flags);
2934 list_add(elist, &md->uevent_list);
2935 spin_unlock_irqrestore(&md->uevent_lock, flags);
2939 * The gendisk is only valid as long as you have a reference
2942 struct gendisk *dm_disk(struct mapped_device *md)
2946 EXPORT_SYMBOL_GPL(dm_disk);
2948 struct kobject *dm_kobject(struct mapped_device *md)
2950 return &md->kobj_holder.kobj;
2953 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2955 struct mapped_device *md;
2957 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2959 spin_lock(&_minor_lock);
2960 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2966 spin_unlock(&_minor_lock);
2971 int dm_suspended_md(struct mapped_device *md)
2973 return test_bit(DMF_SUSPENDED, &md->flags);
2976 int dm_suspended_internally_md(struct mapped_device *md)
2978 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2981 int dm_test_deferred_remove_flag(struct mapped_device *md)
2983 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2986 int dm_suspended(struct dm_target *ti)
2988 return dm_suspended_md(dm_table_get_md(ti->table));
2990 EXPORT_SYMBOL_GPL(dm_suspended);
2992 int dm_noflush_suspending(struct dm_target *ti)
2994 return __noflush_suspending(dm_table_get_md(ti->table));
2996 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2998 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2999 unsigned integrity, unsigned per_io_data_size,
3000 unsigned min_pool_size)
3002 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3003 unsigned int pool_size = 0;
3004 unsigned int front_pad, io_front_pad;
3011 case DM_TYPE_BIO_BASED:
3012 case DM_TYPE_DAX_BIO_BASED:
3013 case DM_TYPE_NVME_BIO_BASED:
3014 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3015 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3016 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3017 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3020 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3023 case DM_TYPE_REQUEST_BASED:
3024 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3025 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3026 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3032 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3036 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3042 dm_free_md_mempools(pools);
3047 void dm_free_md_mempools(struct dm_md_mempools *pools)
3052 bioset_exit(&pools->bs);
3053 bioset_exit(&pools->io_bs);
3065 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3068 struct mapped_device *md = bdev->bd_disk->private_data;
3069 struct dm_table *table;
3070 struct dm_target *ti;
3071 int ret = -ENOTTY, srcu_idx;
3073 table = dm_get_live_table(md, &srcu_idx);
3074 if (!table || !dm_table_get_size(table))
3077 /* We only support devices that have a single target */
3078 if (dm_table_get_num_targets(table) != 1)
3080 ti = dm_table_get_target(table, 0);
3083 if (!ti->type->iterate_devices)
3086 ret = ti->type->iterate_devices(ti, fn, data);
3088 dm_put_live_table(md, srcu_idx);
3093 * For register / unregister we need to manually call out to every path.
3095 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3096 sector_t start, sector_t len, void *data)
3098 struct dm_pr *pr = data;
3099 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3101 if (!ops || !ops->pr_register)
3103 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3106 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3117 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3118 if (ret && new_key) {
3119 /* unregister all paths if we failed to register any path */
3120 pr.old_key = new_key;
3123 pr.fail_early = false;
3124 dm_call_pr(bdev, __dm_pr_register, &pr);
3130 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
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_reserve)
3143 r = ops->pr_reserve(bdev, key, type, flags);
3147 dm_unprepare_ioctl(md, srcu_idx);
3151 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
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_release)
3163 r = ops->pr_release(bdev, key, type);
3167 dm_unprepare_ioctl(md, srcu_idx);
3171 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3172 enum pr_type type, bool abort)
3174 struct mapped_device *md = bdev->bd_disk->private_data;
3175 const struct pr_ops *ops;
3178 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3182 ops = bdev->bd_disk->fops->pr_ops;
3183 if (ops && ops->pr_preempt)
3184 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3188 dm_unprepare_ioctl(md, srcu_idx);
3192 static int dm_pr_clear(struct block_device *bdev, u64 key)
3194 struct mapped_device *md = bdev->bd_disk->private_data;
3195 const struct pr_ops *ops;
3198 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3202 ops = bdev->bd_disk->fops->pr_ops;
3203 if (ops && ops->pr_clear)
3204 r = ops->pr_clear(bdev, key);
3208 dm_unprepare_ioctl(md, srcu_idx);
3212 static const struct pr_ops dm_pr_ops = {
3213 .pr_register = dm_pr_register,
3214 .pr_reserve = dm_pr_reserve,
3215 .pr_release = dm_pr_release,
3216 .pr_preempt = dm_pr_preempt,
3217 .pr_clear = dm_pr_clear,
3220 static const struct block_device_operations dm_blk_dops = {
3221 .open = dm_blk_open,
3222 .release = dm_blk_close,
3223 .ioctl = dm_blk_ioctl,
3224 .getgeo = dm_blk_getgeo,
3225 .report_zones = dm_blk_report_zones,
3226 .pr_ops = &dm_pr_ops,
3227 .owner = THIS_MODULE
3230 static const struct dax_operations dm_dax_ops = {
3231 .direct_access = dm_dax_direct_access,
3232 .dax_supported = dm_dax_supported,
3233 .copy_from_iter = dm_dax_copy_from_iter,
3234 .copy_to_iter = dm_dax_copy_to_iter,
3235 .zero_page_range = dm_dax_zero_page_range,
3241 module_init(dm_init);
3242 module_exit(dm_exit);
3244 module_param(major, uint, 0);
3245 MODULE_PARM_DESC(major, "The major number of the device mapper");
3247 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3248 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3250 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3251 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3253 MODULE_DESCRIPTION(DM_NAME " driver");
3254 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3255 MODULE_LICENSE("GPL");