2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/mm.h>
16 #include <linux/sched/signal.h>
17 #include <linux/blkpg.h>
18 #include <linux/bio.h>
19 #include <linux/mempool.h>
20 #include <linux/dax.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/uio.h>
24 #include <linux/hdreg.h>
25 #include <linux/delay.h>
26 #include <linux/wait.h>
28 #include <linux/refcount.h>
29 #include <linux/part_stat.h>
30 #include <linux/blk-crypto.h>
32 #define DM_MSG_PREFIX "core"
35 * Cookies are numeric values sent with CHANGE and REMOVE
36 * uevents while resuming, removing or renaming the device.
38 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
39 #define DM_COOKIE_LENGTH 24
41 static const char *_name = DM_NAME;
43 static unsigned int major = 0;
44 static unsigned int _major = 0;
46 static DEFINE_IDR(_minor_idr);
48 static DEFINE_SPINLOCK(_minor_lock);
50 static void do_deferred_remove(struct work_struct *w);
52 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
54 static struct workqueue_struct *deferred_remove_workqueue;
56 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
57 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
59 void dm_issue_global_event(void)
61 atomic_inc(&dm_global_event_nr);
62 wake_up(&dm_global_eventq);
66 * One of these is allocated (on-stack) per original bio.
73 unsigned sector_count;
77 * One of these is allocated per clone bio.
79 #define DM_TIO_MAGIC 7282014
84 unsigned target_bio_nr;
91 * One of these is allocated per original bio.
92 * It contains the first clone used for that original.
94 #define DM_IO_MAGIC 5191977
97 struct mapped_device *md;
100 struct bio *orig_bio;
101 unsigned long start_time;
102 spinlock_t endio_lock;
103 struct dm_stats_aux stats_aux;
104 /* last member of dm_target_io is 'struct bio' */
105 struct dm_target_io tio;
108 void *dm_per_bio_data(struct bio *bio, size_t data_size)
110 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
111 if (!tio->inside_dm_io)
112 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
113 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
115 EXPORT_SYMBOL_GPL(dm_per_bio_data);
117 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
119 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
120 if (io->magic == DM_IO_MAGIC)
121 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
122 BUG_ON(io->magic != DM_TIO_MAGIC);
123 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
125 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
127 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
129 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
131 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
133 #define MINOR_ALLOCED ((void *)-1)
136 * Bits for the md->flags field.
138 #define DMF_BLOCK_IO_FOR_SUSPEND 0
139 #define DMF_SUSPENDED 1
141 #define DMF_FREEING 3
142 #define DMF_DELETING 4
143 #define DMF_NOFLUSH_SUSPENDING 5
144 #define DMF_DEFERRED_REMOVE 6
145 #define DMF_SUSPENDED_INTERNALLY 7
146 #define DMF_POST_SUSPENDING 8
148 #define DM_NUMA_NODE NUMA_NO_NODE
149 static int dm_numa_node = DM_NUMA_NODE;
151 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
152 static int swap_bios = DEFAULT_SWAP_BIOS;
153 static int get_swap_bios(void)
155 int latch = READ_ONCE(swap_bios);
156 if (unlikely(latch <= 0))
157 latch = DEFAULT_SWAP_BIOS;
162 * For mempools pre-allocation at the table loading time.
164 struct dm_md_mempools {
166 struct bio_set io_bs;
169 struct table_device {
170 struct list_head list;
172 struct dm_dev dm_dev;
176 * Bio-based DM's mempools' reserved IOs set by the user.
178 #define RESERVED_BIO_BASED_IOS 16
179 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
181 static int __dm_get_module_param_int(int *module_param, int min, int max)
183 int param = READ_ONCE(*module_param);
184 int modified_param = 0;
185 bool modified = true;
188 modified_param = min;
189 else if (param > max)
190 modified_param = max;
195 (void)cmpxchg(module_param, param, modified_param);
196 param = modified_param;
202 unsigned __dm_get_module_param(unsigned *module_param,
203 unsigned def, unsigned max)
205 unsigned param = READ_ONCE(*module_param);
206 unsigned modified_param = 0;
209 modified_param = def;
210 else if (param > max)
211 modified_param = max;
213 if (modified_param) {
214 (void)cmpxchg(module_param, param, modified_param);
215 param = modified_param;
221 unsigned dm_get_reserved_bio_based_ios(void)
223 return __dm_get_module_param(&reserved_bio_based_ios,
224 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
226 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
228 static unsigned dm_get_numa_node(void)
230 return __dm_get_module_param_int(&dm_numa_node,
231 DM_NUMA_NODE, num_online_nodes() - 1);
234 static int __init local_init(void)
238 r = dm_uevent_init();
242 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
243 if (!deferred_remove_workqueue) {
245 goto out_uevent_exit;
249 r = register_blkdev(_major, _name);
251 goto out_free_workqueue;
259 destroy_workqueue(deferred_remove_workqueue);
266 static void local_exit(void)
268 destroy_workqueue(deferred_remove_workqueue);
270 unregister_blkdev(_major, _name);
275 DMINFO("cleaned up");
278 static int (*_inits[])(void) __initdata = {
289 static void (*_exits[])(void) = {
300 static int __init dm_init(void)
302 const int count = ARRAY_SIZE(_inits);
306 for (i = 0; i < count; i++) {
321 static void __exit dm_exit(void)
323 int i = ARRAY_SIZE(_exits);
329 * Should be empty by this point.
331 idr_destroy(&_minor_idr);
335 * Block device functions
337 int dm_deleting_md(struct mapped_device *md)
339 return test_bit(DMF_DELETING, &md->flags);
342 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
344 struct mapped_device *md;
346 spin_lock(&_minor_lock);
348 md = bdev->bd_disk->private_data;
352 if (test_bit(DMF_FREEING, &md->flags) ||
353 dm_deleting_md(md)) {
359 atomic_inc(&md->open_count);
361 spin_unlock(&_minor_lock);
363 return md ? 0 : -ENXIO;
366 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
368 struct mapped_device *md;
370 spin_lock(&_minor_lock);
372 md = disk->private_data;
376 if (atomic_dec_and_test(&md->open_count) &&
377 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
378 queue_work(deferred_remove_workqueue, &deferred_remove_work);
382 spin_unlock(&_minor_lock);
385 int dm_open_count(struct mapped_device *md)
387 return atomic_read(&md->open_count);
391 * Guarantees nothing is using the device before it's deleted.
393 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
397 spin_lock(&_minor_lock);
399 if (dm_open_count(md)) {
402 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
403 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
406 set_bit(DMF_DELETING, &md->flags);
408 spin_unlock(&_minor_lock);
413 int dm_cancel_deferred_remove(struct mapped_device *md)
417 spin_lock(&_minor_lock);
419 if (test_bit(DMF_DELETING, &md->flags))
422 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
424 spin_unlock(&_minor_lock);
429 static void do_deferred_remove(struct work_struct *w)
431 dm_deferred_remove();
434 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
436 struct mapped_device *md = bdev->bd_disk->private_data;
438 return dm_get_geometry(md, geo);
441 #ifdef CONFIG_BLK_DEV_ZONED
442 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
444 struct dm_report_zones_args *args = data;
445 sector_t sector_diff = args->tgt->begin - args->start;
448 * Ignore zones beyond the target range.
450 if (zone->start >= args->start + args->tgt->len)
454 * Remap the start sector and write pointer position of the zone
455 * to match its position in the target range.
457 zone->start += sector_diff;
458 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
459 if (zone->cond == BLK_ZONE_COND_FULL)
460 zone->wp = zone->start + zone->len;
461 else if (zone->cond == BLK_ZONE_COND_EMPTY)
462 zone->wp = zone->start;
464 zone->wp += sector_diff;
467 args->next_sector = zone->start + zone->len;
468 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
470 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
472 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
473 unsigned int nr_zones, report_zones_cb cb, void *data)
475 struct mapped_device *md = disk->private_data;
476 struct dm_table *map;
478 struct dm_report_zones_args args = {
479 .next_sector = sector,
484 if (dm_suspended_md(md))
487 map = dm_get_live_table(md, &srcu_idx);
494 struct dm_target *tgt;
496 tgt = dm_table_find_target(map, args.next_sector);
497 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
503 ret = tgt->type->report_zones(tgt, &args,
504 nr_zones - args.zone_idx);
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)
522 struct dm_target *tgt;
523 struct dm_table *map;
528 map = dm_get_live_table(md, srcu_idx);
529 if (!map || !dm_table_get_size(map))
532 /* We only support devices that have a single target */
533 if (dm_table_get_num_targets(map) != 1)
536 tgt = dm_table_get_target(map, 0);
537 if (!tgt->type->prepare_ioctl)
540 if (dm_suspended_md(md))
543 r = tgt->type->prepare_ioctl(tgt, bdev);
544 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
545 dm_put_live_table(md, *srcu_idx);
553 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
555 dm_put_live_table(md, srcu_idx);
558 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
559 unsigned int cmd, unsigned long arg)
561 struct mapped_device *md = bdev->bd_disk->private_data;
564 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
570 * Target determined this ioctl is being issued against a
571 * subset of the parent bdev; require extra privileges.
573 if (!capable(CAP_SYS_RAWIO)) {
575 "%s: sending ioctl %x to DM device without required privilege.",
582 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
584 dm_unprepare_ioctl(md, srcu_idx);
588 u64 dm_start_time_ns_from_clone(struct bio *bio)
590 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
591 struct dm_io *io = tio->io;
593 return jiffies_to_nsecs(io->start_time);
595 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
597 static void start_io_acct(struct dm_io *io)
599 struct mapped_device *md = io->md;
600 struct bio *bio = io->orig_bio;
602 io->start_time = bio_start_io_acct(bio);
603 if (unlikely(dm_stats_used(&md->stats)))
604 dm_stats_account_io(&md->stats, bio_data_dir(bio),
605 bio->bi_iter.bi_sector, bio_sectors(bio),
606 false, 0, &io->stats_aux);
609 static void end_io_acct(struct mapped_device *md, struct bio *bio,
610 unsigned long start_time, struct dm_stats_aux *stats_aux)
612 unsigned long duration = jiffies - start_time;
614 if (unlikely(dm_stats_used(&md->stats)))
615 dm_stats_account_io(&md->stats, bio_data_dir(bio),
616 bio->bi_iter.bi_sector, bio_sectors(bio),
617 true, duration, stats_aux);
621 bio_end_io_acct(bio, start_time);
623 /* nudge anyone waiting on suspend queue */
624 if (unlikely(wq_has_sleeper(&md->wait)))
628 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
631 struct dm_target_io *tio;
634 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
638 tio = container_of(clone, struct dm_target_io, clone);
639 tio->inside_dm_io = true;
642 io = container_of(tio, struct dm_io, tio);
643 io->magic = DM_IO_MAGIC;
645 atomic_set(&io->io_count, 1);
648 spin_lock_init(&io->endio_lock);
655 static void free_io(struct mapped_device *md, struct dm_io *io)
657 bio_put(&io->tio.clone);
660 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
661 unsigned target_bio_nr, gfp_t gfp_mask)
663 struct dm_target_io *tio;
665 if (!ci->io->tio.io) {
666 /* the dm_target_io embedded in ci->io is available */
669 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
673 tio = container_of(clone, struct dm_target_io, clone);
674 tio->inside_dm_io = false;
677 tio->magic = DM_TIO_MAGIC;
680 tio->target_bio_nr = target_bio_nr;
685 static void free_tio(struct dm_target_io *tio)
687 if (tio->inside_dm_io)
689 bio_put(&tio->clone);
693 * Add the bio to the list of deferred io.
695 static void queue_io(struct mapped_device *md, struct bio *bio)
699 spin_lock_irqsave(&md->deferred_lock, flags);
700 bio_list_add(&md->deferred, bio);
701 spin_unlock_irqrestore(&md->deferred_lock, flags);
702 queue_work(md->wq, &md->work);
706 * Everyone (including functions in this file), should use this
707 * function to access the md->map field, and make sure they call
708 * dm_put_live_table() when finished.
710 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
712 *srcu_idx = srcu_read_lock(&md->io_barrier);
714 return srcu_dereference(md->map, &md->io_barrier);
717 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
719 srcu_read_unlock(&md->io_barrier, srcu_idx);
722 void dm_sync_table(struct mapped_device *md)
724 synchronize_srcu(&md->io_barrier);
725 synchronize_rcu_expedited();
729 * A fast alternative to dm_get_live_table/dm_put_live_table.
730 * The caller must not block between these two functions.
732 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
735 return rcu_dereference(md->map);
738 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
743 static char *_dm_claim_ptr = "I belong to device-mapper";
746 * Open a table device so we can use it as a map destination.
748 static int open_table_device(struct table_device *td, dev_t dev,
749 struct mapped_device *md)
751 struct block_device *bdev;
755 BUG_ON(td->dm_dev.bdev);
757 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
759 return PTR_ERR(bdev);
761 r = bd_link_disk_holder(bdev, dm_disk(md));
763 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
767 td->dm_dev.bdev = bdev;
768 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
773 * Close a table device that we've been using.
775 static void close_table_device(struct table_device *td, struct mapped_device *md)
777 if (!td->dm_dev.bdev)
780 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
781 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
782 put_dax(td->dm_dev.dax_dev);
783 td->dm_dev.bdev = NULL;
784 td->dm_dev.dax_dev = NULL;
787 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
790 struct table_device *td;
792 list_for_each_entry(td, l, list)
793 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
799 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
800 struct dm_dev **result)
803 struct table_device *td;
805 mutex_lock(&md->table_devices_lock);
806 td = find_table_device(&md->table_devices, dev, mode);
808 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
810 mutex_unlock(&md->table_devices_lock);
814 td->dm_dev.mode = mode;
815 td->dm_dev.bdev = NULL;
817 if ((r = open_table_device(td, dev, md))) {
818 mutex_unlock(&md->table_devices_lock);
823 format_dev_t(td->dm_dev.name, dev);
825 refcount_set(&td->count, 1);
826 list_add(&td->list, &md->table_devices);
828 refcount_inc(&td->count);
830 mutex_unlock(&md->table_devices_lock);
832 *result = &td->dm_dev;
835 EXPORT_SYMBOL_GPL(dm_get_table_device);
837 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
839 struct table_device *td = container_of(d, struct table_device, dm_dev);
841 mutex_lock(&md->table_devices_lock);
842 if (refcount_dec_and_test(&td->count)) {
843 close_table_device(td, md);
847 mutex_unlock(&md->table_devices_lock);
849 EXPORT_SYMBOL(dm_put_table_device);
851 static void free_table_devices(struct list_head *devices)
853 struct list_head *tmp, *next;
855 list_for_each_safe(tmp, next, devices) {
856 struct table_device *td = list_entry(tmp, struct table_device, list);
858 DMWARN("dm_destroy: %s still exists with %d references",
859 td->dm_dev.name, refcount_read(&td->count));
865 * Get the geometry associated with a dm device
867 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
875 * Set the geometry of a device.
877 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
879 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
881 if (geo->start > sz) {
882 DMWARN("Start sector is beyond the geometry limits.");
891 static int __noflush_suspending(struct mapped_device *md)
893 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
897 * Decrements the number of outstanding ios that a bio has been
898 * cloned into, completing the original io if necc.
900 static void dec_pending(struct dm_io *io, blk_status_t error)
903 blk_status_t io_error;
905 struct mapped_device *md = io->md;
906 unsigned long start_time = 0;
907 struct dm_stats_aux stats_aux;
909 /* Push-back supersedes any I/O errors */
910 if (unlikely(error)) {
911 spin_lock_irqsave(&io->endio_lock, flags);
912 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
914 spin_unlock_irqrestore(&io->endio_lock, flags);
917 if (atomic_dec_and_test(&io->io_count)) {
918 if (io->status == BLK_STS_DM_REQUEUE) {
920 * Target requested pushing back the I/O.
922 spin_lock_irqsave(&md->deferred_lock, flags);
923 if (__noflush_suspending(md))
924 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
925 bio_list_add_head(&md->deferred, io->orig_bio);
927 /* noflush suspend was interrupted. */
928 io->status = BLK_STS_IOERR;
929 spin_unlock_irqrestore(&md->deferred_lock, flags);
932 io_error = io->status;
934 start_time = io->start_time;
935 stats_aux = io->stats_aux;
937 end_io_acct(md, bio, start_time, &stats_aux);
939 if (io_error == BLK_STS_DM_REQUEUE)
942 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
944 * Preflush done for flush with data, reissue
945 * without REQ_PREFLUSH.
947 bio->bi_opf &= ~REQ_PREFLUSH;
950 /* done with normal IO or empty flush */
952 bio->bi_status = io_error;
958 void disable_discard(struct mapped_device *md)
960 struct queue_limits *limits = dm_get_queue_limits(md);
962 /* device doesn't really support DISCARD, disable it */
963 limits->max_discard_sectors = 0;
964 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
967 void disable_write_same(struct mapped_device *md)
969 struct queue_limits *limits = dm_get_queue_limits(md);
971 /* device doesn't really support WRITE SAME, disable it */
972 limits->max_write_same_sectors = 0;
975 void disable_write_zeroes(struct mapped_device *md)
977 struct queue_limits *limits = dm_get_queue_limits(md);
979 /* device doesn't really support WRITE ZEROES, disable it */
980 limits->max_write_zeroes_sectors = 0;
983 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
985 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
988 static void clone_endio(struct bio *bio)
990 blk_status_t error = bio->bi_status;
991 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
992 struct dm_io *io = tio->io;
993 struct mapped_device *md = tio->io->md;
994 dm_endio_fn endio = tio->ti->type->end_io;
995 struct bio *orig_bio = io->orig_bio;
997 if (unlikely(error == BLK_STS_TARGET)) {
998 if (bio_op(bio) == REQ_OP_DISCARD &&
999 !bio->bi_disk->queue->limits.max_discard_sectors)
1000 disable_discard(md);
1001 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1002 !bio->bi_disk->queue->limits.max_write_same_sectors)
1003 disable_write_same(md);
1004 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1005 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1006 disable_write_zeroes(md);
1010 * For zone-append bios get offset in zone of the written
1011 * sector and add that to the original bio sector pos.
1013 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1014 sector_t written_sector = bio->bi_iter.bi_sector;
1015 struct request_queue *q = orig_bio->bi_disk->queue;
1016 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1018 orig_bio->bi_iter.bi_sector += written_sector & mask;
1022 int r = endio(tio->ti, bio, &error);
1024 case DM_ENDIO_REQUEUE:
1025 error = BLK_STS_DM_REQUEUE;
1029 case DM_ENDIO_INCOMPLETE:
1030 /* The target will handle the io */
1033 DMWARN("unimplemented target endio return value: %d", r);
1038 if (unlikely(swap_bios_limit(tio->ti, bio))) {
1039 struct mapped_device *md = io->md;
1040 up(&md->swap_bios_semaphore);
1044 dec_pending(io, error);
1048 * Return maximum size of I/O possible at the supplied sector up to the current
1051 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1052 sector_t target_offset)
1054 return ti->len - target_offset;
1057 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1059 sector_t target_offset = dm_target_offset(ti, sector);
1060 sector_t len = max_io_len_target_boundary(ti, target_offset);
1064 * Does the target need to split IO even further?
1065 * - varied (per target) IO splitting is a tenet of DM; this
1066 * explains why stacked chunk_sectors based splitting via
1067 * blk_max_size_offset() isn't possible here. So pass in
1068 * ti->max_io_len to override stacked chunk_sectors.
1070 if (ti->max_io_len) {
1071 max_len = blk_max_size_offset(ti->table->md->queue,
1072 target_offset, ti->max_io_len);
1080 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1082 if (len > UINT_MAX) {
1083 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1084 (unsigned long long)len, UINT_MAX);
1085 ti->error = "Maximum size of target IO is too large";
1089 ti->max_io_len = (uint32_t) len;
1093 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1095 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1096 sector_t sector, int *srcu_idx)
1097 __acquires(md->io_barrier)
1099 struct dm_table *map;
1100 struct dm_target *ti;
1102 map = dm_get_live_table(md, srcu_idx);
1106 ti = dm_table_find_target(map, sector);
1113 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1114 long nr_pages, void **kaddr, pfn_t *pfn)
1116 struct mapped_device *md = dax_get_private(dax_dev);
1117 sector_t sector = pgoff * PAGE_SECTORS;
1118 struct dm_target *ti;
1119 long len, ret = -EIO;
1122 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1126 if (!ti->type->direct_access)
1128 len = max_io_len(ti, sector) / PAGE_SECTORS;
1131 nr_pages = min(len, nr_pages);
1132 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1135 dm_put_live_table(md, srcu_idx);
1140 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1141 int blocksize, sector_t start, sector_t len)
1143 struct mapped_device *md = dax_get_private(dax_dev);
1144 struct dm_table *map;
1148 map = dm_get_live_table(md, &srcu_idx);
1152 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1155 dm_put_live_table(md, srcu_idx);
1160 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1161 void *addr, size_t bytes, struct iov_iter *i)
1163 struct mapped_device *md = dax_get_private(dax_dev);
1164 sector_t sector = pgoff * PAGE_SECTORS;
1165 struct dm_target *ti;
1169 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1173 if (!ti->type->dax_copy_from_iter) {
1174 ret = copy_from_iter(addr, bytes, i);
1177 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1179 dm_put_live_table(md, srcu_idx);
1184 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1185 void *addr, size_t bytes, struct iov_iter *i)
1187 struct mapped_device *md = dax_get_private(dax_dev);
1188 sector_t sector = pgoff * PAGE_SECTORS;
1189 struct dm_target *ti;
1193 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1197 if (!ti->type->dax_copy_to_iter) {
1198 ret = copy_to_iter(addr, bytes, i);
1201 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1203 dm_put_live_table(md, srcu_idx);
1208 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1211 struct mapped_device *md = dax_get_private(dax_dev);
1212 sector_t sector = pgoff * PAGE_SECTORS;
1213 struct dm_target *ti;
1217 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1221 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1223 * ->zero_page_range() is mandatory dax operation. If we are
1224 * here, something is wrong.
1228 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1230 dm_put_live_table(md, srcu_idx);
1236 * A target may call dm_accept_partial_bio only from the map routine. It is
1237 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1238 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1240 * dm_accept_partial_bio informs the dm that the target only wants to process
1241 * additional n_sectors sectors of the bio and the rest of the data should be
1242 * sent in a next bio.
1244 * A diagram that explains the arithmetics:
1245 * +--------------------+---------------+-------+
1247 * +--------------------+---------------+-------+
1249 * <-------------- *tio->len_ptr --------------->
1250 * <------- bi_size ------->
1253 * Region 1 was already iterated over with bio_advance or similar function.
1254 * (it may be empty if the target doesn't use bio_advance)
1255 * Region 2 is the remaining bio size that the target wants to process.
1256 * (it may be empty if region 1 is non-empty, although there is no reason
1258 * The target requires that region 3 is to be sent in the next bio.
1260 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1261 * the partially processed part (the sum of regions 1+2) must be the same for all
1262 * copies of the bio.
1264 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1266 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1267 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1269 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1270 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1271 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1272 BUG_ON(bi_size > *tio->len_ptr);
1273 BUG_ON(n_sectors > bi_size);
1275 *tio->len_ptr -= bi_size - n_sectors;
1276 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1278 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1280 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1282 mutex_lock(&md->swap_bios_lock);
1283 while (latch < md->swap_bios) {
1285 down(&md->swap_bios_semaphore);
1288 while (latch > md->swap_bios) {
1290 up(&md->swap_bios_semaphore);
1293 mutex_unlock(&md->swap_bios_lock);
1296 static blk_qc_t __map_bio(struct dm_target_io *tio)
1300 struct bio *clone = &tio->clone;
1301 struct dm_io *io = tio->io;
1302 struct dm_target *ti = tio->ti;
1303 blk_qc_t ret = BLK_QC_T_NONE;
1305 clone->bi_end_io = clone_endio;
1308 * Map the clone. If r == 0 we don't need to do
1309 * anything, the target has assumed ownership of
1312 atomic_inc(&io->io_count);
1313 sector = clone->bi_iter.bi_sector;
1315 if (unlikely(swap_bios_limit(ti, clone))) {
1316 struct mapped_device *md = io->md;
1317 int latch = get_swap_bios();
1318 if (unlikely(latch != md->swap_bios))
1319 __set_swap_bios_limit(md, latch);
1320 down(&md->swap_bios_semaphore);
1323 r = ti->type->map(ti, clone);
1325 case DM_MAPIO_SUBMITTED:
1327 case DM_MAPIO_REMAPPED:
1328 /* the bio has been remapped so dispatch it */
1329 trace_block_bio_remap(clone->bi_disk->queue, clone,
1330 bio_dev(io->orig_bio), sector);
1331 ret = submit_bio_noacct(clone);
1334 if (unlikely(swap_bios_limit(ti, clone))) {
1335 struct mapped_device *md = io->md;
1336 up(&md->swap_bios_semaphore);
1339 dec_pending(io, BLK_STS_IOERR);
1341 case DM_MAPIO_REQUEUE:
1342 if (unlikely(swap_bios_limit(ti, clone))) {
1343 struct mapped_device *md = io->md;
1344 up(&md->swap_bios_semaphore);
1347 dec_pending(io, BLK_STS_DM_REQUEUE);
1350 DMWARN("unimplemented target map return value: %d", r);
1357 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1359 bio->bi_iter.bi_sector = sector;
1360 bio->bi_iter.bi_size = to_bytes(len);
1364 * Creates a bio that consists of range of complete bvecs.
1366 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1367 sector_t sector, unsigned len)
1369 struct bio *clone = &tio->clone;
1372 __bio_clone_fast(clone, bio);
1374 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1378 if (bio_integrity(bio)) {
1379 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1380 !dm_target_passes_integrity(tio->ti->type))) {
1381 DMWARN("%s: the target %s doesn't support integrity data.",
1382 dm_device_name(tio->io->md),
1383 tio->ti->type->name);
1387 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1392 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1393 clone->bi_iter.bi_size = to_bytes(len);
1395 if (bio_integrity(bio))
1396 bio_integrity_trim(clone);
1401 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1402 struct dm_target *ti, unsigned num_bios)
1404 struct dm_target_io *tio;
1410 if (num_bios == 1) {
1411 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1412 bio_list_add(blist, &tio->clone);
1416 for (try = 0; try < 2; try++) {
1421 mutex_lock(&ci->io->md->table_devices_lock);
1422 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1423 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1427 bio_list_add(blist, &tio->clone);
1430 mutex_unlock(&ci->io->md->table_devices_lock);
1431 if (bio_nr == num_bios)
1434 while ((bio = bio_list_pop(blist))) {
1435 tio = container_of(bio, struct dm_target_io, clone);
1441 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1442 struct dm_target_io *tio, unsigned *len)
1444 struct bio *clone = &tio->clone;
1448 __bio_clone_fast(clone, ci->bio);
1450 bio_setup_sector(clone, ci->sector, *len);
1452 return __map_bio(tio);
1455 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1456 unsigned num_bios, unsigned *len)
1458 struct bio_list blist = BIO_EMPTY_LIST;
1460 struct dm_target_io *tio;
1462 alloc_multiple_bios(&blist, ci, ti, num_bios);
1464 while ((bio = bio_list_pop(&blist))) {
1465 tio = container_of(bio, struct dm_target_io, clone);
1466 (void) __clone_and_map_simple_bio(ci, tio, len);
1470 static int __send_empty_flush(struct clone_info *ci)
1472 unsigned target_nr = 0;
1473 struct dm_target *ti;
1474 struct bio flush_bio;
1477 * Use an on-stack bio for this, it's safe since we don't
1478 * need to reference it after submit. It's just used as
1479 * the basis for the clone(s).
1481 bio_init(&flush_bio, NULL, 0);
1482 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1483 ci->bio = &flush_bio;
1484 ci->sector_count = 0;
1487 * Empty flush uses a statically initialized bio, as the base for
1488 * cloning. However, blkg association requires that a bdev is
1489 * associated with a gendisk, which doesn't happen until the bdev is
1490 * opened. So, blkg association is done at issue time of the flush
1491 * rather than when the device is created in alloc_dev().
1493 bio_set_dev(ci->bio, ci->io->md->bdev);
1495 BUG_ON(bio_has_data(ci->bio));
1496 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1497 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1499 bio_uninit(ci->bio);
1503 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1504 sector_t sector, unsigned *len)
1506 struct bio *bio = ci->bio;
1507 struct dm_target_io *tio;
1510 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1512 r = clone_bio(tio, bio, sector, *len);
1517 (void) __map_bio(tio);
1522 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1528 * Even though the device advertised support for this type of
1529 * request, that does not mean every target supports it, and
1530 * reconfiguration might also have changed that since the
1531 * check was performed.
1536 len = min_t(sector_t, ci->sector_count,
1537 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1539 __send_duplicate_bios(ci, ti, num_bios, &len);
1542 ci->sector_count -= len;
1547 static bool is_abnormal_io(struct bio *bio)
1551 switch (bio_op(bio)) {
1552 case REQ_OP_DISCARD:
1553 case REQ_OP_SECURE_ERASE:
1554 case REQ_OP_WRITE_SAME:
1555 case REQ_OP_WRITE_ZEROES:
1563 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1566 struct bio *bio = ci->bio;
1567 unsigned num_bios = 0;
1569 switch (bio_op(bio)) {
1570 case REQ_OP_DISCARD:
1571 num_bios = ti->num_discard_bios;
1573 case REQ_OP_SECURE_ERASE:
1574 num_bios = ti->num_secure_erase_bios;
1576 case REQ_OP_WRITE_SAME:
1577 num_bios = ti->num_write_same_bios;
1579 case REQ_OP_WRITE_ZEROES:
1580 num_bios = ti->num_write_zeroes_bios;
1586 *result = __send_changing_extent_only(ci, ti, num_bios);
1591 * Select the correct strategy for processing a non-flush bio.
1593 static int __split_and_process_non_flush(struct clone_info *ci)
1595 struct dm_target *ti;
1599 ti = dm_table_find_target(ci->map, ci->sector);
1603 if (__process_abnormal_io(ci, ti, &r))
1606 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1608 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1613 ci->sector_count -= len;
1618 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1619 struct dm_table *map, struct bio *bio)
1622 ci->io = alloc_io(md, bio);
1623 ci->sector = bio->bi_iter.bi_sector;
1626 #define __dm_part_stat_sub(part, field, subnd) \
1627 (part_stat_get(part, field) -= (subnd))
1630 * Entry point to split a bio into clones and submit them to the targets.
1632 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1633 struct dm_table *map, struct bio *bio)
1635 struct clone_info ci;
1636 blk_qc_t ret = BLK_QC_T_NONE;
1639 init_clone_info(&ci, md, map, bio);
1641 if (bio->bi_opf & REQ_PREFLUSH) {
1642 error = __send_empty_flush(&ci);
1643 /* dec_pending submits any data associated with flush */
1644 } else if (op_is_zone_mgmt(bio_op(bio))) {
1646 ci.sector_count = 0;
1647 error = __split_and_process_non_flush(&ci);
1650 ci.sector_count = bio_sectors(bio);
1651 while (ci.sector_count && !error) {
1652 error = __split_and_process_non_flush(&ci);
1653 if (current->bio_list && ci.sector_count && !error) {
1655 * Remainder must be passed to submit_bio_noacct()
1656 * so that it gets handled *after* bios already submitted
1657 * have been completely processed.
1658 * We take a clone of the original to store in
1659 * ci.io->orig_bio to be used by end_io_acct() and
1660 * for dec_pending to use for completion handling.
1662 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1663 GFP_NOIO, &md->queue->bio_split);
1664 ci.io->orig_bio = b;
1667 * Adjust IO stats for each split, otherwise upon queue
1668 * reentry there will be redundant IO accounting.
1669 * NOTE: this is a stop-gap fix, a proper fix involves
1670 * significant refactoring of DM core's bio splitting
1671 * (by eliminating DM's splitting and just using bio_split)
1674 __dm_part_stat_sub(&dm_disk(md)->part0,
1675 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1679 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1680 ret = submit_bio_noacct(bio);
1686 /* drop the extra reference count */
1687 dec_pending(ci.io, errno_to_blk_status(error));
1691 static blk_qc_t dm_submit_bio(struct bio *bio)
1693 struct mapped_device *md = bio->bi_disk->private_data;
1694 blk_qc_t ret = BLK_QC_T_NONE;
1696 struct dm_table *map;
1698 map = dm_get_live_table(md, &srcu_idx);
1700 /* If suspended, or map not yet available, queue this IO for later */
1701 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1703 if (bio->bi_opf & REQ_NOWAIT)
1704 bio_wouldblock_error(bio);
1705 else if (bio->bi_opf & REQ_RAHEAD)
1713 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1714 * otherwise associated queue_limits won't be imposed.
1716 if (is_abnormal_io(bio))
1717 blk_queue_split(&bio);
1719 ret = __split_and_process_bio(md, map, bio);
1721 dm_put_live_table(md, srcu_idx);
1725 /*-----------------------------------------------------------------
1726 * An IDR is used to keep track of allocated minor numbers.
1727 *---------------------------------------------------------------*/
1728 static void free_minor(int minor)
1730 spin_lock(&_minor_lock);
1731 idr_remove(&_minor_idr, minor);
1732 spin_unlock(&_minor_lock);
1736 * See if the device with a specific minor # is free.
1738 static int specific_minor(int minor)
1742 if (minor >= (1 << MINORBITS))
1745 idr_preload(GFP_KERNEL);
1746 spin_lock(&_minor_lock);
1748 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1750 spin_unlock(&_minor_lock);
1753 return r == -ENOSPC ? -EBUSY : r;
1757 static int next_free_minor(int *minor)
1761 idr_preload(GFP_KERNEL);
1762 spin_lock(&_minor_lock);
1764 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1766 spin_unlock(&_minor_lock);
1774 static const struct block_device_operations dm_blk_dops;
1775 static const struct block_device_operations dm_rq_blk_dops;
1776 static const struct dax_operations dm_dax_ops;
1778 static void dm_wq_work(struct work_struct *work);
1780 static void cleanup_mapped_device(struct mapped_device *md)
1783 destroy_workqueue(md->wq);
1784 bioset_exit(&md->bs);
1785 bioset_exit(&md->io_bs);
1788 kill_dax(md->dax_dev);
1789 put_dax(md->dax_dev);
1794 spin_lock(&_minor_lock);
1795 md->disk->private_data = NULL;
1796 spin_unlock(&_minor_lock);
1797 del_gendisk(md->disk);
1802 blk_cleanup_queue(md->queue);
1804 cleanup_srcu_struct(&md->io_barrier);
1811 mutex_destroy(&md->suspend_lock);
1812 mutex_destroy(&md->type_lock);
1813 mutex_destroy(&md->table_devices_lock);
1814 mutex_destroy(&md->swap_bios_lock);
1816 dm_mq_cleanup_mapped_device(md);
1820 * Allocate and initialise a blank device with a given minor.
1822 static struct mapped_device *alloc_dev(int minor)
1824 int r, numa_node_id = dm_get_numa_node();
1825 struct mapped_device *md;
1828 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1830 DMWARN("unable to allocate device, out of memory.");
1834 if (!try_module_get(THIS_MODULE))
1835 goto bad_module_get;
1837 /* get a minor number for the dev */
1838 if (minor == DM_ANY_MINOR)
1839 r = next_free_minor(&minor);
1841 r = specific_minor(minor);
1845 r = init_srcu_struct(&md->io_barrier);
1847 goto bad_io_barrier;
1849 md->numa_node_id = numa_node_id;
1850 md->init_tio_pdu = false;
1851 md->type = DM_TYPE_NONE;
1852 mutex_init(&md->suspend_lock);
1853 mutex_init(&md->type_lock);
1854 mutex_init(&md->table_devices_lock);
1855 spin_lock_init(&md->deferred_lock);
1856 atomic_set(&md->holders, 1);
1857 atomic_set(&md->open_count, 0);
1858 atomic_set(&md->event_nr, 0);
1859 atomic_set(&md->uevent_seq, 0);
1860 INIT_LIST_HEAD(&md->uevent_list);
1861 INIT_LIST_HEAD(&md->table_devices);
1862 spin_lock_init(&md->uevent_lock);
1865 * default to bio-based until DM table is loaded and md->type
1866 * established. If request-based table is loaded: blk-mq will
1867 * override accordingly.
1869 md->queue = blk_alloc_queue(numa_node_id);
1873 md->disk = alloc_disk_node(1, md->numa_node_id);
1877 init_waitqueue_head(&md->wait);
1878 INIT_WORK(&md->work, dm_wq_work);
1879 init_waitqueue_head(&md->eventq);
1880 init_completion(&md->kobj_holder.completion);
1882 md->swap_bios = get_swap_bios();
1883 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1884 mutex_init(&md->swap_bios_lock);
1886 md->disk->major = _major;
1887 md->disk->first_minor = minor;
1888 md->disk->fops = &dm_blk_dops;
1889 md->disk->queue = md->queue;
1890 md->disk->private_data = md;
1891 sprintf(md->disk->disk_name, "dm-%d", minor);
1893 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1894 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1896 if (IS_ERR(md->dax_dev)) {
1902 add_disk_no_queue_reg(md->disk);
1903 format_dev_t(md->name, MKDEV(_major, minor));
1905 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1909 md->bdev = bdget_disk(md->disk, 0);
1913 r = dm_stats_init(&md->stats);
1917 /* Populate the mapping, nobody knows we exist yet */
1918 spin_lock(&_minor_lock);
1919 old_md = idr_replace(&_minor_idr, md, minor);
1920 spin_unlock(&_minor_lock);
1922 BUG_ON(old_md != MINOR_ALLOCED);
1927 cleanup_mapped_device(md);
1931 module_put(THIS_MODULE);
1937 static void unlock_fs(struct mapped_device *md);
1939 static void free_dev(struct mapped_device *md)
1941 int minor = MINOR(disk_devt(md->disk));
1945 cleanup_mapped_device(md);
1947 free_table_devices(&md->table_devices);
1948 dm_stats_cleanup(&md->stats);
1951 module_put(THIS_MODULE);
1955 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1957 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1960 if (dm_table_bio_based(t)) {
1962 * The md may already have mempools that need changing.
1963 * If so, reload bioset because front_pad may have changed
1964 * because a different table was loaded.
1966 bioset_exit(&md->bs);
1967 bioset_exit(&md->io_bs);
1969 } else if (bioset_initialized(&md->bs)) {
1971 * There's no need to reload with request-based dm
1972 * because the size of front_pad doesn't change.
1973 * Note for future: If you are to reload bioset,
1974 * prep-ed requests in the queue may refer
1975 * to bio from the old bioset, so you must walk
1976 * through the queue to unprep.
1982 bioset_initialized(&md->bs) ||
1983 bioset_initialized(&md->io_bs));
1985 ret = bioset_init_from_src(&md->bs, &p->bs);
1988 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1990 bioset_exit(&md->bs);
1992 /* mempool bind completed, no longer need any mempools in the table */
1993 dm_table_free_md_mempools(t);
1998 * Bind a table to the device.
2000 static void event_callback(void *context)
2002 unsigned long flags;
2004 struct mapped_device *md = (struct mapped_device *) context;
2006 spin_lock_irqsave(&md->uevent_lock, flags);
2007 list_splice_init(&md->uevent_list, &uevents);
2008 spin_unlock_irqrestore(&md->uevent_lock, flags);
2010 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2012 atomic_inc(&md->event_nr);
2013 wake_up(&md->eventq);
2014 dm_issue_global_event();
2018 * Returns old map, which caller must destroy.
2020 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2021 struct queue_limits *limits)
2023 struct dm_table *old_map;
2024 struct request_queue *q = md->queue;
2025 bool request_based = dm_table_request_based(t);
2029 lockdep_assert_held(&md->suspend_lock);
2031 size = dm_table_get_size(t);
2034 * Wipe any geometry if the size of the table changed.
2036 if (size != dm_get_size(md))
2037 memset(&md->geometry, 0, sizeof(md->geometry));
2039 set_capacity(md->disk, size);
2040 bd_set_nr_sectors(md->bdev, size);
2042 dm_table_event_callback(t, event_callback, md);
2045 * The queue hasn't been stopped yet, if the old table type wasn't
2046 * for request-based during suspension. So stop it to prevent
2047 * I/O mapping before resume.
2048 * This must be done before setting the queue restrictions,
2049 * because request-based dm may be run just after the setting.
2054 if (request_based) {
2056 * Leverage the fact that request-based DM targets are
2057 * immutable singletons - used to optimize dm_mq_queue_rq.
2059 md->immutable_target = dm_table_get_immutable_target(t);
2062 ret = __bind_mempools(md, t);
2064 old_map = ERR_PTR(ret);
2068 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2069 rcu_assign_pointer(md->map, (void *)t);
2070 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2072 dm_table_set_restrictions(t, q, limits);
2081 * Returns unbound table for the caller to free.
2083 static struct dm_table *__unbind(struct mapped_device *md)
2085 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2090 dm_table_event_callback(map, NULL, NULL);
2091 RCU_INIT_POINTER(md->map, NULL);
2098 * Constructor for a new device.
2100 int dm_create(int minor, struct mapped_device **result)
2103 struct mapped_device *md;
2105 md = alloc_dev(minor);
2109 r = dm_sysfs_init(md);
2120 * Functions to manage md->type.
2121 * All are required to hold md->type_lock.
2123 void dm_lock_md_type(struct mapped_device *md)
2125 mutex_lock(&md->type_lock);
2128 void dm_unlock_md_type(struct mapped_device *md)
2130 mutex_unlock(&md->type_lock);
2133 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2135 BUG_ON(!mutex_is_locked(&md->type_lock));
2139 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2144 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2146 return md->immutable_target_type;
2150 * The queue_limits are only valid as long as you have a reference
2153 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2155 BUG_ON(!atomic_read(&md->holders));
2156 return &md->queue->limits;
2158 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2161 * Setup the DM device's queue based on md's type
2163 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2166 struct queue_limits limits;
2167 enum dm_queue_mode type = dm_get_md_type(md);
2170 case DM_TYPE_REQUEST_BASED:
2171 md->disk->fops = &dm_rq_blk_dops;
2172 r = dm_mq_init_request_queue(md, t);
2174 DMERR("Cannot initialize queue for request-based dm mapped device");
2178 case DM_TYPE_BIO_BASED:
2179 case DM_TYPE_DAX_BIO_BASED:
2186 r = dm_calculate_queue_limits(t, &limits);
2188 DMERR("Cannot calculate initial queue limits");
2191 dm_table_set_restrictions(t, md->queue, &limits);
2192 blk_register_queue(md->disk);
2197 struct mapped_device *dm_get_md(dev_t dev)
2199 struct mapped_device *md;
2200 unsigned minor = MINOR(dev);
2202 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2205 spin_lock(&_minor_lock);
2207 md = idr_find(&_minor_idr, minor);
2208 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2209 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2215 spin_unlock(&_minor_lock);
2219 EXPORT_SYMBOL_GPL(dm_get_md);
2221 void *dm_get_mdptr(struct mapped_device *md)
2223 return md->interface_ptr;
2226 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2228 md->interface_ptr = ptr;
2231 void dm_get(struct mapped_device *md)
2233 atomic_inc(&md->holders);
2234 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2237 int dm_hold(struct mapped_device *md)
2239 spin_lock(&_minor_lock);
2240 if (test_bit(DMF_FREEING, &md->flags)) {
2241 spin_unlock(&_minor_lock);
2245 spin_unlock(&_minor_lock);
2248 EXPORT_SYMBOL_GPL(dm_hold);
2250 const char *dm_device_name(struct mapped_device *md)
2254 EXPORT_SYMBOL_GPL(dm_device_name);
2256 static void __dm_destroy(struct mapped_device *md, bool wait)
2258 struct dm_table *map;
2263 spin_lock(&_minor_lock);
2264 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2265 set_bit(DMF_FREEING, &md->flags);
2266 spin_unlock(&_minor_lock);
2268 blk_set_queue_dying(md->queue);
2271 * Take suspend_lock so that presuspend and postsuspend methods
2272 * do not race with internal suspend.
2274 mutex_lock(&md->suspend_lock);
2275 map = dm_get_live_table(md, &srcu_idx);
2276 if (!dm_suspended_md(md)) {
2277 dm_table_presuspend_targets(map);
2278 set_bit(DMF_SUSPENDED, &md->flags);
2279 set_bit(DMF_POST_SUSPENDING, &md->flags);
2280 dm_table_postsuspend_targets(map);
2282 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2283 dm_put_live_table(md, srcu_idx);
2284 mutex_unlock(&md->suspend_lock);
2287 * Rare, but there may be I/O requests still going to complete,
2288 * for example. Wait for all references to disappear.
2289 * No one should increment the reference count of the mapped_device,
2290 * after the mapped_device state becomes DMF_FREEING.
2293 while (atomic_read(&md->holders))
2295 else if (atomic_read(&md->holders))
2296 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2297 dm_device_name(md), atomic_read(&md->holders));
2300 dm_table_destroy(__unbind(md));
2304 void dm_destroy(struct mapped_device *md)
2306 __dm_destroy(md, true);
2309 void dm_destroy_immediate(struct mapped_device *md)
2311 __dm_destroy(md, false);
2314 void dm_put(struct mapped_device *md)
2316 atomic_dec(&md->holders);
2318 EXPORT_SYMBOL_GPL(dm_put);
2320 static bool md_in_flight_bios(struct mapped_device *md)
2323 struct hd_struct *part = &dm_disk(md)->part0;
2326 for_each_possible_cpu(cpu) {
2327 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2328 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2334 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2340 prepare_to_wait(&md->wait, &wait, task_state);
2342 if (!md_in_flight_bios(md))
2345 if (signal_pending_state(task_state, current)) {
2352 finish_wait(&md->wait, &wait);
2359 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2363 if (!queue_is_mq(md->queue))
2364 return dm_wait_for_bios_completion(md, task_state);
2367 if (!blk_mq_queue_inflight(md->queue))
2370 if (signal_pending_state(task_state, current)) {
2382 * Process the deferred bios
2384 static void dm_wq_work(struct work_struct *work)
2386 struct mapped_device *md = container_of(work, struct mapped_device, work);
2389 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2390 spin_lock_irq(&md->deferred_lock);
2391 bio = bio_list_pop(&md->deferred);
2392 spin_unlock_irq(&md->deferred_lock);
2397 submit_bio_noacct(bio);
2402 static void dm_queue_flush(struct mapped_device *md)
2404 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2405 smp_mb__after_atomic();
2406 queue_work(md->wq, &md->work);
2410 * Swap in a new table, returning the old one for the caller to destroy.
2412 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2414 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2415 struct queue_limits limits;
2418 mutex_lock(&md->suspend_lock);
2420 /* device must be suspended */
2421 if (!dm_suspended_md(md))
2425 * If the new table has no data devices, retain the existing limits.
2426 * This helps multipath with queue_if_no_path if all paths disappear,
2427 * then new I/O is queued based on these limits, and then some paths
2430 if (dm_table_has_no_data_devices(table)) {
2431 live_map = dm_get_live_table_fast(md);
2433 limits = md->queue->limits;
2434 dm_put_live_table_fast(md);
2438 r = dm_calculate_queue_limits(table, &limits);
2445 map = __bind(md, table, &limits);
2446 dm_issue_global_event();
2449 mutex_unlock(&md->suspend_lock);
2454 * Functions to lock and unlock any filesystem running on the
2457 static int lock_fs(struct mapped_device *md)
2461 WARN_ON(md->frozen_sb);
2463 md->frozen_sb = freeze_bdev(md->bdev);
2464 if (IS_ERR(md->frozen_sb)) {
2465 r = PTR_ERR(md->frozen_sb);
2466 md->frozen_sb = NULL;
2470 set_bit(DMF_FROZEN, &md->flags);
2475 static void unlock_fs(struct mapped_device *md)
2477 if (!test_bit(DMF_FROZEN, &md->flags))
2480 thaw_bdev(md->bdev, md->frozen_sb);
2481 md->frozen_sb = NULL;
2482 clear_bit(DMF_FROZEN, &md->flags);
2486 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2487 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2488 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2490 * If __dm_suspend returns 0, the device is completely quiescent
2491 * now. There is no request-processing activity. All new requests
2492 * are being added to md->deferred list.
2494 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2495 unsigned suspend_flags, long task_state,
2496 int dmf_suspended_flag)
2498 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2499 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2502 lockdep_assert_held(&md->suspend_lock);
2505 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2506 * This flag is cleared before dm_suspend returns.
2509 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2511 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2514 * This gets reverted if there's an error later and the targets
2515 * provide the .presuspend_undo hook.
2517 dm_table_presuspend_targets(map);
2520 * Flush I/O to the device.
2521 * Any I/O submitted after lock_fs() may not be flushed.
2522 * noflush takes precedence over do_lockfs.
2523 * (lock_fs() flushes I/Os and waits for them to complete.)
2525 if (!noflush && do_lockfs) {
2528 dm_table_presuspend_undo_targets(map);
2534 * Here we must make sure that no processes are submitting requests
2535 * to target drivers i.e. no one may be executing
2536 * __split_and_process_bio from dm_submit_bio.
2538 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2539 * we take the write lock. To prevent any process from reentering
2540 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2541 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2542 * flush_workqueue(md->wq).
2544 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2546 synchronize_srcu(&md->io_barrier);
2549 * Stop md->queue before flushing md->wq in case request-based
2550 * dm defers requests to md->wq from md->queue.
2552 if (dm_request_based(md))
2553 dm_stop_queue(md->queue);
2555 flush_workqueue(md->wq);
2558 * At this point no more requests are entering target request routines.
2559 * We call dm_wait_for_completion to wait for all existing requests
2562 r = dm_wait_for_completion(md, task_state);
2564 set_bit(dmf_suspended_flag, &md->flags);
2567 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2569 synchronize_srcu(&md->io_barrier);
2571 /* were we interrupted ? */
2575 if (dm_request_based(md))
2576 dm_start_queue(md->queue);
2579 dm_table_presuspend_undo_targets(map);
2580 /* pushback list is already flushed, so skip flush */
2587 * We need to be able to change a mapping table under a mounted
2588 * filesystem. For example we might want to move some data in
2589 * the background. Before the table can be swapped with
2590 * dm_bind_table, dm_suspend must be called to flush any in
2591 * flight bios and ensure that any further io gets deferred.
2594 * Suspend mechanism in request-based dm.
2596 * 1. Flush all I/Os by lock_fs() if needed.
2597 * 2. Stop dispatching any I/O by stopping the request_queue.
2598 * 3. Wait for all in-flight I/Os to be completed or requeued.
2600 * To abort suspend, start the request_queue.
2602 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2604 struct dm_table *map = NULL;
2608 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2610 if (dm_suspended_md(md)) {
2615 if (dm_suspended_internally_md(md)) {
2616 /* already internally suspended, wait for internal resume */
2617 mutex_unlock(&md->suspend_lock);
2618 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2624 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2626 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2630 set_bit(DMF_POST_SUSPENDING, &md->flags);
2631 dm_table_postsuspend_targets(map);
2632 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2635 mutex_unlock(&md->suspend_lock);
2639 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2642 int r = dm_table_resume_targets(map);
2650 * Flushing deferred I/Os must be done after targets are resumed
2651 * so that mapping of targets can work correctly.
2652 * Request-based dm is queueing the deferred I/Os in its request_queue.
2654 if (dm_request_based(md))
2655 dm_start_queue(md->queue);
2662 int dm_resume(struct mapped_device *md)
2665 struct dm_table *map = NULL;
2669 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2671 if (!dm_suspended_md(md))
2674 if (dm_suspended_internally_md(md)) {
2675 /* already internally suspended, wait for internal resume */
2676 mutex_unlock(&md->suspend_lock);
2677 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2683 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2684 if (!map || !dm_table_get_size(map))
2687 r = __dm_resume(md, map);
2691 clear_bit(DMF_SUSPENDED, &md->flags);
2693 mutex_unlock(&md->suspend_lock);
2699 * Internal suspend/resume works like userspace-driven suspend. It waits
2700 * until all bios finish and prevents issuing new bios to the target drivers.
2701 * It may be used only from the kernel.
2704 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2706 struct dm_table *map = NULL;
2708 lockdep_assert_held(&md->suspend_lock);
2710 if (md->internal_suspend_count++)
2711 return; /* nested internal suspend */
2713 if (dm_suspended_md(md)) {
2714 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2715 return; /* nest suspend */
2718 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2721 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2722 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2723 * would require changing .presuspend to return an error -- avoid this
2724 * until there is a need for more elaborate variants of internal suspend.
2726 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2727 DMF_SUSPENDED_INTERNALLY);
2729 set_bit(DMF_POST_SUSPENDING, &md->flags);
2730 dm_table_postsuspend_targets(map);
2731 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2734 static void __dm_internal_resume(struct mapped_device *md)
2737 struct dm_table *map;
2739 BUG_ON(!md->internal_suspend_count);
2741 if (--md->internal_suspend_count)
2742 return; /* resume from nested internal suspend */
2744 if (dm_suspended_md(md))
2745 goto done; /* resume from nested suspend */
2747 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2748 r = __dm_resume(md, map);
2751 * If a preresume method of some target failed, we are in a
2752 * tricky situation. We can't return an error to the caller. We
2753 * can't fake success because then the "resume" and
2754 * "postsuspend" methods would not be paired correctly, and it
2755 * would break various targets, for example it would cause list
2756 * corruption in the "origin" target.
2758 * So, we fake normal suspend here, to make sure that the
2759 * "resume" and "postsuspend" methods will be paired correctly.
2761 DMERR("Preresume method failed: %d", r);
2762 set_bit(DMF_SUSPENDED, &md->flags);
2765 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2766 smp_mb__after_atomic();
2767 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2770 void dm_internal_suspend_noflush(struct mapped_device *md)
2772 mutex_lock(&md->suspend_lock);
2773 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2774 mutex_unlock(&md->suspend_lock);
2776 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2778 void dm_internal_resume(struct mapped_device *md)
2780 mutex_lock(&md->suspend_lock);
2781 __dm_internal_resume(md);
2782 mutex_unlock(&md->suspend_lock);
2784 EXPORT_SYMBOL_GPL(dm_internal_resume);
2787 * Fast variants of internal suspend/resume hold md->suspend_lock,
2788 * which prevents interaction with userspace-driven suspend.
2791 void dm_internal_suspend_fast(struct mapped_device *md)
2793 mutex_lock(&md->suspend_lock);
2794 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2797 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2798 synchronize_srcu(&md->io_barrier);
2799 flush_workqueue(md->wq);
2800 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2802 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2804 void dm_internal_resume_fast(struct mapped_device *md)
2806 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2812 mutex_unlock(&md->suspend_lock);
2814 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2816 /*-----------------------------------------------------------------
2817 * Event notification.
2818 *---------------------------------------------------------------*/
2819 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2824 char udev_cookie[DM_COOKIE_LENGTH];
2825 char *envp[] = { udev_cookie, NULL };
2827 noio_flag = memalloc_noio_save();
2830 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2832 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2833 DM_COOKIE_ENV_VAR_NAME, cookie);
2834 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2838 memalloc_noio_restore(noio_flag);
2843 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2845 return atomic_add_return(1, &md->uevent_seq);
2848 uint32_t dm_get_event_nr(struct mapped_device *md)
2850 return atomic_read(&md->event_nr);
2853 int dm_wait_event(struct mapped_device *md, int event_nr)
2855 return wait_event_interruptible(md->eventq,
2856 (event_nr != atomic_read(&md->event_nr)));
2859 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2861 unsigned long flags;
2863 spin_lock_irqsave(&md->uevent_lock, flags);
2864 list_add(elist, &md->uevent_list);
2865 spin_unlock_irqrestore(&md->uevent_lock, flags);
2869 * The gendisk is only valid as long as you have a reference
2872 struct gendisk *dm_disk(struct mapped_device *md)
2876 EXPORT_SYMBOL_GPL(dm_disk);
2878 struct kobject *dm_kobject(struct mapped_device *md)
2880 return &md->kobj_holder.kobj;
2883 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2885 struct mapped_device *md;
2887 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2889 spin_lock(&_minor_lock);
2890 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2896 spin_unlock(&_minor_lock);
2901 int dm_suspended_md(struct mapped_device *md)
2903 return test_bit(DMF_SUSPENDED, &md->flags);
2906 static int dm_post_suspending_md(struct mapped_device *md)
2908 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2911 int dm_suspended_internally_md(struct mapped_device *md)
2913 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2916 int dm_test_deferred_remove_flag(struct mapped_device *md)
2918 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2921 int dm_suspended(struct dm_target *ti)
2923 return dm_suspended_md(ti->table->md);
2925 EXPORT_SYMBOL_GPL(dm_suspended);
2927 int dm_post_suspending(struct dm_target *ti)
2929 return dm_post_suspending_md(ti->table->md);
2931 EXPORT_SYMBOL_GPL(dm_post_suspending);
2933 int dm_noflush_suspending(struct dm_target *ti)
2935 return __noflush_suspending(ti->table->md);
2937 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2939 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2940 unsigned integrity, unsigned per_io_data_size,
2941 unsigned min_pool_size)
2943 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2944 unsigned int pool_size = 0;
2945 unsigned int front_pad, io_front_pad;
2952 case DM_TYPE_BIO_BASED:
2953 case DM_TYPE_DAX_BIO_BASED:
2954 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2955 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2956 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2957 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2960 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2963 case DM_TYPE_REQUEST_BASED:
2964 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2965 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2966 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2972 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2976 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2982 dm_free_md_mempools(pools);
2987 void dm_free_md_mempools(struct dm_md_mempools *pools)
2992 bioset_exit(&pools->bs);
2993 bioset_exit(&pools->io_bs);
3005 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3008 struct mapped_device *md = bdev->bd_disk->private_data;
3009 struct dm_table *table;
3010 struct dm_target *ti;
3011 int ret = -ENOTTY, srcu_idx;
3013 table = dm_get_live_table(md, &srcu_idx);
3014 if (!table || !dm_table_get_size(table))
3017 /* We only support devices that have a single target */
3018 if (dm_table_get_num_targets(table) != 1)
3020 ti = dm_table_get_target(table, 0);
3022 if (dm_suspended_md(md)) {
3028 if (!ti->type->iterate_devices)
3031 ret = ti->type->iterate_devices(ti, fn, data);
3033 dm_put_live_table(md, srcu_idx);
3038 * For register / unregister we need to manually call out to every path.
3040 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3041 sector_t start, sector_t len, void *data)
3043 struct dm_pr *pr = data;
3044 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3046 if (!ops || !ops->pr_register)
3048 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3051 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3062 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3063 if (ret && new_key) {
3064 /* unregister all paths if we failed to register any path */
3065 pr.old_key = new_key;
3068 pr.fail_early = false;
3069 dm_call_pr(bdev, __dm_pr_register, &pr);
3075 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3078 struct mapped_device *md = bdev->bd_disk->private_data;
3079 const struct pr_ops *ops;
3082 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3086 ops = bdev->bd_disk->fops->pr_ops;
3087 if (ops && ops->pr_reserve)
3088 r = ops->pr_reserve(bdev, key, type, flags);
3092 dm_unprepare_ioctl(md, srcu_idx);
3096 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3098 struct mapped_device *md = bdev->bd_disk->private_data;
3099 const struct pr_ops *ops;
3102 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3106 ops = bdev->bd_disk->fops->pr_ops;
3107 if (ops && ops->pr_release)
3108 r = ops->pr_release(bdev, key, type);
3112 dm_unprepare_ioctl(md, srcu_idx);
3116 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3117 enum pr_type type, bool abort)
3119 struct mapped_device *md = bdev->bd_disk->private_data;
3120 const struct pr_ops *ops;
3123 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3127 ops = bdev->bd_disk->fops->pr_ops;
3128 if (ops && ops->pr_preempt)
3129 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3133 dm_unprepare_ioctl(md, srcu_idx);
3137 static int dm_pr_clear(struct block_device *bdev, u64 key)
3139 struct mapped_device *md = bdev->bd_disk->private_data;
3140 const struct pr_ops *ops;
3143 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3147 ops = bdev->bd_disk->fops->pr_ops;
3148 if (ops && ops->pr_clear)
3149 r = ops->pr_clear(bdev, key);
3153 dm_unprepare_ioctl(md, srcu_idx);
3157 static const struct pr_ops dm_pr_ops = {
3158 .pr_register = dm_pr_register,
3159 .pr_reserve = dm_pr_reserve,
3160 .pr_release = dm_pr_release,
3161 .pr_preempt = dm_pr_preempt,
3162 .pr_clear = dm_pr_clear,
3165 static const struct block_device_operations dm_blk_dops = {
3166 .submit_bio = dm_submit_bio,
3167 .open = dm_blk_open,
3168 .release = dm_blk_close,
3169 .ioctl = dm_blk_ioctl,
3170 .getgeo = dm_blk_getgeo,
3171 .report_zones = dm_blk_report_zones,
3172 .pr_ops = &dm_pr_ops,
3173 .owner = THIS_MODULE
3176 static const struct block_device_operations dm_rq_blk_dops = {
3177 .open = dm_blk_open,
3178 .release = dm_blk_close,
3179 .ioctl = dm_blk_ioctl,
3180 .getgeo = dm_blk_getgeo,
3181 .pr_ops = &dm_pr_ops,
3182 .owner = THIS_MODULE
3185 static const struct dax_operations dm_dax_ops = {
3186 .direct_access = dm_dax_direct_access,
3187 .dax_supported = dm_dax_supported,
3188 .copy_from_iter = dm_dax_copy_from_iter,
3189 .copy_to_iter = dm_dax_copy_to_iter,
3190 .zero_page_range = dm_dax_zero_page_range,
3196 module_init(dm_init);
3197 module_exit(dm_exit);
3199 module_param(major, uint, 0);
3200 MODULE_PARM_DESC(major, "The major number of the device mapper");
3202 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3203 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3205 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3206 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3208 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3209 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3211 MODULE_DESCRIPTION(DM_NAME " driver");
3212 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3213 MODULE_LICENSE("GPL");