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 flush_scheduled_work();
269 destroy_workqueue(deferred_remove_workqueue);
271 unregister_blkdev(_major, _name);
276 DMINFO("cleaned up");
279 static int (*_inits[])(void) __initdata = {
290 static void (*_exits[])(void) = {
301 static int __init dm_init(void)
303 const int count = ARRAY_SIZE(_inits);
307 for (i = 0; i < count; i++) {
322 static void __exit dm_exit(void)
324 int i = ARRAY_SIZE(_exits);
330 * Should be empty by this point.
332 idr_destroy(&_minor_idr);
336 * Block device functions
338 int dm_deleting_md(struct mapped_device *md)
340 return test_bit(DMF_DELETING, &md->flags);
343 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
345 struct mapped_device *md;
347 spin_lock(&_minor_lock);
349 md = bdev->bd_disk->private_data;
353 if (test_bit(DMF_FREEING, &md->flags) ||
354 dm_deleting_md(md)) {
360 atomic_inc(&md->open_count);
362 spin_unlock(&_minor_lock);
364 return md ? 0 : -ENXIO;
367 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
369 struct mapped_device *md;
371 spin_lock(&_minor_lock);
373 md = disk->private_data;
377 if (atomic_dec_and_test(&md->open_count) &&
378 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
379 queue_work(deferred_remove_workqueue, &deferred_remove_work);
383 spin_unlock(&_minor_lock);
386 int dm_open_count(struct mapped_device *md)
388 return atomic_read(&md->open_count);
392 * Guarantees nothing is using the device before it's deleted.
394 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
398 spin_lock(&_minor_lock);
400 if (dm_open_count(md)) {
403 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
404 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
407 set_bit(DMF_DELETING, &md->flags);
409 spin_unlock(&_minor_lock);
414 int dm_cancel_deferred_remove(struct mapped_device *md)
418 spin_lock(&_minor_lock);
420 if (test_bit(DMF_DELETING, &md->flags))
423 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
425 spin_unlock(&_minor_lock);
430 static void do_deferred_remove(struct work_struct *w)
432 dm_deferred_remove();
435 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
437 struct mapped_device *md = bdev->bd_disk->private_data;
439 return dm_get_geometry(md, geo);
442 #ifdef CONFIG_BLK_DEV_ZONED
443 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
445 struct dm_report_zones_args *args = data;
446 sector_t sector_diff = args->tgt->begin - args->start;
449 * Ignore zones beyond the target range.
451 if (zone->start >= args->start + args->tgt->len)
455 * Remap the start sector and write pointer position of the zone
456 * to match its position in the target range.
458 zone->start += sector_diff;
459 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
460 if (zone->cond == BLK_ZONE_COND_FULL)
461 zone->wp = zone->start + zone->len;
462 else if (zone->cond == BLK_ZONE_COND_EMPTY)
463 zone->wp = zone->start;
465 zone->wp += sector_diff;
468 args->next_sector = zone->start + zone->len;
469 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
471 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
473 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
474 unsigned int nr_zones, report_zones_cb cb, void *data)
476 struct mapped_device *md = disk->private_data;
477 struct dm_table *map;
479 struct dm_report_zones_args args = {
480 .next_sector = sector,
485 if (dm_suspended_md(md))
488 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,
505 nr_zones - args.zone_idx);
508 } while (args.zone_idx < nr_zones &&
509 args.next_sector < get_capacity(disk));
513 dm_put_live_table(md, srcu_idx);
517 #define dm_blk_report_zones NULL
518 #endif /* CONFIG_BLK_DEV_ZONED */
520 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
521 struct block_device **bdev)
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)
556 dm_put_live_table(md, srcu_idx);
559 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
560 unsigned int cmd, unsigned long arg)
562 struct mapped_device *md = bdev->bd_disk->private_data;
565 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
571 * Target determined this ioctl is being issued against a
572 * subset of the parent bdev; require extra privileges.
574 if (!capable(CAP_SYS_RAWIO)) {
576 "%s: sending ioctl %x to DM device without required privilege.",
583 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
585 dm_unprepare_ioctl(md, srcu_idx);
589 u64 dm_start_time_ns_from_clone(struct bio *bio)
591 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
592 struct dm_io *io = tio->io;
594 return jiffies_to_nsecs(io->start_time);
596 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
598 static void start_io_acct(struct dm_io *io)
600 struct mapped_device *md = io->md;
601 struct bio *bio = io->orig_bio;
603 io->start_time = bio_start_io_acct(bio);
604 if (unlikely(dm_stats_used(&md->stats)))
605 dm_stats_account_io(&md->stats, bio_data_dir(bio),
606 bio->bi_iter.bi_sector, bio_sectors(bio),
607 false, 0, &io->stats_aux);
610 static void end_io_acct(struct mapped_device *md, struct bio *bio,
611 unsigned long start_time, struct dm_stats_aux *stats_aux)
613 unsigned long duration = jiffies - start_time;
615 if (unlikely(dm_stats_used(&md->stats)))
616 dm_stats_account_io(&md->stats, bio_data_dir(bio),
617 bio->bi_iter.bi_sector, bio_sectors(bio),
618 true, duration, stats_aux);
622 bio_end_io_acct(bio, start_time);
624 /* nudge anyone waiting on suspend queue */
625 if (unlikely(wq_has_sleeper(&md->wait)))
629 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
632 struct dm_target_io *tio;
635 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
639 tio = container_of(clone, struct dm_target_io, clone);
640 tio->inside_dm_io = true;
643 io = container_of(tio, struct dm_io, tio);
644 io->magic = DM_IO_MAGIC;
646 atomic_set(&io->io_count, 1);
649 spin_lock_init(&io->endio_lock);
656 static void free_io(struct mapped_device *md, struct dm_io *io)
658 bio_put(&io->tio.clone);
661 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
662 unsigned target_bio_nr, gfp_t gfp_mask)
664 struct dm_target_io *tio;
666 if (!ci->io->tio.io) {
667 /* the dm_target_io embedded in ci->io is available */
670 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
674 tio = container_of(clone, struct dm_target_io, clone);
675 tio->inside_dm_io = false;
678 tio->magic = DM_TIO_MAGIC;
681 tio->target_bio_nr = target_bio_nr;
686 static void free_tio(struct dm_target_io *tio)
688 if (tio->inside_dm_io)
690 bio_put(&tio->clone);
694 * Add the bio to the list of deferred io.
696 static void queue_io(struct mapped_device *md, struct bio *bio)
700 spin_lock_irqsave(&md->deferred_lock, flags);
701 bio_list_add(&md->deferred, bio);
702 spin_unlock_irqrestore(&md->deferred_lock, flags);
703 queue_work(md->wq, &md->work);
707 * Everyone (including functions in this file), should use this
708 * function to access the md->map field, and make sure they call
709 * dm_put_live_table() when finished.
711 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
713 *srcu_idx = srcu_read_lock(&md->io_barrier);
715 return srcu_dereference(md->map, &md->io_barrier);
718 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
720 srcu_read_unlock(&md->io_barrier, srcu_idx);
723 void dm_sync_table(struct mapped_device *md)
725 synchronize_srcu(&md->io_barrier);
726 synchronize_rcu_expedited();
730 * A fast alternative to dm_get_live_table/dm_put_live_table.
731 * The caller must not block between these two functions.
733 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
736 return rcu_dereference(md->map);
739 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
744 static char *_dm_claim_ptr = "I belong to device-mapper";
747 * Open a table device so we can use it as a map destination.
749 static int open_table_device(struct table_device *td, dev_t dev,
750 struct mapped_device *md)
752 struct block_device *bdev;
756 BUG_ON(td->dm_dev.bdev);
758 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
760 return PTR_ERR(bdev);
762 r = bd_link_disk_holder(bdev, dm_disk(md));
764 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
768 td->dm_dev.bdev = bdev;
769 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
774 * Close a table device that we've been using.
776 static void close_table_device(struct table_device *td, struct mapped_device *md)
778 if (!td->dm_dev.bdev)
781 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
782 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
783 put_dax(td->dm_dev.dax_dev);
784 td->dm_dev.bdev = NULL;
785 td->dm_dev.dax_dev = NULL;
788 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
791 struct table_device *td;
793 list_for_each_entry(td, l, list)
794 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
800 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
801 struct dm_dev **result)
804 struct table_device *td;
806 mutex_lock(&md->table_devices_lock);
807 td = find_table_device(&md->table_devices, dev, mode);
809 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
811 mutex_unlock(&md->table_devices_lock);
815 td->dm_dev.mode = mode;
816 td->dm_dev.bdev = NULL;
818 if ((r = open_table_device(td, dev, md))) {
819 mutex_unlock(&md->table_devices_lock);
824 format_dev_t(td->dm_dev.name, dev);
826 refcount_set(&td->count, 1);
827 list_add(&td->list, &md->table_devices);
829 refcount_inc(&td->count);
831 mutex_unlock(&md->table_devices_lock);
833 *result = &td->dm_dev;
836 EXPORT_SYMBOL_GPL(dm_get_table_device);
838 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
840 struct table_device *td = container_of(d, struct table_device, dm_dev);
842 mutex_lock(&md->table_devices_lock);
843 if (refcount_dec_and_test(&td->count)) {
844 close_table_device(td, md);
848 mutex_unlock(&md->table_devices_lock);
850 EXPORT_SYMBOL(dm_put_table_device);
852 static void free_table_devices(struct list_head *devices)
854 struct list_head *tmp, *next;
856 list_for_each_safe(tmp, next, devices) {
857 struct table_device *td = list_entry(tmp, struct table_device, list);
859 DMWARN("dm_destroy: %s still exists with %d references",
860 td->dm_dev.name, refcount_read(&td->count));
866 * Get the geometry associated with a dm device
868 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
876 * Set the geometry of a device.
878 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
880 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
882 if (geo->start > sz) {
883 DMWARN("Start sector is beyond the geometry limits.");
892 static int __noflush_suspending(struct mapped_device *md)
894 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
898 * Decrements the number of outstanding ios that a bio has been
899 * cloned into, completing the original io if necc.
901 static void dec_pending(struct dm_io *io, blk_status_t error)
904 blk_status_t io_error;
906 struct mapped_device *md = io->md;
907 unsigned long start_time = 0;
908 struct dm_stats_aux stats_aux;
910 /* Push-back supersedes any I/O errors */
911 if (unlikely(error)) {
912 spin_lock_irqsave(&io->endio_lock, flags);
913 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
915 spin_unlock_irqrestore(&io->endio_lock, flags);
918 if (atomic_dec_and_test(&io->io_count)) {
919 if (io->status == BLK_STS_DM_REQUEUE) {
921 * Target requested pushing back the I/O.
923 spin_lock_irqsave(&md->deferred_lock, flags);
924 if (__noflush_suspending(md))
925 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
926 bio_list_add_head(&md->deferred, io->orig_bio);
928 /* noflush suspend was interrupted. */
929 io->status = BLK_STS_IOERR;
930 spin_unlock_irqrestore(&md->deferred_lock, flags);
933 io_error = io->status;
935 start_time = io->start_time;
936 stats_aux = io->stats_aux;
938 end_io_acct(md, bio, start_time, &stats_aux);
940 if (io_error == BLK_STS_DM_REQUEUE)
943 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
945 * Preflush done for flush with data, reissue
946 * without REQ_PREFLUSH.
948 bio->bi_opf &= ~REQ_PREFLUSH;
951 /* done with normal IO or empty flush */
953 bio->bi_status = io_error;
959 void disable_discard(struct mapped_device *md)
961 struct queue_limits *limits = dm_get_queue_limits(md);
963 /* device doesn't really support DISCARD, disable it */
964 limits->max_discard_sectors = 0;
965 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
968 void disable_write_same(struct mapped_device *md)
970 struct queue_limits *limits = dm_get_queue_limits(md);
972 /* device doesn't really support WRITE SAME, disable it */
973 limits->max_write_same_sectors = 0;
976 void disable_write_zeroes(struct mapped_device *md)
978 struct queue_limits *limits = dm_get_queue_limits(md);
980 /* device doesn't really support WRITE ZEROES, disable it */
981 limits->max_write_zeroes_sectors = 0;
984 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
986 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
989 static void clone_endio(struct bio *bio)
991 blk_status_t error = bio->bi_status;
992 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
993 struct dm_io *io = tio->io;
994 struct mapped_device *md = tio->io->md;
995 dm_endio_fn endio = tio->ti->type->end_io;
996 struct bio *orig_bio = io->orig_bio;
998 if (unlikely(error == BLK_STS_TARGET)) {
999 if (bio_op(bio) == REQ_OP_DISCARD &&
1000 !bio->bi_disk->queue->limits.max_discard_sectors)
1001 disable_discard(md);
1002 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1003 !bio->bi_disk->queue->limits.max_write_same_sectors)
1004 disable_write_same(md);
1005 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1006 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1007 disable_write_zeroes(md);
1011 * For zone-append bios get offset in zone of the written
1012 * sector and add that to the original bio sector pos.
1014 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1015 sector_t written_sector = bio->bi_iter.bi_sector;
1016 struct request_queue *q = orig_bio->bi_disk->queue;
1017 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1019 orig_bio->bi_iter.bi_sector += written_sector & mask;
1023 int r = endio(tio->ti, bio, &error);
1025 case DM_ENDIO_REQUEUE:
1026 error = BLK_STS_DM_REQUEUE;
1030 case DM_ENDIO_INCOMPLETE:
1031 /* The target will handle the io */
1034 DMWARN("unimplemented target endio return value: %d", r);
1039 if (unlikely(swap_bios_limit(tio->ti, bio))) {
1040 struct mapped_device *md = io->md;
1041 up(&md->swap_bios_semaphore);
1045 dec_pending(io, error);
1049 * Return maximum size of I/O possible at the supplied sector up to the current
1052 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1053 sector_t target_offset)
1055 return ti->len - target_offset;
1058 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1060 sector_t target_offset = dm_target_offset(ti, sector);
1061 sector_t len = max_io_len_target_boundary(ti, target_offset);
1065 * Does the target need to split IO even further?
1066 * - varied (per target) IO splitting is a tenet of DM; this
1067 * explains why stacked chunk_sectors based splitting via
1068 * blk_max_size_offset() isn't possible here. So pass in
1069 * ti->max_io_len to override stacked chunk_sectors.
1071 if (ti->max_io_len) {
1072 max_len = blk_max_size_offset(ti->table->md->queue,
1073 target_offset, ti->max_io_len);
1081 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1083 if (len > UINT_MAX) {
1084 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1085 (unsigned long long)len, UINT_MAX);
1086 ti->error = "Maximum size of target IO is too large";
1090 ti->max_io_len = (uint32_t) len;
1094 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1096 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1097 sector_t sector, int *srcu_idx)
1098 __acquires(md->io_barrier)
1100 struct dm_table *map;
1101 struct dm_target *ti;
1103 map = dm_get_live_table(md, srcu_idx);
1107 ti = dm_table_find_target(map, sector);
1114 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1115 long nr_pages, void **kaddr, pfn_t *pfn)
1117 struct mapped_device *md = dax_get_private(dax_dev);
1118 sector_t sector = pgoff * PAGE_SECTORS;
1119 struct dm_target *ti;
1120 long len, ret = -EIO;
1123 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1127 if (!ti->type->direct_access)
1129 len = max_io_len(ti, sector) / PAGE_SECTORS;
1132 nr_pages = min(len, nr_pages);
1133 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1136 dm_put_live_table(md, srcu_idx);
1141 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1142 int blocksize, sector_t start, sector_t len)
1144 struct mapped_device *md = dax_get_private(dax_dev);
1145 struct dm_table *map;
1149 map = dm_get_live_table(md, &srcu_idx);
1153 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1156 dm_put_live_table(md, srcu_idx);
1161 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1162 void *addr, size_t bytes, struct iov_iter *i)
1164 struct mapped_device *md = dax_get_private(dax_dev);
1165 sector_t sector = pgoff * PAGE_SECTORS;
1166 struct dm_target *ti;
1170 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1174 if (!ti->type->dax_copy_from_iter) {
1175 ret = copy_from_iter(addr, bytes, i);
1178 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1180 dm_put_live_table(md, srcu_idx);
1185 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1186 void *addr, size_t bytes, struct iov_iter *i)
1188 struct mapped_device *md = dax_get_private(dax_dev);
1189 sector_t sector = pgoff * PAGE_SECTORS;
1190 struct dm_target *ti;
1194 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1198 if (!ti->type->dax_copy_to_iter) {
1199 ret = copy_to_iter(addr, bytes, i);
1202 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1204 dm_put_live_table(md, srcu_idx);
1209 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1212 struct mapped_device *md = dax_get_private(dax_dev);
1213 sector_t sector = pgoff * PAGE_SECTORS;
1214 struct dm_target *ti;
1218 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1222 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1224 * ->zero_page_range() is mandatory dax operation. If we are
1225 * here, something is wrong.
1229 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1231 dm_put_live_table(md, srcu_idx);
1237 * A target may call dm_accept_partial_bio only from the map routine. It is
1238 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1239 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1241 * dm_accept_partial_bio informs the dm that the target only wants to process
1242 * additional n_sectors sectors of the bio and the rest of the data should be
1243 * sent in a next bio.
1245 * A diagram that explains the arithmetics:
1246 * +--------------------+---------------+-------+
1248 * +--------------------+---------------+-------+
1250 * <-------------- *tio->len_ptr --------------->
1251 * <------- bi_size ------->
1254 * Region 1 was already iterated over with bio_advance or similar function.
1255 * (it may be empty if the target doesn't use bio_advance)
1256 * Region 2 is the remaining bio size that the target wants to process.
1257 * (it may be empty if region 1 is non-empty, although there is no reason
1259 * The target requires that region 3 is to be sent in the next bio.
1261 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1262 * the partially processed part (the sum of regions 1+2) must be the same for all
1263 * copies of the bio.
1265 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1267 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1268 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1270 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1271 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1272 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1273 BUG_ON(bi_size > *tio->len_ptr);
1274 BUG_ON(n_sectors > bi_size);
1276 *tio->len_ptr -= bi_size - n_sectors;
1277 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1279 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1281 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1283 mutex_lock(&md->swap_bios_lock);
1284 while (latch < md->swap_bios) {
1286 down(&md->swap_bios_semaphore);
1289 while (latch > md->swap_bios) {
1291 up(&md->swap_bios_semaphore);
1294 mutex_unlock(&md->swap_bios_lock);
1297 static blk_qc_t __map_bio(struct dm_target_io *tio)
1301 struct bio *clone = &tio->clone;
1302 struct dm_io *io = tio->io;
1303 struct dm_target *ti = tio->ti;
1304 blk_qc_t ret = BLK_QC_T_NONE;
1306 clone->bi_end_io = clone_endio;
1309 * Map the clone. If r == 0 we don't need to do
1310 * anything, the target has assumed ownership of
1313 atomic_inc(&io->io_count);
1314 sector = clone->bi_iter.bi_sector;
1316 if (unlikely(swap_bios_limit(ti, clone))) {
1317 struct mapped_device *md = io->md;
1318 int latch = get_swap_bios();
1319 if (unlikely(latch != md->swap_bios))
1320 __set_swap_bios_limit(md, latch);
1321 down(&md->swap_bios_semaphore);
1324 r = ti->type->map(ti, clone);
1326 case DM_MAPIO_SUBMITTED:
1328 case DM_MAPIO_REMAPPED:
1329 /* the bio has been remapped so dispatch it */
1330 trace_block_bio_remap(clone->bi_disk->queue, clone,
1331 bio_dev(io->orig_bio), sector);
1332 ret = submit_bio_noacct(clone);
1335 if (unlikely(swap_bios_limit(ti, clone))) {
1336 struct mapped_device *md = io->md;
1337 up(&md->swap_bios_semaphore);
1340 dec_pending(io, BLK_STS_IOERR);
1342 case DM_MAPIO_REQUEUE:
1343 if (unlikely(swap_bios_limit(ti, clone))) {
1344 struct mapped_device *md = io->md;
1345 up(&md->swap_bios_semaphore);
1348 dec_pending(io, BLK_STS_DM_REQUEUE);
1351 DMWARN("unimplemented target map return value: %d", r);
1358 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1360 bio->bi_iter.bi_sector = sector;
1361 bio->bi_iter.bi_size = to_bytes(len);
1365 * Creates a bio that consists of range of complete bvecs.
1367 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1368 sector_t sector, unsigned len)
1370 struct bio *clone = &tio->clone;
1373 __bio_clone_fast(clone, bio);
1375 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1379 if (bio_integrity(bio)) {
1380 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1381 !dm_target_passes_integrity(tio->ti->type))) {
1382 DMWARN("%s: the target %s doesn't support integrity data.",
1383 dm_device_name(tio->io->md),
1384 tio->ti->type->name);
1388 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1393 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1394 clone->bi_iter.bi_size = to_bytes(len);
1396 if (bio_integrity(bio))
1397 bio_integrity_trim(clone);
1402 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1403 struct dm_target *ti, unsigned num_bios)
1405 struct dm_target_io *tio;
1411 if (num_bios == 1) {
1412 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1413 bio_list_add(blist, &tio->clone);
1417 for (try = 0; try < 2; try++) {
1422 mutex_lock(&ci->io->md->table_devices_lock);
1423 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1424 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1428 bio_list_add(blist, &tio->clone);
1431 mutex_unlock(&ci->io->md->table_devices_lock);
1432 if (bio_nr == num_bios)
1435 while ((bio = bio_list_pop(blist))) {
1436 tio = container_of(bio, struct dm_target_io, clone);
1442 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1443 struct dm_target_io *tio, unsigned *len)
1445 struct bio *clone = &tio->clone;
1449 __bio_clone_fast(clone, ci->bio);
1451 bio_setup_sector(clone, ci->sector, *len);
1453 return __map_bio(tio);
1456 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1457 unsigned num_bios, unsigned *len)
1459 struct bio_list blist = BIO_EMPTY_LIST;
1461 struct dm_target_io *tio;
1463 alloc_multiple_bios(&blist, ci, ti, num_bios);
1465 while ((bio = bio_list_pop(&blist))) {
1466 tio = container_of(bio, struct dm_target_io, clone);
1467 (void) __clone_and_map_simple_bio(ci, tio, len);
1471 static int __send_empty_flush(struct clone_info *ci)
1473 unsigned target_nr = 0;
1474 struct dm_target *ti;
1475 struct bio flush_bio;
1478 * Use an on-stack bio for this, it's safe since we don't
1479 * need to reference it after submit. It's just used as
1480 * the basis for the clone(s).
1482 bio_init(&flush_bio, NULL, 0);
1483 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1484 ci->bio = &flush_bio;
1485 ci->sector_count = 0;
1488 * Empty flush uses a statically initialized bio, as the base for
1489 * cloning. However, blkg association requires that a bdev is
1490 * associated with a gendisk, which doesn't happen until the bdev is
1491 * opened. So, blkg association is done at issue time of the flush
1492 * rather than when the device is created in alloc_dev().
1494 bio_set_dev(ci->bio, ci->io->md->bdev);
1496 BUG_ON(bio_has_data(ci->bio));
1497 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1498 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1500 bio_uninit(ci->bio);
1504 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1505 sector_t sector, unsigned *len)
1507 struct bio *bio = ci->bio;
1508 struct dm_target_io *tio;
1511 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1513 r = clone_bio(tio, bio, sector, *len);
1518 (void) __map_bio(tio);
1523 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1529 * Even though the device advertised support for this type of
1530 * request, that does not mean every target supports it, and
1531 * reconfiguration might also have changed that since the
1532 * check was performed.
1537 len = min_t(sector_t, ci->sector_count,
1538 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1540 __send_duplicate_bios(ci, ti, num_bios, &len);
1543 ci->sector_count -= len;
1548 static bool is_abnormal_io(struct bio *bio)
1552 switch (bio_op(bio)) {
1553 case REQ_OP_DISCARD:
1554 case REQ_OP_SECURE_ERASE:
1555 case REQ_OP_WRITE_SAME:
1556 case REQ_OP_WRITE_ZEROES:
1564 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1567 struct bio *bio = ci->bio;
1568 unsigned num_bios = 0;
1570 switch (bio_op(bio)) {
1571 case REQ_OP_DISCARD:
1572 num_bios = ti->num_discard_bios;
1574 case REQ_OP_SECURE_ERASE:
1575 num_bios = ti->num_secure_erase_bios;
1577 case REQ_OP_WRITE_SAME:
1578 num_bios = ti->num_write_same_bios;
1580 case REQ_OP_WRITE_ZEROES:
1581 num_bios = ti->num_write_zeroes_bios;
1587 *result = __send_changing_extent_only(ci, ti, num_bios);
1592 * Select the correct strategy for processing a non-flush bio.
1594 static int __split_and_process_non_flush(struct clone_info *ci)
1596 struct dm_target *ti;
1600 ti = dm_table_find_target(ci->map, ci->sector);
1604 if (__process_abnormal_io(ci, ti, &r))
1607 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1609 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1614 ci->sector_count -= len;
1619 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1620 struct dm_table *map, struct bio *bio)
1623 ci->io = alloc_io(md, bio);
1624 ci->sector = bio->bi_iter.bi_sector;
1627 #define __dm_part_stat_sub(part, field, subnd) \
1628 (part_stat_get(part, field) -= (subnd))
1631 * Entry point to split a bio into clones and submit them to the targets.
1633 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1634 struct dm_table *map, struct bio *bio)
1636 struct clone_info ci;
1637 blk_qc_t ret = BLK_QC_T_NONE;
1640 init_clone_info(&ci, md, map, bio);
1642 if (bio->bi_opf & REQ_PREFLUSH) {
1643 error = __send_empty_flush(&ci);
1644 /* dec_pending submits any data associated with flush */
1645 } else if (op_is_zone_mgmt(bio_op(bio))) {
1647 ci.sector_count = 0;
1648 error = __split_and_process_non_flush(&ci);
1651 ci.sector_count = bio_sectors(bio);
1652 while (ci.sector_count && !error) {
1653 error = __split_and_process_non_flush(&ci);
1654 if (current->bio_list && ci.sector_count && !error) {
1656 * Remainder must be passed to submit_bio_noacct()
1657 * so that it gets handled *after* bios already submitted
1658 * have been completely processed.
1659 * We take a clone of the original to store in
1660 * ci.io->orig_bio to be used by end_io_acct() and
1661 * for dec_pending to use for completion handling.
1663 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1664 GFP_NOIO, &md->queue->bio_split);
1665 ci.io->orig_bio = b;
1668 * Adjust IO stats for each split, otherwise upon queue
1669 * reentry there will be redundant IO accounting.
1670 * NOTE: this is a stop-gap fix, a proper fix involves
1671 * significant refactoring of DM core's bio splitting
1672 * (by eliminating DM's splitting and just using bio_split)
1675 __dm_part_stat_sub(&dm_disk(md)->part0,
1676 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1680 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1681 ret = submit_bio_noacct(bio);
1687 /* drop the extra reference count */
1688 dec_pending(ci.io, errno_to_blk_status(error));
1692 static blk_qc_t dm_submit_bio(struct bio *bio)
1694 struct mapped_device *md = bio->bi_disk->private_data;
1695 blk_qc_t ret = BLK_QC_T_NONE;
1697 struct dm_table *map;
1699 map = dm_get_live_table(md, &srcu_idx);
1701 /* If suspended, or map not yet available, queue this IO for later */
1702 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1704 if (bio->bi_opf & REQ_NOWAIT)
1705 bio_wouldblock_error(bio);
1706 else if (bio->bi_opf & REQ_RAHEAD)
1714 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1715 * otherwise associated queue_limits won't be imposed.
1717 if (is_abnormal_io(bio))
1718 blk_queue_split(&bio);
1720 ret = __split_and_process_bio(md, map, bio);
1722 dm_put_live_table(md, srcu_idx);
1726 /*-----------------------------------------------------------------
1727 * An IDR is used to keep track of allocated minor numbers.
1728 *---------------------------------------------------------------*/
1729 static void free_minor(int minor)
1731 spin_lock(&_minor_lock);
1732 idr_remove(&_minor_idr, minor);
1733 spin_unlock(&_minor_lock);
1737 * See if the device with a specific minor # is free.
1739 static int specific_minor(int minor)
1743 if (minor >= (1 << MINORBITS))
1746 idr_preload(GFP_KERNEL);
1747 spin_lock(&_minor_lock);
1749 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1751 spin_unlock(&_minor_lock);
1754 return r == -ENOSPC ? -EBUSY : r;
1758 static int next_free_minor(int *minor)
1762 idr_preload(GFP_KERNEL);
1763 spin_lock(&_minor_lock);
1765 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1767 spin_unlock(&_minor_lock);
1775 static const struct block_device_operations dm_blk_dops;
1776 static const struct block_device_operations dm_rq_blk_dops;
1777 static const struct dax_operations dm_dax_ops;
1779 static void dm_wq_work(struct work_struct *work);
1781 static void cleanup_mapped_device(struct mapped_device *md)
1784 destroy_workqueue(md->wq);
1785 bioset_exit(&md->bs);
1786 bioset_exit(&md->io_bs);
1789 kill_dax(md->dax_dev);
1790 put_dax(md->dax_dev);
1795 spin_lock(&_minor_lock);
1796 md->disk->private_data = NULL;
1797 spin_unlock(&_minor_lock);
1798 del_gendisk(md->disk);
1803 blk_cleanup_queue(md->queue);
1805 cleanup_srcu_struct(&md->io_barrier);
1812 mutex_destroy(&md->suspend_lock);
1813 mutex_destroy(&md->type_lock);
1814 mutex_destroy(&md->table_devices_lock);
1815 mutex_destroy(&md->swap_bios_lock);
1817 dm_mq_cleanup_mapped_device(md);
1821 * Allocate and initialise a blank device with a given minor.
1823 static struct mapped_device *alloc_dev(int minor)
1825 int r, numa_node_id = dm_get_numa_node();
1826 struct mapped_device *md;
1829 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1831 DMWARN("unable to allocate device, out of memory.");
1835 if (!try_module_get(THIS_MODULE))
1836 goto bad_module_get;
1838 /* get a minor number for the dev */
1839 if (minor == DM_ANY_MINOR)
1840 r = next_free_minor(&minor);
1842 r = specific_minor(minor);
1846 r = init_srcu_struct(&md->io_barrier);
1848 goto bad_io_barrier;
1850 md->numa_node_id = numa_node_id;
1851 md->init_tio_pdu = false;
1852 md->type = DM_TYPE_NONE;
1853 mutex_init(&md->suspend_lock);
1854 mutex_init(&md->type_lock);
1855 mutex_init(&md->table_devices_lock);
1856 spin_lock_init(&md->deferred_lock);
1857 atomic_set(&md->holders, 1);
1858 atomic_set(&md->open_count, 0);
1859 atomic_set(&md->event_nr, 0);
1860 atomic_set(&md->uevent_seq, 0);
1861 INIT_LIST_HEAD(&md->uevent_list);
1862 INIT_LIST_HEAD(&md->table_devices);
1863 spin_lock_init(&md->uevent_lock);
1866 * default to bio-based until DM table is loaded and md->type
1867 * established. If request-based table is loaded: blk-mq will
1868 * override accordingly.
1870 md->queue = blk_alloc_queue(numa_node_id);
1874 md->disk = alloc_disk_node(1, md->numa_node_id);
1878 init_waitqueue_head(&md->wait);
1879 INIT_WORK(&md->work, dm_wq_work);
1880 init_waitqueue_head(&md->eventq);
1881 init_completion(&md->kobj_holder.completion);
1883 md->swap_bios = get_swap_bios();
1884 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1885 mutex_init(&md->swap_bios_lock);
1887 md->disk->major = _major;
1888 md->disk->first_minor = minor;
1889 md->disk->fops = &dm_blk_dops;
1890 md->disk->queue = md->queue;
1891 md->disk->private_data = md;
1892 sprintf(md->disk->disk_name, "dm-%d", minor);
1894 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1895 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1897 if (IS_ERR(md->dax_dev)) {
1903 add_disk_no_queue_reg(md->disk);
1904 format_dev_t(md->name, MKDEV(_major, minor));
1906 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1910 md->bdev = bdget_disk(md->disk, 0);
1914 dm_stats_init(&md->stats);
1916 /* Populate the mapping, nobody knows we exist yet */
1917 spin_lock(&_minor_lock);
1918 old_md = idr_replace(&_minor_idr, md, minor);
1919 spin_unlock(&_minor_lock);
1921 BUG_ON(old_md != MINOR_ALLOCED);
1926 cleanup_mapped_device(md);
1930 module_put(THIS_MODULE);
1936 static void unlock_fs(struct mapped_device *md);
1938 static void free_dev(struct mapped_device *md)
1940 int minor = MINOR(disk_devt(md->disk));
1944 cleanup_mapped_device(md);
1946 free_table_devices(&md->table_devices);
1947 dm_stats_cleanup(&md->stats);
1950 module_put(THIS_MODULE);
1954 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1956 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1959 if (dm_table_bio_based(t)) {
1961 * The md may already have mempools that need changing.
1962 * If so, reload bioset because front_pad may have changed
1963 * because a different table was loaded.
1965 bioset_exit(&md->bs);
1966 bioset_exit(&md->io_bs);
1968 } else if (bioset_initialized(&md->bs)) {
1970 * There's no need to reload with request-based dm
1971 * because the size of front_pad doesn't change.
1972 * Note for future: If you are to reload bioset,
1973 * prep-ed requests in the queue may refer
1974 * to bio from the old bioset, so you must walk
1975 * through the queue to unprep.
1981 bioset_initialized(&md->bs) ||
1982 bioset_initialized(&md->io_bs));
1984 ret = bioset_init_from_src(&md->bs, &p->bs);
1987 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1989 bioset_exit(&md->bs);
1991 /* mempool bind completed, no longer need any mempools in the table */
1992 dm_table_free_md_mempools(t);
1997 * Bind a table to the device.
1999 static void event_callback(void *context)
2001 unsigned long flags;
2003 struct mapped_device *md = (struct mapped_device *) context;
2005 spin_lock_irqsave(&md->uevent_lock, flags);
2006 list_splice_init(&md->uevent_list, &uevents);
2007 spin_unlock_irqrestore(&md->uevent_lock, flags);
2009 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2011 atomic_inc(&md->event_nr);
2012 wake_up(&md->eventq);
2013 dm_issue_global_event();
2017 * Returns old map, which caller must destroy.
2019 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2020 struct queue_limits *limits)
2022 struct dm_table *old_map;
2023 struct request_queue *q = md->queue;
2024 bool request_based = dm_table_request_based(t);
2028 lockdep_assert_held(&md->suspend_lock);
2030 size = dm_table_get_size(t);
2033 * Wipe any geometry if the size of the table changed.
2035 if (size != dm_get_size(md))
2036 memset(&md->geometry, 0, sizeof(md->geometry));
2038 set_capacity(md->disk, size);
2039 bd_set_nr_sectors(md->bdev, size);
2041 dm_table_event_callback(t, event_callback, md);
2044 * The queue hasn't been stopped yet, if the old table type wasn't
2045 * for request-based during suspension. So stop it to prevent
2046 * I/O mapping before resume.
2047 * This must be done before setting the queue restrictions,
2048 * because request-based dm may be run just after the setting.
2053 if (request_based) {
2055 * Leverage the fact that request-based DM targets are
2056 * immutable singletons - used to optimize dm_mq_queue_rq.
2058 md->immutable_target = dm_table_get_immutable_target(t);
2061 ret = __bind_mempools(md, t);
2063 old_map = ERR_PTR(ret);
2067 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2068 rcu_assign_pointer(md->map, (void *)t);
2069 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2071 dm_table_set_restrictions(t, q, limits);
2080 * Returns unbound table for the caller to free.
2082 static struct dm_table *__unbind(struct mapped_device *md)
2084 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2089 dm_table_event_callback(map, NULL, NULL);
2090 RCU_INIT_POINTER(md->map, NULL);
2097 * Constructor for a new device.
2099 int dm_create(int minor, struct mapped_device **result)
2102 struct mapped_device *md;
2104 md = alloc_dev(minor);
2108 r = dm_sysfs_init(md);
2119 * Functions to manage md->type.
2120 * All are required to hold md->type_lock.
2122 void dm_lock_md_type(struct mapped_device *md)
2124 mutex_lock(&md->type_lock);
2127 void dm_unlock_md_type(struct mapped_device *md)
2129 mutex_unlock(&md->type_lock);
2132 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2134 BUG_ON(!mutex_is_locked(&md->type_lock));
2138 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2143 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2145 return md->immutable_target_type;
2149 * The queue_limits are only valid as long as you have a reference
2152 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2154 BUG_ON(!atomic_read(&md->holders));
2155 return &md->queue->limits;
2157 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2160 * Setup the DM device's queue based on md's type
2162 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2165 struct queue_limits limits;
2166 enum dm_queue_mode type = dm_get_md_type(md);
2169 case DM_TYPE_REQUEST_BASED:
2170 md->disk->fops = &dm_rq_blk_dops;
2171 r = dm_mq_init_request_queue(md, t);
2173 DMERR("Cannot initialize queue for request-based dm mapped device");
2177 case DM_TYPE_BIO_BASED:
2178 case DM_TYPE_DAX_BIO_BASED:
2185 r = dm_calculate_queue_limits(t, &limits);
2187 DMERR("Cannot calculate initial queue limits");
2190 dm_table_set_restrictions(t, md->queue, &limits);
2191 blk_register_queue(md->disk);
2196 struct mapped_device *dm_get_md(dev_t dev)
2198 struct mapped_device *md;
2199 unsigned minor = MINOR(dev);
2201 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2204 spin_lock(&_minor_lock);
2206 md = idr_find(&_minor_idr, minor);
2207 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2208 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2214 spin_unlock(&_minor_lock);
2218 EXPORT_SYMBOL_GPL(dm_get_md);
2220 void *dm_get_mdptr(struct mapped_device *md)
2222 return md->interface_ptr;
2225 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2227 md->interface_ptr = ptr;
2230 void dm_get(struct mapped_device *md)
2232 atomic_inc(&md->holders);
2233 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2236 int dm_hold(struct mapped_device *md)
2238 spin_lock(&_minor_lock);
2239 if (test_bit(DMF_FREEING, &md->flags)) {
2240 spin_unlock(&_minor_lock);
2244 spin_unlock(&_minor_lock);
2247 EXPORT_SYMBOL_GPL(dm_hold);
2249 const char *dm_device_name(struct mapped_device *md)
2253 EXPORT_SYMBOL_GPL(dm_device_name);
2255 static void __dm_destroy(struct mapped_device *md, bool wait)
2257 struct dm_table *map;
2262 spin_lock(&_minor_lock);
2263 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2264 set_bit(DMF_FREEING, &md->flags);
2265 spin_unlock(&_minor_lock);
2267 blk_set_queue_dying(md->queue);
2270 * Take suspend_lock so that presuspend and postsuspend methods
2271 * do not race with internal suspend.
2273 mutex_lock(&md->suspend_lock);
2274 map = dm_get_live_table(md, &srcu_idx);
2275 if (!dm_suspended_md(md)) {
2276 dm_table_presuspend_targets(map);
2277 set_bit(DMF_SUSPENDED, &md->flags);
2278 set_bit(DMF_POST_SUSPENDING, &md->flags);
2279 dm_table_postsuspend_targets(map);
2281 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2282 dm_put_live_table(md, srcu_idx);
2283 mutex_unlock(&md->suspend_lock);
2286 * Rare, but there may be I/O requests still going to complete,
2287 * for example. Wait for all references to disappear.
2288 * No one should increment the reference count of the mapped_device,
2289 * after the mapped_device state becomes DMF_FREEING.
2292 while (atomic_read(&md->holders))
2294 else if (atomic_read(&md->holders))
2295 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2296 dm_device_name(md), atomic_read(&md->holders));
2299 dm_table_destroy(__unbind(md));
2303 void dm_destroy(struct mapped_device *md)
2305 __dm_destroy(md, true);
2308 void dm_destroy_immediate(struct mapped_device *md)
2310 __dm_destroy(md, false);
2313 void dm_put(struct mapped_device *md)
2315 atomic_dec(&md->holders);
2317 EXPORT_SYMBOL_GPL(dm_put);
2319 static bool md_in_flight_bios(struct mapped_device *md)
2322 struct hd_struct *part = &dm_disk(md)->part0;
2325 for_each_possible_cpu(cpu) {
2326 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2327 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2333 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2339 prepare_to_wait(&md->wait, &wait, task_state);
2341 if (!md_in_flight_bios(md))
2344 if (signal_pending_state(task_state, current)) {
2351 finish_wait(&md->wait, &wait);
2358 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2362 if (!queue_is_mq(md->queue))
2363 return dm_wait_for_bios_completion(md, task_state);
2366 if (!blk_mq_queue_inflight(md->queue))
2369 if (signal_pending_state(task_state, current)) {
2381 * Process the deferred bios
2383 static void dm_wq_work(struct work_struct *work)
2385 struct mapped_device *md = container_of(work, struct mapped_device, work);
2388 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2389 spin_lock_irq(&md->deferred_lock);
2390 bio = bio_list_pop(&md->deferred);
2391 spin_unlock_irq(&md->deferred_lock);
2396 submit_bio_noacct(bio);
2400 static void dm_queue_flush(struct mapped_device *md)
2402 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2403 smp_mb__after_atomic();
2404 queue_work(md->wq, &md->work);
2408 * Swap in a new table, returning the old one for the caller to destroy.
2410 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2412 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2413 struct queue_limits limits;
2416 mutex_lock(&md->suspend_lock);
2418 /* device must be suspended */
2419 if (!dm_suspended_md(md))
2423 * If the new table has no data devices, retain the existing limits.
2424 * This helps multipath with queue_if_no_path if all paths disappear,
2425 * then new I/O is queued based on these limits, and then some paths
2428 if (dm_table_has_no_data_devices(table)) {
2429 live_map = dm_get_live_table_fast(md);
2431 limits = md->queue->limits;
2432 dm_put_live_table_fast(md);
2436 r = dm_calculate_queue_limits(table, &limits);
2443 map = __bind(md, table, &limits);
2444 dm_issue_global_event();
2447 mutex_unlock(&md->suspend_lock);
2452 * Functions to lock and unlock any filesystem running on the
2455 static int lock_fs(struct mapped_device *md)
2459 WARN_ON(md->frozen_sb);
2461 md->frozen_sb = freeze_bdev(md->bdev);
2462 if (IS_ERR(md->frozen_sb)) {
2463 r = PTR_ERR(md->frozen_sb);
2464 md->frozen_sb = NULL;
2468 set_bit(DMF_FROZEN, &md->flags);
2473 static void unlock_fs(struct mapped_device *md)
2475 if (!test_bit(DMF_FROZEN, &md->flags))
2478 thaw_bdev(md->bdev, md->frozen_sb);
2479 md->frozen_sb = NULL;
2480 clear_bit(DMF_FROZEN, &md->flags);
2484 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2485 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2486 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2488 * If __dm_suspend returns 0, the device is completely quiescent
2489 * now. There is no request-processing activity. All new requests
2490 * are being added to md->deferred list.
2492 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2493 unsigned suspend_flags, long task_state,
2494 int dmf_suspended_flag)
2496 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2497 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2500 lockdep_assert_held(&md->suspend_lock);
2503 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2504 * This flag is cleared before dm_suspend returns.
2507 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2509 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2512 * This gets reverted if there's an error later and the targets
2513 * provide the .presuspend_undo hook.
2515 dm_table_presuspend_targets(map);
2518 * Flush I/O to the device.
2519 * Any I/O submitted after lock_fs() may not be flushed.
2520 * noflush takes precedence over do_lockfs.
2521 * (lock_fs() flushes I/Os and waits for them to complete.)
2523 if (!noflush && do_lockfs) {
2526 dm_table_presuspend_undo_targets(map);
2532 * Here we must make sure that no processes are submitting requests
2533 * to target drivers i.e. no one may be executing
2534 * __split_and_process_bio from dm_submit_bio.
2536 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2537 * we take the write lock. To prevent any process from reentering
2538 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2539 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2540 * flush_workqueue(md->wq).
2542 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2544 synchronize_srcu(&md->io_barrier);
2547 * Stop md->queue before flushing md->wq in case request-based
2548 * dm defers requests to md->wq from md->queue.
2550 if (dm_request_based(md))
2551 dm_stop_queue(md->queue);
2553 flush_workqueue(md->wq);
2556 * At this point no more requests are entering target request routines.
2557 * We call dm_wait_for_completion to wait for all existing requests
2560 r = dm_wait_for_completion(md, task_state);
2562 set_bit(dmf_suspended_flag, &md->flags);
2565 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2567 synchronize_srcu(&md->io_barrier);
2569 /* were we interrupted ? */
2573 if (dm_request_based(md))
2574 dm_start_queue(md->queue);
2577 dm_table_presuspend_undo_targets(map);
2578 /* pushback list is already flushed, so skip flush */
2585 * We need to be able to change a mapping table under a mounted
2586 * filesystem. For example we might want to move some data in
2587 * the background. Before the table can be swapped with
2588 * dm_bind_table, dm_suspend must be called to flush any in
2589 * flight bios and ensure that any further io gets deferred.
2592 * Suspend mechanism in request-based dm.
2594 * 1. Flush all I/Os by lock_fs() if needed.
2595 * 2. Stop dispatching any I/O by stopping the request_queue.
2596 * 3. Wait for all in-flight I/Os to be completed or requeued.
2598 * To abort suspend, start the request_queue.
2600 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2602 struct dm_table *map = NULL;
2606 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2608 if (dm_suspended_md(md)) {
2613 if (dm_suspended_internally_md(md)) {
2614 /* already internally suspended, wait for internal resume */
2615 mutex_unlock(&md->suspend_lock);
2616 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2622 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2624 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2628 set_bit(DMF_POST_SUSPENDING, &md->flags);
2629 dm_table_postsuspend_targets(map);
2630 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2633 mutex_unlock(&md->suspend_lock);
2637 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2640 int r = dm_table_resume_targets(map);
2648 * Flushing deferred I/Os must be done after targets are resumed
2649 * so that mapping of targets can work correctly.
2650 * Request-based dm is queueing the deferred I/Os in its request_queue.
2652 if (dm_request_based(md))
2653 dm_start_queue(md->queue);
2660 int dm_resume(struct mapped_device *md)
2663 struct dm_table *map = NULL;
2667 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2669 if (!dm_suspended_md(md))
2672 if (dm_suspended_internally_md(md)) {
2673 /* already internally suspended, wait for internal resume */
2674 mutex_unlock(&md->suspend_lock);
2675 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2681 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2682 if (!map || !dm_table_get_size(map))
2685 r = __dm_resume(md, map);
2689 clear_bit(DMF_SUSPENDED, &md->flags);
2691 mutex_unlock(&md->suspend_lock);
2697 * Internal suspend/resume works like userspace-driven suspend. It waits
2698 * until all bios finish and prevents issuing new bios to the target drivers.
2699 * It may be used only from the kernel.
2702 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2704 struct dm_table *map = NULL;
2706 lockdep_assert_held(&md->suspend_lock);
2708 if (md->internal_suspend_count++)
2709 return; /* nested internal suspend */
2711 if (dm_suspended_md(md)) {
2712 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2713 return; /* nest suspend */
2716 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2719 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2720 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2721 * would require changing .presuspend to return an error -- avoid this
2722 * until there is a need for more elaborate variants of internal suspend.
2724 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2725 DMF_SUSPENDED_INTERNALLY);
2727 set_bit(DMF_POST_SUSPENDING, &md->flags);
2728 dm_table_postsuspend_targets(map);
2729 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2732 static void __dm_internal_resume(struct mapped_device *md)
2734 BUG_ON(!md->internal_suspend_count);
2736 if (--md->internal_suspend_count)
2737 return; /* resume from nested internal suspend */
2739 if (dm_suspended_md(md))
2740 goto done; /* resume from nested suspend */
2743 * NOTE: existing callers don't need to call dm_table_resume_targets
2744 * (which may fail -- so best to avoid it for now by passing NULL map)
2746 (void) __dm_resume(md, NULL);
2749 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2750 smp_mb__after_atomic();
2751 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2754 void dm_internal_suspend_noflush(struct mapped_device *md)
2756 mutex_lock(&md->suspend_lock);
2757 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2758 mutex_unlock(&md->suspend_lock);
2760 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2762 void dm_internal_resume(struct mapped_device *md)
2764 mutex_lock(&md->suspend_lock);
2765 __dm_internal_resume(md);
2766 mutex_unlock(&md->suspend_lock);
2768 EXPORT_SYMBOL_GPL(dm_internal_resume);
2771 * Fast variants of internal suspend/resume hold md->suspend_lock,
2772 * which prevents interaction with userspace-driven suspend.
2775 void dm_internal_suspend_fast(struct mapped_device *md)
2777 mutex_lock(&md->suspend_lock);
2778 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2781 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2782 synchronize_srcu(&md->io_barrier);
2783 flush_workqueue(md->wq);
2784 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2786 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2788 void dm_internal_resume_fast(struct mapped_device *md)
2790 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2796 mutex_unlock(&md->suspend_lock);
2798 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2800 /*-----------------------------------------------------------------
2801 * Event notification.
2802 *---------------------------------------------------------------*/
2803 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2808 char udev_cookie[DM_COOKIE_LENGTH];
2809 char *envp[] = { udev_cookie, NULL };
2811 noio_flag = memalloc_noio_save();
2814 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2816 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2817 DM_COOKIE_ENV_VAR_NAME, cookie);
2818 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2822 memalloc_noio_restore(noio_flag);
2827 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2829 return atomic_add_return(1, &md->uevent_seq);
2832 uint32_t dm_get_event_nr(struct mapped_device *md)
2834 return atomic_read(&md->event_nr);
2837 int dm_wait_event(struct mapped_device *md, int event_nr)
2839 return wait_event_interruptible(md->eventq,
2840 (event_nr != atomic_read(&md->event_nr)));
2843 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2845 unsigned long flags;
2847 spin_lock_irqsave(&md->uevent_lock, flags);
2848 list_add(elist, &md->uevent_list);
2849 spin_unlock_irqrestore(&md->uevent_lock, flags);
2853 * The gendisk is only valid as long as you have a reference
2856 struct gendisk *dm_disk(struct mapped_device *md)
2860 EXPORT_SYMBOL_GPL(dm_disk);
2862 struct kobject *dm_kobject(struct mapped_device *md)
2864 return &md->kobj_holder.kobj;
2867 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2869 struct mapped_device *md;
2871 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2873 spin_lock(&_minor_lock);
2874 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2880 spin_unlock(&_minor_lock);
2885 int dm_suspended_md(struct mapped_device *md)
2887 return test_bit(DMF_SUSPENDED, &md->flags);
2890 static int dm_post_suspending_md(struct mapped_device *md)
2892 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2895 int dm_suspended_internally_md(struct mapped_device *md)
2897 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2900 int dm_test_deferred_remove_flag(struct mapped_device *md)
2902 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2905 int dm_suspended(struct dm_target *ti)
2907 return dm_suspended_md(ti->table->md);
2909 EXPORT_SYMBOL_GPL(dm_suspended);
2911 int dm_post_suspending(struct dm_target *ti)
2913 return dm_post_suspending_md(ti->table->md);
2915 EXPORT_SYMBOL_GPL(dm_post_suspending);
2917 int dm_noflush_suspending(struct dm_target *ti)
2919 return __noflush_suspending(ti->table->md);
2921 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2923 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2924 unsigned integrity, unsigned per_io_data_size,
2925 unsigned min_pool_size)
2927 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2928 unsigned int pool_size = 0;
2929 unsigned int front_pad, io_front_pad;
2936 case DM_TYPE_BIO_BASED:
2937 case DM_TYPE_DAX_BIO_BASED:
2938 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2939 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2940 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2941 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2944 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2947 case DM_TYPE_REQUEST_BASED:
2948 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2949 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2950 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2956 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2960 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2966 dm_free_md_mempools(pools);
2971 void dm_free_md_mempools(struct dm_md_mempools *pools)
2976 bioset_exit(&pools->bs);
2977 bioset_exit(&pools->io_bs);
2989 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2992 struct mapped_device *md = bdev->bd_disk->private_data;
2993 struct dm_table *table;
2994 struct dm_target *ti;
2995 int ret = -ENOTTY, srcu_idx;
2997 table = dm_get_live_table(md, &srcu_idx);
2998 if (!table || !dm_table_get_size(table))
3001 /* We only support devices that have a single target */
3002 if (dm_table_get_num_targets(table) != 1)
3004 ti = dm_table_get_target(table, 0);
3006 if (dm_suspended_md(md)) {
3012 if (!ti->type->iterate_devices)
3015 ret = ti->type->iterate_devices(ti, fn, data);
3017 dm_put_live_table(md, srcu_idx);
3022 * For register / unregister we need to manually call out to every path.
3024 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3025 sector_t start, sector_t len, void *data)
3027 struct dm_pr *pr = data;
3028 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3030 if (!ops || !ops->pr_register)
3032 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3035 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3046 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3047 if (ret && new_key) {
3048 /* unregister all paths if we failed to register any path */
3049 pr.old_key = new_key;
3052 pr.fail_early = false;
3053 dm_call_pr(bdev, __dm_pr_register, &pr);
3059 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3062 struct mapped_device *md = bdev->bd_disk->private_data;
3063 const struct pr_ops *ops;
3066 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3070 ops = bdev->bd_disk->fops->pr_ops;
3071 if (ops && ops->pr_reserve)
3072 r = ops->pr_reserve(bdev, key, type, flags);
3076 dm_unprepare_ioctl(md, srcu_idx);
3080 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3082 struct mapped_device *md = bdev->bd_disk->private_data;
3083 const struct pr_ops *ops;
3086 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3090 ops = bdev->bd_disk->fops->pr_ops;
3091 if (ops && ops->pr_release)
3092 r = ops->pr_release(bdev, key, type);
3096 dm_unprepare_ioctl(md, srcu_idx);
3100 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3101 enum pr_type type, bool abort)
3103 struct mapped_device *md = bdev->bd_disk->private_data;
3104 const struct pr_ops *ops;
3107 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3111 ops = bdev->bd_disk->fops->pr_ops;
3112 if (ops && ops->pr_preempt)
3113 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3117 dm_unprepare_ioctl(md, srcu_idx);
3121 static int dm_pr_clear(struct block_device *bdev, u64 key)
3123 struct mapped_device *md = bdev->bd_disk->private_data;
3124 const struct pr_ops *ops;
3127 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3131 ops = bdev->bd_disk->fops->pr_ops;
3132 if (ops && ops->pr_clear)
3133 r = ops->pr_clear(bdev, key);
3137 dm_unprepare_ioctl(md, srcu_idx);
3141 static const struct pr_ops dm_pr_ops = {
3142 .pr_register = dm_pr_register,
3143 .pr_reserve = dm_pr_reserve,
3144 .pr_release = dm_pr_release,
3145 .pr_preempt = dm_pr_preempt,
3146 .pr_clear = dm_pr_clear,
3149 static const struct block_device_operations dm_blk_dops = {
3150 .submit_bio = dm_submit_bio,
3151 .open = dm_blk_open,
3152 .release = dm_blk_close,
3153 .ioctl = dm_blk_ioctl,
3154 .getgeo = dm_blk_getgeo,
3155 .report_zones = dm_blk_report_zones,
3156 .pr_ops = &dm_pr_ops,
3157 .owner = THIS_MODULE
3160 static const struct block_device_operations dm_rq_blk_dops = {
3161 .open = dm_blk_open,
3162 .release = dm_blk_close,
3163 .ioctl = dm_blk_ioctl,
3164 .getgeo = dm_blk_getgeo,
3165 .pr_ops = &dm_pr_ops,
3166 .owner = THIS_MODULE
3169 static const struct dax_operations dm_dax_ops = {
3170 .direct_access = dm_dax_direct_access,
3171 .dax_supported = dm_dax_supported,
3172 .copy_from_iter = dm_dax_copy_from_iter,
3173 .copy_to_iter = dm_dax_copy_to_iter,
3174 .zero_page_range = dm_dax_zero_page_range,
3180 module_init(dm_init);
3181 module_exit(dm_exit);
3183 module_param(major, uint, 0);
3184 MODULE_PARM_DESC(major, "The major number of the device mapper");
3186 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3187 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3189 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3190 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3192 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3193 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3195 MODULE_DESCRIPTION(DM_NAME " driver");
3196 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3197 MODULE_LICENSE("GPL");