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;
152 * For mempools pre-allocation at the table loading time.
154 struct dm_md_mempools {
156 struct bio_set io_bs;
159 struct table_device {
160 struct list_head list;
162 struct dm_dev dm_dev;
166 * Bio-based DM's mempools' reserved IOs set by the user.
168 #define RESERVED_BIO_BASED_IOS 16
169 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
171 static int __dm_get_module_param_int(int *module_param, int min, int max)
173 int param = READ_ONCE(*module_param);
174 int modified_param = 0;
175 bool modified = true;
178 modified_param = min;
179 else if (param > max)
180 modified_param = max;
185 (void)cmpxchg(module_param, param, modified_param);
186 param = modified_param;
192 unsigned __dm_get_module_param(unsigned *module_param,
193 unsigned def, unsigned max)
195 unsigned param = READ_ONCE(*module_param);
196 unsigned modified_param = 0;
199 modified_param = def;
200 else if (param > max)
201 modified_param = max;
203 if (modified_param) {
204 (void)cmpxchg(module_param, param, modified_param);
205 param = modified_param;
211 unsigned dm_get_reserved_bio_based_ios(void)
213 return __dm_get_module_param(&reserved_bio_based_ios,
214 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
216 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
218 static unsigned dm_get_numa_node(void)
220 return __dm_get_module_param_int(&dm_numa_node,
221 DM_NUMA_NODE, num_online_nodes() - 1);
224 static int __init local_init(void)
228 r = dm_uevent_init();
232 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
233 if (!deferred_remove_workqueue) {
235 goto out_uevent_exit;
239 r = register_blkdev(_major, _name);
241 goto out_free_workqueue;
249 destroy_workqueue(deferred_remove_workqueue);
256 static void local_exit(void)
258 flush_scheduled_work();
259 destroy_workqueue(deferred_remove_workqueue);
261 unregister_blkdev(_major, _name);
266 DMINFO("cleaned up");
269 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
291 static int __init dm_init(void)
293 const int count = ARRAY_SIZE(_inits);
297 for (i = 0; i < count; i++) {
312 static void __exit dm_exit(void)
314 int i = ARRAY_SIZE(_exits);
320 * Should be empty by this point.
322 idr_destroy(&_minor_idr);
326 * Block device functions
328 int dm_deleting_md(struct mapped_device *md)
330 return test_bit(DMF_DELETING, &md->flags);
333 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
335 struct mapped_device *md;
337 spin_lock(&_minor_lock);
339 md = bdev->bd_disk->private_data;
343 if (test_bit(DMF_FREEING, &md->flags) ||
344 dm_deleting_md(md)) {
350 atomic_inc(&md->open_count);
352 spin_unlock(&_minor_lock);
354 return md ? 0 : -ENXIO;
357 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
359 struct mapped_device *md;
361 spin_lock(&_minor_lock);
363 md = disk->private_data;
367 if (atomic_dec_and_test(&md->open_count) &&
368 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
369 queue_work(deferred_remove_workqueue, &deferred_remove_work);
373 spin_unlock(&_minor_lock);
376 int dm_open_count(struct mapped_device *md)
378 return atomic_read(&md->open_count);
382 * Guarantees nothing is using the device before it's deleted.
384 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
388 spin_lock(&_minor_lock);
390 if (dm_open_count(md)) {
393 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
394 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
397 set_bit(DMF_DELETING, &md->flags);
399 spin_unlock(&_minor_lock);
404 int dm_cancel_deferred_remove(struct mapped_device *md)
408 spin_lock(&_minor_lock);
410 if (test_bit(DMF_DELETING, &md->flags))
413 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
415 spin_unlock(&_minor_lock);
420 static void do_deferred_remove(struct work_struct *w)
422 dm_deferred_remove();
425 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
427 struct mapped_device *md = bdev->bd_disk->private_data;
429 return dm_get_geometry(md, geo);
432 #ifdef CONFIG_BLK_DEV_ZONED
433 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
435 struct dm_report_zones_args *args = data;
436 sector_t sector_diff = args->tgt->begin - args->start;
439 * Ignore zones beyond the target range.
441 if (zone->start >= args->start + args->tgt->len)
445 * Remap the start sector and write pointer position of the zone
446 * to match its position in the target range.
448 zone->start += sector_diff;
449 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
450 if (zone->cond == BLK_ZONE_COND_FULL)
451 zone->wp = zone->start + zone->len;
452 else if (zone->cond == BLK_ZONE_COND_EMPTY)
453 zone->wp = zone->start;
455 zone->wp += sector_diff;
458 args->next_sector = zone->start + zone->len;
459 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
461 EXPORT_SYMBOL_GPL(dm_report_zones_cb);
463 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
464 unsigned int nr_zones, report_zones_cb cb, void *data)
466 struct mapped_device *md = disk->private_data;
467 struct dm_table *map;
469 struct dm_report_zones_args args = {
470 .next_sector = sector,
475 if (dm_suspended_md(md))
478 map = dm_get_live_table(md, &srcu_idx);
485 struct dm_target *tgt;
487 tgt = dm_table_find_target(map, args.next_sector);
488 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
494 ret = tgt->type->report_zones(tgt, &args,
495 nr_zones - args.zone_idx);
498 } while (args.zone_idx < nr_zones &&
499 args.next_sector < get_capacity(disk));
503 dm_put_live_table(md, srcu_idx);
507 #define dm_blk_report_zones NULL
508 #endif /* CONFIG_BLK_DEV_ZONED */
510 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
511 struct block_device **bdev)
513 struct dm_target *tgt;
514 struct dm_table *map;
519 map = dm_get_live_table(md, srcu_idx);
520 if (!map || !dm_table_get_size(map))
523 /* We only support devices that have a single target */
524 if (dm_table_get_num_targets(map) != 1)
527 tgt = dm_table_get_target(map, 0);
528 if (!tgt->type->prepare_ioctl)
531 if (dm_suspended_md(md))
534 r = tgt->type->prepare_ioctl(tgt, bdev);
535 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
536 dm_put_live_table(md, *srcu_idx);
544 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
546 dm_put_live_table(md, srcu_idx);
549 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
550 unsigned int cmd, unsigned long arg)
552 struct mapped_device *md = bdev->bd_disk->private_data;
555 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
561 * Target determined this ioctl is being issued against a
562 * subset of the parent bdev; require extra privileges.
564 if (!capable(CAP_SYS_RAWIO)) {
566 "%s: sending ioctl %x to DM device without required privilege.",
573 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
575 dm_unprepare_ioctl(md, srcu_idx);
579 u64 dm_start_time_ns_from_clone(struct bio *bio)
581 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
582 struct dm_io *io = tio->io;
584 return jiffies_to_nsecs(io->start_time);
586 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
588 static void start_io_acct(struct dm_io *io)
590 struct mapped_device *md = io->md;
591 struct bio *bio = io->orig_bio;
593 io->start_time = bio_start_io_acct(bio);
594 if (unlikely(dm_stats_used(&md->stats)))
595 dm_stats_account_io(&md->stats, bio_data_dir(bio),
596 bio->bi_iter.bi_sector, bio_sectors(bio),
597 false, 0, &io->stats_aux);
600 static void end_io_acct(struct dm_io *io)
602 struct mapped_device *md = io->md;
603 struct bio *bio = io->orig_bio;
604 unsigned long duration = jiffies - io->start_time;
606 bio_end_io_acct(bio, io->start_time);
608 if (unlikely(dm_stats_used(&md->stats)))
609 dm_stats_account_io(&md->stats, bio_data_dir(bio),
610 bio->bi_iter.bi_sector, bio_sectors(bio),
611 true, duration, &io->stats_aux);
613 /* nudge anyone waiting on suspend queue */
614 if (unlikely(wq_has_sleeper(&md->wait)))
618 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
621 struct dm_target_io *tio;
624 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
628 tio = container_of(clone, struct dm_target_io, clone);
629 tio->inside_dm_io = true;
632 io = container_of(tio, struct dm_io, tio);
633 io->magic = DM_IO_MAGIC;
635 atomic_set(&io->io_count, 1);
638 spin_lock_init(&io->endio_lock);
645 static void free_io(struct mapped_device *md, struct dm_io *io)
647 bio_put(&io->tio.clone);
650 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
651 unsigned target_bio_nr, gfp_t gfp_mask)
653 struct dm_target_io *tio;
655 if (!ci->io->tio.io) {
656 /* the dm_target_io embedded in ci->io is available */
659 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
663 tio = container_of(clone, struct dm_target_io, clone);
664 tio->inside_dm_io = false;
667 tio->magic = DM_TIO_MAGIC;
670 tio->target_bio_nr = target_bio_nr;
675 static void free_tio(struct dm_target_io *tio)
677 if (tio->inside_dm_io)
679 bio_put(&tio->clone);
683 * Add the bio to the list of deferred io.
685 static void queue_io(struct mapped_device *md, struct bio *bio)
689 spin_lock_irqsave(&md->deferred_lock, flags);
690 bio_list_add(&md->deferred, bio);
691 spin_unlock_irqrestore(&md->deferred_lock, flags);
692 queue_work(md->wq, &md->work);
696 * Everyone (including functions in this file), should use this
697 * function to access the md->map field, and make sure they call
698 * dm_put_live_table() when finished.
700 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
702 *srcu_idx = srcu_read_lock(&md->io_barrier);
704 return srcu_dereference(md->map, &md->io_barrier);
707 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
709 srcu_read_unlock(&md->io_barrier, srcu_idx);
712 void dm_sync_table(struct mapped_device *md)
714 synchronize_srcu(&md->io_barrier);
715 synchronize_rcu_expedited();
719 * A fast alternative to dm_get_live_table/dm_put_live_table.
720 * The caller must not block between these two functions.
722 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
725 return rcu_dereference(md->map);
728 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
733 static char *_dm_claim_ptr = "I belong to device-mapper";
736 * Open a table device so we can use it as a map destination.
738 static int open_table_device(struct table_device *td, dev_t dev,
739 struct mapped_device *md)
741 struct block_device *bdev;
745 BUG_ON(td->dm_dev.bdev);
747 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
749 return PTR_ERR(bdev);
751 r = bd_link_disk_holder(bdev, dm_disk(md));
753 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
757 td->dm_dev.bdev = bdev;
758 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
763 * Close a table device that we've been using.
765 static void close_table_device(struct table_device *td, struct mapped_device *md)
767 if (!td->dm_dev.bdev)
770 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
771 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
772 put_dax(td->dm_dev.dax_dev);
773 td->dm_dev.bdev = NULL;
774 td->dm_dev.dax_dev = NULL;
777 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
780 struct table_device *td;
782 list_for_each_entry(td, l, list)
783 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
789 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
790 struct dm_dev **result)
793 struct table_device *td;
795 mutex_lock(&md->table_devices_lock);
796 td = find_table_device(&md->table_devices, dev, mode);
798 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
800 mutex_unlock(&md->table_devices_lock);
804 td->dm_dev.mode = mode;
805 td->dm_dev.bdev = NULL;
807 if ((r = open_table_device(td, dev, md))) {
808 mutex_unlock(&md->table_devices_lock);
813 format_dev_t(td->dm_dev.name, dev);
815 refcount_set(&td->count, 1);
816 list_add(&td->list, &md->table_devices);
818 refcount_inc(&td->count);
820 mutex_unlock(&md->table_devices_lock);
822 *result = &td->dm_dev;
825 EXPORT_SYMBOL_GPL(dm_get_table_device);
827 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
829 struct table_device *td = container_of(d, struct table_device, dm_dev);
831 mutex_lock(&md->table_devices_lock);
832 if (refcount_dec_and_test(&td->count)) {
833 close_table_device(td, md);
837 mutex_unlock(&md->table_devices_lock);
839 EXPORT_SYMBOL(dm_put_table_device);
841 static void free_table_devices(struct list_head *devices)
843 struct list_head *tmp, *next;
845 list_for_each_safe(tmp, next, devices) {
846 struct table_device *td = list_entry(tmp, struct table_device, list);
848 DMWARN("dm_destroy: %s still exists with %d references",
849 td->dm_dev.name, refcount_read(&td->count));
855 * Get the geometry associated with a dm device
857 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
865 * Set the geometry of a device.
867 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
869 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
871 if (geo->start > sz) {
872 DMWARN("Start sector is beyond the geometry limits.");
881 static int __noflush_suspending(struct mapped_device *md)
883 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
887 * Decrements the number of outstanding ios that a bio has been
888 * cloned into, completing the original io if necc.
890 static void dec_pending(struct dm_io *io, blk_status_t error)
893 blk_status_t io_error;
895 struct mapped_device *md = io->md;
897 /* Push-back supersedes any I/O errors */
898 if (unlikely(error)) {
899 spin_lock_irqsave(&io->endio_lock, flags);
900 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
902 spin_unlock_irqrestore(&io->endio_lock, flags);
905 if (atomic_dec_and_test(&io->io_count)) {
906 if (io->status == BLK_STS_DM_REQUEUE) {
908 * Target requested pushing back the I/O.
910 spin_lock_irqsave(&md->deferred_lock, flags);
911 if (__noflush_suspending(md))
912 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
913 bio_list_add_head(&md->deferred, io->orig_bio);
915 /* noflush suspend was interrupted. */
916 io->status = BLK_STS_IOERR;
917 spin_unlock_irqrestore(&md->deferred_lock, flags);
920 io_error = io->status;
925 if (io_error == BLK_STS_DM_REQUEUE)
928 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
930 * Preflush done for flush with data, reissue
931 * without REQ_PREFLUSH.
933 bio->bi_opf &= ~REQ_PREFLUSH;
936 /* done with normal IO or empty flush */
938 bio->bi_status = io_error;
944 void disable_discard(struct mapped_device *md)
946 struct queue_limits *limits = dm_get_queue_limits(md);
948 /* device doesn't really support DISCARD, disable it */
949 limits->max_discard_sectors = 0;
950 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
953 void disable_write_same(struct mapped_device *md)
955 struct queue_limits *limits = dm_get_queue_limits(md);
957 /* device doesn't really support WRITE SAME, disable it */
958 limits->max_write_same_sectors = 0;
961 void disable_write_zeroes(struct mapped_device *md)
963 struct queue_limits *limits = dm_get_queue_limits(md);
965 /* device doesn't really support WRITE ZEROES, disable it */
966 limits->max_write_zeroes_sectors = 0;
969 static void clone_endio(struct bio *bio)
971 blk_status_t error = bio->bi_status;
972 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
973 struct dm_io *io = tio->io;
974 struct mapped_device *md = tio->io->md;
975 dm_endio_fn endio = tio->ti->type->end_io;
976 struct bio *orig_bio = io->orig_bio;
978 if (unlikely(error == BLK_STS_TARGET)) {
979 if (bio_op(bio) == REQ_OP_DISCARD &&
980 !bio->bi_disk->queue->limits.max_discard_sectors)
982 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
983 !bio->bi_disk->queue->limits.max_write_same_sectors)
984 disable_write_same(md);
985 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
986 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
987 disable_write_zeroes(md);
991 * For zone-append bios get offset in zone of the written
992 * sector and add that to the original bio sector pos.
994 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
995 sector_t written_sector = bio->bi_iter.bi_sector;
996 struct request_queue *q = orig_bio->bi_disk->queue;
997 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
999 orig_bio->bi_iter.bi_sector += written_sector & mask;
1003 int r = endio(tio->ti, bio, &error);
1005 case DM_ENDIO_REQUEUE:
1006 error = BLK_STS_DM_REQUEUE;
1010 case DM_ENDIO_INCOMPLETE:
1011 /* The target will handle the io */
1014 DMWARN("unimplemented target endio return value: %d", r);
1020 dec_pending(io, error);
1024 * Return maximum size of I/O possible at the supplied sector up to the current
1027 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1028 sector_t target_offset)
1030 return ti->len - target_offset;
1033 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1035 sector_t target_offset = dm_target_offset(ti, sector);
1036 sector_t len = max_io_len_target_boundary(ti, target_offset);
1040 * Does the target need to split IO even further?
1041 * - varied (per target) IO splitting is a tenet of DM; this
1042 * explains why stacked chunk_sectors based splitting via
1043 * blk_max_size_offset() isn't possible here. So pass in
1044 * ti->max_io_len to override stacked chunk_sectors.
1046 if (ti->max_io_len) {
1047 max_len = blk_max_size_offset(ti->table->md->queue,
1048 target_offset, ti->max_io_len);
1056 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1058 if (len > UINT_MAX) {
1059 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1060 (unsigned long long)len, UINT_MAX);
1061 ti->error = "Maximum size of target IO is too large";
1065 ti->max_io_len = (uint32_t) len;
1069 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1071 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1072 sector_t sector, int *srcu_idx)
1073 __acquires(md->io_barrier)
1075 struct dm_table *map;
1076 struct dm_target *ti;
1078 map = dm_get_live_table(md, srcu_idx);
1082 ti = dm_table_find_target(map, sector);
1089 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1090 long nr_pages, void **kaddr, pfn_t *pfn)
1092 struct mapped_device *md = dax_get_private(dax_dev);
1093 sector_t sector = pgoff * PAGE_SECTORS;
1094 struct dm_target *ti;
1095 long len, ret = -EIO;
1098 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1102 if (!ti->type->direct_access)
1104 len = max_io_len(ti, sector) / PAGE_SECTORS;
1107 nr_pages = min(len, nr_pages);
1108 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1111 dm_put_live_table(md, srcu_idx);
1116 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1117 int blocksize, sector_t start, sector_t len)
1119 struct mapped_device *md = dax_get_private(dax_dev);
1120 struct dm_table *map;
1124 map = dm_get_live_table(md, &srcu_idx);
1128 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1131 dm_put_live_table(md, srcu_idx);
1136 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1137 void *addr, size_t bytes, struct iov_iter *i)
1139 struct mapped_device *md = dax_get_private(dax_dev);
1140 sector_t sector = pgoff * PAGE_SECTORS;
1141 struct dm_target *ti;
1145 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1149 if (!ti->type->dax_copy_from_iter) {
1150 ret = copy_from_iter(addr, bytes, i);
1153 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1155 dm_put_live_table(md, srcu_idx);
1160 static size_t dm_dax_copy_to_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_to_iter) {
1174 ret = copy_to_iter(addr, bytes, i);
1177 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1179 dm_put_live_table(md, srcu_idx);
1184 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
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 (WARN_ON(!ti->type->dax_zero_page_range)) {
1199 * ->zero_page_range() is mandatory dax operation. If we are
1200 * here, something is wrong.
1204 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1206 dm_put_live_table(md, srcu_idx);
1212 * A target may call dm_accept_partial_bio only from the map routine. It is
1213 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1214 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1216 * dm_accept_partial_bio informs the dm that the target only wants to process
1217 * additional n_sectors sectors of the bio and the rest of the data should be
1218 * sent in a next bio.
1220 * A diagram that explains the arithmetics:
1221 * +--------------------+---------------+-------+
1223 * +--------------------+---------------+-------+
1225 * <-------------- *tio->len_ptr --------------->
1226 * <------- bi_size ------->
1229 * Region 1 was already iterated over with bio_advance or similar function.
1230 * (it may be empty if the target doesn't use bio_advance)
1231 * Region 2 is the remaining bio size that the target wants to process.
1232 * (it may be empty if region 1 is non-empty, although there is no reason
1234 * The target requires that region 3 is to be sent in the next bio.
1236 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1237 * the partially processed part (the sum of regions 1+2) must be the same for all
1238 * copies of the bio.
1240 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1242 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1243 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1244 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1245 BUG_ON(bi_size > *tio->len_ptr);
1246 BUG_ON(n_sectors > bi_size);
1247 *tio->len_ptr -= bi_size - n_sectors;
1248 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1250 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1252 static blk_qc_t __map_bio(struct dm_target_io *tio)
1256 struct bio *clone = &tio->clone;
1257 struct dm_io *io = tio->io;
1258 struct dm_target *ti = tio->ti;
1259 blk_qc_t ret = BLK_QC_T_NONE;
1261 clone->bi_end_io = clone_endio;
1264 * Map the clone. If r == 0 we don't need to do
1265 * anything, the target has assumed ownership of
1268 atomic_inc(&io->io_count);
1269 sector = clone->bi_iter.bi_sector;
1271 r = ti->type->map(ti, clone);
1273 case DM_MAPIO_SUBMITTED:
1275 case DM_MAPIO_REMAPPED:
1276 /* the bio has been remapped so dispatch it */
1277 trace_block_bio_remap(clone->bi_disk->queue, clone,
1278 bio_dev(io->orig_bio), sector);
1279 ret = submit_bio_noacct(clone);
1283 dec_pending(io, BLK_STS_IOERR);
1285 case DM_MAPIO_REQUEUE:
1287 dec_pending(io, BLK_STS_DM_REQUEUE);
1290 DMWARN("unimplemented target map return value: %d", r);
1297 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1299 bio->bi_iter.bi_sector = sector;
1300 bio->bi_iter.bi_size = to_bytes(len);
1304 * Creates a bio that consists of range of complete bvecs.
1306 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1307 sector_t sector, unsigned len)
1309 struct bio *clone = &tio->clone;
1312 __bio_clone_fast(clone, bio);
1314 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1318 if (bio_integrity(bio)) {
1319 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1320 !dm_target_passes_integrity(tio->ti->type))) {
1321 DMWARN("%s: the target %s doesn't support integrity data.",
1322 dm_device_name(tio->io->md),
1323 tio->ti->type->name);
1327 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1332 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1333 clone->bi_iter.bi_size = to_bytes(len);
1335 if (bio_integrity(bio))
1336 bio_integrity_trim(clone);
1341 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1342 struct dm_target *ti, unsigned num_bios)
1344 struct dm_target_io *tio;
1350 if (num_bios == 1) {
1351 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1352 bio_list_add(blist, &tio->clone);
1356 for (try = 0; try < 2; try++) {
1361 mutex_lock(&ci->io->md->table_devices_lock);
1362 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1363 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1367 bio_list_add(blist, &tio->clone);
1370 mutex_unlock(&ci->io->md->table_devices_lock);
1371 if (bio_nr == num_bios)
1374 while ((bio = bio_list_pop(blist))) {
1375 tio = container_of(bio, struct dm_target_io, clone);
1381 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1382 struct dm_target_io *tio, unsigned *len)
1384 struct bio *clone = &tio->clone;
1388 __bio_clone_fast(clone, ci->bio);
1390 bio_setup_sector(clone, ci->sector, *len);
1392 return __map_bio(tio);
1395 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1396 unsigned num_bios, unsigned *len)
1398 struct bio_list blist = BIO_EMPTY_LIST;
1400 struct dm_target_io *tio;
1402 alloc_multiple_bios(&blist, ci, ti, num_bios);
1404 while ((bio = bio_list_pop(&blist))) {
1405 tio = container_of(bio, struct dm_target_io, clone);
1406 (void) __clone_and_map_simple_bio(ci, tio, len);
1410 static int __send_empty_flush(struct clone_info *ci)
1412 unsigned target_nr = 0;
1413 struct dm_target *ti;
1414 struct bio flush_bio;
1417 * Use an on-stack bio for this, it's safe since we don't
1418 * need to reference it after submit. It's just used as
1419 * the basis for the clone(s).
1421 bio_init(&flush_bio, NULL, 0);
1422 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1423 ci->bio = &flush_bio;
1424 ci->sector_count = 0;
1427 * Empty flush uses a statically initialized bio, as the base for
1428 * cloning. However, blkg association requires that a bdev is
1429 * associated with a gendisk, which doesn't happen until the bdev is
1430 * opened. So, blkg association is done at issue time of the flush
1431 * rather than when the device is created in alloc_dev().
1433 bio_set_dev(ci->bio, ci->io->md->bdev);
1435 BUG_ON(bio_has_data(ci->bio));
1436 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1437 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1439 bio_uninit(ci->bio);
1443 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1444 sector_t sector, unsigned *len)
1446 struct bio *bio = ci->bio;
1447 struct dm_target_io *tio;
1450 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1452 r = clone_bio(tio, bio, sector, *len);
1457 (void) __map_bio(tio);
1462 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1468 * Even though the device advertised support for this type of
1469 * request, that does not mean every target supports it, and
1470 * reconfiguration might also have changed that since the
1471 * check was performed.
1476 len = min_t(sector_t, ci->sector_count,
1477 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1479 __send_duplicate_bios(ci, ti, num_bios, &len);
1482 ci->sector_count -= len;
1487 static bool is_abnormal_io(struct bio *bio)
1491 switch (bio_op(bio)) {
1492 case REQ_OP_DISCARD:
1493 case REQ_OP_SECURE_ERASE:
1494 case REQ_OP_WRITE_SAME:
1495 case REQ_OP_WRITE_ZEROES:
1503 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1506 struct bio *bio = ci->bio;
1507 unsigned num_bios = 0;
1509 switch (bio_op(bio)) {
1510 case REQ_OP_DISCARD:
1511 num_bios = ti->num_discard_bios;
1513 case REQ_OP_SECURE_ERASE:
1514 num_bios = ti->num_secure_erase_bios;
1516 case REQ_OP_WRITE_SAME:
1517 num_bios = ti->num_write_same_bios;
1519 case REQ_OP_WRITE_ZEROES:
1520 num_bios = ti->num_write_zeroes_bios;
1526 *result = __send_changing_extent_only(ci, ti, num_bios);
1531 * Select the correct strategy for processing a non-flush bio.
1533 static int __split_and_process_non_flush(struct clone_info *ci)
1535 struct dm_target *ti;
1539 ti = dm_table_find_target(ci->map, ci->sector);
1543 if (__process_abnormal_io(ci, ti, &r))
1546 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1548 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1553 ci->sector_count -= len;
1558 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1559 struct dm_table *map, struct bio *bio)
1562 ci->io = alloc_io(md, bio);
1563 ci->sector = bio->bi_iter.bi_sector;
1566 #define __dm_part_stat_sub(part, field, subnd) \
1567 (part_stat_get(part, field) -= (subnd))
1570 * Entry point to split a bio into clones and submit them to the targets.
1572 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1573 struct dm_table *map, struct bio *bio)
1575 struct clone_info ci;
1576 blk_qc_t ret = BLK_QC_T_NONE;
1579 init_clone_info(&ci, md, map, bio);
1581 if (bio->bi_opf & REQ_PREFLUSH) {
1582 error = __send_empty_flush(&ci);
1583 /* dec_pending submits any data associated with flush */
1584 } else if (op_is_zone_mgmt(bio_op(bio))) {
1586 ci.sector_count = 0;
1587 error = __split_and_process_non_flush(&ci);
1590 ci.sector_count = bio_sectors(bio);
1591 while (ci.sector_count && !error) {
1592 error = __split_and_process_non_flush(&ci);
1593 if (current->bio_list && ci.sector_count && !error) {
1595 * Remainder must be passed to submit_bio_noacct()
1596 * so that it gets handled *after* bios already submitted
1597 * have been completely processed.
1598 * We take a clone of the original to store in
1599 * ci.io->orig_bio to be used by end_io_acct() and
1600 * for dec_pending to use for completion handling.
1602 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1603 GFP_NOIO, &md->queue->bio_split);
1604 ci.io->orig_bio = b;
1607 * Adjust IO stats for each split, otherwise upon queue
1608 * reentry there will be redundant IO accounting.
1609 * NOTE: this is a stop-gap fix, a proper fix involves
1610 * significant refactoring of DM core's bio splitting
1611 * (by eliminating DM's splitting and just using bio_split)
1614 __dm_part_stat_sub(&dm_disk(md)->part0,
1615 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1619 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1620 ret = submit_bio_noacct(bio);
1626 /* drop the extra reference count */
1627 dec_pending(ci.io, errno_to_blk_status(error));
1631 static blk_qc_t dm_submit_bio(struct bio *bio)
1633 struct mapped_device *md = bio->bi_disk->private_data;
1634 blk_qc_t ret = BLK_QC_T_NONE;
1636 struct dm_table *map;
1638 map = dm_get_live_table(md, &srcu_idx);
1639 if (unlikely(!map)) {
1640 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1641 dm_device_name(md));
1646 /* If suspended, queue this IO for later */
1647 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1648 if (bio->bi_opf & REQ_NOWAIT)
1649 bio_wouldblock_error(bio);
1650 else if (bio->bi_opf & REQ_RAHEAD)
1658 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1659 * otherwise associated queue_limits won't be imposed.
1661 if (is_abnormal_io(bio))
1662 blk_queue_split(&bio);
1664 ret = __split_and_process_bio(md, map, bio);
1666 dm_put_live_table(md, srcu_idx);
1670 /*-----------------------------------------------------------------
1671 * An IDR is used to keep track of allocated minor numbers.
1672 *---------------------------------------------------------------*/
1673 static void free_minor(int minor)
1675 spin_lock(&_minor_lock);
1676 idr_remove(&_minor_idr, minor);
1677 spin_unlock(&_minor_lock);
1681 * See if the device with a specific minor # is free.
1683 static int specific_minor(int minor)
1687 if (minor >= (1 << MINORBITS))
1690 idr_preload(GFP_KERNEL);
1691 spin_lock(&_minor_lock);
1693 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1695 spin_unlock(&_minor_lock);
1698 return r == -ENOSPC ? -EBUSY : r;
1702 static int next_free_minor(int *minor)
1706 idr_preload(GFP_KERNEL);
1707 spin_lock(&_minor_lock);
1709 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1711 spin_unlock(&_minor_lock);
1719 static const struct block_device_operations dm_blk_dops;
1720 static const struct block_device_operations dm_rq_blk_dops;
1721 static const struct dax_operations dm_dax_ops;
1723 static void dm_wq_work(struct work_struct *work);
1725 static void cleanup_mapped_device(struct mapped_device *md)
1728 destroy_workqueue(md->wq);
1729 bioset_exit(&md->bs);
1730 bioset_exit(&md->io_bs);
1733 kill_dax(md->dax_dev);
1734 put_dax(md->dax_dev);
1739 spin_lock(&_minor_lock);
1740 md->disk->private_data = NULL;
1741 spin_unlock(&_minor_lock);
1742 del_gendisk(md->disk);
1747 blk_cleanup_queue(md->queue);
1749 cleanup_srcu_struct(&md->io_barrier);
1756 mutex_destroy(&md->suspend_lock);
1757 mutex_destroy(&md->type_lock);
1758 mutex_destroy(&md->table_devices_lock);
1760 dm_mq_cleanup_mapped_device(md);
1764 * Allocate and initialise a blank device with a given minor.
1766 static struct mapped_device *alloc_dev(int minor)
1768 int r, numa_node_id = dm_get_numa_node();
1769 struct mapped_device *md;
1772 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1774 DMWARN("unable to allocate device, out of memory.");
1778 if (!try_module_get(THIS_MODULE))
1779 goto bad_module_get;
1781 /* get a minor number for the dev */
1782 if (minor == DM_ANY_MINOR)
1783 r = next_free_minor(&minor);
1785 r = specific_minor(minor);
1789 r = init_srcu_struct(&md->io_barrier);
1791 goto bad_io_barrier;
1793 md->numa_node_id = numa_node_id;
1794 md->init_tio_pdu = false;
1795 md->type = DM_TYPE_NONE;
1796 mutex_init(&md->suspend_lock);
1797 mutex_init(&md->type_lock);
1798 mutex_init(&md->table_devices_lock);
1799 spin_lock_init(&md->deferred_lock);
1800 atomic_set(&md->holders, 1);
1801 atomic_set(&md->open_count, 0);
1802 atomic_set(&md->event_nr, 0);
1803 atomic_set(&md->uevent_seq, 0);
1804 INIT_LIST_HEAD(&md->uevent_list);
1805 INIT_LIST_HEAD(&md->table_devices);
1806 spin_lock_init(&md->uevent_lock);
1809 * default to bio-based until DM table is loaded and md->type
1810 * established. If request-based table is loaded: blk-mq will
1811 * override accordingly.
1813 md->queue = blk_alloc_queue(numa_node_id);
1817 md->disk = alloc_disk_node(1, md->numa_node_id);
1821 init_waitqueue_head(&md->wait);
1822 INIT_WORK(&md->work, dm_wq_work);
1823 init_waitqueue_head(&md->eventq);
1824 init_completion(&md->kobj_holder.completion);
1826 md->disk->major = _major;
1827 md->disk->first_minor = minor;
1828 md->disk->fops = &dm_blk_dops;
1829 md->disk->queue = md->queue;
1830 md->disk->private_data = md;
1831 sprintf(md->disk->disk_name, "dm-%d", minor);
1833 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1834 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1836 if (IS_ERR(md->dax_dev))
1840 add_disk_no_queue_reg(md->disk);
1841 format_dev_t(md->name, MKDEV(_major, minor));
1843 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1847 md->bdev = bdget_disk(md->disk, 0);
1851 dm_stats_init(&md->stats);
1853 /* Populate the mapping, nobody knows we exist yet */
1854 spin_lock(&_minor_lock);
1855 old_md = idr_replace(&_minor_idr, md, minor);
1856 spin_unlock(&_minor_lock);
1858 BUG_ON(old_md != MINOR_ALLOCED);
1863 cleanup_mapped_device(md);
1867 module_put(THIS_MODULE);
1873 static void unlock_fs(struct mapped_device *md);
1875 static void free_dev(struct mapped_device *md)
1877 int minor = MINOR(disk_devt(md->disk));
1881 cleanup_mapped_device(md);
1883 free_table_devices(&md->table_devices);
1884 dm_stats_cleanup(&md->stats);
1887 module_put(THIS_MODULE);
1891 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1893 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1896 if (dm_table_bio_based(t)) {
1898 * The md may already have mempools that need changing.
1899 * If so, reload bioset because front_pad may have changed
1900 * because a different table was loaded.
1902 bioset_exit(&md->bs);
1903 bioset_exit(&md->io_bs);
1905 } else if (bioset_initialized(&md->bs)) {
1907 * There's no need to reload with request-based dm
1908 * because the size of front_pad doesn't change.
1909 * Note for future: If you are to reload bioset,
1910 * prep-ed requests in the queue may refer
1911 * to bio from the old bioset, so you must walk
1912 * through the queue to unprep.
1918 bioset_initialized(&md->bs) ||
1919 bioset_initialized(&md->io_bs));
1921 ret = bioset_init_from_src(&md->bs, &p->bs);
1924 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1926 bioset_exit(&md->bs);
1928 /* mempool bind completed, no longer need any mempools in the table */
1929 dm_table_free_md_mempools(t);
1934 * Bind a table to the device.
1936 static void event_callback(void *context)
1938 unsigned long flags;
1940 struct mapped_device *md = (struct mapped_device *) context;
1942 spin_lock_irqsave(&md->uevent_lock, flags);
1943 list_splice_init(&md->uevent_list, &uevents);
1944 spin_unlock_irqrestore(&md->uevent_lock, flags);
1946 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1948 atomic_inc(&md->event_nr);
1949 wake_up(&md->eventq);
1950 dm_issue_global_event();
1954 * Returns old map, which caller must destroy.
1956 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1957 struct queue_limits *limits)
1959 struct dm_table *old_map;
1960 struct request_queue *q = md->queue;
1961 bool request_based = dm_table_request_based(t);
1965 lockdep_assert_held(&md->suspend_lock);
1967 size = dm_table_get_size(t);
1970 * Wipe any geometry if the size of the table changed.
1972 if (size != dm_get_size(md))
1973 memset(&md->geometry, 0, sizeof(md->geometry));
1975 set_capacity(md->disk, size);
1976 bd_set_nr_sectors(md->bdev, size);
1978 dm_table_event_callback(t, event_callback, md);
1981 * The queue hasn't been stopped yet, if the old table type wasn't
1982 * for request-based during suspension. So stop it to prevent
1983 * I/O mapping before resume.
1984 * This must be done before setting the queue restrictions,
1985 * because request-based dm may be run just after the setting.
1990 if (request_based) {
1992 * Leverage the fact that request-based DM targets are
1993 * immutable singletons - used to optimize dm_mq_queue_rq.
1995 md->immutable_target = dm_table_get_immutable_target(t);
1998 ret = __bind_mempools(md, t);
2000 old_map = ERR_PTR(ret);
2004 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2005 rcu_assign_pointer(md->map, (void *)t);
2006 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2008 dm_table_set_restrictions(t, q, limits);
2017 * Returns unbound table for the caller to free.
2019 static struct dm_table *__unbind(struct mapped_device *md)
2021 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2026 dm_table_event_callback(map, NULL, NULL);
2027 RCU_INIT_POINTER(md->map, NULL);
2034 * Constructor for a new device.
2036 int dm_create(int minor, struct mapped_device **result)
2039 struct mapped_device *md;
2041 md = alloc_dev(minor);
2045 r = dm_sysfs_init(md);
2056 * Functions to manage md->type.
2057 * All are required to hold md->type_lock.
2059 void dm_lock_md_type(struct mapped_device *md)
2061 mutex_lock(&md->type_lock);
2064 void dm_unlock_md_type(struct mapped_device *md)
2066 mutex_unlock(&md->type_lock);
2069 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2071 BUG_ON(!mutex_is_locked(&md->type_lock));
2075 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2080 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2082 return md->immutable_target_type;
2086 * The queue_limits are only valid as long as you have a reference
2089 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2091 BUG_ON(!atomic_read(&md->holders));
2092 return &md->queue->limits;
2094 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2097 * Setup the DM device's queue based on md's type
2099 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2102 struct queue_limits limits;
2103 enum dm_queue_mode type = dm_get_md_type(md);
2106 case DM_TYPE_REQUEST_BASED:
2107 md->disk->fops = &dm_rq_blk_dops;
2108 r = dm_mq_init_request_queue(md, t);
2110 DMERR("Cannot initialize queue for request-based dm mapped device");
2114 case DM_TYPE_BIO_BASED:
2115 case DM_TYPE_DAX_BIO_BASED:
2122 r = dm_calculate_queue_limits(t, &limits);
2124 DMERR("Cannot calculate initial queue limits");
2127 dm_table_set_restrictions(t, md->queue, &limits);
2128 blk_register_queue(md->disk);
2133 struct mapped_device *dm_get_md(dev_t dev)
2135 struct mapped_device *md;
2136 unsigned minor = MINOR(dev);
2138 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2141 spin_lock(&_minor_lock);
2143 md = idr_find(&_minor_idr, minor);
2144 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2145 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2151 spin_unlock(&_minor_lock);
2155 EXPORT_SYMBOL_GPL(dm_get_md);
2157 void *dm_get_mdptr(struct mapped_device *md)
2159 return md->interface_ptr;
2162 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2164 md->interface_ptr = ptr;
2167 void dm_get(struct mapped_device *md)
2169 atomic_inc(&md->holders);
2170 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2173 int dm_hold(struct mapped_device *md)
2175 spin_lock(&_minor_lock);
2176 if (test_bit(DMF_FREEING, &md->flags)) {
2177 spin_unlock(&_minor_lock);
2181 spin_unlock(&_minor_lock);
2184 EXPORT_SYMBOL_GPL(dm_hold);
2186 const char *dm_device_name(struct mapped_device *md)
2190 EXPORT_SYMBOL_GPL(dm_device_name);
2192 static void __dm_destroy(struct mapped_device *md, bool wait)
2194 struct dm_table *map;
2199 spin_lock(&_minor_lock);
2200 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2201 set_bit(DMF_FREEING, &md->flags);
2202 spin_unlock(&_minor_lock);
2204 blk_set_queue_dying(md->queue);
2207 * Take suspend_lock so that presuspend and postsuspend methods
2208 * do not race with internal suspend.
2210 mutex_lock(&md->suspend_lock);
2211 map = dm_get_live_table(md, &srcu_idx);
2212 if (!dm_suspended_md(md)) {
2213 dm_table_presuspend_targets(map);
2214 set_bit(DMF_SUSPENDED, &md->flags);
2215 set_bit(DMF_POST_SUSPENDING, &md->flags);
2216 dm_table_postsuspend_targets(map);
2218 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2219 dm_put_live_table(md, srcu_idx);
2220 mutex_unlock(&md->suspend_lock);
2223 * Rare, but there may be I/O requests still going to complete,
2224 * for example. Wait for all references to disappear.
2225 * No one should increment the reference count of the mapped_device,
2226 * after the mapped_device state becomes DMF_FREEING.
2229 while (atomic_read(&md->holders))
2231 else if (atomic_read(&md->holders))
2232 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2233 dm_device_name(md), atomic_read(&md->holders));
2236 dm_table_destroy(__unbind(md));
2240 void dm_destroy(struct mapped_device *md)
2242 __dm_destroy(md, true);
2245 void dm_destroy_immediate(struct mapped_device *md)
2247 __dm_destroy(md, false);
2250 void dm_put(struct mapped_device *md)
2252 atomic_dec(&md->holders);
2254 EXPORT_SYMBOL_GPL(dm_put);
2256 static bool md_in_flight_bios(struct mapped_device *md)
2259 struct hd_struct *part = &dm_disk(md)->part0;
2262 for_each_possible_cpu(cpu) {
2263 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2264 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2270 static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2276 prepare_to_wait(&md->wait, &wait, task_state);
2278 if (!md_in_flight_bios(md))
2281 if (signal_pending_state(task_state, current)) {
2288 finish_wait(&md->wait, &wait);
2293 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2297 if (!queue_is_mq(md->queue))
2298 return dm_wait_for_bios_completion(md, task_state);
2301 if (!blk_mq_queue_inflight(md->queue))
2304 if (signal_pending_state(task_state, current)) {
2316 * Process the deferred bios
2318 static void dm_wq_work(struct work_struct *work)
2320 struct mapped_device *md = container_of(work, struct mapped_device, work);
2323 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2324 spin_lock_irq(&md->deferred_lock);
2325 bio = bio_list_pop(&md->deferred);
2326 spin_unlock_irq(&md->deferred_lock);
2331 submit_bio_noacct(bio);
2335 static void dm_queue_flush(struct mapped_device *md)
2337 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2338 smp_mb__after_atomic();
2339 queue_work(md->wq, &md->work);
2343 * Swap in a new table, returning the old one for the caller to destroy.
2345 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2347 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2348 struct queue_limits limits;
2351 mutex_lock(&md->suspend_lock);
2353 /* device must be suspended */
2354 if (!dm_suspended_md(md))
2358 * If the new table has no data devices, retain the existing limits.
2359 * This helps multipath with queue_if_no_path if all paths disappear,
2360 * then new I/O is queued based on these limits, and then some paths
2363 if (dm_table_has_no_data_devices(table)) {
2364 live_map = dm_get_live_table_fast(md);
2366 limits = md->queue->limits;
2367 dm_put_live_table_fast(md);
2371 r = dm_calculate_queue_limits(table, &limits);
2378 map = __bind(md, table, &limits);
2379 dm_issue_global_event();
2382 mutex_unlock(&md->suspend_lock);
2387 * Functions to lock and unlock any filesystem running on the
2390 static int lock_fs(struct mapped_device *md)
2394 WARN_ON(md->frozen_sb);
2396 md->frozen_sb = freeze_bdev(md->bdev);
2397 if (IS_ERR(md->frozen_sb)) {
2398 r = PTR_ERR(md->frozen_sb);
2399 md->frozen_sb = NULL;
2403 set_bit(DMF_FROZEN, &md->flags);
2408 static void unlock_fs(struct mapped_device *md)
2410 if (!test_bit(DMF_FROZEN, &md->flags))
2413 thaw_bdev(md->bdev, md->frozen_sb);
2414 md->frozen_sb = NULL;
2415 clear_bit(DMF_FROZEN, &md->flags);
2419 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2420 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2421 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2423 * If __dm_suspend returns 0, the device is completely quiescent
2424 * now. There is no request-processing activity. All new requests
2425 * are being added to md->deferred list.
2427 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2428 unsigned suspend_flags, long task_state,
2429 int dmf_suspended_flag)
2431 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2432 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2435 lockdep_assert_held(&md->suspend_lock);
2438 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2439 * This flag is cleared before dm_suspend returns.
2442 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2444 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2447 * This gets reverted if there's an error later and the targets
2448 * provide the .presuspend_undo hook.
2450 dm_table_presuspend_targets(map);
2453 * Flush I/O to the device.
2454 * Any I/O submitted after lock_fs() may not be flushed.
2455 * noflush takes precedence over do_lockfs.
2456 * (lock_fs() flushes I/Os and waits for them to complete.)
2458 if (!noflush && do_lockfs) {
2461 dm_table_presuspend_undo_targets(map);
2467 * Here we must make sure that no processes are submitting requests
2468 * to target drivers i.e. no one may be executing
2469 * __split_and_process_bio from dm_submit_bio.
2471 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2472 * we take the write lock. To prevent any process from reentering
2473 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2474 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2475 * flush_workqueue(md->wq).
2477 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2479 synchronize_srcu(&md->io_barrier);
2482 * Stop md->queue before flushing md->wq in case request-based
2483 * dm defers requests to md->wq from md->queue.
2485 if (dm_request_based(md))
2486 dm_stop_queue(md->queue);
2488 flush_workqueue(md->wq);
2491 * At this point no more requests are entering target request routines.
2492 * We call dm_wait_for_completion to wait for all existing requests
2495 r = dm_wait_for_completion(md, task_state);
2497 set_bit(dmf_suspended_flag, &md->flags);
2500 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2502 synchronize_srcu(&md->io_barrier);
2504 /* were we interrupted ? */
2508 if (dm_request_based(md))
2509 dm_start_queue(md->queue);
2512 dm_table_presuspend_undo_targets(map);
2513 /* pushback list is already flushed, so skip flush */
2520 * We need to be able to change a mapping table under a mounted
2521 * filesystem. For example we might want to move some data in
2522 * the background. Before the table can be swapped with
2523 * dm_bind_table, dm_suspend must be called to flush any in
2524 * flight bios and ensure that any further io gets deferred.
2527 * Suspend mechanism in request-based dm.
2529 * 1. Flush all I/Os by lock_fs() if needed.
2530 * 2. Stop dispatching any I/O by stopping the request_queue.
2531 * 3. Wait for all in-flight I/Os to be completed or requeued.
2533 * To abort suspend, start the request_queue.
2535 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2537 struct dm_table *map = NULL;
2541 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2543 if (dm_suspended_md(md)) {
2548 if (dm_suspended_internally_md(md)) {
2549 /* already internally suspended, wait for internal resume */
2550 mutex_unlock(&md->suspend_lock);
2551 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2557 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2559 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2563 set_bit(DMF_POST_SUSPENDING, &md->flags);
2564 dm_table_postsuspend_targets(map);
2565 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2568 mutex_unlock(&md->suspend_lock);
2572 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2575 int r = dm_table_resume_targets(map);
2583 * Flushing deferred I/Os must be done after targets are resumed
2584 * so that mapping of targets can work correctly.
2585 * Request-based dm is queueing the deferred I/Os in its request_queue.
2587 if (dm_request_based(md))
2588 dm_start_queue(md->queue);
2595 int dm_resume(struct mapped_device *md)
2598 struct dm_table *map = NULL;
2602 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2604 if (!dm_suspended_md(md))
2607 if (dm_suspended_internally_md(md)) {
2608 /* already internally suspended, wait for internal resume */
2609 mutex_unlock(&md->suspend_lock);
2610 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2616 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2617 if (!map || !dm_table_get_size(map))
2620 r = __dm_resume(md, map);
2624 clear_bit(DMF_SUSPENDED, &md->flags);
2626 mutex_unlock(&md->suspend_lock);
2632 * Internal suspend/resume works like userspace-driven suspend. It waits
2633 * until all bios finish and prevents issuing new bios to the target drivers.
2634 * It may be used only from the kernel.
2637 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2639 struct dm_table *map = NULL;
2641 lockdep_assert_held(&md->suspend_lock);
2643 if (md->internal_suspend_count++)
2644 return; /* nested internal suspend */
2646 if (dm_suspended_md(md)) {
2647 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2648 return; /* nest suspend */
2651 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2654 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2655 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2656 * would require changing .presuspend to return an error -- avoid this
2657 * until there is a need for more elaborate variants of internal suspend.
2659 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2660 DMF_SUSPENDED_INTERNALLY);
2662 set_bit(DMF_POST_SUSPENDING, &md->flags);
2663 dm_table_postsuspend_targets(map);
2664 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2667 static void __dm_internal_resume(struct mapped_device *md)
2669 BUG_ON(!md->internal_suspend_count);
2671 if (--md->internal_suspend_count)
2672 return; /* resume from nested internal suspend */
2674 if (dm_suspended_md(md))
2675 goto done; /* resume from nested suspend */
2678 * NOTE: existing callers don't need to call dm_table_resume_targets
2679 * (which may fail -- so best to avoid it for now by passing NULL map)
2681 (void) __dm_resume(md, NULL);
2684 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2685 smp_mb__after_atomic();
2686 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2689 void dm_internal_suspend_noflush(struct mapped_device *md)
2691 mutex_lock(&md->suspend_lock);
2692 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2693 mutex_unlock(&md->suspend_lock);
2695 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2697 void dm_internal_resume(struct mapped_device *md)
2699 mutex_lock(&md->suspend_lock);
2700 __dm_internal_resume(md);
2701 mutex_unlock(&md->suspend_lock);
2703 EXPORT_SYMBOL_GPL(dm_internal_resume);
2706 * Fast variants of internal suspend/resume hold md->suspend_lock,
2707 * which prevents interaction with userspace-driven suspend.
2710 void dm_internal_suspend_fast(struct mapped_device *md)
2712 mutex_lock(&md->suspend_lock);
2713 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2716 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2717 synchronize_srcu(&md->io_barrier);
2718 flush_workqueue(md->wq);
2719 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2721 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2723 void dm_internal_resume_fast(struct mapped_device *md)
2725 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2731 mutex_unlock(&md->suspend_lock);
2733 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2735 /*-----------------------------------------------------------------
2736 * Event notification.
2737 *---------------------------------------------------------------*/
2738 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2743 char udev_cookie[DM_COOKIE_LENGTH];
2744 char *envp[] = { udev_cookie, NULL };
2746 noio_flag = memalloc_noio_save();
2749 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2751 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2752 DM_COOKIE_ENV_VAR_NAME, cookie);
2753 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2757 memalloc_noio_restore(noio_flag);
2762 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2764 return atomic_add_return(1, &md->uevent_seq);
2767 uint32_t dm_get_event_nr(struct mapped_device *md)
2769 return atomic_read(&md->event_nr);
2772 int dm_wait_event(struct mapped_device *md, int event_nr)
2774 return wait_event_interruptible(md->eventq,
2775 (event_nr != atomic_read(&md->event_nr)));
2778 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2780 unsigned long flags;
2782 spin_lock_irqsave(&md->uevent_lock, flags);
2783 list_add(elist, &md->uevent_list);
2784 spin_unlock_irqrestore(&md->uevent_lock, flags);
2788 * The gendisk is only valid as long as you have a reference
2791 struct gendisk *dm_disk(struct mapped_device *md)
2795 EXPORT_SYMBOL_GPL(dm_disk);
2797 struct kobject *dm_kobject(struct mapped_device *md)
2799 return &md->kobj_holder.kobj;
2802 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2804 struct mapped_device *md;
2806 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2808 spin_lock(&_minor_lock);
2809 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2815 spin_unlock(&_minor_lock);
2820 int dm_suspended_md(struct mapped_device *md)
2822 return test_bit(DMF_SUSPENDED, &md->flags);
2825 static int dm_post_suspending_md(struct mapped_device *md)
2827 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2830 int dm_suspended_internally_md(struct mapped_device *md)
2832 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2835 int dm_test_deferred_remove_flag(struct mapped_device *md)
2837 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2840 int dm_suspended(struct dm_target *ti)
2842 return dm_suspended_md(ti->table->md);
2844 EXPORT_SYMBOL_GPL(dm_suspended);
2846 int dm_post_suspending(struct dm_target *ti)
2848 return dm_post_suspending_md(ti->table->md);
2850 EXPORT_SYMBOL_GPL(dm_post_suspending);
2852 int dm_noflush_suspending(struct dm_target *ti)
2854 return __noflush_suspending(ti->table->md);
2856 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2858 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2859 unsigned integrity, unsigned per_io_data_size,
2860 unsigned min_pool_size)
2862 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2863 unsigned int pool_size = 0;
2864 unsigned int front_pad, io_front_pad;
2871 case DM_TYPE_BIO_BASED:
2872 case DM_TYPE_DAX_BIO_BASED:
2873 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2874 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2875 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2876 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2879 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2882 case DM_TYPE_REQUEST_BASED:
2883 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2884 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2885 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2891 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2895 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2901 dm_free_md_mempools(pools);
2906 void dm_free_md_mempools(struct dm_md_mempools *pools)
2911 bioset_exit(&pools->bs);
2912 bioset_exit(&pools->io_bs);
2924 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2927 struct mapped_device *md = bdev->bd_disk->private_data;
2928 struct dm_table *table;
2929 struct dm_target *ti;
2930 int ret = -ENOTTY, srcu_idx;
2932 table = dm_get_live_table(md, &srcu_idx);
2933 if (!table || !dm_table_get_size(table))
2936 /* We only support devices that have a single target */
2937 if (dm_table_get_num_targets(table) != 1)
2939 ti = dm_table_get_target(table, 0);
2942 if (!ti->type->iterate_devices)
2945 ret = ti->type->iterate_devices(ti, fn, data);
2947 dm_put_live_table(md, srcu_idx);
2952 * For register / unregister we need to manually call out to every path.
2954 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2955 sector_t start, sector_t len, void *data)
2957 struct dm_pr *pr = data;
2958 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2960 if (!ops || !ops->pr_register)
2962 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2965 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2976 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2977 if (ret && new_key) {
2978 /* unregister all paths if we failed to register any path */
2979 pr.old_key = new_key;
2982 pr.fail_early = false;
2983 dm_call_pr(bdev, __dm_pr_register, &pr);
2989 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2992 struct mapped_device *md = bdev->bd_disk->private_data;
2993 const struct pr_ops *ops;
2996 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3000 ops = bdev->bd_disk->fops->pr_ops;
3001 if (ops && ops->pr_reserve)
3002 r = ops->pr_reserve(bdev, key, type, flags);
3006 dm_unprepare_ioctl(md, srcu_idx);
3010 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3012 struct mapped_device *md = bdev->bd_disk->private_data;
3013 const struct pr_ops *ops;
3016 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3020 ops = bdev->bd_disk->fops->pr_ops;
3021 if (ops && ops->pr_release)
3022 r = ops->pr_release(bdev, key, type);
3026 dm_unprepare_ioctl(md, srcu_idx);
3030 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3031 enum pr_type type, bool abort)
3033 struct mapped_device *md = bdev->bd_disk->private_data;
3034 const struct pr_ops *ops;
3037 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3041 ops = bdev->bd_disk->fops->pr_ops;
3042 if (ops && ops->pr_preempt)
3043 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3047 dm_unprepare_ioctl(md, srcu_idx);
3051 static int dm_pr_clear(struct block_device *bdev, u64 key)
3053 struct mapped_device *md = bdev->bd_disk->private_data;
3054 const struct pr_ops *ops;
3057 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3061 ops = bdev->bd_disk->fops->pr_ops;
3062 if (ops && ops->pr_clear)
3063 r = ops->pr_clear(bdev, key);
3067 dm_unprepare_ioctl(md, srcu_idx);
3071 static const struct pr_ops dm_pr_ops = {
3072 .pr_register = dm_pr_register,
3073 .pr_reserve = dm_pr_reserve,
3074 .pr_release = dm_pr_release,
3075 .pr_preempt = dm_pr_preempt,
3076 .pr_clear = dm_pr_clear,
3079 static const struct block_device_operations dm_blk_dops = {
3080 .submit_bio = dm_submit_bio,
3081 .open = dm_blk_open,
3082 .release = dm_blk_close,
3083 .ioctl = dm_blk_ioctl,
3084 .getgeo = dm_blk_getgeo,
3085 .report_zones = dm_blk_report_zones,
3086 .pr_ops = &dm_pr_ops,
3087 .owner = THIS_MODULE
3090 static const struct block_device_operations dm_rq_blk_dops = {
3091 .open = dm_blk_open,
3092 .release = dm_blk_close,
3093 .ioctl = dm_blk_ioctl,
3094 .getgeo = dm_blk_getgeo,
3095 .pr_ops = &dm_pr_ops,
3096 .owner = THIS_MODULE
3099 static const struct dax_operations dm_dax_ops = {
3100 .direct_access = dm_dax_direct_access,
3101 .dax_supported = dm_dax_supported,
3102 .copy_from_iter = dm_dax_copy_from_iter,
3103 .copy_to_iter = dm_dax_copy_to_iter,
3104 .zero_page_range = dm_dax_zero_page_range,
3110 module_init(dm_init);
3111 module_exit(dm_exit);
3113 module_param(major, uint, 0);
3114 MODULE_PARM_DESC(major, "The major number of the device mapper");
3116 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3117 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3119 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3120 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3122 MODULE_DESCRIPTION(DM_NAME " driver");
3123 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3124 MODULE_LICENSE("GPL");