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>
30 #define DM_MSG_PREFIX "core"
33 * Cookies are numeric values sent with CHANGE and REMOVE
34 * uevents while resuming, removing or renaming the device.
36 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
37 #define DM_COOKIE_LENGTH 24
39 static const char *_name = DM_NAME;
41 static unsigned int major = 0;
42 static unsigned int _major = 0;
44 static DEFINE_IDR(_minor_idr);
46 static DEFINE_SPINLOCK(_minor_lock);
48 static void do_deferred_remove(struct work_struct *w);
50 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
52 static struct workqueue_struct *deferred_remove_workqueue;
54 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
55 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
57 void dm_issue_global_event(void)
59 atomic_inc(&dm_global_event_nr);
60 wake_up(&dm_global_eventq);
64 * One of these is allocated (on-stack) per original bio.
71 unsigned sector_count;
75 * One of these is allocated per clone bio.
77 #define DM_TIO_MAGIC 7282014
82 unsigned target_bio_nr;
89 * One of these is allocated per original bio.
90 * It contains the first clone used for that original.
92 #define DM_IO_MAGIC 5191977
95 struct mapped_device *md;
99 unsigned long start_time;
100 spinlock_t endio_lock;
101 struct dm_stats_aux stats_aux;
102 /* last member of dm_target_io is 'struct bio' */
103 struct dm_target_io tio;
106 void *dm_per_bio_data(struct bio *bio, size_t data_size)
108 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
109 if (!tio->inside_dm_io)
110 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
111 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
113 EXPORT_SYMBOL_GPL(dm_per_bio_data);
115 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
117 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
118 if (io->magic == DM_IO_MAGIC)
119 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
120 BUG_ON(io->magic != DM_TIO_MAGIC);
121 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
123 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
125 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
127 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
129 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
131 #define MINOR_ALLOCED ((void *)-1)
134 * Bits for the md->flags field.
136 #define DMF_BLOCK_IO_FOR_SUSPEND 0
137 #define DMF_SUSPENDED 1
139 #define DMF_FREEING 3
140 #define DMF_DELETING 4
141 #define DMF_NOFLUSH_SUSPENDING 5
142 #define DMF_DEFERRED_REMOVE 6
143 #define DMF_SUSPENDED_INTERNALLY 7
144 #define DMF_POST_SUSPENDING 8
146 #define DM_NUMA_NODE NUMA_NO_NODE
147 static int dm_numa_node = DM_NUMA_NODE;
149 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
150 static int swap_bios = DEFAULT_SWAP_BIOS;
151 static int get_swap_bios(void)
153 int latch = READ_ONCE(swap_bios);
154 if (unlikely(latch <= 0))
155 latch = DEFAULT_SWAP_BIOS;
160 * For mempools pre-allocation at the table loading time.
162 struct dm_md_mempools {
164 struct bio_set io_bs;
167 struct table_device {
168 struct list_head list;
170 struct dm_dev dm_dev;
173 static struct kmem_cache *_rq_tio_cache;
174 static struct kmem_cache *_rq_cache;
177 * Bio-based DM's mempools' reserved IOs set by the user.
179 #define RESERVED_BIO_BASED_IOS 16
180 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
182 static int __dm_get_module_param_int(int *module_param, int min, int max)
184 int param = READ_ONCE(*module_param);
185 int modified_param = 0;
186 bool modified = true;
189 modified_param = min;
190 else if (param > max)
191 modified_param = max;
196 (void)cmpxchg(module_param, param, modified_param);
197 param = modified_param;
203 unsigned __dm_get_module_param(unsigned *module_param,
204 unsigned def, unsigned max)
206 unsigned param = READ_ONCE(*module_param);
207 unsigned modified_param = 0;
210 modified_param = def;
211 else if (param > max)
212 modified_param = max;
214 if (modified_param) {
215 (void)cmpxchg(module_param, param, modified_param);
216 param = modified_param;
222 unsigned dm_get_reserved_bio_based_ios(void)
224 return __dm_get_module_param(&reserved_bio_based_ios,
225 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
227 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
229 static unsigned dm_get_numa_node(void)
231 return __dm_get_module_param_int(&dm_numa_node,
232 DM_NUMA_NODE, num_online_nodes() - 1);
235 static int __init local_init(void)
239 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
243 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
244 __alignof__(struct request), 0, NULL);
246 goto out_free_rq_tio_cache;
248 r = dm_uevent_init();
250 goto out_free_rq_cache;
252 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
253 if (!deferred_remove_workqueue) {
255 goto out_uevent_exit;
259 r = register_blkdev(_major, _name);
261 goto out_free_workqueue;
269 destroy_workqueue(deferred_remove_workqueue);
273 kmem_cache_destroy(_rq_cache);
274 out_free_rq_tio_cache:
275 kmem_cache_destroy(_rq_tio_cache);
280 static void local_exit(void)
282 flush_scheduled_work();
283 destroy_workqueue(deferred_remove_workqueue);
285 kmem_cache_destroy(_rq_cache);
286 kmem_cache_destroy(_rq_tio_cache);
287 unregister_blkdev(_major, _name);
292 DMINFO("cleaned up");
295 static int (*_inits[])(void) __initdata = {
306 static void (*_exits[])(void) = {
317 static int __init dm_init(void)
319 const int count = ARRAY_SIZE(_inits);
323 for (i = 0; i < count; i++) {
338 static void __exit dm_exit(void)
340 int i = ARRAY_SIZE(_exits);
346 * Should be empty by this point.
348 idr_destroy(&_minor_idr);
352 * Block device functions
354 int dm_deleting_md(struct mapped_device *md)
356 return test_bit(DMF_DELETING, &md->flags);
359 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
361 struct mapped_device *md;
363 spin_lock(&_minor_lock);
365 md = bdev->bd_disk->private_data;
369 if (test_bit(DMF_FREEING, &md->flags) ||
370 dm_deleting_md(md)) {
376 atomic_inc(&md->open_count);
378 spin_unlock(&_minor_lock);
380 return md ? 0 : -ENXIO;
383 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
385 struct mapped_device *md;
387 spin_lock(&_minor_lock);
389 md = disk->private_data;
393 if (atomic_dec_and_test(&md->open_count) &&
394 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
395 queue_work(deferred_remove_workqueue, &deferred_remove_work);
399 spin_unlock(&_minor_lock);
402 int dm_open_count(struct mapped_device *md)
404 return atomic_read(&md->open_count);
408 * Guarantees nothing is using the device before it's deleted.
410 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
414 spin_lock(&_minor_lock);
416 if (dm_open_count(md)) {
419 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
420 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
423 set_bit(DMF_DELETING, &md->flags);
425 spin_unlock(&_minor_lock);
430 int dm_cancel_deferred_remove(struct mapped_device *md)
434 spin_lock(&_minor_lock);
436 if (test_bit(DMF_DELETING, &md->flags))
439 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
441 spin_unlock(&_minor_lock);
446 static void do_deferred_remove(struct work_struct *w)
448 dm_deferred_remove();
451 sector_t dm_get_size(struct mapped_device *md)
453 return get_capacity(md->disk);
456 struct request_queue *dm_get_md_queue(struct mapped_device *md)
461 struct dm_stats *dm_get_stats(struct mapped_device *md)
466 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
468 struct mapped_device *md = bdev->bd_disk->private_data;
470 return dm_get_geometry(md, geo);
473 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
474 struct block_device **bdev)
476 struct dm_target *tgt;
477 struct dm_table *map;
482 map = dm_get_live_table(md, srcu_idx);
483 if (!map || !dm_table_get_size(map))
486 /* We only support devices that have a single target */
487 if (dm_table_get_num_targets(map) != 1)
490 tgt = dm_table_get_target(map, 0);
491 if (!tgt->type->prepare_ioctl)
494 if (dm_suspended_md(md))
497 r = tgt->type->prepare_ioctl(tgt, bdev);
498 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
499 dm_put_live_table(md, *srcu_idx);
507 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
509 dm_put_live_table(md, srcu_idx);
512 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
513 unsigned int cmd, unsigned long arg)
515 struct mapped_device *md = bdev->bd_disk->private_data;
518 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
524 * Target determined this ioctl is being issued against a
525 * subset of the parent bdev; require extra privileges.
527 if (!capable(CAP_SYS_RAWIO)) {
529 "%s: sending ioctl %x to DM device without required privilege.",
536 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
538 dm_unprepare_ioctl(md, srcu_idx);
542 static void start_io_acct(struct dm_io *io);
544 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
547 struct dm_target_io *tio;
550 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
554 tio = container_of(clone, struct dm_target_io, clone);
555 tio->inside_dm_io = true;
558 io = container_of(tio, struct dm_io, tio);
559 io->magic = DM_IO_MAGIC;
561 atomic_set(&io->io_count, 1);
564 spin_lock_init(&io->endio_lock);
571 static void free_io(struct mapped_device *md, struct dm_io *io)
573 bio_put(&io->tio.clone);
576 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
577 unsigned target_bio_nr, gfp_t gfp_mask)
579 struct dm_target_io *tio;
581 if (!ci->io->tio.io) {
582 /* the dm_target_io embedded in ci->io is available */
585 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
589 tio = container_of(clone, struct dm_target_io, clone);
590 tio->inside_dm_io = false;
593 tio->magic = DM_TIO_MAGIC;
596 tio->target_bio_nr = target_bio_nr;
601 static void free_tio(struct dm_target_io *tio)
603 if (tio->inside_dm_io)
605 bio_put(&tio->clone);
608 int md_in_flight(struct mapped_device *md)
610 return atomic_read(&md->pending[READ]) +
611 atomic_read(&md->pending[WRITE]);
614 static void start_io_acct(struct dm_io *io)
616 struct mapped_device *md = io->md;
617 struct bio *bio = io->orig_bio;
618 int rw = bio_data_dir(bio);
620 io->start_time = jiffies;
622 generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
623 &dm_disk(md)->part0);
625 atomic_set(&dm_disk(md)->part0.in_flight[rw],
626 atomic_inc_return(&md->pending[rw]));
628 if (unlikely(dm_stats_used(&md->stats)))
629 dm_stats_account_io(&md->stats, bio_data_dir(bio),
630 bio->bi_iter.bi_sector, bio_sectors(bio),
631 false, 0, &io->stats_aux);
634 static void end_io_acct(struct dm_io *io)
636 struct mapped_device *md = io->md;
637 struct bio *bio = io->orig_bio;
638 unsigned long duration = jiffies - io->start_time;
640 int rw = bio_data_dir(bio);
642 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
645 if (unlikely(dm_stats_used(&md->stats)))
646 dm_stats_account_io(&md->stats, bio_data_dir(bio),
647 bio->bi_iter.bi_sector, bio_sectors(bio),
648 true, duration, &io->stats_aux);
651 * After this is decremented the bio must not be touched if it is
654 pending = atomic_dec_return(&md->pending[rw]);
655 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
656 pending += atomic_read(&md->pending[rw^0x1]);
658 /* nudge anyone waiting on suspend queue */
664 * Add the bio to the list of deferred io.
666 static void queue_io(struct mapped_device *md, struct bio *bio)
670 spin_lock_irqsave(&md->deferred_lock, flags);
671 bio_list_add(&md->deferred, bio);
672 spin_unlock_irqrestore(&md->deferred_lock, flags);
673 queue_work(md->wq, &md->work);
677 * Everyone (including functions in this file), should use this
678 * function to access the md->map field, and make sure they call
679 * dm_put_live_table() when finished.
681 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
683 *srcu_idx = srcu_read_lock(&md->io_barrier);
685 return srcu_dereference(md->map, &md->io_barrier);
688 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
690 srcu_read_unlock(&md->io_barrier, srcu_idx);
693 void dm_sync_table(struct mapped_device *md)
695 synchronize_srcu(&md->io_barrier);
696 synchronize_rcu_expedited();
700 * A fast alternative to dm_get_live_table/dm_put_live_table.
701 * The caller must not block between these two functions.
703 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
706 return rcu_dereference(md->map);
709 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
714 static char *_dm_claim_ptr = "I belong to device-mapper";
717 * Open a table device so we can use it as a map destination.
719 static int open_table_device(struct table_device *td, dev_t dev,
720 struct mapped_device *md)
722 struct block_device *bdev;
726 BUG_ON(td->dm_dev.bdev);
728 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
730 return PTR_ERR(bdev);
732 r = bd_link_disk_holder(bdev, dm_disk(md));
734 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
738 td->dm_dev.bdev = bdev;
739 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
744 * Close a table device that we've been using.
746 static void close_table_device(struct table_device *td, struct mapped_device *md)
748 if (!td->dm_dev.bdev)
751 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
752 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
753 put_dax(td->dm_dev.dax_dev);
754 td->dm_dev.bdev = NULL;
755 td->dm_dev.dax_dev = NULL;
758 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
760 struct table_device *td;
762 list_for_each_entry(td, l, list)
763 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
769 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
770 struct dm_dev **result) {
772 struct table_device *td;
774 mutex_lock(&md->table_devices_lock);
775 td = find_table_device(&md->table_devices, dev, mode);
777 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
779 mutex_unlock(&md->table_devices_lock);
783 td->dm_dev.mode = mode;
784 td->dm_dev.bdev = NULL;
786 if ((r = open_table_device(td, dev, md))) {
787 mutex_unlock(&md->table_devices_lock);
792 format_dev_t(td->dm_dev.name, dev);
794 refcount_set(&td->count, 1);
795 list_add(&td->list, &md->table_devices);
797 refcount_inc(&td->count);
799 mutex_unlock(&md->table_devices_lock);
801 *result = &td->dm_dev;
804 EXPORT_SYMBOL_GPL(dm_get_table_device);
806 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
808 struct table_device *td = container_of(d, struct table_device, dm_dev);
810 mutex_lock(&md->table_devices_lock);
811 if (refcount_dec_and_test(&td->count)) {
812 close_table_device(td, md);
816 mutex_unlock(&md->table_devices_lock);
818 EXPORT_SYMBOL(dm_put_table_device);
820 static void free_table_devices(struct list_head *devices)
822 struct list_head *tmp, *next;
824 list_for_each_safe(tmp, next, devices) {
825 struct table_device *td = list_entry(tmp, struct table_device, list);
827 DMWARN("dm_destroy: %s still exists with %d references",
828 td->dm_dev.name, refcount_read(&td->count));
834 * Get the geometry associated with a dm device
836 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
844 * Set the geometry of a device.
846 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
848 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
850 if (geo->start > sz) {
851 DMWARN("Start sector is beyond the geometry limits.");
860 static int __noflush_suspending(struct mapped_device *md)
862 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
866 * Decrements the number of outstanding ios that a bio has been
867 * cloned into, completing the original io if necc.
869 static void dec_pending(struct dm_io *io, blk_status_t error)
872 blk_status_t io_error;
874 struct mapped_device *md = io->md;
876 /* Push-back supersedes any I/O errors */
877 if (unlikely(error)) {
878 spin_lock_irqsave(&io->endio_lock, flags);
879 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
881 spin_unlock_irqrestore(&io->endio_lock, flags);
884 if (atomic_dec_and_test(&io->io_count)) {
885 if (io->status == BLK_STS_DM_REQUEUE) {
887 * Target requested pushing back the I/O.
889 spin_lock_irqsave(&md->deferred_lock, flags);
890 if (__noflush_suspending(md))
891 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
892 bio_list_add_head(&md->deferred, io->orig_bio);
894 /* noflush suspend was interrupted. */
895 io->status = BLK_STS_IOERR;
896 spin_unlock_irqrestore(&md->deferred_lock, flags);
899 io_error = io->status;
904 if (io_error == BLK_STS_DM_REQUEUE)
907 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
909 * Preflush done for flush with data, reissue
910 * without REQ_PREFLUSH.
912 bio->bi_opf &= ~REQ_PREFLUSH;
915 /* done with normal IO or empty flush */
917 bio->bi_status = io_error;
923 void disable_discard(struct mapped_device *md)
925 struct queue_limits *limits = dm_get_queue_limits(md);
927 /* device doesn't really support DISCARD, disable it */
928 limits->max_discard_sectors = 0;
929 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
932 void disable_write_same(struct mapped_device *md)
934 struct queue_limits *limits = dm_get_queue_limits(md);
936 /* device doesn't really support WRITE SAME, disable it */
937 limits->max_write_same_sectors = 0;
940 void disable_write_zeroes(struct mapped_device *md)
942 struct queue_limits *limits = dm_get_queue_limits(md);
944 /* device doesn't really support WRITE ZEROES, disable it */
945 limits->max_write_zeroes_sectors = 0;
948 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
950 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
953 static void clone_endio(struct bio *bio)
955 blk_status_t error = bio->bi_status;
956 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
957 struct dm_io *io = tio->io;
958 struct mapped_device *md = tio->io->md;
959 dm_endio_fn endio = tio->ti->type->end_io;
961 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
962 if (bio_op(bio) == REQ_OP_DISCARD &&
963 !bio->bi_disk->queue->limits.max_discard_sectors)
965 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
966 !bio->bi_disk->queue->limits.max_write_same_sectors)
967 disable_write_same(md);
968 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
969 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
970 disable_write_zeroes(md);
974 int r = endio(tio->ti, bio, &error);
976 case DM_ENDIO_REQUEUE:
977 error = BLK_STS_DM_REQUEUE;
981 case DM_ENDIO_INCOMPLETE:
982 /* The target will handle the io */
985 DMWARN("unimplemented target endio return value: %d", r);
990 if (unlikely(swap_bios_limit(tio->ti, bio))) {
991 struct mapped_device *md = io->md;
992 up(&md->swap_bios_semaphore);
996 dec_pending(io, error);
1000 * Return maximum size of I/O possible at the supplied sector up to the current
1003 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1005 sector_t target_offset = dm_target_offset(ti, sector);
1007 return ti->len - target_offset;
1010 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1012 sector_t len = max_io_len_target_boundary(sector, ti);
1013 sector_t offset, max_len;
1016 * Does the target need to split even further?
1018 if (ti->max_io_len) {
1019 offset = dm_target_offset(ti, sector);
1020 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1021 max_len = sector_div(offset, ti->max_io_len);
1023 max_len = offset & (ti->max_io_len - 1);
1024 max_len = ti->max_io_len - max_len;
1033 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1035 if (len > UINT_MAX) {
1036 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1037 (unsigned long long)len, UINT_MAX);
1038 ti->error = "Maximum size of target IO is too large";
1042 ti->max_io_len = (uint32_t) len;
1046 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1048 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1049 sector_t sector, int *srcu_idx)
1050 __acquires(md->io_barrier)
1052 struct dm_table *map;
1053 struct dm_target *ti;
1055 map = dm_get_live_table(md, srcu_idx);
1059 ti = dm_table_find_target(map, sector);
1060 if (!dm_target_is_valid(ti))
1066 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1067 long nr_pages, void **kaddr, pfn_t *pfn)
1069 struct mapped_device *md = dax_get_private(dax_dev);
1070 sector_t sector = pgoff * PAGE_SECTORS;
1071 struct dm_target *ti;
1072 long len, ret = -EIO;
1075 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1079 if (!ti->type->direct_access)
1081 len = max_io_len(sector, ti) / PAGE_SECTORS;
1084 nr_pages = min(len, nr_pages);
1085 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1088 dm_put_live_table(md, srcu_idx);
1093 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1094 void *addr, size_t bytes, struct iov_iter *i)
1096 struct mapped_device *md = dax_get_private(dax_dev);
1097 sector_t sector = pgoff * PAGE_SECTORS;
1098 struct dm_target *ti;
1102 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1106 if (!ti->type->dax_copy_from_iter) {
1107 ret = copy_from_iter(addr, bytes, i);
1110 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1112 dm_put_live_table(md, srcu_idx);
1117 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1118 void *addr, size_t bytes, struct iov_iter *i)
1120 struct mapped_device *md = dax_get_private(dax_dev);
1121 sector_t sector = pgoff * PAGE_SECTORS;
1122 struct dm_target *ti;
1126 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1130 if (!ti->type->dax_copy_to_iter) {
1131 ret = copy_to_iter(addr, bytes, i);
1134 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1136 dm_put_live_table(md, srcu_idx);
1142 * A target may call dm_accept_partial_bio only from the map routine. It is
1143 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1145 * dm_accept_partial_bio informs the dm that the target only wants to process
1146 * additional n_sectors sectors of the bio and the rest of the data should be
1147 * sent in a next bio.
1149 * A diagram that explains the arithmetics:
1150 * +--------------------+---------------+-------+
1152 * +--------------------+---------------+-------+
1154 * <-------------- *tio->len_ptr --------------->
1155 * <------- bi_size ------->
1158 * Region 1 was already iterated over with bio_advance or similar function.
1159 * (it may be empty if the target doesn't use bio_advance)
1160 * Region 2 is the remaining bio size that the target wants to process.
1161 * (it may be empty if region 1 is non-empty, although there is no reason
1163 * The target requires that region 3 is to be sent in the next bio.
1165 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1166 * the partially processed part (the sum of regions 1+2) must be the same for all
1167 * copies of the bio.
1169 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1171 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1172 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1173 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1174 BUG_ON(bi_size > *tio->len_ptr);
1175 BUG_ON(n_sectors > bi_size);
1176 *tio->len_ptr -= bi_size - n_sectors;
1177 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1179 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1182 * The zone descriptors obtained with a zone report indicate zone positions
1183 * within the target backing device, regardless of that device is a partition
1184 * and regardless of the target mapping start sector on the device or partition.
1185 * The zone descriptors start sector and write pointer position must be adjusted
1186 * to match their relative position within the dm device.
1187 * A target may call dm_remap_zone_report() after completion of a
1188 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1191 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1193 #ifdef CONFIG_BLK_DEV_ZONED
1194 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1195 struct bio *report_bio = tio->io->orig_bio;
1196 struct blk_zone_report_hdr *hdr = NULL;
1197 struct blk_zone *zone;
1198 unsigned int nr_rep = 0;
1200 sector_t part_offset;
1201 struct bio_vec bvec;
1202 struct bvec_iter iter;
1209 * bio sector was incremented by the request size on completion. Taking
1210 * into account the original request sector, the target start offset on
1211 * the backing device and the target mapping offset (ti->begin), the
1212 * start sector of the backing device. The partition offset is always 0
1213 * if the target uses a whole device.
1215 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1218 * Remap the start sector of the reported zones. For sequential zones,
1219 * also remap the write pointer position.
1221 bio_for_each_segment(bvec, report_bio, iter) {
1222 addr = kmap_atomic(bvec.bv_page);
1224 /* Remember the report header in the first page */
1227 ofst = sizeof(struct blk_zone_report_hdr);
1231 /* Set zones start sector */
1232 while (hdr->nr_zones && ofst < bvec.bv_len) {
1234 zone->start -= part_offset;
1235 if (zone->start >= start + ti->len) {
1239 zone->start = zone->start + ti->begin - start;
1240 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1241 if (zone->cond == BLK_ZONE_COND_FULL)
1242 zone->wp = zone->start + zone->len;
1243 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1244 zone->wp = zone->start;
1246 zone->wp = zone->wp + ti->begin - start - part_offset;
1248 ofst += sizeof(struct blk_zone);
1254 kunmap_atomic(addr);
1261 hdr->nr_zones = nr_rep;
1265 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1267 #else /* !CONFIG_BLK_DEV_ZONED */
1268 bio->bi_status = BLK_STS_NOTSUPP;
1271 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1273 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1275 mutex_lock(&md->swap_bios_lock);
1276 while (latch < md->swap_bios) {
1278 down(&md->swap_bios_semaphore);
1281 while (latch > md->swap_bios) {
1283 up(&md->swap_bios_semaphore);
1286 mutex_unlock(&md->swap_bios_lock);
1289 static blk_qc_t __map_bio(struct dm_target_io *tio)
1293 struct bio *clone = &tio->clone;
1294 struct dm_io *io = tio->io;
1295 struct mapped_device *md = io->md;
1296 struct dm_target *ti = tio->ti;
1297 blk_qc_t ret = BLK_QC_T_NONE;
1299 clone->bi_end_io = clone_endio;
1302 * Map the clone. If r == 0 we don't need to do
1303 * anything, the target has assumed ownership of
1306 atomic_inc(&io->io_count);
1307 sector = clone->bi_iter.bi_sector;
1309 if (unlikely(swap_bios_limit(ti, clone))) {
1310 struct mapped_device *md = io->md;
1311 int latch = get_swap_bios();
1312 if (unlikely(latch != md->swap_bios))
1313 __set_swap_bios_limit(md, latch);
1314 down(&md->swap_bios_semaphore);
1317 r = ti->type->map(ti, clone);
1319 case DM_MAPIO_SUBMITTED:
1321 case DM_MAPIO_REMAPPED:
1322 /* the bio has been remapped so dispatch it */
1323 trace_block_bio_remap(clone->bi_disk->queue, clone,
1324 bio_dev(io->orig_bio), sector);
1325 if (md->type == DM_TYPE_NVME_BIO_BASED)
1326 ret = direct_make_request(clone);
1328 ret = generic_make_request(clone);
1331 if (unlikely(swap_bios_limit(ti, clone))) {
1332 struct mapped_device *md = io->md;
1333 up(&md->swap_bios_semaphore);
1336 dec_pending(io, BLK_STS_IOERR);
1338 case DM_MAPIO_REQUEUE:
1339 if (unlikely(swap_bios_limit(ti, clone))) {
1340 struct mapped_device *md = io->md;
1341 up(&md->swap_bios_semaphore);
1344 dec_pending(io, BLK_STS_DM_REQUEUE);
1347 DMWARN("unimplemented target map return value: %d", r);
1354 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1356 bio->bi_iter.bi_sector = sector;
1357 bio->bi_iter.bi_size = to_bytes(len);
1361 * Creates a bio that consists of range of complete bvecs.
1363 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1364 sector_t sector, unsigned len)
1366 struct bio *clone = &tio->clone;
1368 __bio_clone_fast(clone, bio);
1370 if (unlikely(bio_integrity(bio) != NULL)) {
1373 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1374 !dm_target_passes_integrity(tio->ti->type))) {
1375 DMWARN("%s: the target %s doesn't support integrity data.",
1376 dm_device_name(tio->io->md),
1377 tio->ti->type->name);
1381 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1386 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1387 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1388 clone->bi_iter.bi_size = to_bytes(len);
1390 if (unlikely(bio_integrity(bio) != NULL))
1391 bio_integrity_trim(clone);
1396 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1397 struct dm_target *ti, unsigned num_bios)
1399 struct dm_target_io *tio;
1405 if (num_bios == 1) {
1406 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1407 bio_list_add(blist, &tio->clone);
1411 for (try = 0; try < 2; try++) {
1416 mutex_lock(&ci->io->md->table_devices_lock);
1417 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1418 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1422 bio_list_add(blist, &tio->clone);
1425 mutex_unlock(&ci->io->md->table_devices_lock);
1426 if (bio_nr == num_bios)
1429 while ((bio = bio_list_pop(blist))) {
1430 tio = container_of(bio, struct dm_target_io, clone);
1436 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1437 struct dm_target_io *tio, unsigned *len)
1439 struct bio *clone = &tio->clone;
1443 __bio_clone_fast(clone, ci->bio);
1445 bio_setup_sector(clone, ci->sector, *len);
1447 return __map_bio(tio);
1450 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1451 unsigned num_bios, unsigned *len)
1453 struct bio_list blist = BIO_EMPTY_LIST;
1455 struct dm_target_io *tio;
1457 alloc_multiple_bios(&blist, ci, ti, num_bios);
1459 while ((bio = bio_list_pop(&blist))) {
1460 tio = container_of(bio, struct dm_target_io, clone);
1461 (void) __clone_and_map_simple_bio(ci, tio, len);
1465 static int __send_empty_flush(struct clone_info *ci)
1467 unsigned target_nr = 0;
1468 struct dm_target *ti;
1470 BUG_ON(bio_has_data(ci->bio));
1471 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1472 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1477 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1478 sector_t sector, unsigned *len)
1480 struct bio *bio = ci->bio;
1481 struct dm_target_io *tio;
1484 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1486 r = clone_bio(tio, bio, sector, *len);
1491 (void) __map_bio(tio);
1496 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1498 static unsigned get_num_discard_bios(struct dm_target *ti)
1500 return ti->num_discard_bios;
1503 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1505 return ti->num_secure_erase_bios;
1508 static unsigned get_num_write_same_bios(struct dm_target *ti)
1510 return ti->num_write_same_bios;
1513 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1515 return ti->num_write_zeroes_bios;
1518 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1520 static bool is_split_required_for_discard(struct dm_target *ti)
1522 return ti->split_discard_bios;
1525 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1526 get_num_bios_fn get_num_bios,
1527 is_split_required_fn is_split_required)
1533 * Even though the device advertised support for this type of
1534 * request, that does not mean every target supports it, and
1535 * reconfiguration might also have changed that since the
1536 * check was performed.
1538 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1542 if (is_split_required && !is_split_required(ti))
1543 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1545 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1547 __send_duplicate_bios(ci, ti, num_bios, &len);
1550 ci->sector_count -= len;
1555 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1557 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1558 is_split_required_for_discard);
1561 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1563 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1566 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1568 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1571 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1573 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1576 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1579 struct bio *bio = ci->bio;
1581 if (bio_op(bio) == REQ_OP_DISCARD)
1582 *result = __send_discard(ci, ti);
1583 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1584 *result = __send_secure_erase(ci, ti);
1585 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1586 *result = __send_write_same(ci, ti);
1587 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1588 *result = __send_write_zeroes(ci, ti);
1596 * Select the correct strategy for processing a non-flush bio.
1598 static int __split_and_process_non_flush(struct clone_info *ci)
1600 struct bio *bio = ci->bio;
1601 struct dm_target *ti;
1605 ti = dm_table_find_target(ci->map, ci->sector);
1606 if (!dm_target_is_valid(ti))
1609 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1612 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1613 len = ci->sector_count;
1615 len = min_t(sector_t, max_io_len(ci->sector, ti),
1618 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1623 ci->sector_count -= len;
1628 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1629 struct dm_table *map, struct bio *bio)
1632 ci->io = alloc_io(md, bio);
1633 ci->sector = bio->bi_iter.bi_sector;
1637 * Entry point to split a bio into clones and submit them to the targets.
1639 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1640 struct dm_table *map, struct bio *bio)
1642 struct clone_info ci;
1643 blk_qc_t ret = BLK_QC_T_NONE;
1646 if (unlikely(!map)) {
1651 blk_queue_split(md->queue, &bio);
1653 init_clone_info(&ci, md, map, bio);
1655 if (bio->bi_opf & REQ_PREFLUSH) {
1656 ci.bio = &ci.io->md->flush_bio;
1657 ci.sector_count = 0;
1658 error = __send_empty_flush(&ci);
1659 /* dec_pending submits any data associated with flush */
1660 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1662 ci.sector_count = 0;
1663 error = __split_and_process_non_flush(&ci);
1666 ci.sector_count = bio_sectors(bio);
1667 while (ci.sector_count && !error) {
1668 error = __split_and_process_non_flush(&ci);
1669 if (current->bio_list && ci.sector_count && !error) {
1671 * Remainder must be passed to generic_make_request()
1672 * so that it gets handled *after* bios already submitted
1673 * have been completely processed.
1674 * We take a clone of the original to store in
1675 * ci.io->orig_bio to be used by end_io_acct() and
1676 * for dec_pending to use for completion handling.
1677 * As this path is not used for REQ_OP_ZONE_REPORT,
1678 * the usage of io->orig_bio in dm_remap_zone_report()
1679 * won't be affected by this reassignment.
1681 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1682 GFP_NOIO, &md->queue->bio_split);
1683 ci.io->orig_bio = b;
1685 ret = generic_make_request(bio);
1691 /* drop the extra reference count */
1692 dec_pending(ci.io, errno_to_blk_status(error));
1697 * Optimized variant of __split_and_process_bio that leverages the
1698 * fact that targets that use it do _not_ have a need to split bios.
1700 static blk_qc_t __process_bio(struct mapped_device *md,
1701 struct dm_table *map, struct bio *bio)
1703 struct clone_info ci;
1704 blk_qc_t ret = BLK_QC_T_NONE;
1707 if (unlikely(!map)) {
1712 init_clone_info(&ci, md, map, bio);
1714 if (bio->bi_opf & REQ_PREFLUSH) {
1715 ci.bio = &ci.io->md->flush_bio;
1716 ci.sector_count = 0;
1717 error = __send_empty_flush(&ci);
1718 /* dec_pending submits any data associated with flush */
1720 struct dm_target *ti = md->immutable_target;
1721 struct dm_target_io *tio;
1724 * Defend against IO still getting in during teardown
1725 * - as was seen for a time with nvme-fcloop
1727 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1733 ci.sector_count = bio_sectors(bio);
1734 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1737 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1738 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1741 /* drop the extra reference count */
1742 dec_pending(ci.io, errno_to_blk_status(error));
1746 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1748 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1749 process_bio_fn process_bio)
1751 struct mapped_device *md = q->queuedata;
1752 blk_qc_t ret = BLK_QC_T_NONE;
1754 struct dm_table *map;
1756 map = dm_get_live_table(md, &srcu_idx);
1758 /* if we're suspended, we have to queue this io for later */
1759 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1760 dm_put_live_table(md, srcu_idx);
1762 if (!(bio->bi_opf & REQ_RAHEAD))
1769 ret = process_bio(md, map, bio);
1771 dm_put_live_table(md, srcu_idx);
1776 * The request function that remaps the bio to one target and
1777 * splits off any remainder.
1779 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1781 return __dm_make_request(q, bio, __split_and_process_bio);
1784 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1786 return __dm_make_request(q, bio, __process_bio);
1789 static int dm_any_congested(void *congested_data, int bdi_bits)
1792 struct mapped_device *md = congested_data;
1793 struct dm_table *map;
1795 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1796 if (dm_request_based(md)) {
1798 * With request-based DM we only need to check the
1799 * top-level queue for congestion.
1801 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1803 map = dm_get_live_table_fast(md);
1805 r = dm_table_any_congested(map, bdi_bits);
1806 dm_put_live_table_fast(md);
1813 /*-----------------------------------------------------------------
1814 * An IDR is used to keep track of allocated minor numbers.
1815 *---------------------------------------------------------------*/
1816 static void free_minor(int minor)
1818 spin_lock(&_minor_lock);
1819 idr_remove(&_minor_idr, minor);
1820 spin_unlock(&_minor_lock);
1824 * See if the device with a specific minor # is free.
1826 static int specific_minor(int minor)
1830 if (minor >= (1 << MINORBITS))
1833 idr_preload(GFP_KERNEL);
1834 spin_lock(&_minor_lock);
1836 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1838 spin_unlock(&_minor_lock);
1841 return r == -ENOSPC ? -EBUSY : r;
1845 static int next_free_minor(int *minor)
1849 idr_preload(GFP_KERNEL);
1850 spin_lock(&_minor_lock);
1852 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1854 spin_unlock(&_minor_lock);
1862 static const struct block_device_operations dm_blk_dops;
1863 static const struct dax_operations dm_dax_ops;
1865 static void dm_wq_work(struct work_struct *work);
1867 static void dm_init_normal_md_queue(struct mapped_device *md)
1869 md->use_blk_mq = false;
1872 * Initialize aspects of queue that aren't relevant for blk-mq
1874 md->queue->backing_dev_info->congested_data = md;
1875 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1878 static void cleanup_mapped_device(struct mapped_device *md)
1881 destroy_workqueue(md->wq);
1882 if (md->kworker_task)
1883 kthread_stop(md->kworker_task);
1884 bioset_exit(&md->bs);
1885 bioset_exit(&md->io_bs);
1888 kill_dax(md->dax_dev);
1889 put_dax(md->dax_dev);
1894 spin_lock(&_minor_lock);
1895 md->disk->private_data = NULL;
1896 spin_unlock(&_minor_lock);
1897 del_gendisk(md->disk);
1902 blk_cleanup_queue(md->queue);
1904 cleanup_srcu_struct(&md->io_barrier);
1911 mutex_destroy(&md->suspend_lock);
1912 mutex_destroy(&md->type_lock);
1913 mutex_destroy(&md->table_devices_lock);
1914 mutex_destroy(&md->swap_bios_lock);
1916 dm_mq_cleanup_mapped_device(md);
1920 * Allocate and initialise a blank device with a given minor.
1922 static struct mapped_device *alloc_dev(int minor)
1924 int r, numa_node_id = dm_get_numa_node();
1925 struct dax_device *dax_dev = NULL;
1926 struct mapped_device *md;
1929 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1931 DMWARN("unable to allocate device, out of memory.");
1935 if (!try_module_get(THIS_MODULE))
1936 goto bad_module_get;
1938 /* get a minor number for the dev */
1939 if (minor == DM_ANY_MINOR)
1940 r = next_free_minor(&minor);
1942 r = specific_minor(minor);
1946 r = init_srcu_struct(&md->io_barrier);
1948 goto bad_io_barrier;
1950 md->numa_node_id = numa_node_id;
1951 md->use_blk_mq = dm_use_blk_mq_default();
1952 md->init_tio_pdu = false;
1953 md->type = DM_TYPE_NONE;
1954 mutex_init(&md->suspend_lock);
1955 mutex_init(&md->type_lock);
1956 mutex_init(&md->table_devices_lock);
1957 spin_lock_init(&md->deferred_lock);
1958 atomic_set(&md->holders, 1);
1959 atomic_set(&md->open_count, 0);
1960 atomic_set(&md->event_nr, 0);
1961 atomic_set(&md->uevent_seq, 0);
1962 INIT_LIST_HEAD(&md->uevent_list);
1963 INIT_LIST_HEAD(&md->table_devices);
1964 spin_lock_init(&md->uevent_lock);
1966 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1969 md->queue->queuedata = md;
1971 * default to bio-based required ->make_request_fn until DM
1972 * table is loaded and md->type established. If request-based
1973 * table is loaded: blk-mq will override accordingly.
1975 blk_queue_make_request(md->queue, dm_make_request);
1977 md->disk = alloc_disk_node(1, md->numa_node_id);
1981 atomic_set(&md->pending[0], 0);
1982 atomic_set(&md->pending[1], 0);
1983 init_waitqueue_head(&md->wait);
1984 INIT_WORK(&md->work, dm_wq_work);
1985 init_waitqueue_head(&md->eventq);
1986 init_completion(&md->kobj_holder.completion);
1987 md->kworker_task = NULL;
1989 md->swap_bios = get_swap_bios();
1990 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1991 mutex_init(&md->swap_bios_lock);
1993 md->disk->major = _major;
1994 md->disk->first_minor = minor;
1995 md->disk->fops = &dm_blk_dops;
1996 md->disk->queue = md->queue;
1997 md->disk->private_data = md;
1998 sprintf(md->disk->disk_name, "dm-%d", minor);
2000 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
2001 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
2005 md->dax_dev = dax_dev;
2007 add_disk_no_queue_reg(md->disk);
2008 format_dev_t(md->name, MKDEV(_major, minor));
2010 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2014 md->bdev = bdget_disk(md->disk, 0);
2018 bio_init(&md->flush_bio, NULL, 0);
2019 bio_set_dev(&md->flush_bio, md->bdev);
2020 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
2022 dm_stats_init(&md->stats);
2024 /* Populate the mapping, nobody knows we exist yet */
2025 spin_lock(&_minor_lock);
2026 old_md = idr_replace(&_minor_idr, md, minor);
2027 spin_unlock(&_minor_lock);
2029 BUG_ON(old_md != MINOR_ALLOCED);
2034 cleanup_mapped_device(md);
2038 module_put(THIS_MODULE);
2044 static void unlock_fs(struct mapped_device *md);
2046 static void free_dev(struct mapped_device *md)
2048 int minor = MINOR(disk_devt(md->disk));
2052 cleanup_mapped_device(md);
2054 free_table_devices(&md->table_devices);
2055 dm_stats_cleanup(&md->stats);
2058 module_put(THIS_MODULE);
2062 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2064 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2067 if (dm_table_bio_based(t)) {
2069 * The md may already have mempools that need changing.
2070 * If so, reload bioset because front_pad may have changed
2071 * because a different table was loaded.
2073 bioset_exit(&md->bs);
2074 bioset_exit(&md->io_bs);
2076 } else if (bioset_initialized(&md->bs)) {
2078 * There's no need to reload with request-based dm
2079 * because the size of front_pad doesn't change.
2080 * Note for future: If you are to reload bioset,
2081 * prep-ed requests in the queue may refer
2082 * to bio from the old bioset, so you must walk
2083 * through the queue to unprep.
2089 bioset_initialized(&md->bs) ||
2090 bioset_initialized(&md->io_bs));
2092 ret = bioset_init_from_src(&md->bs, &p->bs);
2095 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2097 bioset_exit(&md->bs);
2099 /* mempool bind completed, no longer need any mempools in the table */
2100 dm_table_free_md_mempools(t);
2105 * Bind a table to the device.
2107 static void event_callback(void *context)
2109 unsigned long flags;
2111 struct mapped_device *md = (struct mapped_device *) context;
2113 spin_lock_irqsave(&md->uevent_lock, flags);
2114 list_splice_init(&md->uevent_list, &uevents);
2115 spin_unlock_irqrestore(&md->uevent_lock, flags);
2117 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2119 atomic_inc(&md->event_nr);
2120 wake_up(&md->eventq);
2121 dm_issue_global_event();
2125 * Protected by md->suspend_lock obtained by dm_swap_table().
2127 static void __set_size(struct mapped_device *md, sector_t size)
2129 lockdep_assert_held(&md->suspend_lock);
2131 set_capacity(md->disk, size);
2133 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2137 * Returns old map, which caller must destroy.
2139 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2140 struct queue_limits *limits)
2142 struct dm_table *old_map;
2143 struct request_queue *q = md->queue;
2144 bool request_based = dm_table_request_based(t);
2148 lockdep_assert_held(&md->suspend_lock);
2150 size = dm_table_get_size(t);
2153 * Wipe any geometry if the size of the table changed.
2155 if (size != dm_get_size(md))
2156 memset(&md->geometry, 0, sizeof(md->geometry));
2158 __set_size(md, size);
2160 dm_table_event_callback(t, event_callback, md);
2163 * The queue hasn't been stopped yet, if the old table type wasn't
2164 * for request-based during suspension. So stop it to prevent
2165 * I/O mapping before resume.
2166 * This must be done before setting the queue restrictions,
2167 * because request-based dm may be run just after the setting.
2172 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2174 * Leverage the fact that request-based DM targets and
2175 * NVMe bio based targets are immutable singletons
2176 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2177 * and __process_bio.
2179 md->immutable_target = dm_table_get_immutable_target(t);
2182 ret = __bind_mempools(md, t);
2184 old_map = ERR_PTR(ret);
2188 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2189 rcu_assign_pointer(md->map, (void *)t);
2190 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2192 dm_table_set_restrictions(t, q, limits);
2201 * Returns unbound table for the caller to free.
2203 static struct dm_table *__unbind(struct mapped_device *md)
2205 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2210 dm_table_event_callback(map, NULL, NULL);
2211 RCU_INIT_POINTER(md->map, NULL);
2218 * Constructor for a new device.
2220 int dm_create(int minor, struct mapped_device **result)
2223 struct mapped_device *md;
2225 md = alloc_dev(minor);
2229 r = dm_sysfs_init(md);
2240 * Functions to manage md->type.
2241 * All are required to hold md->type_lock.
2243 void dm_lock_md_type(struct mapped_device *md)
2245 mutex_lock(&md->type_lock);
2248 void dm_unlock_md_type(struct mapped_device *md)
2250 mutex_unlock(&md->type_lock);
2253 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2255 BUG_ON(!mutex_is_locked(&md->type_lock));
2259 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2264 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2266 return md->immutable_target_type;
2270 * The queue_limits are only valid as long as you have a reference
2273 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2275 BUG_ON(!atomic_read(&md->holders));
2276 return &md->queue->limits;
2278 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2281 * Setup the DM device's queue based on md's type
2283 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2286 struct queue_limits limits;
2287 enum dm_queue_mode type = dm_get_md_type(md);
2290 case DM_TYPE_REQUEST_BASED:
2291 dm_init_normal_md_queue(md);
2292 r = dm_old_init_request_queue(md, t);
2294 DMERR("Cannot initialize queue for request-based mapped device");
2298 case DM_TYPE_MQ_REQUEST_BASED:
2299 r = dm_mq_init_request_queue(md, t);
2301 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2305 case DM_TYPE_BIO_BASED:
2306 case DM_TYPE_DAX_BIO_BASED:
2307 dm_init_normal_md_queue(md);
2309 case DM_TYPE_NVME_BIO_BASED:
2310 dm_init_normal_md_queue(md);
2311 blk_queue_make_request(md->queue, dm_make_request_nvme);
2318 r = dm_calculate_queue_limits(t, &limits);
2320 DMERR("Cannot calculate initial queue limits");
2323 dm_table_set_restrictions(t, md->queue, &limits);
2324 blk_register_queue(md->disk);
2329 struct mapped_device *dm_get_md(dev_t dev)
2331 struct mapped_device *md;
2332 unsigned minor = MINOR(dev);
2334 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2337 spin_lock(&_minor_lock);
2339 md = idr_find(&_minor_idr, minor);
2340 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2341 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2347 spin_unlock(&_minor_lock);
2351 EXPORT_SYMBOL_GPL(dm_get_md);
2353 void *dm_get_mdptr(struct mapped_device *md)
2355 return md->interface_ptr;
2358 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2360 md->interface_ptr = ptr;
2363 void dm_get(struct mapped_device *md)
2365 atomic_inc(&md->holders);
2366 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2369 int dm_hold(struct mapped_device *md)
2371 spin_lock(&_minor_lock);
2372 if (test_bit(DMF_FREEING, &md->flags)) {
2373 spin_unlock(&_minor_lock);
2377 spin_unlock(&_minor_lock);
2380 EXPORT_SYMBOL_GPL(dm_hold);
2382 const char *dm_device_name(struct mapped_device *md)
2386 EXPORT_SYMBOL_GPL(dm_device_name);
2388 static void __dm_destroy(struct mapped_device *md, bool wait)
2390 struct dm_table *map;
2395 spin_lock(&_minor_lock);
2396 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2397 set_bit(DMF_FREEING, &md->flags);
2398 spin_unlock(&_minor_lock);
2400 blk_set_queue_dying(md->queue);
2402 if (dm_request_based(md) && md->kworker_task)
2403 kthread_flush_worker(&md->kworker);
2406 * Take suspend_lock so that presuspend and postsuspend methods
2407 * do not race with internal suspend.
2409 mutex_lock(&md->suspend_lock);
2410 map = dm_get_live_table(md, &srcu_idx);
2411 if (!dm_suspended_md(md)) {
2412 dm_table_presuspend_targets(map);
2413 set_bit(DMF_SUSPENDED, &md->flags);
2414 set_bit(DMF_POST_SUSPENDING, &md->flags);
2415 dm_table_postsuspend_targets(map);
2417 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2418 dm_put_live_table(md, srcu_idx);
2419 mutex_unlock(&md->suspend_lock);
2422 * Rare, but there may be I/O requests still going to complete,
2423 * for example. Wait for all references to disappear.
2424 * No one should increment the reference count of the mapped_device,
2425 * after the mapped_device state becomes DMF_FREEING.
2428 while (atomic_read(&md->holders))
2430 else if (atomic_read(&md->holders))
2431 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2432 dm_device_name(md), atomic_read(&md->holders));
2435 dm_table_destroy(__unbind(md));
2439 void dm_destroy(struct mapped_device *md)
2441 __dm_destroy(md, true);
2444 void dm_destroy_immediate(struct mapped_device *md)
2446 __dm_destroy(md, false);
2449 void dm_put(struct mapped_device *md)
2451 atomic_dec(&md->holders);
2453 EXPORT_SYMBOL_GPL(dm_put);
2455 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2461 prepare_to_wait(&md->wait, &wait, task_state);
2463 if (!md_in_flight(md))
2466 if (signal_pending_state(task_state, current)) {
2473 finish_wait(&md->wait, &wait);
2479 * Process the deferred bios
2481 static void dm_wq_work(struct work_struct *work)
2483 struct mapped_device *md = container_of(work, struct mapped_device,
2487 struct dm_table *map;
2489 map = dm_get_live_table(md, &srcu_idx);
2491 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2492 spin_lock_irq(&md->deferred_lock);
2493 c = bio_list_pop(&md->deferred);
2494 spin_unlock_irq(&md->deferred_lock);
2499 if (dm_request_based(md))
2500 generic_make_request(c);
2502 __split_and_process_bio(md, map, c);
2505 dm_put_live_table(md, srcu_idx);
2508 static void dm_queue_flush(struct mapped_device *md)
2510 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2511 smp_mb__after_atomic();
2512 queue_work(md->wq, &md->work);
2516 * Swap in a new table, returning the old one for the caller to destroy.
2518 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2520 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2521 struct queue_limits limits;
2524 mutex_lock(&md->suspend_lock);
2526 /* device must be suspended */
2527 if (!dm_suspended_md(md))
2531 * If the new table has no data devices, retain the existing limits.
2532 * This helps multipath with queue_if_no_path if all paths disappear,
2533 * then new I/O is queued based on these limits, and then some paths
2536 if (dm_table_has_no_data_devices(table)) {
2537 live_map = dm_get_live_table_fast(md);
2539 limits = md->queue->limits;
2540 dm_put_live_table_fast(md);
2544 r = dm_calculate_queue_limits(table, &limits);
2551 map = __bind(md, table, &limits);
2552 dm_issue_global_event();
2555 mutex_unlock(&md->suspend_lock);
2560 * Functions to lock and unlock any filesystem running on the
2563 static int lock_fs(struct mapped_device *md)
2567 WARN_ON(md->frozen_sb);
2569 md->frozen_sb = freeze_bdev(md->bdev);
2570 if (IS_ERR(md->frozen_sb)) {
2571 r = PTR_ERR(md->frozen_sb);
2572 md->frozen_sb = NULL;
2576 set_bit(DMF_FROZEN, &md->flags);
2581 static void unlock_fs(struct mapped_device *md)
2583 if (!test_bit(DMF_FROZEN, &md->flags))
2586 thaw_bdev(md->bdev, md->frozen_sb);
2587 md->frozen_sb = NULL;
2588 clear_bit(DMF_FROZEN, &md->flags);
2592 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2593 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2594 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2596 * If __dm_suspend returns 0, the device is completely quiescent
2597 * now. There is no request-processing activity. All new requests
2598 * are being added to md->deferred list.
2600 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2601 unsigned suspend_flags, long task_state,
2602 int dmf_suspended_flag)
2604 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2605 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2608 lockdep_assert_held(&md->suspend_lock);
2611 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2612 * This flag is cleared before dm_suspend returns.
2615 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2617 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2620 * This gets reverted if there's an error later and the targets
2621 * provide the .presuspend_undo hook.
2623 dm_table_presuspend_targets(map);
2626 * Flush I/O to the device.
2627 * Any I/O submitted after lock_fs() may not be flushed.
2628 * noflush takes precedence over do_lockfs.
2629 * (lock_fs() flushes I/Os and waits for them to complete.)
2631 if (!noflush && do_lockfs) {
2634 dm_table_presuspend_undo_targets(map);
2640 * Here we must make sure that no processes are submitting requests
2641 * to target drivers i.e. no one may be executing
2642 * __split_and_process_bio. This is called from dm_request and
2645 * To get all processes out of __split_and_process_bio in dm_request,
2646 * we take the write lock. To prevent any process from reentering
2647 * __split_and_process_bio from dm_request and quiesce the thread
2648 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2649 * flush_workqueue(md->wq).
2651 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2653 synchronize_srcu(&md->io_barrier);
2656 * Stop md->queue before flushing md->wq in case request-based
2657 * dm defers requests to md->wq from md->queue.
2659 if (dm_request_based(md)) {
2660 dm_stop_queue(md->queue);
2661 if (md->kworker_task)
2662 kthread_flush_worker(&md->kworker);
2665 flush_workqueue(md->wq);
2668 * At this point no more requests are entering target request routines.
2669 * We call dm_wait_for_completion to wait for all existing requests
2672 r = dm_wait_for_completion(md, task_state);
2674 set_bit(dmf_suspended_flag, &md->flags);
2677 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2679 synchronize_srcu(&md->io_barrier);
2681 /* were we interrupted ? */
2685 if (dm_request_based(md))
2686 dm_start_queue(md->queue);
2689 dm_table_presuspend_undo_targets(map);
2690 /* pushback list is already flushed, so skip flush */
2697 * We need to be able to change a mapping table under a mounted
2698 * filesystem. For example we might want to move some data in
2699 * the background. Before the table can be swapped with
2700 * dm_bind_table, dm_suspend must be called to flush any in
2701 * flight bios and ensure that any further io gets deferred.
2704 * Suspend mechanism in request-based dm.
2706 * 1. Flush all I/Os by lock_fs() if needed.
2707 * 2. Stop dispatching any I/O by stopping the request_queue.
2708 * 3. Wait for all in-flight I/Os to be completed or requeued.
2710 * To abort suspend, start the request_queue.
2712 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2714 struct dm_table *map = NULL;
2718 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2720 if (dm_suspended_md(md)) {
2725 if (dm_suspended_internally_md(md)) {
2726 /* already internally suspended, wait for internal resume */
2727 mutex_unlock(&md->suspend_lock);
2728 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2734 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2736 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2740 set_bit(DMF_POST_SUSPENDING, &md->flags);
2741 dm_table_postsuspend_targets(map);
2742 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2745 mutex_unlock(&md->suspend_lock);
2749 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2752 int r = dm_table_resume_targets(map);
2760 * Flushing deferred I/Os must be done after targets are resumed
2761 * so that mapping of targets can work correctly.
2762 * Request-based dm is queueing the deferred I/Os in its request_queue.
2764 if (dm_request_based(md))
2765 dm_start_queue(md->queue);
2772 int dm_resume(struct mapped_device *md)
2775 struct dm_table *map = NULL;
2779 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2781 if (!dm_suspended_md(md))
2784 if (dm_suspended_internally_md(md)) {
2785 /* already internally suspended, wait for internal resume */
2786 mutex_unlock(&md->suspend_lock);
2787 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2793 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2794 if (!map || !dm_table_get_size(map))
2797 r = __dm_resume(md, map);
2801 clear_bit(DMF_SUSPENDED, &md->flags);
2803 mutex_unlock(&md->suspend_lock);
2809 * Internal suspend/resume works like userspace-driven suspend. It waits
2810 * until all bios finish and prevents issuing new bios to the target drivers.
2811 * It may be used only from the kernel.
2814 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2816 struct dm_table *map = NULL;
2818 lockdep_assert_held(&md->suspend_lock);
2820 if (md->internal_suspend_count++)
2821 return; /* nested internal suspend */
2823 if (dm_suspended_md(md)) {
2824 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2825 return; /* nest suspend */
2828 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2831 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2832 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2833 * would require changing .presuspend to return an error -- avoid this
2834 * until there is a need for more elaborate variants of internal suspend.
2836 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2837 DMF_SUSPENDED_INTERNALLY);
2839 set_bit(DMF_POST_SUSPENDING, &md->flags);
2840 dm_table_postsuspend_targets(map);
2841 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2844 static void __dm_internal_resume(struct mapped_device *md)
2846 BUG_ON(!md->internal_suspend_count);
2848 if (--md->internal_suspend_count)
2849 return; /* resume from nested internal suspend */
2851 if (dm_suspended_md(md))
2852 goto done; /* resume from nested suspend */
2855 * NOTE: existing callers don't need to call dm_table_resume_targets
2856 * (which may fail -- so best to avoid it for now by passing NULL map)
2858 (void) __dm_resume(md, NULL);
2861 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2862 smp_mb__after_atomic();
2863 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2866 void dm_internal_suspend_noflush(struct mapped_device *md)
2868 mutex_lock(&md->suspend_lock);
2869 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2870 mutex_unlock(&md->suspend_lock);
2872 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2874 void dm_internal_resume(struct mapped_device *md)
2876 mutex_lock(&md->suspend_lock);
2877 __dm_internal_resume(md);
2878 mutex_unlock(&md->suspend_lock);
2880 EXPORT_SYMBOL_GPL(dm_internal_resume);
2883 * Fast variants of internal suspend/resume hold md->suspend_lock,
2884 * which prevents interaction with userspace-driven suspend.
2887 void dm_internal_suspend_fast(struct mapped_device *md)
2889 mutex_lock(&md->suspend_lock);
2890 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2893 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2894 synchronize_srcu(&md->io_barrier);
2895 flush_workqueue(md->wq);
2896 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2898 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2900 void dm_internal_resume_fast(struct mapped_device *md)
2902 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2908 mutex_unlock(&md->suspend_lock);
2910 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2912 /*-----------------------------------------------------------------
2913 * Event notification.
2914 *---------------------------------------------------------------*/
2915 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2920 char udev_cookie[DM_COOKIE_LENGTH];
2921 char *envp[] = { udev_cookie, NULL };
2923 noio_flag = memalloc_noio_save();
2926 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2928 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2929 DM_COOKIE_ENV_VAR_NAME, cookie);
2930 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2934 memalloc_noio_restore(noio_flag);
2939 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2941 return atomic_add_return(1, &md->uevent_seq);
2944 uint32_t dm_get_event_nr(struct mapped_device *md)
2946 return atomic_read(&md->event_nr);
2949 int dm_wait_event(struct mapped_device *md, int event_nr)
2951 return wait_event_interruptible(md->eventq,
2952 (event_nr != atomic_read(&md->event_nr)));
2955 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2957 unsigned long flags;
2959 spin_lock_irqsave(&md->uevent_lock, flags);
2960 list_add(elist, &md->uevent_list);
2961 spin_unlock_irqrestore(&md->uevent_lock, flags);
2965 * The gendisk is only valid as long as you have a reference
2968 struct gendisk *dm_disk(struct mapped_device *md)
2972 EXPORT_SYMBOL_GPL(dm_disk);
2974 struct kobject *dm_kobject(struct mapped_device *md)
2976 return &md->kobj_holder.kobj;
2979 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2981 struct mapped_device *md;
2983 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2985 spin_lock(&_minor_lock);
2986 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2992 spin_unlock(&_minor_lock);
2997 int dm_suspended_md(struct mapped_device *md)
2999 return test_bit(DMF_SUSPENDED, &md->flags);
3002 static int dm_post_suspending_md(struct mapped_device *md)
3004 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3007 int dm_suspended_internally_md(struct mapped_device *md)
3009 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3012 int dm_test_deferred_remove_flag(struct mapped_device *md)
3014 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3017 int dm_suspended(struct dm_target *ti)
3019 return dm_suspended_md(dm_table_get_md(ti->table));
3021 EXPORT_SYMBOL_GPL(dm_suspended);
3023 int dm_post_suspending(struct dm_target *ti)
3025 return dm_post_suspending_md(dm_table_get_md(ti->table));
3027 EXPORT_SYMBOL_GPL(dm_post_suspending);
3029 int dm_noflush_suspending(struct dm_target *ti)
3031 return __noflush_suspending(dm_table_get_md(ti->table));
3033 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3035 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3036 unsigned integrity, unsigned per_io_data_size,
3037 unsigned min_pool_size)
3039 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3040 unsigned int pool_size = 0;
3041 unsigned int front_pad, io_front_pad;
3048 case DM_TYPE_BIO_BASED:
3049 case DM_TYPE_DAX_BIO_BASED:
3050 case DM_TYPE_NVME_BIO_BASED:
3051 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3052 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3053 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3054 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3057 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3060 case DM_TYPE_REQUEST_BASED:
3061 case DM_TYPE_MQ_REQUEST_BASED:
3062 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3063 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3064 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3070 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3074 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3080 dm_free_md_mempools(pools);
3085 void dm_free_md_mempools(struct dm_md_mempools *pools)
3090 bioset_exit(&pools->bs);
3091 bioset_exit(&pools->io_bs);
3103 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3106 struct mapped_device *md = bdev->bd_disk->private_data;
3107 struct dm_table *table;
3108 struct dm_target *ti;
3109 int ret = -ENOTTY, srcu_idx;
3111 table = dm_get_live_table(md, &srcu_idx);
3112 if (!table || !dm_table_get_size(table))
3115 /* We only support devices that have a single target */
3116 if (dm_table_get_num_targets(table) != 1)
3118 ti = dm_table_get_target(table, 0);
3121 if (!ti->type->iterate_devices)
3124 ret = ti->type->iterate_devices(ti, fn, data);
3126 dm_put_live_table(md, srcu_idx);
3131 * For register / unregister we need to manually call out to every path.
3133 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3134 sector_t start, sector_t len, void *data)
3136 struct dm_pr *pr = data;
3137 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3139 if (!ops || !ops->pr_register)
3141 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3144 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3155 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3156 if (ret && new_key) {
3157 /* unregister all paths if we failed to register any path */
3158 pr.old_key = new_key;
3161 pr.fail_early = false;
3162 dm_call_pr(bdev, __dm_pr_register, &pr);
3168 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3171 struct mapped_device *md = bdev->bd_disk->private_data;
3172 const struct pr_ops *ops;
3175 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3179 ops = bdev->bd_disk->fops->pr_ops;
3180 if (ops && ops->pr_reserve)
3181 r = ops->pr_reserve(bdev, key, type, flags);
3185 dm_unprepare_ioctl(md, srcu_idx);
3189 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3191 struct mapped_device *md = bdev->bd_disk->private_data;
3192 const struct pr_ops *ops;
3195 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3199 ops = bdev->bd_disk->fops->pr_ops;
3200 if (ops && ops->pr_release)
3201 r = ops->pr_release(bdev, key, type);
3205 dm_unprepare_ioctl(md, srcu_idx);
3209 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3210 enum pr_type type, bool abort)
3212 struct mapped_device *md = bdev->bd_disk->private_data;
3213 const struct pr_ops *ops;
3216 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3220 ops = bdev->bd_disk->fops->pr_ops;
3221 if (ops && ops->pr_preempt)
3222 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3226 dm_unprepare_ioctl(md, srcu_idx);
3230 static int dm_pr_clear(struct block_device *bdev, u64 key)
3232 struct mapped_device *md = bdev->bd_disk->private_data;
3233 const struct pr_ops *ops;
3236 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3240 ops = bdev->bd_disk->fops->pr_ops;
3241 if (ops && ops->pr_clear)
3242 r = ops->pr_clear(bdev, key);
3246 dm_unprepare_ioctl(md, srcu_idx);
3250 static const struct pr_ops dm_pr_ops = {
3251 .pr_register = dm_pr_register,
3252 .pr_reserve = dm_pr_reserve,
3253 .pr_release = dm_pr_release,
3254 .pr_preempt = dm_pr_preempt,
3255 .pr_clear = dm_pr_clear,
3258 static const struct block_device_operations dm_blk_dops = {
3259 .open = dm_blk_open,
3260 .release = dm_blk_close,
3261 .ioctl = dm_blk_ioctl,
3262 .getgeo = dm_blk_getgeo,
3263 .pr_ops = &dm_pr_ops,
3264 .owner = THIS_MODULE
3267 static const struct dax_operations dm_dax_ops = {
3268 .direct_access = dm_dax_direct_access,
3269 .copy_from_iter = dm_dax_copy_from_iter,
3270 .copy_to_iter = dm_dax_copy_to_iter,
3276 module_init(dm_init);
3277 module_exit(dm_exit);
3279 module_param(major, uint, 0);
3280 MODULE_PARM_DESC(major, "The major number of the device mapper");
3282 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3283 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3285 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3286 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3288 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3289 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3291 MODULE_DESCRIPTION(DM_NAME " driver");
3292 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3293 MODULE_LICENSE("GPL");