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 mapped_device *md, struct bio *bio,
635 unsigned long start_time, struct dm_stats_aux *stats_aux)
637 unsigned long duration = jiffies - start_time;
639 int rw = bio_data_dir(bio);
641 generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
644 if (unlikely(dm_stats_used(&md->stats)))
645 dm_stats_account_io(&md->stats, bio_data_dir(bio),
646 bio->bi_iter.bi_sector, bio_sectors(bio),
647 true, duration, stats_aux);
650 * After this is decremented the bio must not be touched if it is
653 pending = atomic_dec_return(&md->pending[rw]);
654 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
655 pending += atomic_read(&md->pending[rw^0x1]);
657 /* nudge anyone waiting on suspend queue */
663 * Add the bio to the list of deferred io.
665 static void queue_io(struct mapped_device *md, struct bio *bio)
669 spin_lock_irqsave(&md->deferred_lock, flags);
670 bio_list_add(&md->deferred, bio);
671 spin_unlock_irqrestore(&md->deferred_lock, flags);
672 queue_work(md->wq, &md->work);
676 * Everyone (including functions in this file), should use this
677 * function to access the md->map field, and make sure they call
678 * dm_put_live_table() when finished.
680 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
682 *srcu_idx = srcu_read_lock(&md->io_barrier);
684 return srcu_dereference(md->map, &md->io_barrier);
687 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
689 srcu_read_unlock(&md->io_barrier, srcu_idx);
692 void dm_sync_table(struct mapped_device *md)
694 synchronize_srcu(&md->io_barrier);
695 synchronize_rcu_expedited();
699 * A fast alternative to dm_get_live_table/dm_put_live_table.
700 * The caller must not block between these two functions.
702 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
705 return rcu_dereference(md->map);
708 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
713 static char *_dm_claim_ptr = "I belong to device-mapper";
716 * Open a table device so we can use it as a map destination.
718 static int open_table_device(struct table_device *td, dev_t dev,
719 struct mapped_device *md)
721 struct block_device *bdev;
725 BUG_ON(td->dm_dev.bdev);
727 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
729 return PTR_ERR(bdev);
731 r = bd_link_disk_holder(bdev, dm_disk(md));
733 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
737 td->dm_dev.bdev = bdev;
738 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
743 * Close a table device that we've been using.
745 static void close_table_device(struct table_device *td, struct mapped_device *md)
747 if (!td->dm_dev.bdev)
750 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
751 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
752 put_dax(td->dm_dev.dax_dev);
753 td->dm_dev.bdev = NULL;
754 td->dm_dev.dax_dev = NULL;
757 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
759 struct table_device *td;
761 list_for_each_entry(td, l, list)
762 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
768 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
769 struct dm_dev **result) {
771 struct table_device *td;
773 mutex_lock(&md->table_devices_lock);
774 td = find_table_device(&md->table_devices, dev, mode);
776 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
778 mutex_unlock(&md->table_devices_lock);
782 td->dm_dev.mode = mode;
783 td->dm_dev.bdev = NULL;
785 if ((r = open_table_device(td, dev, md))) {
786 mutex_unlock(&md->table_devices_lock);
791 format_dev_t(td->dm_dev.name, dev);
793 refcount_set(&td->count, 1);
794 list_add(&td->list, &md->table_devices);
796 refcount_inc(&td->count);
798 mutex_unlock(&md->table_devices_lock);
800 *result = &td->dm_dev;
803 EXPORT_SYMBOL_GPL(dm_get_table_device);
805 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
807 struct table_device *td = container_of(d, struct table_device, dm_dev);
809 mutex_lock(&md->table_devices_lock);
810 if (refcount_dec_and_test(&td->count)) {
811 close_table_device(td, md);
815 mutex_unlock(&md->table_devices_lock);
817 EXPORT_SYMBOL(dm_put_table_device);
819 static void free_table_devices(struct list_head *devices)
821 struct list_head *tmp, *next;
823 list_for_each_safe(tmp, next, devices) {
824 struct table_device *td = list_entry(tmp, struct table_device, list);
826 DMWARN("dm_destroy: %s still exists with %d references",
827 td->dm_dev.name, refcount_read(&td->count));
833 * Get the geometry associated with a dm device
835 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
843 * Set the geometry of a device.
845 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
847 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
849 if (geo->start > sz) {
850 DMWARN("Start sector is beyond the geometry limits.");
859 static int __noflush_suspending(struct mapped_device *md)
861 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
865 * Decrements the number of outstanding ios that a bio has been
866 * cloned into, completing the original io if necc.
868 static void dec_pending(struct dm_io *io, blk_status_t error)
871 blk_status_t io_error;
873 struct mapped_device *md = io->md;
874 unsigned long start_time = 0;
875 struct dm_stats_aux stats_aux;
877 /* Push-back supersedes any I/O errors */
878 if (unlikely(error)) {
879 spin_lock_irqsave(&io->endio_lock, flags);
880 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
882 spin_unlock_irqrestore(&io->endio_lock, flags);
885 if (atomic_dec_and_test(&io->io_count)) {
886 if (io->status == BLK_STS_DM_REQUEUE) {
888 * Target requested pushing back the I/O.
890 spin_lock_irqsave(&md->deferred_lock, flags);
891 if (__noflush_suspending(md))
892 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
893 bio_list_add_head(&md->deferred, io->orig_bio);
895 /* noflush suspend was interrupted. */
896 io->status = BLK_STS_IOERR;
897 spin_unlock_irqrestore(&md->deferred_lock, flags);
900 io_error = io->status;
902 start_time = io->start_time;
903 stats_aux = io->stats_aux;
905 end_io_acct(md, bio, start_time, &stats_aux);
907 if (io_error == BLK_STS_DM_REQUEUE)
910 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
912 * Preflush done for flush with data, reissue
913 * without REQ_PREFLUSH.
915 bio->bi_opf &= ~REQ_PREFLUSH;
918 /* done with normal IO or empty flush */
920 bio->bi_status = io_error;
926 void disable_discard(struct mapped_device *md)
928 struct queue_limits *limits = dm_get_queue_limits(md);
930 /* device doesn't really support DISCARD, disable it */
931 limits->max_discard_sectors = 0;
932 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
935 void disable_write_same(struct mapped_device *md)
937 struct queue_limits *limits = dm_get_queue_limits(md);
939 /* device doesn't really support WRITE SAME, disable it */
940 limits->max_write_same_sectors = 0;
943 void disable_write_zeroes(struct mapped_device *md)
945 struct queue_limits *limits = dm_get_queue_limits(md);
947 /* device doesn't really support WRITE ZEROES, disable it */
948 limits->max_write_zeroes_sectors = 0;
951 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
953 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
956 static void clone_endio(struct bio *bio)
958 blk_status_t error = bio->bi_status;
959 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
960 struct dm_io *io = tio->io;
961 struct mapped_device *md = tio->io->md;
962 dm_endio_fn endio = tio->ti->type->end_io;
964 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
965 if (bio_op(bio) == REQ_OP_DISCARD &&
966 !bio->bi_disk->queue->limits.max_discard_sectors)
968 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
969 !bio->bi_disk->queue->limits.max_write_same_sectors)
970 disable_write_same(md);
971 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
972 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
973 disable_write_zeroes(md);
977 int r = endio(tio->ti, bio, &error);
979 case DM_ENDIO_REQUEUE:
980 error = BLK_STS_DM_REQUEUE;
984 case DM_ENDIO_INCOMPLETE:
985 /* The target will handle the io */
988 DMWARN("unimplemented target endio return value: %d", r);
993 if (unlikely(swap_bios_limit(tio->ti, bio))) {
994 struct mapped_device *md = io->md;
995 up(&md->swap_bios_semaphore);
999 dec_pending(io, error);
1003 * Return maximum size of I/O possible at the supplied sector up to the current
1006 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1008 sector_t target_offset = dm_target_offset(ti, sector);
1010 return ti->len - target_offset;
1013 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1015 sector_t len = max_io_len_target_boundary(sector, ti);
1016 sector_t offset, max_len;
1019 * Does the target need to split even further?
1021 if (ti->max_io_len) {
1022 offset = dm_target_offset(ti, sector);
1023 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1024 max_len = sector_div(offset, ti->max_io_len);
1026 max_len = offset & (ti->max_io_len - 1);
1027 max_len = ti->max_io_len - max_len;
1036 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1038 if (len > UINT_MAX) {
1039 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1040 (unsigned long long)len, UINT_MAX);
1041 ti->error = "Maximum size of target IO is too large";
1045 ti->max_io_len = (uint32_t) len;
1049 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1051 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1052 sector_t sector, int *srcu_idx)
1053 __acquires(md->io_barrier)
1055 struct dm_table *map;
1056 struct dm_target *ti;
1058 map = dm_get_live_table(md, srcu_idx);
1062 ti = dm_table_find_target(map, sector);
1063 if (!dm_target_is_valid(ti))
1069 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1070 long nr_pages, void **kaddr, pfn_t *pfn)
1072 struct mapped_device *md = dax_get_private(dax_dev);
1073 sector_t sector = pgoff * PAGE_SECTORS;
1074 struct dm_target *ti;
1075 long len, ret = -EIO;
1078 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1082 if (!ti->type->direct_access)
1084 len = max_io_len(sector, ti) / PAGE_SECTORS;
1087 nr_pages = min(len, nr_pages);
1088 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1091 dm_put_live_table(md, srcu_idx);
1096 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1097 void *addr, size_t bytes, struct iov_iter *i)
1099 struct mapped_device *md = dax_get_private(dax_dev);
1100 sector_t sector = pgoff * PAGE_SECTORS;
1101 struct dm_target *ti;
1105 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1109 if (!ti->type->dax_copy_from_iter) {
1110 ret = copy_from_iter(addr, bytes, i);
1113 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1115 dm_put_live_table(md, srcu_idx);
1120 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1121 void *addr, size_t bytes, struct iov_iter *i)
1123 struct mapped_device *md = dax_get_private(dax_dev);
1124 sector_t sector = pgoff * PAGE_SECTORS;
1125 struct dm_target *ti;
1129 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1133 if (!ti->type->dax_copy_to_iter) {
1134 ret = copy_to_iter(addr, bytes, i);
1137 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1139 dm_put_live_table(md, srcu_idx);
1145 * A target may call dm_accept_partial_bio only from the map routine. It is
1146 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1148 * dm_accept_partial_bio informs the dm that the target only wants to process
1149 * additional n_sectors sectors of the bio and the rest of the data should be
1150 * sent in a next bio.
1152 * A diagram that explains the arithmetics:
1153 * +--------------------+---------------+-------+
1155 * +--------------------+---------------+-------+
1157 * <-------------- *tio->len_ptr --------------->
1158 * <------- bi_size ------->
1161 * Region 1 was already iterated over with bio_advance or similar function.
1162 * (it may be empty if the target doesn't use bio_advance)
1163 * Region 2 is the remaining bio size that the target wants to process.
1164 * (it may be empty if region 1 is non-empty, although there is no reason
1166 * The target requires that region 3 is to be sent in the next bio.
1168 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1169 * the partially processed part (the sum of regions 1+2) must be the same for all
1170 * copies of the bio.
1172 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1174 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1175 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1176 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1177 BUG_ON(bi_size > *tio->len_ptr);
1178 BUG_ON(n_sectors > bi_size);
1179 *tio->len_ptr -= bi_size - n_sectors;
1180 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1182 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1185 * The zone descriptors obtained with a zone report indicate zone positions
1186 * within the target backing device, regardless of that device is a partition
1187 * and regardless of the target mapping start sector on the device or partition.
1188 * The zone descriptors start sector and write pointer position must be adjusted
1189 * to match their relative position within the dm device.
1190 * A target may call dm_remap_zone_report() after completion of a
1191 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1194 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1196 #ifdef CONFIG_BLK_DEV_ZONED
1197 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1198 struct bio *report_bio = tio->io->orig_bio;
1199 struct blk_zone_report_hdr *hdr = NULL;
1200 struct blk_zone *zone;
1201 unsigned int nr_rep = 0;
1203 sector_t part_offset;
1204 struct bio_vec bvec;
1205 struct bvec_iter iter;
1212 * bio sector was incremented by the request size on completion. Taking
1213 * into account the original request sector, the target start offset on
1214 * the backing device and the target mapping offset (ti->begin), the
1215 * start sector of the backing device. The partition offset is always 0
1216 * if the target uses a whole device.
1218 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1221 * Remap the start sector of the reported zones. For sequential zones,
1222 * also remap the write pointer position.
1224 bio_for_each_segment(bvec, report_bio, iter) {
1225 addr = kmap_atomic(bvec.bv_page);
1227 /* Remember the report header in the first page */
1230 ofst = sizeof(struct blk_zone_report_hdr);
1234 /* Set zones start sector */
1235 while (hdr->nr_zones && ofst < bvec.bv_len) {
1237 zone->start -= part_offset;
1238 if (zone->start >= start + ti->len) {
1242 zone->start = zone->start + ti->begin - start;
1243 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1244 if (zone->cond == BLK_ZONE_COND_FULL)
1245 zone->wp = zone->start + zone->len;
1246 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1247 zone->wp = zone->start;
1249 zone->wp = zone->wp + ti->begin - start - part_offset;
1251 ofst += sizeof(struct blk_zone);
1257 kunmap_atomic(addr);
1264 hdr->nr_zones = nr_rep;
1268 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1270 #else /* !CONFIG_BLK_DEV_ZONED */
1271 bio->bi_status = BLK_STS_NOTSUPP;
1274 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1276 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1278 mutex_lock(&md->swap_bios_lock);
1279 while (latch < md->swap_bios) {
1281 down(&md->swap_bios_semaphore);
1284 while (latch > md->swap_bios) {
1286 up(&md->swap_bios_semaphore);
1289 mutex_unlock(&md->swap_bios_lock);
1292 static blk_qc_t __map_bio(struct dm_target_io *tio)
1296 struct bio *clone = &tio->clone;
1297 struct dm_io *io = tio->io;
1298 struct mapped_device *md = io->md;
1299 struct dm_target *ti = tio->ti;
1300 blk_qc_t ret = BLK_QC_T_NONE;
1302 clone->bi_end_io = clone_endio;
1305 * Map the clone. If r == 0 we don't need to do
1306 * anything, the target has assumed ownership of
1309 atomic_inc(&io->io_count);
1310 sector = clone->bi_iter.bi_sector;
1312 if (unlikely(swap_bios_limit(ti, clone))) {
1313 struct mapped_device *md = io->md;
1314 int latch = get_swap_bios();
1315 if (unlikely(latch != md->swap_bios))
1316 __set_swap_bios_limit(md, latch);
1317 down(&md->swap_bios_semaphore);
1320 r = ti->type->map(ti, clone);
1322 case DM_MAPIO_SUBMITTED:
1324 case DM_MAPIO_REMAPPED:
1325 /* the bio has been remapped so dispatch it */
1326 trace_block_bio_remap(clone->bi_disk->queue, clone,
1327 bio_dev(io->orig_bio), sector);
1328 if (md->type == DM_TYPE_NVME_BIO_BASED)
1329 ret = direct_make_request(clone);
1331 ret = generic_make_request(clone);
1334 if (unlikely(swap_bios_limit(ti, clone))) {
1335 struct mapped_device *md = io->md;
1336 up(&md->swap_bios_semaphore);
1339 dec_pending(io, BLK_STS_IOERR);
1341 case DM_MAPIO_REQUEUE:
1342 if (unlikely(swap_bios_limit(ti, clone))) {
1343 struct mapped_device *md = io->md;
1344 up(&md->swap_bios_semaphore);
1347 dec_pending(io, BLK_STS_DM_REQUEUE);
1350 DMWARN("unimplemented target map return value: %d", r);
1357 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1359 bio->bi_iter.bi_sector = sector;
1360 bio->bi_iter.bi_size = to_bytes(len);
1364 * Creates a bio that consists of range of complete bvecs.
1366 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1367 sector_t sector, unsigned len)
1369 struct bio *clone = &tio->clone;
1371 __bio_clone_fast(clone, bio);
1373 if (unlikely(bio_integrity(bio) != NULL)) {
1376 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1377 !dm_target_passes_integrity(tio->ti->type))) {
1378 DMWARN("%s: the target %s doesn't support integrity data.",
1379 dm_device_name(tio->io->md),
1380 tio->ti->type->name);
1384 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1389 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1390 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1391 clone->bi_iter.bi_size = to_bytes(len);
1393 if (unlikely(bio_integrity(bio) != NULL))
1394 bio_integrity_trim(clone);
1399 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1400 struct dm_target *ti, unsigned num_bios)
1402 struct dm_target_io *tio;
1408 if (num_bios == 1) {
1409 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1410 bio_list_add(blist, &tio->clone);
1414 for (try = 0; try < 2; try++) {
1419 mutex_lock(&ci->io->md->table_devices_lock);
1420 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1421 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1425 bio_list_add(blist, &tio->clone);
1428 mutex_unlock(&ci->io->md->table_devices_lock);
1429 if (bio_nr == num_bios)
1432 while ((bio = bio_list_pop(blist))) {
1433 tio = container_of(bio, struct dm_target_io, clone);
1439 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1440 struct dm_target_io *tio, unsigned *len)
1442 struct bio *clone = &tio->clone;
1446 __bio_clone_fast(clone, ci->bio);
1448 bio_setup_sector(clone, ci->sector, *len);
1450 return __map_bio(tio);
1453 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1454 unsigned num_bios, unsigned *len)
1456 struct bio_list blist = BIO_EMPTY_LIST;
1458 struct dm_target_io *tio;
1460 alloc_multiple_bios(&blist, ci, ti, num_bios);
1462 while ((bio = bio_list_pop(&blist))) {
1463 tio = container_of(bio, struct dm_target_io, clone);
1464 (void) __clone_and_map_simple_bio(ci, tio, len);
1468 static int __send_empty_flush(struct clone_info *ci)
1470 unsigned target_nr = 0;
1471 struct dm_target *ti;
1473 BUG_ON(bio_has_data(ci->bio));
1474 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1475 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1480 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1481 sector_t sector, unsigned *len)
1483 struct bio *bio = ci->bio;
1484 struct dm_target_io *tio;
1487 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1489 r = clone_bio(tio, bio, sector, *len);
1494 (void) __map_bio(tio);
1499 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1501 static unsigned get_num_discard_bios(struct dm_target *ti)
1503 return ti->num_discard_bios;
1506 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1508 return ti->num_secure_erase_bios;
1511 static unsigned get_num_write_same_bios(struct dm_target *ti)
1513 return ti->num_write_same_bios;
1516 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1518 return ti->num_write_zeroes_bios;
1521 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1523 static bool is_split_required_for_discard(struct dm_target *ti)
1525 return ti->split_discard_bios;
1528 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1529 get_num_bios_fn get_num_bios,
1530 is_split_required_fn is_split_required)
1536 * Even though the device advertised support for this type of
1537 * request, that does not mean every target supports it, and
1538 * reconfiguration might also have changed that since the
1539 * check was performed.
1541 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1545 if (is_split_required && !is_split_required(ti))
1546 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1548 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1550 __send_duplicate_bios(ci, ti, num_bios, &len);
1553 ci->sector_count -= len;
1558 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1560 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1561 is_split_required_for_discard);
1564 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1566 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1569 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1571 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1574 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1576 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1579 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1582 struct bio *bio = ci->bio;
1584 if (bio_op(bio) == REQ_OP_DISCARD)
1585 *result = __send_discard(ci, ti);
1586 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1587 *result = __send_secure_erase(ci, ti);
1588 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1589 *result = __send_write_same(ci, ti);
1590 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1591 *result = __send_write_zeroes(ci, ti);
1599 * Select the correct strategy for processing a non-flush bio.
1601 static int __split_and_process_non_flush(struct clone_info *ci)
1603 struct bio *bio = ci->bio;
1604 struct dm_target *ti;
1608 ti = dm_table_find_target(ci->map, ci->sector);
1609 if (!dm_target_is_valid(ti))
1612 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1615 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1616 len = ci->sector_count;
1618 len = min_t(sector_t, max_io_len(ci->sector, ti),
1621 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1626 ci->sector_count -= len;
1631 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1632 struct dm_table *map, struct bio *bio)
1635 ci->io = alloc_io(md, bio);
1636 ci->sector = bio->bi_iter.bi_sector;
1640 * Entry point to split a bio into clones and submit them to the targets.
1642 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1643 struct dm_table *map, struct bio *bio)
1645 struct clone_info ci;
1646 blk_qc_t ret = BLK_QC_T_NONE;
1649 if (unlikely(!map)) {
1654 blk_queue_split(md->queue, &bio);
1656 init_clone_info(&ci, md, map, bio);
1658 if (bio->bi_opf & REQ_PREFLUSH) {
1659 ci.bio = &ci.io->md->flush_bio;
1660 ci.sector_count = 0;
1661 error = __send_empty_flush(&ci);
1662 /* dec_pending submits any data associated with flush */
1663 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1665 ci.sector_count = 0;
1666 error = __split_and_process_non_flush(&ci);
1669 ci.sector_count = bio_sectors(bio);
1670 while (ci.sector_count && !error) {
1671 error = __split_and_process_non_flush(&ci);
1672 if (current->bio_list && ci.sector_count && !error) {
1674 * Remainder must be passed to generic_make_request()
1675 * so that it gets handled *after* bios already submitted
1676 * have been completely processed.
1677 * We take a clone of the original to store in
1678 * ci.io->orig_bio to be used by end_io_acct() and
1679 * for dec_pending to use for completion handling.
1680 * As this path is not used for REQ_OP_ZONE_REPORT,
1681 * the usage of io->orig_bio in dm_remap_zone_report()
1682 * won't be affected by this reassignment.
1684 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1685 GFP_NOIO, &md->queue->bio_split);
1686 ci.io->orig_bio = b;
1688 ret = generic_make_request(bio);
1694 /* drop the extra reference count */
1695 dec_pending(ci.io, errno_to_blk_status(error));
1700 * Optimized variant of __split_and_process_bio that leverages the
1701 * fact that targets that use it do _not_ have a need to split bios.
1703 static blk_qc_t __process_bio(struct mapped_device *md,
1704 struct dm_table *map, struct bio *bio)
1706 struct clone_info ci;
1707 blk_qc_t ret = BLK_QC_T_NONE;
1710 if (unlikely(!map)) {
1715 init_clone_info(&ci, md, map, bio);
1717 if (bio->bi_opf & REQ_PREFLUSH) {
1718 ci.bio = &ci.io->md->flush_bio;
1719 ci.sector_count = 0;
1720 error = __send_empty_flush(&ci);
1721 /* dec_pending submits any data associated with flush */
1723 struct dm_target *ti = md->immutable_target;
1724 struct dm_target_io *tio;
1727 * Defend against IO still getting in during teardown
1728 * - as was seen for a time with nvme-fcloop
1730 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1736 ci.sector_count = bio_sectors(bio);
1737 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1740 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1741 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1744 /* drop the extra reference count */
1745 dec_pending(ci.io, errno_to_blk_status(error));
1749 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1751 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1752 process_bio_fn process_bio)
1754 struct mapped_device *md = q->queuedata;
1755 blk_qc_t ret = BLK_QC_T_NONE;
1757 struct dm_table *map;
1759 map = dm_get_live_table(md, &srcu_idx);
1761 /* if we're suspended, we have to queue this io for later */
1762 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1763 dm_put_live_table(md, srcu_idx);
1765 if (!(bio->bi_opf & REQ_RAHEAD))
1772 ret = process_bio(md, map, bio);
1774 dm_put_live_table(md, srcu_idx);
1779 * The request function that remaps the bio to one target and
1780 * splits off any remainder.
1782 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1784 return __dm_make_request(q, bio, __split_and_process_bio);
1787 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1789 return __dm_make_request(q, bio, __process_bio);
1792 static int dm_any_congested(void *congested_data, int bdi_bits)
1795 struct mapped_device *md = congested_data;
1796 struct dm_table *map;
1798 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1799 if (dm_request_based(md)) {
1801 * With request-based DM we only need to check the
1802 * top-level queue for congestion.
1804 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1806 map = dm_get_live_table_fast(md);
1808 r = dm_table_any_congested(map, bdi_bits);
1809 dm_put_live_table_fast(md);
1816 /*-----------------------------------------------------------------
1817 * An IDR is used to keep track of allocated minor numbers.
1818 *---------------------------------------------------------------*/
1819 static void free_minor(int minor)
1821 spin_lock(&_minor_lock);
1822 idr_remove(&_minor_idr, minor);
1823 spin_unlock(&_minor_lock);
1827 * See if the device with a specific minor # is free.
1829 static int specific_minor(int minor)
1833 if (minor >= (1 << MINORBITS))
1836 idr_preload(GFP_KERNEL);
1837 spin_lock(&_minor_lock);
1839 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1841 spin_unlock(&_minor_lock);
1844 return r == -ENOSPC ? -EBUSY : r;
1848 static int next_free_minor(int *minor)
1852 idr_preload(GFP_KERNEL);
1853 spin_lock(&_minor_lock);
1855 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1857 spin_unlock(&_minor_lock);
1865 static const struct block_device_operations dm_blk_dops;
1866 static const struct dax_operations dm_dax_ops;
1868 static void dm_wq_work(struct work_struct *work);
1870 static void dm_init_normal_md_queue(struct mapped_device *md)
1872 md->use_blk_mq = false;
1875 * Initialize aspects of queue that aren't relevant for blk-mq
1877 md->queue->backing_dev_info->congested_data = md;
1878 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1881 static void cleanup_mapped_device(struct mapped_device *md)
1884 destroy_workqueue(md->wq);
1885 if (md->kworker_task)
1886 kthread_stop(md->kworker_task);
1887 bioset_exit(&md->bs);
1888 bioset_exit(&md->io_bs);
1891 kill_dax(md->dax_dev);
1892 put_dax(md->dax_dev);
1897 spin_lock(&_minor_lock);
1898 md->disk->private_data = NULL;
1899 spin_unlock(&_minor_lock);
1900 del_gendisk(md->disk);
1905 blk_cleanup_queue(md->queue);
1907 cleanup_srcu_struct(&md->io_barrier);
1914 mutex_destroy(&md->suspend_lock);
1915 mutex_destroy(&md->type_lock);
1916 mutex_destroy(&md->table_devices_lock);
1917 mutex_destroy(&md->swap_bios_lock);
1919 dm_mq_cleanup_mapped_device(md);
1923 * Allocate and initialise a blank device with a given minor.
1925 static struct mapped_device *alloc_dev(int minor)
1927 int r, numa_node_id = dm_get_numa_node();
1928 struct dax_device *dax_dev = NULL;
1929 struct mapped_device *md;
1932 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1934 DMWARN("unable to allocate device, out of memory.");
1938 if (!try_module_get(THIS_MODULE))
1939 goto bad_module_get;
1941 /* get a minor number for the dev */
1942 if (minor == DM_ANY_MINOR)
1943 r = next_free_minor(&minor);
1945 r = specific_minor(minor);
1949 r = init_srcu_struct(&md->io_barrier);
1951 goto bad_io_barrier;
1953 md->numa_node_id = numa_node_id;
1954 md->use_blk_mq = dm_use_blk_mq_default();
1955 md->init_tio_pdu = false;
1956 md->type = DM_TYPE_NONE;
1957 mutex_init(&md->suspend_lock);
1958 mutex_init(&md->type_lock);
1959 mutex_init(&md->table_devices_lock);
1960 spin_lock_init(&md->deferred_lock);
1961 atomic_set(&md->holders, 1);
1962 atomic_set(&md->open_count, 0);
1963 atomic_set(&md->event_nr, 0);
1964 atomic_set(&md->uevent_seq, 0);
1965 INIT_LIST_HEAD(&md->uevent_list);
1966 INIT_LIST_HEAD(&md->table_devices);
1967 spin_lock_init(&md->uevent_lock);
1969 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1972 md->queue->queuedata = md;
1974 * default to bio-based required ->make_request_fn until DM
1975 * table is loaded and md->type established. If request-based
1976 * table is loaded: blk-mq will override accordingly.
1978 blk_queue_make_request(md->queue, dm_make_request);
1980 md->disk = alloc_disk_node(1, md->numa_node_id);
1984 atomic_set(&md->pending[0], 0);
1985 atomic_set(&md->pending[1], 0);
1986 init_waitqueue_head(&md->wait);
1987 INIT_WORK(&md->work, dm_wq_work);
1988 init_waitqueue_head(&md->eventq);
1989 init_completion(&md->kobj_holder.completion);
1990 md->kworker_task = NULL;
1992 md->swap_bios = get_swap_bios();
1993 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1994 mutex_init(&md->swap_bios_lock);
1996 md->disk->major = _major;
1997 md->disk->first_minor = minor;
1998 md->disk->fops = &dm_blk_dops;
1999 md->disk->queue = md->queue;
2000 md->disk->private_data = md;
2001 sprintf(md->disk->disk_name, "dm-%d", minor);
2003 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
2004 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
2008 md->dax_dev = dax_dev;
2010 add_disk_no_queue_reg(md->disk);
2011 format_dev_t(md->name, MKDEV(_major, minor));
2013 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2017 md->bdev = bdget_disk(md->disk, 0);
2021 bio_init(&md->flush_bio, NULL, 0);
2022 bio_set_dev(&md->flush_bio, md->bdev);
2023 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
2025 dm_stats_init(&md->stats);
2027 /* Populate the mapping, nobody knows we exist yet */
2028 spin_lock(&_minor_lock);
2029 old_md = idr_replace(&_minor_idr, md, minor);
2030 spin_unlock(&_minor_lock);
2032 BUG_ON(old_md != MINOR_ALLOCED);
2037 cleanup_mapped_device(md);
2041 module_put(THIS_MODULE);
2047 static void unlock_fs(struct mapped_device *md);
2049 static void free_dev(struct mapped_device *md)
2051 int minor = MINOR(disk_devt(md->disk));
2055 cleanup_mapped_device(md);
2057 free_table_devices(&md->table_devices);
2058 dm_stats_cleanup(&md->stats);
2061 module_put(THIS_MODULE);
2065 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2067 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2070 if (dm_table_bio_based(t)) {
2072 * The md may already have mempools that need changing.
2073 * If so, reload bioset because front_pad may have changed
2074 * because a different table was loaded.
2076 bioset_exit(&md->bs);
2077 bioset_exit(&md->io_bs);
2079 } else if (bioset_initialized(&md->bs)) {
2081 * There's no need to reload with request-based dm
2082 * because the size of front_pad doesn't change.
2083 * Note for future: If you are to reload bioset,
2084 * prep-ed requests in the queue may refer
2085 * to bio from the old bioset, so you must walk
2086 * through the queue to unprep.
2092 bioset_initialized(&md->bs) ||
2093 bioset_initialized(&md->io_bs));
2095 ret = bioset_init_from_src(&md->bs, &p->bs);
2098 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2100 bioset_exit(&md->bs);
2102 /* mempool bind completed, no longer need any mempools in the table */
2103 dm_table_free_md_mempools(t);
2108 * Bind a table to the device.
2110 static void event_callback(void *context)
2112 unsigned long flags;
2114 struct mapped_device *md = (struct mapped_device *) context;
2116 spin_lock_irqsave(&md->uevent_lock, flags);
2117 list_splice_init(&md->uevent_list, &uevents);
2118 spin_unlock_irqrestore(&md->uevent_lock, flags);
2120 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2122 atomic_inc(&md->event_nr);
2123 wake_up(&md->eventq);
2124 dm_issue_global_event();
2128 * Protected by md->suspend_lock obtained by dm_swap_table().
2130 static void __set_size(struct mapped_device *md, sector_t size)
2132 lockdep_assert_held(&md->suspend_lock);
2134 set_capacity(md->disk, size);
2136 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2140 * Returns old map, which caller must destroy.
2142 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2143 struct queue_limits *limits)
2145 struct dm_table *old_map;
2146 struct request_queue *q = md->queue;
2147 bool request_based = dm_table_request_based(t);
2151 lockdep_assert_held(&md->suspend_lock);
2153 size = dm_table_get_size(t);
2156 * Wipe any geometry if the size of the table changed.
2158 if (size != dm_get_size(md))
2159 memset(&md->geometry, 0, sizeof(md->geometry));
2161 __set_size(md, size);
2163 dm_table_event_callback(t, event_callback, md);
2166 * The queue hasn't been stopped yet, if the old table type wasn't
2167 * for request-based during suspension. So stop it to prevent
2168 * I/O mapping before resume.
2169 * This must be done before setting the queue restrictions,
2170 * because request-based dm may be run just after the setting.
2175 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2177 * Leverage the fact that request-based DM targets and
2178 * NVMe bio based targets are immutable singletons
2179 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2180 * and __process_bio.
2182 md->immutable_target = dm_table_get_immutable_target(t);
2185 ret = __bind_mempools(md, t);
2187 old_map = ERR_PTR(ret);
2191 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2192 rcu_assign_pointer(md->map, (void *)t);
2193 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2195 dm_table_set_restrictions(t, q, limits);
2204 * Returns unbound table for the caller to free.
2206 static struct dm_table *__unbind(struct mapped_device *md)
2208 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2213 dm_table_event_callback(map, NULL, NULL);
2214 RCU_INIT_POINTER(md->map, NULL);
2221 * Constructor for a new device.
2223 int dm_create(int minor, struct mapped_device **result)
2226 struct mapped_device *md;
2228 md = alloc_dev(minor);
2232 r = dm_sysfs_init(md);
2243 * Functions to manage md->type.
2244 * All are required to hold md->type_lock.
2246 void dm_lock_md_type(struct mapped_device *md)
2248 mutex_lock(&md->type_lock);
2251 void dm_unlock_md_type(struct mapped_device *md)
2253 mutex_unlock(&md->type_lock);
2256 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2258 BUG_ON(!mutex_is_locked(&md->type_lock));
2262 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2267 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2269 return md->immutable_target_type;
2273 * The queue_limits are only valid as long as you have a reference
2276 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2278 BUG_ON(!atomic_read(&md->holders));
2279 return &md->queue->limits;
2281 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2284 * Setup the DM device's queue based on md's type
2286 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2289 struct queue_limits limits;
2290 enum dm_queue_mode type = dm_get_md_type(md);
2293 case DM_TYPE_REQUEST_BASED:
2294 dm_init_normal_md_queue(md);
2295 r = dm_old_init_request_queue(md, t);
2297 DMERR("Cannot initialize queue for request-based mapped device");
2301 case DM_TYPE_MQ_REQUEST_BASED:
2302 r = dm_mq_init_request_queue(md, t);
2304 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2308 case DM_TYPE_BIO_BASED:
2309 case DM_TYPE_DAX_BIO_BASED:
2310 dm_init_normal_md_queue(md);
2312 case DM_TYPE_NVME_BIO_BASED:
2313 dm_init_normal_md_queue(md);
2314 blk_queue_make_request(md->queue, dm_make_request_nvme);
2321 r = dm_calculate_queue_limits(t, &limits);
2323 DMERR("Cannot calculate initial queue limits");
2326 dm_table_set_restrictions(t, md->queue, &limits);
2327 blk_register_queue(md->disk);
2332 struct mapped_device *dm_get_md(dev_t dev)
2334 struct mapped_device *md;
2335 unsigned minor = MINOR(dev);
2337 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2340 spin_lock(&_minor_lock);
2342 md = idr_find(&_minor_idr, minor);
2343 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2344 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2350 spin_unlock(&_minor_lock);
2354 EXPORT_SYMBOL_GPL(dm_get_md);
2356 void *dm_get_mdptr(struct mapped_device *md)
2358 return md->interface_ptr;
2361 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2363 md->interface_ptr = ptr;
2366 void dm_get(struct mapped_device *md)
2368 atomic_inc(&md->holders);
2369 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2372 int dm_hold(struct mapped_device *md)
2374 spin_lock(&_minor_lock);
2375 if (test_bit(DMF_FREEING, &md->flags)) {
2376 spin_unlock(&_minor_lock);
2380 spin_unlock(&_minor_lock);
2383 EXPORT_SYMBOL_GPL(dm_hold);
2385 const char *dm_device_name(struct mapped_device *md)
2389 EXPORT_SYMBOL_GPL(dm_device_name);
2391 static void __dm_destroy(struct mapped_device *md, bool wait)
2393 struct dm_table *map;
2398 spin_lock(&_minor_lock);
2399 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2400 set_bit(DMF_FREEING, &md->flags);
2401 spin_unlock(&_minor_lock);
2403 blk_set_queue_dying(md->queue);
2405 if (dm_request_based(md) && md->kworker_task)
2406 kthread_flush_worker(&md->kworker);
2409 * Take suspend_lock so that presuspend and postsuspend methods
2410 * do not race with internal suspend.
2412 mutex_lock(&md->suspend_lock);
2413 map = dm_get_live_table(md, &srcu_idx);
2414 if (!dm_suspended_md(md)) {
2415 dm_table_presuspend_targets(map);
2416 set_bit(DMF_SUSPENDED, &md->flags);
2417 set_bit(DMF_POST_SUSPENDING, &md->flags);
2418 dm_table_postsuspend_targets(map);
2420 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2421 dm_put_live_table(md, srcu_idx);
2422 mutex_unlock(&md->suspend_lock);
2425 * Rare, but there may be I/O requests still going to complete,
2426 * for example. Wait for all references to disappear.
2427 * No one should increment the reference count of the mapped_device,
2428 * after the mapped_device state becomes DMF_FREEING.
2431 while (atomic_read(&md->holders))
2433 else if (atomic_read(&md->holders))
2434 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2435 dm_device_name(md), atomic_read(&md->holders));
2438 dm_table_destroy(__unbind(md));
2442 void dm_destroy(struct mapped_device *md)
2444 __dm_destroy(md, true);
2447 void dm_destroy_immediate(struct mapped_device *md)
2449 __dm_destroy(md, false);
2452 void dm_put(struct mapped_device *md)
2454 atomic_dec(&md->holders);
2456 EXPORT_SYMBOL_GPL(dm_put);
2458 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2464 prepare_to_wait(&md->wait, &wait, task_state);
2466 if (!md_in_flight(md))
2469 if (signal_pending_state(task_state, current)) {
2476 finish_wait(&md->wait, &wait);
2478 smp_rmb(); /* paired with atomic_dec_return in end_io_acct */
2484 * Process the deferred bios
2486 static void dm_wq_work(struct work_struct *work)
2488 struct mapped_device *md = container_of(work, struct mapped_device,
2492 struct dm_table *map;
2494 map = dm_get_live_table(md, &srcu_idx);
2496 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2497 spin_lock_irq(&md->deferred_lock);
2498 c = bio_list_pop(&md->deferred);
2499 spin_unlock_irq(&md->deferred_lock);
2504 if (dm_request_based(md))
2505 generic_make_request(c);
2507 __split_and_process_bio(md, map, c);
2510 dm_put_live_table(md, srcu_idx);
2513 static void dm_queue_flush(struct mapped_device *md)
2515 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2516 smp_mb__after_atomic();
2517 queue_work(md->wq, &md->work);
2521 * Swap in a new table, returning the old one for the caller to destroy.
2523 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2525 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2526 struct queue_limits limits;
2529 mutex_lock(&md->suspend_lock);
2531 /* device must be suspended */
2532 if (!dm_suspended_md(md))
2536 * If the new table has no data devices, retain the existing limits.
2537 * This helps multipath with queue_if_no_path if all paths disappear,
2538 * then new I/O is queued based on these limits, and then some paths
2541 if (dm_table_has_no_data_devices(table)) {
2542 live_map = dm_get_live_table_fast(md);
2544 limits = md->queue->limits;
2545 dm_put_live_table_fast(md);
2549 r = dm_calculate_queue_limits(table, &limits);
2556 map = __bind(md, table, &limits);
2557 dm_issue_global_event();
2560 mutex_unlock(&md->suspend_lock);
2565 * Functions to lock and unlock any filesystem running on the
2568 static int lock_fs(struct mapped_device *md)
2572 WARN_ON(md->frozen_sb);
2574 md->frozen_sb = freeze_bdev(md->bdev);
2575 if (IS_ERR(md->frozen_sb)) {
2576 r = PTR_ERR(md->frozen_sb);
2577 md->frozen_sb = NULL;
2581 set_bit(DMF_FROZEN, &md->flags);
2586 static void unlock_fs(struct mapped_device *md)
2588 if (!test_bit(DMF_FROZEN, &md->flags))
2591 thaw_bdev(md->bdev, md->frozen_sb);
2592 md->frozen_sb = NULL;
2593 clear_bit(DMF_FROZEN, &md->flags);
2597 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2598 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2599 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2601 * If __dm_suspend returns 0, the device is completely quiescent
2602 * now. There is no request-processing activity. All new requests
2603 * are being added to md->deferred list.
2605 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2606 unsigned suspend_flags, long task_state,
2607 int dmf_suspended_flag)
2609 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2610 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2613 lockdep_assert_held(&md->suspend_lock);
2616 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2617 * This flag is cleared before dm_suspend returns.
2620 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2622 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2625 * This gets reverted if there's an error later and the targets
2626 * provide the .presuspend_undo hook.
2628 dm_table_presuspend_targets(map);
2631 * Flush I/O to the device.
2632 * Any I/O submitted after lock_fs() may not be flushed.
2633 * noflush takes precedence over do_lockfs.
2634 * (lock_fs() flushes I/Os and waits for them to complete.)
2636 if (!noflush && do_lockfs) {
2639 dm_table_presuspend_undo_targets(map);
2645 * Here we must make sure that no processes are submitting requests
2646 * to target drivers i.e. no one may be executing
2647 * __split_and_process_bio. This is called from dm_request and
2650 * To get all processes out of __split_and_process_bio in dm_request,
2651 * we take the write lock. To prevent any process from reentering
2652 * __split_and_process_bio from dm_request and quiesce the thread
2653 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2654 * flush_workqueue(md->wq).
2656 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2658 synchronize_srcu(&md->io_barrier);
2661 * Stop md->queue before flushing md->wq in case request-based
2662 * dm defers requests to md->wq from md->queue.
2664 if (dm_request_based(md)) {
2665 dm_stop_queue(md->queue);
2666 if (md->kworker_task)
2667 kthread_flush_worker(&md->kworker);
2670 flush_workqueue(md->wq);
2673 * At this point no more requests are entering target request routines.
2674 * We call dm_wait_for_completion to wait for all existing requests
2677 r = dm_wait_for_completion(md, task_state);
2679 set_bit(dmf_suspended_flag, &md->flags);
2682 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2684 synchronize_srcu(&md->io_barrier);
2686 /* were we interrupted ? */
2690 if (dm_request_based(md))
2691 dm_start_queue(md->queue);
2694 dm_table_presuspend_undo_targets(map);
2695 /* pushback list is already flushed, so skip flush */
2702 * We need to be able to change a mapping table under a mounted
2703 * filesystem. For example we might want to move some data in
2704 * the background. Before the table can be swapped with
2705 * dm_bind_table, dm_suspend must be called to flush any in
2706 * flight bios and ensure that any further io gets deferred.
2709 * Suspend mechanism in request-based dm.
2711 * 1. Flush all I/Os by lock_fs() if needed.
2712 * 2. Stop dispatching any I/O by stopping the request_queue.
2713 * 3. Wait for all in-flight I/Os to be completed or requeued.
2715 * To abort suspend, start the request_queue.
2717 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2719 struct dm_table *map = NULL;
2723 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2725 if (dm_suspended_md(md)) {
2730 if (dm_suspended_internally_md(md)) {
2731 /* already internally suspended, wait for internal resume */
2732 mutex_unlock(&md->suspend_lock);
2733 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2739 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2741 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2745 set_bit(DMF_POST_SUSPENDING, &md->flags);
2746 dm_table_postsuspend_targets(map);
2747 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2750 mutex_unlock(&md->suspend_lock);
2754 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2757 int r = dm_table_resume_targets(map);
2765 * Flushing deferred I/Os must be done after targets are resumed
2766 * so that mapping of targets can work correctly.
2767 * Request-based dm is queueing the deferred I/Os in its request_queue.
2769 if (dm_request_based(md))
2770 dm_start_queue(md->queue);
2777 int dm_resume(struct mapped_device *md)
2780 struct dm_table *map = NULL;
2784 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2786 if (!dm_suspended_md(md))
2789 if (dm_suspended_internally_md(md)) {
2790 /* already internally suspended, wait for internal resume */
2791 mutex_unlock(&md->suspend_lock);
2792 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2798 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2799 if (!map || !dm_table_get_size(map))
2802 r = __dm_resume(md, map);
2806 clear_bit(DMF_SUSPENDED, &md->flags);
2808 mutex_unlock(&md->suspend_lock);
2814 * Internal suspend/resume works like userspace-driven suspend. It waits
2815 * until all bios finish and prevents issuing new bios to the target drivers.
2816 * It may be used only from the kernel.
2819 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2821 struct dm_table *map = NULL;
2823 lockdep_assert_held(&md->suspend_lock);
2825 if (md->internal_suspend_count++)
2826 return; /* nested internal suspend */
2828 if (dm_suspended_md(md)) {
2829 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2830 return; /* nest suspend */
2833 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2836 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2837 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2838 * would require changing .presuspend to return an error -- avoid this
2839 * until there is a need for more elaborate variants of internal suspend.
2841 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2842 DMF_SUSPENDED_INTERNALLY);
2844 set_bit(DMF_POST_SUSPENDING, &md->flags);
2845 dm_table_postsuspend_targets(map);
2846 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2849 static void __dm_internal_resume(struct mapped_device *md)
2851 BUG_ON(!md->internal_suspend_count);
2853 if (--md->internal_suspend_count)
2854 return; /* resume from nested internal suspend */
2856 if (dm_suspended_md(md))
2857 goto done; /* resume from nested suspend */
2860 * NOTE: existing callers don't need to call dm_table_resume_targets
2861 * (which may fail -- so best to avoid it for now by passing NULL map)
2863 (void) __dm_resume(md, NULL);
2866 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2867 smp_mb__after_atomic();
2868 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2871 void dm_internal_suspend_noflush(struct mapped_device *md)
2873 mutex_lock(&md->suspend_lock);
2874 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2875 mutex_unlock(&md->suspend_lock);
2877 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2879 void dm_internal_resume(struct mapped_device *md)
2881 mutex_lock(&md->suspend_lock);
2882 __dm_internal_resume(md);
2883 mutex_unlock(&md->suspend_lock);
2885 EXPORT_SYMBOL_GPL(dm_internal_resume);
2888 * Fast variants of internal suspend/resume hold md->suspend_lock,
2889 * which prevents interaction with userspace-driven suspend.
2892 void dm_internal_suspend_fast(struct mapped_device *md)
2894 mutex_lock(&md->suspend_lock);
2895 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2898 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2899 synchronize_srcu(&md->io_barrier);
2900 flush_workqueue(md->wq);
2901 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2903 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2905 void dm_internal_resume_fast(struct mapped_device *md)
2907 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2913 mutex_unlock(&md->suspend_lock);
2915 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2917 /*-----------------------------------------------------------------
2918 * Event notification.
2919 *---------------------------------------------------------------*/
2920 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2925 char udev_cookie[DM_COOKIE_LENGTH];
2926 char *envp[] = { udev_cookie, NULL };
2928 noio_flag = memalloc_noio_save();
2931 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2933 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2934 DM_COOKIE_ENV_VAR_NAME, cookie);
2935 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2939 memalloc_noio_restore(noio_flag);
2944 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2946 return atomic_add_return(1, &md->uevent_seq);
2949 uint32_t dm_get_event_nr(struct mapped_device *md)
2951 return atomic_read(&md->event_nr);
2954 int dm_wait_event(struct mapped_device *md, int event_nr)
2956 return wait_event_interruptible(md->eventq,
2957 (event_nr != atomic_read(&md->event_nr)));
2960 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2962 unsigned long flags;
2964 spin_lock_irqsave(&md->uevent_lock, flags);
2965 list_add(elist, &md->uevent_list);
2966 spin_unlock_irqrestore(&md->uevent_lock, flags);
2970 * The gendisk is only valid as long as you have a reference
2973 struct gendisk *dm_disk(struct mapped_device *md)
2977 EXPORT_SYMBOL_GPL(dm_disk);
2979 struct kobject *dm_kobject(struct mapped_device *md)
2981 return &md->kobj_holder.kobj;
2984 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2986 struct mapped_device *md;
2988 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2990 spin_lock(&_minor_lock);
2991 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2997 spin_unlock(&_minor_lock);
3002 int dm_suspended_md(struct mapped_device *md)
3004 return test_bit(DMF_SUSPENDED, &md->flags);
3007 static int dm_post_suspending_md(struct mapped_device *md)
3009 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3012 int dm_suspended_internally_md(struct mapped_device *md)
3014 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3017 int dm_test_deferred_remove_flag(struct mapped_device *md)
3019 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3022 int dm_suspended(struct dm_target *ti)
3024 return dm_suspended_md(dm_table_get_md(ti->table));
3026 EXPORT_SYMBOL_GPL(dm_suspended);
3028 int dm_post_suspending(struct dm_target *ti)
3030 return dm_post_suspending_md(dm_table_get_md(ti->table));
3032 EXPORT_SYMBOL_GPL(dm_post_suspending);
3034 int dm_noflush_suspending(struct dm_target *ti)
3036 return __noflush_suspending(dm_table_get_md(ti->table));
3038 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3040 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3041 unsigned integrity, unsigned per_io_data_size,
3042 unsigned min_pool_size)
3044 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3045 unsigned int pool_size = 0;
3046 unsigned int front_pad, io_front_pad;
3053 case DM_TYPE_BIO_BASED:
3054 case DM_TYPE_DAX_BIO_BASED:
3055 case DM_TYPE_NVME_BIO_BASED:
3056 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3057 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3058 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3059 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3062 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3065 case DM_TYPE_REQUEST_BASED:
3066 case DM_TYPE_MQ_REQUEST_BASED:
3067 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3068 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3069 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3075 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3079 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3085 dm_free_md_mempools(pools);
3090 void dm_free_md_mempools(struct dm_md_mempools *pools)
3095 bioset_exit(&pools->bs);
3096 bioset_exit(&pools->io_bs);
3108 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3111 struct mapped_device *md = bdev->bd_disk->private_data;
3112 struct dm_table *table;
3113 struct dm_target *ti;
3114 int ret = -ENOTTY, srcu_idx;
3116 table = dm_get_live_table(md, &srcu_idx);
3117 if (!table || !dm_table_get_size(table))
3120 /* We only support devices that have a single target */
3121 if (dm_table_get_num_targets(table) != 1)
3123 ti = dm_table_get_target(table, 0);
3125 if (dm_suspended_md(md)) {
3131 if (!ti->type->iterate_devices)
3134 ret = ti->type->iterate_devices(ti, fn, data);
3136 dm_put_live_table(md, srcu_idx);
3141 * For register / unregister we need to manually call out to every path.
3143 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3144 sector_t start, sector_t len, void *data)
3146 struct dm_pr *pr = data;
3147 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3149 if (!ops || !ops->pr_register)
3151 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3154 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3165 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3166 if (ret && new_key) {
3167 /* unregister all paths if we failed to register any path */
3168 pr.old_key = new_key;
3171 pr.fail_early = false;
3172 dm_call_pr(bdev, __dm_pr_register, &pr);
3178 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3181 struct mapped_device *md = bdev->bd_disk->private_data;
3182 const struct pr_ops *ops;
3185 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3189 ops = bdev->bd_disk->fops->pr_ops;
3190 if (ops && ops->pr_reserve)
3191 r = ops->pr_reserve(bdev, key, type, flags);
3195 dm_unprepare_ioctl(md, srcu_idx);
3199 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3201 struct mapped_device *md = bdev->bd_disk->private_data;
3202 const struct pr_ops *ops;
3205 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3209 ops = bdev->bd_disk->fops->pr_ops;
3210 if (ops && ops->pr_release)
3211 r = ops->pr_release(bdev, key, type);
3215 dm_unprepare_ioctl(md, srcu_idx);
3219 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3220 enum pr_type type, bool abort)
3222 struct mapped_device *md = bdev->bd_disk->private_data;
3223 const struct pr_ops *ops;
3226 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3230 ops = bdev->bd_disk->fops->pr_ops;
3231 if (ops && ops->pr_preempt)
3232 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3236 dm_unprepare_ioctl(md, srcu_idx);
3240 static int dm_pr_clear(struct block_device *bdev, u64 key)
3242 struct mapped_device *md = bdev->bd_disk->private_data;
3243 const struct pr_ops *ops;
3246 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3250 ops = bdev->bd_disk->fops->pr_ops;
3251 if (ops && ops->pr_clear)
3252 r = ops->pr_clear(bdev, key);
3256 dm_unprepare_ioctl(md, srcu_idx);
3260 static const struct pr_ops dm_pr_ops = {
3261 .pr_register = dm_pr_register,
3262 .pr_reserve = dm_pr_reserve,
3263 .pr_release = dm_pr_release,
3264 .pr_preempt = dm_pr_preempt,
3265 .pr_clear = dm_pr_clear,
3268 static const struct block_device_operations dm_blk_dops = {
3269 .open = dm_blk_open,
3270 .release = dm_blk_close,
3271 .ioctl = dm_blk_ioctl,
3272 .getgeo = dm_blk_getgeo,
3273 .pr_ops = &dm_pr_ops,
3274 .owner = THIS_MODULE
3277 static const struct dax_operations dm_dax_ops = {
3278 .direct_access = dm_dax_direct_access,
3279 .copy_from_iter = dm_dax_copy_from_iter,
3280 .copy_to_iter = dm_dax_copy_to_iter,
3286 module_init(dm_init);
3287 module_exit(dm_exit);
3289 module_param(major, uint, 0);
3290 MODULE_PARM_DESC(major, "The major number of the device mapper");
3292 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3293 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3295 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3296 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3298 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3299 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3301 MODULE_DESCRIPTION(DM_NAME " driver");
3302 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3303 MODULE_LICENSE("GPL");