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/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
27 #include <linux/refcount.h>
29 #define DM_MSG_PREFIX "core"
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name = DM_NAME;
40 static unsigned int major = 0;
41 static unsigned int _major = 0;
43 static DEFINE_IDR(_minor_idr);
45 static DEFINE_SPINLOCK(_minor_lock);
47 static void do_deferred_remove(struct work_struct *w);
49 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
51 static struct workqueue_struct *deferred_remove_workqueue;
53 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
56 void dm_issue_global_event(void)
58 atomic_inc(&dm_global_event_nr);
59 wake_up(&dm_global_eventq);
63 * One of these is allocated (on-stack) per original bio.
70 unsigned sector_count;
74 * One of these is allocated per clone bio.
76 #define DM_TIO_MAGIC 7282014
81 unsigned target_bio_nr;
88 * One of these is allocated per original bio.
89 * It contains the first clone used for that original.
91 #define DM_IO_MAGIC 5191977
94 struct mapped_device *md;
98 unsigned long start_time;
99 spinlock_t endio_lock;
100 struct dm_stats_aux stats_aux;
101 /* last member of dm_target_io is 'struct bio' */
102 struct dm_target_io tio;
105 void *dm_per_bio_data(struct bio *bio, size_t data_size)
107 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
108 if (!tio->inside_dm_io)
109 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
110 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
112 EXPORT_SYMBOL_GPL(dm_per_bio_data);
114 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
116 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
117 if (io->magic == DM_IO_MAGIC)
118 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
119 BUG_ON(io->magic != DM_TIO_MAGIC);
120 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
122 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
124 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
126 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
128 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
130 #define MINOR_ALLOCED ((void *)-1)
133 * Bits for the md->flags field.
135 #define DMF_BLOCK_IO_FOR_SUSPEND 0
136 #define DMF_SUSPENDED 1
138 #define DMF_FREEING 3
139 #define DMF_DELETING 4
140 #define DMF_NOFLUSH_SUSPENDING 5
141 #define DMF_DEFERRED_REMOVE 6
142 #define DMF_SUSPENDED_INTERNALLY 7
144 #define DM_NUMA_NODE NUMA_NO_NODE
145 static int dm_numa_node = DM_NUMA_NODE;
148 * For mempools pre-allocation at the table loading time.
150 struct dm_md_mempools {
152 struct bio_set io_bs;
155 struct table_device {
156 struct list_head list;
158 struct dm_dev dm_dev;
161 static struct kmem_cache *_rq_tio_cache;
162 static struct kmem_cache *_rq_cache;
165 * Bio-based DM's mempools' reserved IOs set by the user.
167 #define RESERVED_BIO_BASED_IOS 16
168 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
170 static int __dm_get_module_param_int(int *module_param, int min, int max)
172 int param = READ_ONCE(*module_param);
173 int modified_param = 0;
174 bool modified = true;
177 modified_param = min;
178 else if (param > max)
179 modified_param = max;
184 (void)cmpxchg(module_param, param, modified_param);
185 param = modified_param;
191 unsigned __dm_get_module_param(unsigned *module_param,
192 unsigned def, unsigned max)
194 unsigned param = READ_ONCE(*module_param);
195 unsigned modified_param = 0;
198 modified_param = def;
199 else if (param > max)
200 modified_param = max;
202 if (modified_param) {
203 (void)cmpxchg(module_param, param, modified_param);
204 param = modified_param;
210 unsigned dm_get_reserved_bio_based_ios(void)
212 return __dm_get_module_param(&reserved_bio_based_ios,
213 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
215 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
217 static unsigned dm_get_numa_node(void)
219 return __dm_get_module_param_int(&dm_numa_node,
220 DM_NUMA_NODE, num_online_nodes() - 1);
223 static int __init local_init(void)
227 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
231 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
232 __alignof__(struct request), 0, NULL);
234 goto out_free_rq_tio_cache;
236 r = dm_uevent_init();
238 goto out_free_rq_cache;
240 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
241 if (!deferred_remove_workqueue) {
243 goto out_uevent_exit;
247 r = register_blkdev(_major, _name);
249 goto out_free_workqueue;
257 destroy_workqueue(deferred_remove_workqueue);
261 kmem_cache_destroy(_rq_cache);
262 out_free_rq_tio_cache:
263 kmem_cache_destroy(_rq_tio_cache);
268 static void local_exit(void)
270 flush_scheduled_work();
271 destroy_workqueue(deferred_remove_workqueue);
273 kmem_cache_destroy(_rq_cache);
274 kmem_cache_destroy(_rq_tio_cache);
275 unregister_blkdev(_major, _name);
280 DMINFO("cleaned up");
283 static int (*_inits[])(void) __initdata = {
294 static void (*_exits[])(void) = {
305 static int __init dm_init(void)
307 const int count = ARRAY_SIZE(_inits);
311 for (i = 0; i < count; i++) {
326 static void __exit dm_exit(void)
328 int i = ARRAY_SIZE(_exits);
334 * Should be empty by this point.
336 idr_destroy(&_minor_idr);
340 * Block device functions
342 int dm_deleting_md(struct mapped_device *md)
344 return test_bit(DMF_DELETING, &md->flags);
347 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
349 struct mapped_device *md;
351 spin_lock(&_minor_lock);
353 md = bdev->bd_disk->private_data;
357 if (test_bit(DMF_FREEING, &md->flags) ||
358 dm_deleting_md(md)) {
364 atomic_inc(&md->open_count);
366 spin_unlock(&_minor_lock);
368 return md ? 0 : -ENXIO;
371 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
373 struct mapped_device *md;
375 spin_lock(&_minor_lock);
377 md = disk->private_data;
381 if (atomic_dec_and_test(&md->open_count) &&
382 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
383 queue_work(deferred_remove_workqueue, &deferred_remove_work);
387 spin_unlock(&_minor_lock);
390 int dm_open_count(struct mapped_device *md)
392 return atomic_read(&md->open_count);
396 * Guarantees nothing is using the device before it's deleted.
398 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
402 spin_lock(&_minor_lock);
404 if (dm_open_count(md)) {
407 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
408 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
411 set_bit(DMF_DELETING, &md->flags);
413 spin_unlock(&_minor_lock);
418 int dm_cancel_deferred_remove(struct mapped_device *md)
422 spin_lock(&_minor_lock);
424 if (test_bit(DMF_DELETING, &md->flags))
427 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
429 spin_unlock(&_minor_lock);
434 static void do_deferred_remove(struct work_struct *w)
436 dm_deferred_remove();
439 sector_t dm_get_size(struct mapped_device *md)
441 return get_capacity(md->disk);
444 struct request_queue *dm_get_md_queue(struct mapped_device *md)
449 struct dm_stats *dm_get_stats(struct mapped_device *md)
454 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
456 struct mapped_device *md = bdev->bd_disk->private_data;
458 return dm_get_geometry(md, geo);
461 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
462 struct block_device **bdev)
463 __acquires(md->io_barrier)
465 struct dm_target *tgt;
466 struct dm_table *map;
471 map = dm_get_live_table(md, srcu_idx);
472 if (!map || !dm_table_get_size(map))
475 /* We only support devices that have a single target */
476 if (dm_table_get_num_targets(map) != 1)
479 tgt = dm_table_get_target(map, 0);
480 if (!tgt->type->prepare_ioctl)
483 if (dm_suspended_md(md))
486 r = tgt->type->prepare_ioctl(tgt, bdev);
487 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
488 dm_put_live_table(md, *srcu_idx);
496 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
497 __releases(md->io_barrier)
499 dm_put_live_table(md, srcu_idx);
502 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
503 unsigned int cmd, unsigned long arg)
505 struct mapped_device *md = bdev->bd_disk->private_data;
508 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
514 * Target determined this ioctl is being issued against a
515 * subset of the parent bdev; require extra privileges.
517 if (!capable(CAP_SYS_RAWIO)) {
519 "%s: sending ioctl %x to DM device without required privilege.",
526 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
528 dm_unprepare_ioctl(md, srcu_idx);
532 static void start_io_acct(struct dm_io *io);
534 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
537 struct dm_target_io *tio;
540 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
544 tio = container_of(clone, struct dm_target_io, clone);
545 tio->inside_dm_io = true;
548 io = container_of(tio, struct dm_io, tio);
549 io->magic = DM_IO_MAGIC;
551 atomic_set(&io->io_count, 1);
554 spin_lock_init(&io->endio_lock);
561 static void free_io(struct mapped_device *md, struct dm_io *io)
563 bio_put(&io->tio.clone);
566 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
567 unsigned target_bio_nr, gfp_t gfp_mask)
569 struct dm_target_io *tio;
571 if (!ci->io->tio.io) {
572 /* the dm_target_io embedded in ci->io is available */
575 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
579 tio = container_of(clone, struct dm_target_io, clone);
580 tio->inside_dm_io = false;
583 tio->magic = DM_TIO_MAGIC;
586 tio->target_bio_nr = target_bio_nr;
591 static void free_tio(struct dm_target_io *tio)
593 if (tio->inside_dm_io)
595 bio_put(&tio->clone);
598 int md_in_flight(struct mapped_device *md)
600 return atomic_read(&md->pending[READ]) +
601 atomic_read(&md->pending[WRITE]);
604 static void start_io_acct(struct dm_io *io)
606 struct mapped_device *md = io->md;
607 struct bio *bio = io->orig_bio;
608 int rw = bio_data_dir(bio);
610 io->start_time = jiffies;
612 generic_start_io_acct(md->queue, rw, bio_sectors(bio), &dm_disk(md)->part0);
614 atomic_set(&dm_disk(md)->part0.in_flight[rw],
615 atomic_inc_return(&md->pending[rw]));
617 if (unlikely(dm_stats_used(&md->stats)))
618 dm_stats_account_io(&md->stats, bio_data_dir(bio),
619 bio->bi_iter.bi_sector, bio_sectors(bio),
620 false, 0, &io->stats_aux);
623 static void end_io_acct(struct dm_io *io)
625 struct mapped_device *md = io->md;
626 struct bio *bio = io->orig_bio;
627 unsigned long duration = jiffies - io->start_time;
629 int rw = bio_data_dir(bio);
631 generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time);
633 if (unlikely(dm_stats_used(&md->stats)))
634 dm_stats_account_io(&md->stats, bio_data_dir(bio),
635 bio->bi_iter.bi_sector, bio_sectors(bio),
636 true, duration, &io->stats_aux);
639 * After this is decremented the bio must not be touched if it is
642 pending = atomic_dec_return(&md->pending[rw]);
643 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
644 pending += atomic_read(&md->pending[rw^0x1]);
646 /* nudge anyone waiting on suspend queue */
652 * Add the bio to the list of deferred io.
654 static void queue_io(struct mapped_device *md, struct bio *bio)
658 spin_lock_irqsave(&md->deferred_lock, flags);
659 bio_list_add(&md->deferred, bio);
660 spin_unlock_irqrestore(&md->deferred_lock, flags);
661 queue_work(md->wq, &md->work);
665 * Everyone (including functions in this file), should use this
666 * function to access the md->map field, and make sure they call
667 * dm_put_live_table() when finished.
669 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
671 *srcu_idx = srcu_read_lock(&md->io_barrier);
673 return srcu_dereference(md->map, &md->io_barrier);
676 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
678 srcu_read_unlock(&md->io_barrier, srcu_idx);
681 void dm_sync_table(struct mapped_device *md)
683 synchronize_srcu(&md->io_barrier);
684 synchronize_rcu_expedited();
688 * A fast alternative to dm_get_live_table/dm_put_live_table.
689 * The caller must not block between these two functions.
691 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
694 return rcu_dereference(md->map);
697 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
702 static char *_dm_claim_ptr = "I belong to device-mapper";
705 * Open a table device so we can use it as a map destination.
707 static int open_table_device(struct table_device *td, dev_t dev,
708 struct mapped_device *md)
710 struct block_device *bdev;
714 BUG_ON(td->dm_dev.bdev);
716 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
718 return PTR_ERR(bdev);
720 r = bd_link_disk_holder(bdev, dm_disk(md));
722 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
726 td->dm_dev.bdev = bdev;
727 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
732 * Close a table device that we've been using.
734 static void close_table_device(struct table_device *td, struct mapped_device *md)
736 if (!td->dm_dev.bdev)
739 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
740 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
741 put_dax(td->dm_dev.dax_dev);
742 td->dm_dev.bdev = NULL;
743 td->dm_dev.dax_dev = NULL;
746 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
748 struct table_device *td;
750 list_for_each_entry(td, l, list)
751 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
757 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
758 struct dm_dev **result) {
760 struct table_device *td;
762 mutex_lock(&md->table_devices_lock);
763 td = find_table_device(&md->table_devices, dev, mode);
765 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
767 mutex_unlock(&md->table_devices_lock);
771 td->dm_dev.mode = mode;
772 td->dm_dev.bdev = NULL;
774 if ((r = open_table_device(td, dev, md))) {
775 mutex_unlock(&md->table_devices_lock);
780 format_dev_t(td->dm_dev.name, dev);
782 refcount_set(&td->count, 1);
783 list_add(&td->list, &md->table_devices);
785 refcount_inc(&td->count);
787 mutex_unlock(&md->table_devices_lock);
789 *result = &td->dm_dev;
792 EXPORT_SYMBOL_GPL(dm_get_table_device);
794 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
796 struct table_device *td = container_of(d, struct table_device, dm_dev);
798 mutex_lock(&md->table_devices_lock);
799 if (refcount_dec_and_test(&td->count)) {
800 close_table_device(td, md);
804 mutex_unlock(&md->table_devices_lock);
806 EXPORT_SYMBOL(dm_put_table_device);
808 static void free_table_devices(struct list_head *devices)
810 struct list_head *tmp, *next;
812 list_for_each_safe(tmp, next, devices) {
813 struct table_device *td = list_entry(tmp, struct table_device, list);
815 DMWARN("dm_destroy: %s still exists with %d references",
816 td->dm_dev.name, refcount_read(&td->count));
822 * Get the geometry associated with a dm device
824 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
832 * Set the geometry of a device.
834 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
836 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
838 if (geo->start > sz) {
839 DMWARN("Start sector is beyond the geometry limits.");
848 static int __noflush_suspending(struct mapped_device *md)
850 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
854 * Decrements the number of outstanding ios that a bio has been
855 * cloned into, completing the original io if necc.
857 static void dec_pending(struct dm_io *io, blk_status_t error)
860 blk_status_t io_error;
862 struct mapped_device *md = io->md;
864 /* Push-back supersedes any I/O errors */
865 if (unlikely(error)) {
866 spin_lock_irqsave(&io->endio_lock, flags);
867 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
869 spin_unlock_irqrestore(&io->endio_lock, flags);
872 if (atomic_dec_and_test(&io->io_count)) {
873 if (io->status == BLK_STS_DM_REQUEUE) {
875 * Target requested pushing back the I/O.
877 spin_lock_irqsave(&md->deferred_lock, flags);
878 if (__noflush_suspending(md))
879 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
880 bio_list_add_head(&md->deferred, io->orig_bio);
882 /* noflush suspend was interrupted. */
883 io->status = BLK_STS_IOERR;
884 spin_unlock_irqrestore(&md->deferred_lock, flags);
887 io_error = io->status;
892 if (io_error == BLK_STS_DM_REQUEUE)
895 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
897 * Preflush done for flush with data, reissue
898 * without REQ_PREFLUSH.
900 bio->bi_opf &= ~REQ_PREFLUSH;
903 /* done with normal IO or empty flush */
905 bio->bi_status = io_error;
911 void disable_write_same(struct mapped_device *md)
913 struct queue_limits *limits = dm_get_queue_limits(md);
915 /* device doesn't really support WRITE SAME, disable it */
916 limits->max_write_same_sectors = 0;
919 void disable_write_zeroes(struct mapped_device *md)
921 struct queue_limits *limits = dm_get_queue_limits(md);
923 /* device doesn't really support WRITE ZEROES, disable it */
924 limits->max_write_zeroes_sectors = 0;
927 static void clone_endio(struct bio *bio)
929 blk_status_t error = bio->bi_status;
930 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
931 struct dm_io *io = tio->io;
932 struct mapped_device *md = tio->io->md;
933 dm_endio_fn endio = tio->ti->type->end_io;
935 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
936 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
937 !bio->bi_disk->queue->limits.max_write_same_sectors)
938 disable_write_same(md);
939 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
940 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
941 disable_write_zeroes(md);
945 int r = endio(tio->ti, bio, &error);
947 case DM_ENDIO_REQUEUE:
948 error = BLK_STS_DM_REQUEUE;
952 case DM_ENDIO_INCOMPLETE:
953 /* The target will handle the io */
956 DMWARN("unimplemented target endio return value: %d", r);
962 dec_pending(io, error);
966 * Return maximum size of I/O possible at the supplied sector up to the current
969 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
971 sector_t target_offset = dm_target_offset(ti, sector);
973 return ti->len - target_offset;
976 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
978 sector_t len = max_io_len_target_boundary(sector, ti);
979 sector_t offset, max_len;
982 * Does the target need to split even further?
984 if (ti->max_io_len) {
985 offset = dm_target_offset(ti, sector);
986 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
987 max_len = sector_div(offset, ti->max_io_len);
989 max_len = offset & (ti->max_io_len - 1);
990 max_len = ti->max_io_len - max_len;
999 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1001 if (len > UINT_MAX) {
1002 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1003 (unsigned long long)len, UINT_MAX);
1004 ti->error = "Maximum size of target IO is too large";
1009 * BIO based queue uses its own splitting. When multipage bvecs
1010 * is switched on, size of the incoming bio may be too big to
1011 * be handled in some targets, such as crypt.
1013 * When these targets are ready for the big bio, we can remove
1016 ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE);
1020 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1022 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1023 sector_t sector, int *srcu_idx)
1024 __acquires(md->io_barrier)
1026 struct dm_table *map;
1027 struct dm_target *ti;
1029 map = dm_get_live_table(md, srcu_idx);
1033 ti = dm_table_find_target(map, sector);
1034 if (!dm_target_is_valid(ti))
1040 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1041 long nr_pages, void **kaddr, pfn_t *pfn)
1043 struct mapped_device *md = dax_get_private(dax_dev);
1044 sector_t sector = pgoff * PAGE_SECTORS;
1045 struct dm_target *ti;
1046 long len, ret = -EIO;
1049 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1053 if (!ti->type->direct_access)
1055 len = max_io_len(sector, ti) / PAGE_SECTORS;
1058 nr_pages = min(len, nr_pages);
1059 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1062 dm_put_live_table(md, srcu_idx);
1067 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1068 void *addr, size_t bytes, struct iov_iter *i)
1070 struct mapped_device *md = dax_get_private(dax_dev);
1071 sector_t sector = pgoff * PAGE_SECTORS;
1072 struct dm_target *ti;
1076 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1080 if (!ti->type->dax_copy_from_iter) {
1081 ret = copy_from_iter(addr, bytes, i);
1084 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1086 dm_put_live_table(md, srcu_idx);
1091 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1092 void *addr, size_t bytes, struct iov_iter *i)
1094 struct mapped_device *md = dax_get_private(dax_dev);
1095 sector_t sector = pgoff * PAGE_SECTORS;
1096 struct dm_target *ti;
1100 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1104 if (!ti->type->dax_copy_to_iter) {
1105 ret = copy_to_iter(addr, bytes, i);
1108 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1110 dm_put_live_table(md, srcu_idx);
1116 * A target may call dm_accept_partial_bio only from the map routine. It is
1117 * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1119 * dm_accept_partial_bio informs the dm that the target only wants to process
1120 * additional n_sectors sectors of the bio and the rest of the data should be
1121 * sent in a next bio.
1123 * A diagram that explains the arithmetics:
1124 * +--------------------+---------------+-------+
1126 * +--------------------+---------------+-------+
1128 * <-------------- *tio->len_ptr --------------->
1129 * <------- bi_size ------->
1132 * Region 1 was already iterated over with bio_advance or similar function.
1133 * (it may be empty if the target doesn't use bio_advance)
1134 * Region 2 is the remaining bio size that the target wants to process.
1135 * (it may be empty if region 1 is non-empty, although there is no reason
1137 * The target requires that region 3 is to be sent in the next bio.
1139 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1140 * the partially processed part (the sum of regions 1+2) must be the same for all
1141 * copies of the bio.
1143 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1145 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1146 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1147 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1148 BUG_ON(bi_size > *tio->len_ptr);
1149 BUG_ON(n_sectors > bi_size);
1150 *tio->len_ptr -= bi_size - n_sectors;
1151 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1153 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1156 * The zone descriptors obtained with a zone report indicate
1157 * zone positions within the target device. The zone descriptors
1158 * must be remapped to match their position within the dm device.
1159 * A target may call dm_remap_zone_report after completion of a
1160 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained
1161 * from the target device mapping to the dm device.
1163 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1165 #ifdef CONFIG_BLK_DEV_ZONED
1166 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1167 struct bio *report_bio = tio->io->orig_bio;
1168 struct blk_zone_report_hdr *hdr = NULL;
1169 struct blk_zone *zone;
1170 unsigned int nr_rep = 0;
1172 struct bio_vec bvec;
1173 struct bvec_iter iter;
1180 * Remap the start sector of the reported zones. For sequential zones,
1181 * also remap the write pointer position.
1183 bio_for_each_segment(bvec, report_bio, iter) {
1184 addr = kmap_atomic(bvec.bv_page);
1186 /* Remember the report header in the first page */
1189 ofst = sizeof(struct blk_zone_report_hdr);
1193 /* Set zones start sector */
1194 while (hdr->nr_zones && ofst < bvec.bv_len) {
1196 if (zone->start >= start + ti->len) {
1200 zone->start = zone->start + ti->begin - start;
1201 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1202 if (zone->cond == BLK_ZONE_COND_FULL)
1203 zone->wp = zone->start + zone->len;
1204 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1205 zone->wp = zone->start;
1207 zone->wp = zone->wp + ti->begin - start;
1209 ofst += sizeof(struct blk_zone);
1215 kunmap_atomic(addr);
1222 hdr->nr_zones = nr_rep;
1226 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1228 #else /* !CONFIG_BLK_DEV_ZONED */
1229 bio->bi_status = BLK_STS_NOTSUPP;
1232 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1234 static blk_qc_t __map_bio(struct dm_target_io *tio)
1238 struct bio *clone = &tio->clone;
1239 struct dm_io *io = tio->io;
1240 struct mapped_device *md = io->md;
1241 struct dm_target *ti = tio->ti;
1242 blk_qc_t ret = BLK_QC_T_NONE;
1244 clone->bi_end_io = clone_endio;
1247 * Map the clone. If r == 0 we don't need to do
1248 * anything, the target has assumed ownership of
1251 atomic_inc(&io->io_count);
1252 sector = clone->bi_iter.bi_sector;
1254 r = ti->type->map(ti, clone);
1256 case DM_MAPIO_SUBMITTED:
1258 case DM_MAPIO_REMAPPED:
1259 /* the bio has been remapped so dispatch it */
1260 trace_block_bio_remap(clone->bi_disk->queue, clone,
1261 bio_dev(io->orig_bio), sector);
1262 if (md->type == DM_TYPE_NVME_BIO_BASED)
1263 ret = direct_make_request(clone);
1265 ret = generic_make_request(clone);
1269 dec_pending(io, BLK_STS_IOERR);
1271 case DM_MAPIO_REQUEUE:
1273 dec_pending(io, BLK_STS_DM_REQUEUE);
1276 DMWARN("unimplemented target map return value: %d", r);
1283 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1285 bio->bi_iter.bi_sector = sector;
1286 bio->bi_iter.bi_size = to_bytes(len);
1290 * Creates a bio that consists of range of complete bvecs.
1292 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1293 sector_t sector, unsigned len)
1295 struct bio *clone = &tio->clone;
1297 __bio_clone_fast(clone, bio);
1299 if (unlikely(bio_integrity(bio) != NULL)) {
1302 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1303 !dm_target_passes_integrity(tio->ti->type))) {
1304 DMWARN("%s: the target %s doesn't support integrity data.",
1305 dm_device_name(tio->io->md),
1306 tio->ti->type->name);
1310 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1315 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1316 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1317 clone->bi_iter.bi_size = to_bytes(len);
1319 if (unlikely(bio_integrity(bio) != NULL))
1320 bio_integrity_trim(clone);
1325 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1326 struct dm_target *ti, unsigned num_bios)
1328 struct dm_target_io *tio;
1334 if (num_bios == 1) {
1335 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1336 bio_list_add(blist, &tio->clone);
1340 for (try = 0; try < 2; try++) {
1345 mutex_lock(&ci->io->md->table_devices_lock);
1346 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1347 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1351 bio_list_add(blist, &tio->clone);
1354 mutex_unlock(&ci->io->md->table_devices_lock);
1355 if (bio_nr == num_bios)
1358 while ((bio = bio_list_pop(blist))) {
1359 tio = container_of(bio, struct dm_target_io, clone);
1365 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1366 struct dm_target_io *tio, unsigned *len)
1368 struct bio *clone = &tio->clone;
1372 __bio_clone_fast(clone, ci->bio);
1374 bio_setup_sector(clone, ci->sector, *len);
1376 return __map_bio(tio);
1379 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1380 unsigned num_bios, unsigned *len)
1382 struct bio_list blist = BIO_EMPTY_LIST;
1384 struct dm_target_io *tio;
1386 alloc_multiple_bios(&blist, ci, ti, num_bios);
1388 while ((bio = bio_list_pop(&blist))) {
1389 tio = container_of(bio, struct dm_target_io, clone);
1390 (void) __clone_and_map_simple_bio(ci, tio, len);
1394 static int __send_empty_flush(struct clone_info *ci)
1396 unsigned target_nr = 0;
1397 struct dm_target *ti;
1399 BUG_ON(bio_has_data(ci->bio));
1400 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1401 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1406 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1407 sector_t sector, unsigned *len)
1409 struct bio *bio = ci->bio;
1410 struct dm_target_io *tio;
1413 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1415 r = clone_bio(tio, bio, sector, *len);
1420 (void) __map_bio(tio);
1425 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1427 static unsigned get_num_discard_bios(struct dm_target *ti)
1429 return ti->num_discard_bios;
1432 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1434 return ti->num_secure_erase_bios;
1437 static unsigned get_num_write_same_bios(struct dm_target *ti)
1439 return ti->num_write_same_bios;
1442 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1444 return ti->num_write_zeroes_bios;
1447 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1449 static bool is_split_required_for_discard(struct dm_target *ti)
1451 return ti->split_discard_bios;
1454 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1455 get_num_bios_fn get_num_bios,
1456 is_split_required_fn is_split_required)
1462 * Even though the device advertised support for this type of
1463 * request, that does not mean every target supports it, and
1464 * reconfiguration might also have changed that since the
1465 * check was performed.
1467 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1471 if (is_split_required && !is_split_required(ti))
1472 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1474 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1476 __send_duplicate_bios(ci, ti, num_bios, &len);
1479 ci->sector_count -= len;
1484 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1486 return __send_changing_extent_only(ci, ti, get_num_discard_bios,
1487 is_split_required_for_discard);
1490 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1492 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL);
1495 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1497 return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL);
1500 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1502 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL);
1505 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1508 struct bio *bio = ci->bio;
1510 if (bio_op(bio) == REQ_OP_DISCARD)
1511 *result = __send_discard(ci, ti);
1512 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1513 *result = __send_secure_erase(ci, ti);
1514 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1515 *result = __send_write_same(ci, ti);
1516 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1517 *result = __send_write_zeroes(ci, ti);
1525 * Select the correct strategy for processing a non-flush bio.
1527 static int __split_and_process_non_flush(struct clone_info *ci)
1529 struct bio *bio = ci->bio;
1530 struct dm_target *ti;
1534 ti = dm_table_find_target(ci->map, ci->sector);
1535 if (!dm_target_is_valid(ti))
1538 if (unlikely(__process_abnormal_io(ci, ti, &r)))
1541 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1542 len = ci->sector_count;
1544 len = min_t(sector_t, max_io_len(ci->sector, ti),
1547 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1552 ci->sector_count -= len;
1557 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1558 struct dm_table *map, struct bio *bio)
1561 ci->io = alloc_io(md, bio);
1562 ci->sector = bio->bi_iter.bi_sector;
1566 * Entry point to split a bio into clones and submit them to the targets.
1568 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1569 struct dm_table *map, struct bio *bio)
1571 struct clone_info ci;
1572 blk_qc_t ret = BLK_QC_T_NONE;
1575 if (unlikely(!map)) {
1580 init_clone_info(&ci, md, map, bio);
1582 if (bio->bi_opf & REQ_PREFLUSH) {
1583 ci.bio = &ci.io->md->flush_bio;
1584 ci.sector_count = 0;
1585 error = __send_empty_flush(&ci);
1586 /* dec_pending submits any data associated with flush */
1587 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1589 ci.sector_count = 0;
1590 error = __split_and_process_non_flush(&ci);
1593 ci.sector_count = bio_sectors(bio);
1594 while (ci.sector_count && !error) {
1595 error = __split_and_process_non_flush(&ci);
1596 if (current->bio_list && ci.sector_count && !error) {
1598 * Remainder must be passed to generic_make_request()
1599 * so that it gets handled *after* bios already submitted
1600 * have been completely processed.
1601 * We take a clone of the original to store in
1602 * ci.io->orig_bio to be used by end_io_acct() and
1603 * for dec_pending to use for completion handling.
1604 * As this path is not used for REQ_OP_ZONE_REPORT,
1605 * the usage of io->orig_bio in dm_remap_zone_report()
1606 * won't be affected by this reassignment.
1608 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1609 GFP_NOIO, &md->queue->bio_split);
1610 ci.io->orig_bio = b;
1612 ret = generic_make_request(bio);
1618 /* drop the extra reference count */
1619 dec_pending(ci.io, errno_to_blk_status(error));
1624 * Optimized variant of __split_and_process_bio that leverages the
1625 * fact that targets that use it do _not_ have a need to split bios.
1627 static blk_qc_t __process_bio(struct mapped_device *md,
1628 struct dm_table *map, struct bio *bio)
1630 struct clone_info ci;
1631 blk_qc_t ret = BLK_QC_T_NONE;
1634 if (unlikely(!map)) {
1639 init_clone_info(&ci, md, map, bio);
1641 if (bio->bi_opf & REQ_PREFLUSH) {
1642 ci.bio = &ci.io->md->flush_bio;
1643 ci.sector_count = 0;
1644 error = __send_empty_flush(&ci);
1645 /* dec_pending submits any data associated with flush */
1647 struct dm_target *ti = md->immutable_target;
1648 struct dm_target_io *tio;
1651 * Defend against IO still getting in during teardown
1652 * - as was seen for a time with nvme-fcloop
1654 if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) {
1660 ci.sector_count = bio_sectors(bio);
1661 if (unlikely(__process_abnormal_io(&ci, ti, &error)))
1664 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1665 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1668 /* drop the extra reference count */
1669 dec_pending(ci.io, errno_to_blk_status(error));
1673 typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *);
1675 static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio,
1676 process_bio_fn process_bio)
1678 struct mapped_device *md = q->queuedata;
1679 blk_qc_t ret = BLK_QC_T_NONE;
1681 struct dm_table *map;
1683 map = dm_get_live_table(md, &srcu_idx);
1685 /* if we're suspended, we have to queue this io for later */
1686 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1687 dm_put_live_table(md, srcu_idx);
1689 if (!(bio->bi_opf & REQ_RAHEAD))
1696 ret = process_bio(md, map, bio);
1698 dm_put_live_table(md, srcu_idx);
1703 * The request function that remaps the bio to one target and
1704 * splits off any remainder.
1706 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1708 return __dm_make_request(q, bio, __split_and_process_bio);
1711 static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio)
1713 return __dm_make_request(q, bio, __process_bio);
1716 static int dm_any_congested(void *congested_data, int bdi_bits)
1719 struct mapped_device *md = congested_data;
1720 struct dm_table *map;
1722 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1723 if (dm_request_based(md)) {
1725 * With request-based DM we only need to check the
1726 * top-level queue for congestion.
1728 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1730 map = dm_get_live_table_fast(md);
1732 r = dm_table_any_congested(map, bdi_bits);
1733 dm_put_live_table_fast(md);
1740 /*-----------------------------------------------------------------
1741 * An IDR is used to keep track of allocated minor numbers.
1742 *---------------------------------------------------------------*/
1743 static void free_minor(int minor)
1745 spin_lock(&_minor_lock);
1746 idr_remove(&_minor_idr, minor);
1747 spin_unlock(&_minor_lock);
1751 * See if the device with a specific minor # is free.
1753 static int specific_minor(int minor)
1757 if (minor >= (1 << MINORBITS))
1760 idr_preload(GFP_KERNEL);
1761 spin_lock(&_minor_lock);
1763 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1765 spin_unlock(&_minor_lock);
1768 return r == -ENOSPC ? -EBUSY : r;
1772 static int next_free_minor(int *minor)
1776 idr_preload(GFP_KERNEL);
1777 spin_lock(&_minor_lock);
1779 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1781 spin_unlock(&_minor_lock);
1789 static const struct block_device_operations dm_blk_dops;
1790 static const struct dax_operations dm_dax_ops;
1792 static void dm_wq_work(struct work_struct *work);
1794 static void dm_init_normal_md_queue(struct mapped_device *md)
1796 md->use_blk_mq = false;
1799 * Initialize aspects of queue that aren't relevant for blk-mq
1801 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1804 static void cleanup_mapped_device(struct mapped_device *md)
1807 destroy_workqueue(md->wq);
1808 if (md->kworker_task)
1809 kthread_stop(md->kworker_task);
1810 bioset_exit(&md->bs);
1811 bioset_exit(&md->io_bs);
1814 kill_dax(md->dax_dev);
1815 put_dax(md->dax_dev);
1820 spin_lock(&_minor_lock);
1821 md->disk->private_data = NULL;
1822 spin_unlock(&_minor_lock);
1823 del_gendisk(md->disk);
1828 blk_cleanup_queue(md->queue);
1830 cleanup_srcu_struct(&md->io_barrier);
1837 mutex_destroy(&md->suspend_lock);
1838 mutex_destroy(&md->type_lock);
1839 mutex_destroy(&md->table_devices_lock);
1841 dm_mq_cleanup_mapped_device(md);
1845 * Allocate and initialise a blank device with a given minor.
1847 static struct mapped_device *alloc_dev(int minor)
1849 int r, numa_node_id = dm_get_numa_node();
1850 struct dax_device *dax_dev = NULL;
1851 struct mapped_device *md;
1854 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1856 DMWARN("unable to allocate device, out of memory.");
1860 if (!try_module_get(THIS_MODULE))
1861 goto bad_module_get;
1863 /* get a minor number for the dev */
1864 if (minor == DM_ANY_MINOR)
1865 r = next_free_minor(&minor);
1867 r = specific_minor(minor);
1871 r = init_srcu_struct(&md->io_barrier);
1873 goto bad_io_barrier;
1875 md->numa_node_id = numa_node_id;
1876 md->use_blk_mq = dm_use_blk_mq_default();
1877 md->init_tio_pdu = false;
1878 md->type = DM_TYPE_NONE;
1879 mutex_init(&md->suspend_lock);
1880 mutex_init(&md->type_lock);
1881 mutex_init(&md->table_devices_lock);
1882 spin_lock_init(&md->deferred_lock);
1883 atomic_set(&md->holders, 1);
1884 atomic_set(&md->open_count, 0);
1885 atomic_set(&md->event_nr, 0);
1886 atomic_set(&md->uevent_seq, 0);
1887 INIT_LIST_HEAD(&md->uevent_list);
1888 INIT_LIST_HEAD(&md->table_devices);
1889 spin_lock_init(&md->uevent_lock);
1891 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL);
1894 md->queue->queuedata = md;
1895 md->queue->backing_dev_info->congested_data = md;
1897 md->disk = alloc_disk_node(1, md->numa_node_id);
1901 atomic_set(&md->pending[0], 0);
1902 atomic_set(&md->pending[1], 0);
1903 init_waitqueue_head(&md->wait);
1904 INIT_WORK(&md->work, dm_wq_work);
1905 init_waitqueue_head(&md->eventq);
1906 init_completion(&md->kobj_holder.completion);
1907 md->kworker_task = NULL;
1909 md->disk->major = _major;
1910 md->disk->first_minor = minor;
1911 md->disk->fops = &dm_blk_dops;
1912 md->disk->queue = md->queue;
1913 md->disk->private_data = md;
1914 sprintf(md->disk->disk_name, "dm-%d", minor);
1916 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1917 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1921 md->dax_dev = dax_dev;
1923 add_disk_no_queue_reg(md->disk);
1924 format_dev_t(md->name, MKDEV(_major, minor));
1926 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1930 md->bdev = bdget_disk(md->disk, 0);
1934 bio_init(&md->flush_bio, NULL, 0);
1935 bio_set_dev(&md->flush_bio, md->bdev);
1936 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1938 dm_stats_init(&md->stats);
1940 /* Populate the mapping, nobody knows we exist yet */
1941 spin_lock(&_minor_lock);
1942 old_md = idr_replace(&_minor_idr, md, minor);
1943 spin_unlock(&_minor_lock);
1945 BUG_ON(old_md != MINOR_ALLOCED);
1950 cleanup_mapped_device(md);
1954 module_put(THIS_MODULE);
1960 static void unlock_fs(struct mapped_device *md);
1962 static void free_dev(struct mapped_device *md)
1964 int minor = MINOR(disk_devt(md->disk));
1968 cleanup_mapped_device(md);
1970 free_table_devices(&md->table_devices);
1971 dm_stats_cleanup(&md->stats);
1974 module_put(THIS_MODULE);
1978 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1980 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1983 if (dm_table_bio_based(t)) {
1985 * The md may already have mempools that need changing.
1986 * If so, reload bioset because front_pad may have changed
1987 * because a different table was loaded.
1989 bioset_exit(&md->bs);
1990 bioset_exit(&md->io_bs);
1992 } else if (bioset_initialized(&md->bs)) {
1994 * There's no need to reload with request-based dm
1995 * because the size of front_pad doesn't change.
1996 * Note for future: If you are to reload bioset,
1997 * prep-ed requests in the queue may refer
1998 * to bio from the old bioset, so you must walk
1999 * through the queue to unprep.
2005 bioset_initialized(&md->bs) ||
2006 bioset_initialized(&md->io_bs));
2008 ret = bioset_init_from_src(&md->bs, &p->bs);
2011 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2013 bioset_exit(&md->bs);
2015 /* mempool bind completed, no longer need any mempools in the table */
2016 dm_table_free_md_mempools(t);
2021 * Bind a table to the device.
2023 static void event_callback(void *context)
2025 unsigned long flags;
2027 struct mapped_device *md = (struct mapped_device *) context;
2029 spin_lock_irqsave(&md->uevent_lock, flags);
2030 list_splice_init(&md->uevent_list, &uevents);
2031 spin_unlock_irqrestore(&md->uevent_lock, flags);
2033 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2035 atomic_inc(&md->event_nr);
2036 wake_up(&md->eventq);
2037 dm_issue_global_event();
2041 * Protected by md->suspend_lock obtained by dm_swap_table().
2043 static void __set_size(struct mapped_device *md, sector_t size)
2045 lockdep_assert_held(&md->suspend_lock);
2047 set_capacity(md->disk, size);
2049 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2053 * Returns old map, which caller must destroy.
2055 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2056 struct queue_limits *limits)
2058 struct dm_table *old_map;
2059 struct request_queue *q = md->queue;
2060 bool request_based = dm_table_request_based(t);
2064 lockdep_assert_held(&md->suspend_lock);
2066 size = dm_table_get_size(t);
2069 * Wipe any geometry if the size of the table changed.
2071 if (size != dm_get_size(md))
2072 memset(&md->geometry, 0, sizeof(md->geometry));
2074 __set_size(md, size);
2076 dm_table_event_callback(t, event_callback, md);
2079 * The queue hasn't been stopped yet, if the old table type wasn't
2080 * for request-based during suspension. So stop it to prevent
2081 * I/O mapping before resume.
2082 * This must be done before setting the queue restrictions,
2083 * because request-based dm may be run just after the setting.
2088 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2090 * Leverage the fact that request-based DM targets and
2091 * NVMe bio based targets are immutable singletons
2092 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2093 * and __process_bio.
2095 md->immutable_target = dm_table_get_immutable_target(t);
2098 ret = __bind_mempools(md, t);
2100 old_map = ERR_PTR(ret);
2104 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2105 rcu_assign_pointer(md->map, (void *)t);
2106 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2108 dm_table_set_restrictions(t, q, limits);
2117 * Returns unbound table for the caller to free.
2119 static struct dm_table *__unbind(struct mapped_device *md)
2121 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2126 dm_table_event_callback(map, NULL, NULL);
2127 RCU_INIT_POINTER(md->map, NULL);
2134 * Constructor for a new device.
2136 int dm_create(int minor, struct mapped_device **result)
2139 struct mapped_device *md;
2141 md = alloc_dev(minor);
2145 r = dm_sysfs_init(md);
2156 * Functions to manage md->type.
2157 * All are required to hold md->type_lock.
2159 void dm_lock_md_type(struct mapped_device *md)
2161 mutex_lock(&md->type_lock);
2164 void dm_unlock_md_type(struct mapped_device *md)
2166 mutex_unlock(&md->type_lock);
2169 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2171 BUG_ON(!mutex_is_locked(&md->type_lock));
2175 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2180 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2182 return md->immutable_target_type;
2186 * The queue_limits are only valid as long as you have a reference
2189 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2191 BUG_ON(!atomic_read(&md->holders));
2192 return &md->queue->limits;
2194 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2197 * Setup the DM device's queue based on md's type
2199 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2202 struct queue_limits limits;
2203 enum dm_queue_mode type = dm_get_md_type(md);
2206 case DM_TYPE_REQUEST_BASED:
2207 dm_init_normal_md_queue(md);
2208 r = dm_old_init_request_queue(md, t);
2210 DMERR("Cannot initialize queue for request-based mapped device");
2214 case DM_TYPE_MQ_REQUEST_BASED:
2215 r = dm_mq_init_request_queue(md, t);
2217 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2221 case DM_TYPE_BIO_BASED:
2222 case DM_TYPE_DAX_BIO_BASED:
2223 dm_init_normal_md_queue(md);
2224 blk_queue_make_request(md->queue, dm_make_request);
2226 case DM_TYPE_NVME_BIO_BASED:
2227 dm_init_normal_md_queue(md);
2228 blk_queue_make_request(md->queue, dm_make_request_nvme);
2235 r = dm_calculate_queue_limits(t, &limits);
2237 DMERR("Cannot calculate initial queue limits");
2240 dm_table_set_restrictions(t, md->queue, &limits);
2241 blk_register_queue(md->disk);
2246 struct mapped_device *dm_get_md(dev_t dev)
2248 struct mapped_device *md;
2249 unsigned minor = MINOR(dev);
2251 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2254 spin_lock(&_minor_lock);
2256 md = idr_find(&_minor_idr, minor);
2257 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2258 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2264 spin_unlock(&_minor_lock);
2268 EXPORT_SYMBOL_GPL(dm_get_md);
2270 void *dm_get_mdptr(struct mapped_device *md)
2272 return md->interface_ptr;
2275 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2277 md->interface_ptr = ptr;
2280 void dm_get(struct mapped_device *md)
2282 atomic_inc(&md->holders);
2283 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2286 int dm_hold(struct mapped_device *md)
2288 spin_lock(&_minor_lock);
2289 if (test_bit(DMF_FREEING, &md->flags)) {
2290 spin_unlock(&_minor_lock);
2294 spin_unlock(&_minor_lock);
2297 EXPORT_SYMBOL_GPL(dm_hold);
2299 const char *dm_device_name(struct mapped_device *md)
2303 EXPORT_SYMBOL_GPL(dm_device_name);
2305 static void __dm_destroy(struct mapped_device *md, bool wait)
2307 struct dm_table *map;
2312 spin_lock(&_minor_lock);
2313 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2314 set_bit(DMF_FREEING, &md->flags);
2315 spin_unlock(&_minor_lock);
2317 blk_set_queue_dying(md->queue);
2319 if (dm_request_based(md) && md->kworker_task)
2320 kthread_flush_worker(&md->kworker);
2323 * Take suspend_lock so that presuspend and postsuspend methods
2324 * do not race with internal suspend.
2326 mutex_lock(&md->suspend_lock);
2327 map = dm_get_live_table(md, &srcu_idx);
2328 if (!dm_suspended_md(md)) {
2329 dm_table_presuspend_targets(map);
2330 dm_table_postsuspend_targets(map);
2332 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2333 dm_put_live_table(md, srcu_idx);
2334 mutex_unlock(&md->suspend_lock);
2337 * Rare, but there may be I/O requests still going to complete,
2338 * for example. Wait for all references to disappear.
2339 * No one should increment the reference count of the mapped_device,
2340 * after the mapped_device state becomes DMF_FREEING.
2343 while (atomic_read(&md->holders))
2345 else if (atomic_read(&md->holders))
2346 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2347 dm_device_name(md), atomic_read(&md->holders));
2350 dm_table_destroy(__unbind(md));
2354 void dm_destroy(struct mapped_device *md)
2356 __dm_destroy(md, true);
2359 void dm_destroy_immediate(struct mapped_device *md)
2361 __dm_destroy(md, false);
2364 void dm_put(struct mapped_device *md)
2366 atomic_dec(&md->holders);
2368 EXPORT_SYMBOL_GPL(dm_put);
2370 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2376 prepare_to_wait(&md->wait, &wait, task_state);
2378 if (!md_in_flight(md))
2381 if (signal_pending_state(task_state, current)) {
2388 finish_wait(&md->wait, &wait);
2394 * Process the deferred bios
2396 static void dm_wq_work(struct work_struct *work)
2398 struct mapped_device *md = container_of(work, struct mapped_device,
2402 struct dm_table *map;
2404 map = dm_get_live_table(md, &srcu_idx);
2406 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2407 spin_lock_irq(&md->deferred_lock);
2408 c = bio_list_pop(&md->deferred);
2409 spin_unlock_irq(&md->deferred_lock);
2414 if (dm_request_based(md))
2415 generic_make_request(c);
2417 __split_and_process_bio(md, map, c);
2420 dm_put_live_table(md, srcu_idx);
2423 static void dm_queue_flush(struct mapped_device *md)
2425 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2426 smp_mb__after_atomic();
2427 queue_work(md->wq, &md->work);
2431 * Swap in a new table, returning the old one for the caller to destroy.
2433 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2435 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2436 struct queue_limits limits;
2439 mutex_lock(&md->suspend_lock);
2441 /* device must be suspended */
2442 if (!dm_suspended_md(md))
2446 * If the new table has no data devices, retain the existing limits.
2447 * This helps multipath with queue_if_no_path if all paths disappear,
2448 * then new I/O is queued based on these limits, and then some paths
2451 if (dm_table_has_no_data_devices(table)) {
2452 live_map = dm_get_live_table_fast(md);
2454 limits = md->queue->limits;
2455 dm_put_live_table_fast(md);
2459 r = dm_calculate_queue_limits(table, &limits);
2466 map = __bind(md, table, &limits);
2467 dm_issue_global_event();
2470 mutex_unlock(&md->suspend_lock);
2475 * Functions to lock and unlock any filesystem running on the
2478 static int lock_fs(struct mapped_device *md)
2482 WARN_ON(md->frozen_sb);
2484 md->frozen_sb = freeze_bdev(md->bdev);
2485 if (IS_ERR(md->frozen_sb)) {
2486 r = PTR_ERR(md->frozen_sb);
2487 md->frozen_sb = NULL;
2491 set_bit(DMF_FROZEN, &md->flags);
2496 static void unlock_fs(struct mapped_device *md)
2498 if (!test_bit(DMF_FROZEN, &md->flags))
2501 thaw_bdev(md->bdev, md->frozen_sb);
2502 md->frozen_sb = NULL;
2503 clear_bit(DMF_FROZEN, &md->flags);
2507 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2508 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2509 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2511 * If __dm_suspend returns 0, the device is completely quiescent
2512 * now. There is no request-processing activity. All new requests
2513 * are being added to md->deferred list.
2515 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2516 unsigned suspend_flags, long task_state,
2517 int dmf_suspended_flag)
2519 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2520 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2523 lockdep_assert_held(&md->suspend_lock);
2526 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2527 * This flag is cleared before dm_suspend returns.
2530 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2532 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2535 * This gets reverted if there's an error later and the targets
2536 * provide the .presuspend_undo hook.
2538 dm_table_presuspend_targets(map);
2541 * Flush I/O to the device.
2542 * Any I/O submitted after lock_fs() may not be flushed.
2543 * noflush takes precedence over do_lockfs.
2544 * (lock_fs() flushes I/Os and waits for them to complete.)
2546 if (!noflush && do_lockfs) {
2549 dm_table_presuspend_undo_targets(map);
2555 * Here we must make sure that no processes are submitting requests
2556 * to target drivers i.e. no one may be executing
2557 * __split_and_process_bio. This is called from dm_request and
2560 * To get all processes out of __split_and_process_bio in dm_request,
2561 * we take the write lock. To prevent any process from reentering
2562 * __split_and_process_bio from dm_request and quiesce the thread
2563 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2564 * flush_workqueue(md->wq).
2566 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2568 synchronize_srcu(&md->io_barrier);
2571 * Stop md->queue before flushing md->wq in case request-based
2572 * dm defers requests to md->wq from md->queue.
2574 if (dm_request_based(md)) {
2575 dm_stop_queue(md->queue);
2576 if (md->kworker_task)
2577 kthread_flush_worker(&md->kworker);
2580 flush_workqueue(md->wq);
2583 * At this point no more requests are entering target request routines.
2584 * We call dm_wait_for_completion to wait for all existing requests
2587 r = dm_wait_for_completion(md, task_state);
2589 set_bit(dmf_suspended_flag, &md->flags);
2592 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2594 synchronize_srcu(&md->io_barrier);
2596 /* were we interrupted ? */
2600 if (dm_request_based(md))
2601 dm_start_queue(md->queue);
2604 dm_table_presuspend_undo_targets(map);
2605 /* pushback list is already flushed, so skip flush */
2612 * We need to be able to change a mapping table under a mounted
2613 * filesystem. For example we might want to move some data in
2614 * the background. Before the table can be swapped with
2615 * dm_bind_table, dm_suspend must be called to flush any in
2616 * flight bios and ensure that any further io gets deferred.
2619 * Suspend mechanism in request-based dm.
2621 * 1. Flush all I/Os by lock_fs() if needed.
2622 * 2. Stop dispatching any I/O by stopping the request_queue.
2623 * 3. Wait for all in-flight I/Os to be completed or requeued.
2625 * To abort suspend, start the request_queue.
2627 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2629 struct dm_table *map = NULL;
2633 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2635 if (dm_suspended_md(md)) {
2640 if (dm_suspended_internally_md(md)) {
2641 /* already internally suspended, wait for internal resume */
2642 mutex_unlock(&md->suspend_lock);
2643 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2649 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2651 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2655 dm_table_postsuspend_targets(map);
2658 mutex_unlock(&md->suspend_lock);
2662 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2665 int r = dm_table_resume_targets(map);
2673 * Flushing deferred I/Os must be done after targets are resumed
2674 * so that mapping of targets can work correctly.
2675 * Request-based dm is queueing the deferred I/Os in its request_queue.
2677 if (dm_request_based(md))
2678 dm_start_queue(md->queue);
2685 int dm_resume(struct mapped_device *md)
2688 struct dm_table *map = NULL;
2692 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2694 if (!dm_suspended_md(md))
2697 if (dm_suspended_internally_md(md)) {
2698 /* already internally suspended, wait for internal resume */
2699 mutex_unlock(&md->suspend_lock);
2700 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2706 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2707 if (!map || !dm_table_get_size(map))
2710 r = __dm_resume(md, map);
2714 clear_bit(DMF_SUSPENDED, &md->flags);
2716 mutex_unlock(&md->suspend_lock);
2722 * Internal suspend/resume works like userspace-driven suspend. It waits
2723 * until all bios finish and prevents issuing new bios to the target drivers.
2724 * It may be used only from the kernel.
2727 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2729 struct dm_table *map = NULL;
2731 lockdep_assert_held(&md->suspend_lock);
2733 if (md->internal_suspend_count++)
2734 return; /* nested internal suspend */
2736 if (dm_suspended_md(md)) {
2737 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2738 return; /* nest suspend */
2741 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2744 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2745 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2746 * would require changing .presuspend to return an error -- avoid this
2747 * until there is a need for more elaborate variants of internal suspend.
2749 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2750 DMF_SUSPENDED_INTERNALLY);
2752 dm_table_postsuspend_targets(map);
2755 static void __dm_internal_resume(struct mapped_device *md)
2757 BUG_ON(!md->internal_suspend_count);
2759 if (--md->internal_suspend_count)
2760 return; /* resume from nested internal suspend */
2762 if (dm_suspended_md(md))
2763 goto done; /* resume from nested suspend */
2766 * NOTE: existing callers don't need to call dm_table_resume_targets
2767 * (which may fail -- so best to avoid it for now by passing NULL map)
2769 (void) __dm_resume(md, NULL);
2772 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2773 smp_mb__after_atomic();
2774 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2777 void dm_internal_suspend_noflush(struct mapped_device *md)
2779 mutex_lock(&md->suspend_lock);
2780 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2781 mutex_unlock(&md->suspend_lock);
2783 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2785 void dm_internal_resume(struct mapped_device *md)
2787 mutex_lock(&md->suspend_lock);
2788 __dm_internal_resume(md);
2789 mutex_unlock(&md->suspend_lock);
2791 EXPORT_SYMBOL_GPL(dm_internal_resume);
2794 * Fast variants of internal suspend/resume hold md->suspend_lock,
2795 * which prevents interaction with userspace-driven suspend.
2798 void dm_internal_suspend_fast(struct mapped_device *md)
2800 mutex_lock(&md->suspend_lock);
2801 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2804 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2805 synchronize_srcu(&md->io_barrier);
2806 flush_workqueue(md->wq);
2807 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2809 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2811 void dm_internal_resume_fast(struct mapped_device *md)
2813 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2819 mutex_unlock(&md->suspend_lock);
2821 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2823 /*-----------------------------------------------------------------
2824 * Event notification.
2825 *---------------------------------------------------------------*/
2826 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2829 char udev_cookie[DM_COOKIE_LENGTH];
2830 char *envp[] = { udev_cookie, NULL };
2833 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2835 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2836 DM_COOKIE_ENV_VAR_NAME, cookie);
2837 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2842 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2844 return atomic_add_return(1, &md->uevent_seq);
2847 uint32_t dm_get_event_nr(struct mapped_device *md)
2849 return atomic_read(&md->event_nr);
2852 int dm_wait_event(struct mapped_device *md, int event_nr)
2854 return wait_event_interruptible(md->eventq,
2855 (event_nr != atomic_read(&md->event_nr)));
2858 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2860 unsigned long flags;
2862 spin_lock_irqsave(&md->uevent_lock, flags);
2863 list_add(elist, &md->uevent_list);
2864 spin_unlock_irqrestore(&md->uevent_lock, flags);
2868 * The gendisk is only valid as long as you have a reference
2871 struct gendisk *dm_disk(struct mapped_device *md)
2875 EXPORT_SYMBOL_GPL(dm_disk);
2877 struct kobject *dm_kobject(struct mapped_device *md)
2879 return &md->kobj_holder.kobj;
2882 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2884 struct mapped_device *md;
2886 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2888 spin_lock(&_minor_lock);
2889 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2895 spin_unlock(&_minor_lock);
2900 int dm_suspended_md(struct mapped_device *md)
2902 return test_bit(DMF_SUSPENDED, &md->flags);
2905 int dm_suspended_internally_md(struct mapped_device *md)
2907 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2910 int dm_test_deferred_remove_flag(struct mapped_device *md)
2912 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2915 int dm_suspended(struct dm_target *ti)
2917 return dm_suspended_md(dm_table_get_md(ti->table));
2919 EXPORT_SYMBOL_GPL(dm_suspended);
2921 int dm_noflush_suspending(struct dm_target *ti)
2923 return __noflush_suspending(dm_table_get_md(ti->table));
2925 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2927 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2928 unsigned integrity, unsigned per_io_data_size,
2929 unsigned min_pool_size)
2931 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2932 unsigned int pool_size = 0;
2933 unsigned int front_pad, io_front_pad;
2940 case DM_TYPE_BIO_BASED:
2941 case DM_TYPE_DAX_BIO_BASED:
2942 case DM_TYPE_NVME_BIO_BASED:
2943 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2944 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2945 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
2946 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2949 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2952 case DM_TYPE_REQUEST_BASED:
2953 case DM_TYPE_MQ_REQUEST_BASED:
2954 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2955 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2956 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2962 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2966 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2972 dm_free_md_mempools(pools);
2977 void dm_free_md_mempools(struct dm_md_mempools *pools)
2982 bioset_exit(&pools->bs);
2983 bioset_exit(&pools->io_bs);
2995 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2998 struct mapped_device *md = bdev->bd_disk->private_data;
2999 struct dm_table *table;
3000 struct dm_target *ti;
3001 int ret = -ENOTTY, srcu_idx;
3003 table = dm_get_live_table(md, &srcu_idx);
3004 if (!table || !dm_table_get_size(table))
3007 /* We only support devices that have a single target */
3008 if (dm_table_get_num_targets(table) != 1)
3010 ti = dm_table_get_target(table, 0);
3013 if (!ti->type->iterate_devices)
3016 ret = ti->type->iterate_devices(ti, fn, data);
3018 dm_put_live_table(md, srcu_idx);
3023 * For register / unregister we need to manually call out to every path.
3025 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3026 sector_t start, sector_t len, void *data)
3028 struct dm_pr *pr = data;
3029 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3031 if (!ops || !ops->pr_register)
3033 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3036 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3047 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3048 if (ret && new_key) {
3049 /* unregister all paths if we failed to register any path */
3050 pr.old_key = new_key;
3053 pr.fail_early = false;
3054 dm_call_pr(bdev, __dm_pr_register, &pr);
3060 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3063 struct mapped_device *md = bdev->bd_disk->private_data;
3064 const struct pr_ops *ops;
3067 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3071 ops = bdev->bd_disk->fops->pr_ops;
3072 if (ops && ops->pr_reserve)
3073 r = ops->pr_reserve(bdev, key, type, flags);
3077 dm_unprepare_ioctl(md, srcu_idx);
3081 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3083 struct mapped_device *md = bdev->bd_disk->private_data;
3084 const struct pr_ops *ops;
3087 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3091 ops = bdev->bd_disk->fops->pr_ops;
3092 if (ops && ops->pr_release)
3093 r = ops->pr_release(bdev, key, type);
3097 dm_unprepare_ioctl(md, srcu_idx);
3101 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3102 enum pr_type type, bool abort)
3104 struct mapped_device *md = bdev->bd_disk->private_data;
3105 const struct pr_ops *ops;
3108 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3112 ops = bdev->bd_disk->fops->pr_ops;
3113 if (ops && ops->pr_preempt)
3114 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3118 dm_unprepare_ioctl(md, srcu_idx);
3122 static int dm_pr_clear(struct block_device *bdev, u64 key)
3124 struct mapped_device *md = bdev->bd_disk->private_data;
3125 const struct pr_ops *ops;
3128 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3132 ops = bdev->bd_disk->fops->pr_ops;
3133 if (ops && ops->pr_clear)
3134 r = ops->pr_clear(bdev, key);
3138 dm_unprepare_ioctl(md, srcu_idx);
3142 static const struct pr_ops dm_pr_ops = {
3143 .pr_register = dm_pr_register,
3144 .pr_reserve = dm_pr_reserve,
3145 .pr_release = dm_pr_release,
3146 .pr_preempt = dm_pr_preempt,
3147 .pr_clear = dm_pr_clear,
3150 static const struct block_device_operations dm_blk_dops = {
3151 .open = dm_blk_open,
3152 .release = dm_blk_close,
3153 .ioctl = dm_blk_ioctl,
3154 .getgeo = dm_blk_getgeo,
3155 .pr_ops = &dm_pr_ops,
3156 .owner = THIS_MODULE
3159 static const struct dax_operations dm_dax_ops = {
3160 .direct_access = dm_dax_direct_access,
3161 .copy_from_iter = dm_dax_copy_from_iter,
3162 .copy_to_iter = dm_dax_copy_to_iter,
3168 module_init(dm_init);
3169 module_exit(dm_exit);
3171 module_param(major, uint, 0);
3172 MODULE_PARM_DESC(major, "The major number of the device mapper");
3174 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3175 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3177 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3178 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3180 MODULE_DESCRIPTION(DM_NAME " driver");
3181 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3182 MODULE_LICENSE("GPL");