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"
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/mutex.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/signal.h>
18 #include <linux/blkpg.h>
19 #include <linux/bio.h>
20 #include <linux/mempool.h>
21 #include <linux/dax.h>
22 #include <linux/slab.h>
23 #include <linux/idr.h>
24 #include <linux/uio.h>
25 #include <linux/hdreg.h>
26 #include <linux/delay.h>
27 #include <linux/wait.h>
29 #include <linux/refcount.h>
30 #include <linux/part_stat.h>
31 #include <linux/blk-crypto.h>
32 #include <linux/keyslot-manager.h>
34 #define DM_MSG_PREFIX "core"
37 * Cookies are numeric values sent with CHANGE and REMOVE
38 * uevents while resuming, removing or renaming the device.
40 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41 #define DM_COOKIE_LENGTH 24
43 static const char *_name = DM_NAME;
45 static unsigned int major = 0;
46 static unsigned int _major = 0;
48 static DEFINE_IDR(_minor_idr);
50 static DEFINE_SPINLOCK(_minor_lock);
52 static void do_deferred_remove(struct work_struct *w);
54 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
56 static struct workqueue_struct *deferred_remove_workqueue;
58 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
59 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
61 void dm_issue_global_event(void)
63 atomic_inc(&dm_global_event_nr);
64 wake_up(&dm_global_eventq);
68 * One of these is allocated (on-stack) per original bio.
75 unsigned sector_count;
78 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
79 #define DM_IO_BIO_OFFSET \
80 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
82 void *dm_per_bio_data(struct bio *bio, size_t data_size)
84 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
85 if (!tio->inside_dm_io)
86 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
87 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
89 EXPORT_SYMBOL_GPL(dm_per_bio_data);
91 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
93 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
94 if (io->magic == DM_IO_MAGIC)
95 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
96 BUG_ON(io->magic != DM_TIO_MAGIC);
97 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
99 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
101 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
103 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
105 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
107 #define MINOR_ALLOCED ((void *)-1)
109 #define DM_NUMA_NODE NUMA_NO_NODE
110 static int dm_numa_node = DM_NUMA_NODE;
112 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
113 static int swap_bios = DEFAULT_SWAP_BIOS;
114 static int get_swap_bios(void)
116 int latch = READ_ONCE(swap_bios);
117 if (unlikely(latch <= 0))
118 latch = DEFAULT_SWAP_BIOS;
123 * For mempools pre-allocation at the table loading time.
125 struct dm_md_mempools {
127 struct bio_set io_bs;
130 struct table_device {
131 struct list_head list;
133 struct dm_dev dm_dev;
137 * Bio-based DM's mempools' reserved IOs set by the user.
139 #define RESERVED_BIO_BASED_IOS 16
140 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
142 static int __dm_get_module_param_int(int *module_param, int min, int max)
144 int param = READ_ONCE(*module_param);
145 int modified_param = 0;
146 bool modified = true;
149 modified_param = min;
150 else if (param > max)
151 modified_param = max;
156 (void)cmpxchg(module_param, param, modified_param);
157 param = modified_param;
163 unsigned __dm_get_module_param(unsigned *module_param,
164 unsigned def, unsigned max)
166 unsigned param = READ_ONCE(*module_param);
167 unsigned modified_param = 0;
170 modified_param = def;
171 else if (param > max)
172 modified_param = max;
174 if (modified_param) {
175 (void)cmpxchg(module_param, param, modified_param);
176 param = modified_param;
182 unsigned dm_get_reserved_bio_based_ios(void)
184 return __dm_get_module_param(&reserved_bio_based_ios,
185 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
187 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
189 static unsigned dm_get_numa_node(void)
191 return __dm_get_module_param_int(&dm_numa_node,
192 DM_NUMA_NODE, num_online_nodes() - 1);
195 static int __init local_init(void)
199 r = dm_uevent_init();
203 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
204 if (!deferred_remove_workqueue) {
206 goto out_uevent_exit;
210 r = register_blkdev(_major, _name);
212 goto out_free_workqueue;
220 destroy_workqueue(deferred_remove_workqueue);
227 static void local_exit(void)
229 flush_scheduled_work();
230 destroy_workqueue(deferred_remove_workqueue);
232 unregister_blkdev(_major, _name);
237 DMINFO("cleaned up");
240 static int (*_inits[])(void) __initdata = {
251 static void (*_exits[])(void) = {
262 static int __init dm_init(void)
264 const int count = ARRAY_SIZE(_inits);
267 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
268 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
269 " Duplicate IMA measurements will not be recorded in the IMA log.");
272 for (i = 0; i < count; i++) {
286 static void __exit dm_exit(void)
288 int i = ARRAY_SIZE(_exits);
294 * Should be empty by this point.
296 idr_destroy(&_minor_idr);
300 * Block device functions
302 int dm_deleting_md(struct mapped_device *md)
304 return test_bit(DMF_DELETING, &md->flags);
307 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
309 struct mapped_device *md;
311 spin_lock(&_minor_lock);
313 md = bdev->bd_disk->private_data;
317 if (test_bit(DMF_FREEING, &md->flags) ||
318 dm_deleting_md(md)) {
324 atomic_inc(&md->open_count);
326 spin_unlock(&_minor_lock);
328 return md ? 0 : -ENXIO;
331 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
333 struct mapped_device *md;
335 spin_lock(&_minor_lock);
337 md = disk->private_data;
341 if (atomic_dec_and_test(&md->open_count) &&
342 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
343 queue_work(deferred_remove_workqueue, &deferred_remove_work);
347 spin_unlock(&_minor_lock);
350 int dm_open_count(struct mapped_device *md)
352 return atomic_read(&md->open_count);
356 * Guarantees nothing is using the device before it's deleted.
358 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
362 spin_lock(&_minor_lock);
364 if (dm_open_count(md)) {
367 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
368 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
371 set_bit(DMF_DELETING, &md->flags);
373 spin_unlock(&_minor_lock);
378 int dm_cancel_deferred_remove(struct mapped_device *md)
382 spin_lock(&_minor_lock);
384 if (test_bit(DMF_DELETING, &md->flags))
387 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
389 spin_unlock(&_minor_lock);
394 static void do_deferred_remove(struct work_struct *w)
396 dm_deferred_remove();
399 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
401 struct mapped_device *md = bdev->bd_disk->private_data;
403 return dm_get_geometry(md, geo);
406 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
407 struct block_device **bdev)
409 struct dm_target *tgt;
410 struct dm_table *map;
415 map = dm_get_live_table(md, srcu_idx);
416 if (!map || !dm_table_get_size(map))
419 /* We only support devices that have a single target */
420 if (dm_table_get_num_targets(map) != 1)
423 tgt = dm_table_get_target(map, 0);
424 if (!tgt->type->prepare_ioctl)
427 if (dm_suspended_md(md))
430 r = tgt->type->prepare_ioctl(tgt, bdev);
431 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
432 dm_put_live_table(md, *srcu_idx);
440 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
442 dm_put_live_table(md, srcu_idx);
445 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
446 unsigned int cmd, unsigned long arg)
448 struct mapped_device *md = bdev->bd_disk->private_data;
451 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
457 * Target determined this ioctl is being issued against a
458 * subset of the parent bdev; require extra privileges.
460 if (!capable(CAP_SYS_RAWIO)) {
462 "%s: sending ioctl %x to DM device without required privilege.",
469 if (!bdev->bd_disk->fops->ioctl)
472 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
474 dm_unprepare_ioctl(md, srcu_idx);
478 u64 dm_start_time_ns_from_clone(struct bio *bio)
480 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
481 struct dm_io *io = tio->io;
483 return jiffies_to_nsecs(io->start_time);
485 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
487 static void start_io_acct(struct dm_io *io)
489 struct mapped_device *md = io->md;
490 struct bio *bio = io->orig_bio;
492 io->start_time = bio_start_io_acct(bio);
493 if (unlikely(dm_stats_used(&md->stats)))
494 dm_stats_account_io(&md->stats, bio_data_dir(bio),
495 bio->bi_iter.bi_sector, bio_sectors(bio),
496 false, 0, &io->stats_aux);
499 static void end_io_acct(struct mapped_device *md, struct bio *bio,
500 unsigned long start_time, struct dm_stats_aux *stats_aux)
502 unsigned long duration = jiffies - start_time;
504 bio_end_io_acct(bio, start_time);
506 if (unlikely(dm_stats_used(&md->stats)))
507 dm_stats_account_io(&md->stats, bio_data_dir(bio),
508 bio->bi_iter.bi_sector, bio_sectors(bio),
509 true, duration, stats_aux);
511 /* nudge anyone waiting on suspend queue */
512 if (unlikely(wq_has_sleeper(&md->wait)))
516 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
519 struct dm_target_io *tio;
522 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
526 tio = container_of(clone, struct dm_target_io, clone);
527 tio->inside_dm_io = true;
530 io = container_of(tio, struct dm_io, tio);
531 io->magic = DM_IO_MAGIC;
533 atomic_set(&io->io_count, 1);
536 spin_lock_init(&io->endio_lock);
543 static void free_io(struct mapped_device *md, struct dm_io *io)
545 bio_put(&io->tio.clone);
548 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
549 unsigned target_bio_nr, gfp_t gfp_mask)
551 struct dm_target_io *tio;
553 if (!ci->io->tio.io) {
554 /* the dm_target_io embedded in ci->io is available */
557 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
561 tio = container_of(clone, struct dm_target_io, clone);
562 tio->inside_dm_io = false;
565 tio->magic = DM_TIO_MAGIC;
568 tio->target_bio_nr = target_bio_nr;
573 static void free_tio(struct dm_target_io *tio)
575 if (tio->inside_dm_io)
577 bio_put(&tio->clone);
581 * Add the bio to the list of deferred io.
583 static void queue_io(struct mapped_device *md, struct bio *bio)
587 spin_lock_irqsave(&md->deferred_lock, flags);
588 bio_list_add(&md->deferred, bio);
589 spin_unlock_irqrestore(&md->deferred_lock, flags);
590 queue_work(md->wq, &md->work);
594 * Everyone (including functions in this file), should use this
595 * function to access the md->map field, and make sure they call
596 * dm_put_live_table() when finished.
598 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
600 *srcu_idx = srcu_read_lock(&md->io_barrier);
602 return srcu_dereference(md->map, &md->io_barrier);
605 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
607 srcu_read_unlock(&md->io_barrier, srcu_idx);
610 void dm_sync_table(struct mapped_device *md)
612 synchronize_srcu(&md->io_barrier);
613 synchronize_rcu_expedited();
617 * A fast alternative to dm_get_live_table/dm_put_live_table.
618 * The caller must not block between these two functions.
620 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
623 return rcu_dereference(md->map);
626 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
631 static char *_dm_claim_ptr = "I belong to device-mapper";
634 * Open a table device so we can use it as a map destination.
636 static int open_table_device(struct table_device *td, dev_t dev,
637 struct mapped_device *md)
639 struct block_device *bdev;
643 BUG_ON(td->dm_dev.bdev);
645 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
647 return PTR_ERR(bdev);
649 r = bd_link_disk_holder(bdev, dm_disk(md));
651 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
655 td->dm_dev.bdev = bdev;
656 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev);
661 * Close a table device that we've been using.
663 static void close_table_device(struct table_device *td, struct mapped_device *md)
665 if (!td->dm_dev.bdev)
668 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
669 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
670 put_dax(td->dm_dev.dax_dev);
671 td->dm_dev.bdev = NULL;
672 td->dm_dev.dax_dev = NULL;
675 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
678 struct table_device *td;
680 list_for_each_entry(td, l, list)
681 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
687 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
688 struct dm_dev **result)
691 struct table_device *td;
693 mutex_lock(&md->table_devices_lock);
694 td = find_table_device(&md->table_devices, dev, mode);
696 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
698 mutex_unlock(&md->table_devices_lock);
702 td->dm_dev.mode = mode;
703 td->dm_dev.bdev = NULL;
705 if ((r = open_table_device(td, dev, md))) {
706 mutex_unlock(&md->table_devices_lock);
711 format_dev_t(td->dm_dev.name, dev);
713 refcount_set(&td->count, 1);
714 list_add(&td->list, &md->table_devices);
716 refcount_inc(&td->count);
718 mutex_unlock(&md->table_devices_lock);
720 *result = &td->dm_dev;
724 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
726 struct table_device *td = container_of(d, struct table_device, dm_dev);
728 mutex_lock(&md->table_devices_lock);
729 if (refcount_dec_and_test(&td->count)) {
730 close_table_device(td, md);
734 mutex_unlock(&md->table_devices_lock);
737 static void free_table_devices(struct list_head *devices)
739 struct list_head *tmp, *next;
741 list_for_each_safe(tmp, next, devices) {
742 struct table_device *td = list_entry(tmp, struct table_device, list);
744 DMWARN("dm_destroy: %s still exists with %d references",
745 td->dm_dev.name, refcount_read(&td->count));
751 * Get the geometry associated with a dm device
753 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
761 * Set the geometry of a device.
763 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
765 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
767 if (geo->start > sz) {
768 DMWARN("Start sector is beyond the geometry limits.");
777 static int __noflush_suspending(struct mapped_device *md)
779 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
783 * Decrements the number of outstanding ios that a bio has been
784 * cloned into, completing the original io if necc.
786 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
789 blk_status_t io_error;
791 struct mapped_device *md = io->md;
792 unsigned long start_time = 0;
793 struct dm_stats_aux stats_aux;
795 /* Push-back supersedes any I/O errors */
796 if (unlikely(error)) {
797 spin_lock_irqsave(&io->endio_lock, flags);
798 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
800 spin_unlock_irqrestore(&io->endio_lock, flags);
803 if (atomic_dec_and_test(&io->io_count)) {
805 if (io->status == BLK_STS_DM_REQUEUE) {
807 * Target requested pushing back the I/O.
809 spin_lock_irqsave(&md->deferred_lock, flags);
810 if (__noflush_suspending(md) &&
811 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
812 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
813 bio_list_add_head(&md->deferred, bio);
816 * noflush suspend was interrupted or this is
817 * a write to a zoned target.
819 io->status = BLK_STS_IOERR;
821 spin_unlock_irqrestore(&md->deferred_lock, flags);
824 io_error = io->status;
825 start_time = io->start_time;
826 stats_aux = io->stats_aux;
828 end_io_acct(md, bio, start_time, &stats_aux);
830 if (io_error == BLK_STS_DM_REQUEUE)
833 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
835 * Preflush done for flush with data, reissue
836 * without REQ_PREFLUSH.
838 bio->bi_opf &= ~REQ_PREFLUSH;
841 /* done with normal IO or empty flush */
843 bio->bi_status = io_error;
849 void disable_discard(struct mapped_device *md)
851 struct queue_limits *limits = dm_get_queue_limits(md);
853 /* device doesn't really support DISCARD, disable it */
854 limits->max_discard_sectors = 0;
855 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
858 void disable_write_same(struct mapped_device *md)
860 struct queue_limits *limits = dm_get_queue_limits(md);
862 /* device doesn't really support WRITE SAME, disable it */
863 limits->max_write_same_sectors = 0;
866 void disable_write_zeroes(struct mapped_device *md)
868 struct queue_limits *limits = dm_get_queue_limits(md);
870 /* device doesn't really support WRITE ZEROES, disable it */
871 limits->max_write_zeroes_sectors = 0;
874 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
876 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
879 static void clone_endio(struct bio *bio)
881 blk_status_t error = bio->bi_status;
882 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
883 struct dm_io *io = tio->io;
884 struct mapped_device *md = tio->io->md;
885 dm_endio_fn endio = tio->ti->type->end_io;
886 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
888 if (unlikely(error == BLK_STS_TARGET)) {
889 if (bio_op(bio) == REQ_OP_DISCARD &&
890 !q->limits.max_discard_sectors)
892 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
893 !q->limits.max_write_same_sectors)
894 disable_write_same(md);
895 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
896 !q->limits.max_write_zeroes_sectors)
897 disable_write_zeroes(md);
900 if (blk_queue_is_zoned(q))
901 dm_zone_endio(io, bio);
904 int r = endio(tio->ti, bio, &error);
906 case DM_ENDIO_REQUEUE:
908 * Requeuing writes to a sequential zone of a zoned
909 * target will break the sequential write pattern:
912 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
913 error = BLK_STS_IOERR;
915 error = BLK_STS_DM_REQUEUE;
919 case DM_ENDIO_INCOMPLETE:
920 /* The target will handle the io */
923 DMWARN("unimplemented target endio return value: %d", r);
928 if (unlikely(swap_bios_limit(tio->ti, bio))) {
929 struct mapped_device *md = io->md;
930 up(&md->swap_bios_semaphore);
934 dm_io_dec_pending(io, error);
938 * Return maximum size of I/O possible at the supplied sector up to the current
941 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
942 sector_t target_offset)
944 return ti->len - target_offset;
947 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
949 sector_t target_offset = dm_target_offset(ti, sector);
950 sector_t len = max_io_len_target_boundary(ti, target_offset);
954 * Does the target need to split IO even further?
955 * - varied (per target) IO splitting is a tenet of DM; this
956 * explains why stacked chunk_sectors based splitting via
957 * blk_max_size_offset() isn't possible here. So pass in
958 * ti->max_io_len to override stacked chunk_sectors.
960 if (ti->max_io_len) {
961 max_len = blk_max_size_offset(ti->table->md->queue,
962 target_offset, ti->max_io_len);
970 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
972 if (len > UINT_MAX) {
973 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
974 (unsigned long long)len, UINT_MAX);
975 ti->error = "Maximum size of target IO is too large";
979 ti->max_io_len = (uint32_t) len;
983 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
985 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
986 sector_t sector, int *srcu_idx)
987 __acquires(md->io_barrier)
989 struct dm_table *map;
990 struct dm_target *ti;
992 map = dm_get_live_table(md, srcu_idx);
996 ti = dm_table_find_target(map, sector);
1003 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1004 long nr_pages, void **kaddr, pfn_t *pfn)
1006 struct mapped_device *md = dax_get_private(dax_dev);
1007 sector_t sector = pgoff * PAGE_SECTORS;
1008 struct dm_target *ti;
1009 long len, ret = -EIO;
1012 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1016 if (!ti->type->direct_access)
1018 len = max_io_len(ti, sector) / PAGE_SECTORS;
1021 nr_pages = min(len, nr_pages);
1022 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1025 dm_put_live_table(md, srcu_idx);
1030 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1031 int blocksize, sector_t start, sector_t len)
1033 struct mapped_device *md = dax_get_private(dax_dev);
1034 struct dm_table *map;
1038 map = dm_get_live_table(md, &srcu_idx);
1042 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1045 dm_put_live_table(md, srcu_idx);
1050 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1051 void *addr, size_t bytes, struct iov_iter *i)
1053 struct mapped_device *md = dax_get_private(dax_dev);
1054 sector_t sector = pgoff * PAGE_SECTORS;
1055 struct dm_target *ti;
1059 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1063 if (!ti->type->dax_copy_from_iter) {
1064 ret = copy_from_iter(addr, bytes, i);
1067 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1069 dm_put_live_table(md, srcu_idx);
1074 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1075 void *addr, size_t bytes, struct iov_iter *i)
1077 struct mapped_device *md = dax_get_private(dax_dev);
1078 sector_t sector = pgoff * PAGE_SECTORS;
1079 struct dm_target *ti;
1083 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1087 if (!ti->type->dax_copy_to_iter) {
1088 ret = copy_to_iter(addr, bytes, i);
1091 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1093 dm_put_live_table(md, srcu_idx);
1098 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1101 struct mapped_device *md = dax_get_private(dax_dev);
1102 sector_t sector = pgoff * PAGE_SECTORS;
1103 struct dm_target *ti;
1107 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1111 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1113 * ->zero_page_range() is mandatory dax operation. If we are
1114 * here, something is wrong.
1118 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1120 dm_put_live_table(md, srcu_idx);
1126 * A target may call dm_accept_partial_bio only from the map routine. It is
1127 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1128 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1130 * dm_accept_partial_bio informs the dm that the target only wants to process
1131 * additional n_sectors sectors of the bio and the rest of the data should be
1132 * sent in a next bio.
1134 * A diagram that explains the arithmetics:
1135 * +--------------------+---------------+-------+
1137 * +--------------------+---------------+-------+
1139 * <-------------- *tio->len_ptr --------------->
1140 * <------- bi_size ------->
1143 * Region 1 was already iterated over with bio_advance or similar function.
1144 * (it may be empty if the target doesn't use bio_advance)
1145 * Region 2 is the remaining bio size that the target wants to process.
1146 * (it may be empty if region 1 is non-empty, although there is no reason
1148 * The target requires that region 3 is to be sent in the next bio.
1150 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1151 * the partially processed part (the sum of regions 1+2) must be the same for all
1152 * copies of the bio.
1154 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1156 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1157 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1159 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1160 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1161 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1162 BUG_ON(bi_size > *tio->len_ptr);
1163 BUG_ON(n_sectors > bi_size);
1165 *tio->len_ptr -= bi_size - n_sectors;
1166 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1168 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1170 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1172 mutex_lock(&md->swap_bios_lock);
1173 while (latch < md->swap_bios) {
1175 down(&md->swap_bios_semaphore);
1178 while (latch > md->swap_bios) {
1180 up(&md->swap_bios_semaphore);
1183 mutex_unlock(&md->swap_bios_lock);
1186 static blk_qc_t __map_bio(struct dm_target_io *tio)
1190 struct bio *clone = &tio->clone;
1191 struct dm_io *io = tio->io;
1192 struct dm_target *ti = tio->ti;
1193 blk_qc_t ret = BLK_QC_T_NONE;
1195 clone->bi_end_io = clone_endio;
1198 * Map the clone. If r == 0 we don't need to do
1199 * anything, the target has assumed ownership of
1202 dm_io_inc_pending(io);
1203 sector = clone->bi_iter.bi_sector;
1205 if (unlikely(swap_bios_limit(ti, clone))) {
1206 struct mapped_device *md = io->md;
1207 int latch = get_swap_bios();
1208 if (unlikely(latch != md->swap_bios))
1209 __set_swap_bios_limit(md, latch);
1210 down(&md->swap_bios_semaphore);
1214 * Check if the IO needs a special mapping due to zone append emulation
1215 * on zoned target. In this case, dm_zone_map_bio() calls the target
1218 if (dm_emulate_zone_append(io->md))
1219 r = dm_zone_map_bio(tio);
1221 r = ti->type->map(ti, clone);
1224 case DM_MAPIO_SUBMITTED:
1226 case DM_MAPIO_REMAPPED:
1227 /* the bio has been remapped so dispatch it */
1228 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1229 ret = submit_bio_noacct(clone);
1232 if (unlikely(swap_bios_limit(ti, clone))) {
1233 struct mapped_device *md = io->md;
1234 up(&md->swap_bios_semaphore);
1237 dm_io_dec_pending(io, BLK_STS_IOERR);
1239 case DM_MAPIO_REQUEUE:
1240 if (unlikely(swap_bios_limit(ti, clone))) {
1241 struct mapped_device *md = io->md;
1242 up(&md->swap_bios_semaphore);
1245 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1248 DMWARN("unimplemented target map return value: %d", r);
1255 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1257 bio->bi_iter.bi_sector = sector;
1258 bio->bi_iter.bi_size = to_bytes(len);
1262 * Creates a bio that consists of range of complete bvecs.
1264 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1265 sector_t sector, unsigned len)
1267 struct bio *clone = &tio->clone;
1270 __bio_clone_fast(clone, bio);
1272 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1276 if (bio_integrity(bio)) {
1277 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1278 !dm_target_passes_integrity(tio->ti->type))) {
1279 DMWARN("%s: the target %s doesn't support integrity data.",
1280 dm_device_name(tio->io->md),
1281 tio->ti->type->name);
1285 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1290 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1291 clone->bi_iter.bi_size = to_bytes(len);
1293 if (bio_integrity(bio))
1294 bio_integrity_trim(clone);
1299 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1300 struct dm_target *ti, unsigned num_bios)
1302 struct dm_target_io *tio;
1308 if (num_bios == 1) {
1309 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1310 bio_list_add(blist, &tio->clone);
1314 for (try = 0; try < 2; try++) {
1319 mutex_lock(&ci->io->md->table_devices_lock);
1320 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1321 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1325 bio_list_add(blist, &tio->clone);
1328 mutex_unlock(&ci->io->md->table_devices_lock);
1329 if (bio_nr == num_bios)
1332 while ((bio = bio_list_pop(blist))) {
1333 tio = container_of(bio, struct dm_target_io, clone);
1339 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1340 struct dm_target_io *tio, unsigned *len)
1342 struct bio *clone = &tio->clone;
1346 __bio_clone_fast(clone, ci->bio);
1348 bio_setup_sector(clone, ci->sector, *len);
1350 return __map_bio(tio);
1353 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1354 unsigned num_bios, unsigned *len)
1356 struct bio_list blist = BIO_EMPTY_LIST;
1358 struct dm_target_io *tio;
1360 alloc_multiple_bios(&blist, ci, ti, num_bios);
1362 while ((bio = bio_list_pop(&blist))) {
1363 tio = container_of(bio, struct dm_target_io, clone);
1364 (void) __clone_and_map_simple_bio(ci, tio, len);
1368 static int __send_empty_flush(struct clone_info *ci)
1370 unsigned target_nr = 0;
1371 struct dm_target *ti;
1372 struct bio flush_bio;
1375 * Use an on-stack bio for this, it's safe since we don't
1376 * need to reference it after submit. It's just used as
1377 * the basis for the clone(s).
1379 bio_init(&flush_bio, NULL, 0);
1380 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1381 bio_set_dev(&flush_bio, ci->io->md->disk->part0);
1383 ci->bio = &flush_bio;
1384 ci->sector_count = 0;
1386 BUG_ON(bio_has_data(ci->bio));
1387 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1388 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1390 bio_uninit(ci->bio);
1394 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1395 sector_t sector, unsigned *len)
1397 struct bio *bio = ci->bio;
1398 struct dm_target_io *tio;
1401 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1403 r = clone_bio(tio, bio, sector, *len);
1408 (void) __map_bio(tio);
1413 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1419 * Even though the device advertised support for this type of
1420 * request, that does not mean every target supports it, and
1421 * reconfiguration might also have changed that since the
1422 * check was performed.
1427 len = min_t(sector_t, ci->sector_count,
1428 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1430 __send_duplicate_bios(ci, ti, num_bios, &len);
1433 ci->sector_count -= len;
1438 static bool is_abnormal_io(struct bio *bio)
1442 switch (bio_op(bio)) {
1443 case REQ_OP_DISCARD:
1444 case REQ_OP_SECURE_ERASE:
1445 case REQ_OP_WRITE_SAME:
1446 case REQ_OP_WRITE_ZEROES:
1454 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1457 struct bio *bio = ci->bio;
1458 unsigned num_bios = 0;
1460 switch (bio_op(bio)) {
1461 case REQ_OP_DISCARD:
1462 num_bios = ti->num_discard_bios;
1464 case REQ_OP_SECURE_ERASE:
1465 num_bios = ti->num_secure_erase_bios;
1467 case REQ_OP_WRITE_SAME:
1468 num_bios = ti->num_write_same_bios;
1470 case REQ_OP_WRITE_ZEROES:
1471 num_bios = ti->num_write_zeroes_bios;
1477 *result = __send_changing_extent_only(ci, ti, num_bios);
1482 * Select the correct strategy for processing a non-flush bio.
1484 static int __split_and_process_non_flush(struct clone_info *ci)
1486 struct dm_target *ti;
1490 ti = dm_table_find_target(ci->map, ci->sector);
1494 if (__process_abnormal_io(ci, ti, &r))
1497 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1499 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1504 ci->sector_count -= len;
1509 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1510 struct dm_table *map, struct bio *bio)
1513 ci->io = alloc_io(md, bio);
1514 ci->sector = bio->bi_iter.bi_sector;
1517 #define __dm_part_stat_sub(part, field, subnd) \
1518 (part_stat_get(part, field) -= (subnd))
1521 * Entry point to split a bio into clones and submit them to the targets.
1523 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1524 struct dm_table *map, struct bio *bio)
1526 struct clone_info ci;
1527 blk_qc_t ret = BLK_QC_T_NONE;
1530 init_clone_info(&ci, md, map, bio);
1532 if (bio->bi_opf & REQ_PREFLUSH) {
1533 error = __send_empty_flush(&ci);
1534 /* dm_io_dec_pending submits any data associated with flush */
1535 } else if (op_is_zone_mgmt(bio_op(bio))) {
1537 ci.sector_count = 0;
1538 error = __split_and_process_non_flush(&ci);
1541 ci.sector_count = bio_sectors(bio);
1542 error = __split_and_process_non_flush(&ci);
1543 if (ci.sector_count && !error) {
1545 * Remainder must be passed to submit_bio_noacct()
1546 * so that it gets handled *after* bios already submitted
1547 * have been completely processed.
1548 * We take a clone of the original to store in
1549 * ci.io->orig_bio to be used by end_io_acct() and
1550 * for dec_pending to use for completion handling.
1552 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1553 GFP_NOIO, &md->queue->bio_split);
1554 ci.io->orig_bio = b;
1557 * Adjust IO stats for each split, otherwise upon queue
1558 * reentry there will be redundant IO accounting.
1559 * NOTE: this is a stop-gap fix, a proper fix involves
1560 * significant refactoring of DM core's bio splitting
1561 * (by eliminating DM's splitting and just using bio_split)
1564 __dm_part_stat_sub(dm_disk(md)->part0,
1565 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1569 trace_block_split(b, bio->bi_iter.bi_sector);
1570 ret = submit_bio_noacct(bio);
1574 /* drop the extra reference count */
1575 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1579 static blk_qc_t dm_submit_bio(struct bio *bio)
1581 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1582 blk_qc_t ret = BLK_QC_T_NONE;
1584 struct dm_table *map;
1586 map = dm_get_live_table(md, &srcu_idx);
1587 if (unlikely(!map)) {
1588 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1589 dm_device_name(md));
1594 /* If suspended, queue this IO for later */
1595 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1596 if (bio->bi_opf & REQ_NOWAIT)
1597 bio_wouldblock_error(bio);
1598 else if (bio->bi_opf & REQ_RAHEAD)
1606 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1607 * otherwise associated queue_limits won't be imposed.
1609 if (is_abnormal_io(bio))
1610 blk_queue_split(&bio);
1612 ret = __split_and_process_bio(md, map, bio);
1614 dm_put_live_table(md, srcu_idx);
1618 /*-----------------------------------------------------------------
1619 * An IDR is used to keep track of allocated minor numbers.
1620 *---------------------------------------------------------------*/
1621 static void free_minor(int minor)
1623 spin_lock(&_minor_lock);
1624 idr_remove(&_minor_idr, minor);
1625 spin_unlock(&_minor_lock);
1629 * See if the device with a specific minor # is free.
1631 static int specific_minor(int minor)
1635 if (minor >= (1 << MINORBITS))
1638 idr_preload(GFP_KERNEL);
1639 spin_lock(&_minor_lock);
1641 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1643 spin_unlock(&_minor_lock);
1646 return r == -ENOSPC ? -EBUSY : r;
1650 static int next_free_minor(int *minor)
1654 idr_preload(GFP_KERNEL);
1655 spin_lock(&_minor_lock);
1657 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1659 spin_unlock(&_minor_lock);
1667 static const struct block_device_operations dm_blk_dops;
1668 static const struct block_device_operations dm_rq_blk_dops;
1669 static const struct dax_operations dm_dax_ops;
1671 static void dm_wq_work(struct work_struct *work);
1673 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1674 static void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1676 dm_destroy_keyslot_manager(q->ksm);
1679 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1681 static inline void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1684 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1686 static void cleanup_mapped_device(struct mapped_device *md)
1689 destroy_workqueue(md->wq);
1690 bioset_exit(&md->bs);
1691 bioset_exit(&md->io_bs);
1694 kill_dax(md->dax_dev);
1695 put_dax(md->dax_dev);
1700 spin_lock(&_minor_lock);
1701 md->disk->private_data = NULL;
1702 spin_unlock(&_minor_lock);
1703 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1705 del_gendisk(md->disk);
1707 dm_queue_destroy_keyslot_manager(md->queue);
1708 blk_cleanup_disk(md->disk);
1711 cleanup_srcu_struct(&md->io_barrier);
1713 mutex_destroy(&md->suspend_lock);
1714 mutex_destroy(&md->type_lock);
1715 mutex_destroy(&md->table_devices_lock);
1716 mutex_destroy(&md->swap_bios_lock);
1718 dm_mq_cleanup_mapped_device(md);
1719 dm_cleanup_zoned_dev(md);
1723 * Allocate and initialise a blank device with a given minor.
1725 static struct mapped_device *alloc_dev(int minor)
1727 int r, numa_node_id = dm_get_numa_node();
1728 struct mapped_device *md;
1731 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1733 DMWARN("unable to allocate device, out of memory.");
1737 if (!try_module_get(THIS_MODULE))
1738 goto bad_module_get;
1740 /* get a minor number for the dev */
1741 if (minor == DM_ANY_MINOR)
1742 r = next_free_minor(&minor);
1744 r = specific_minor(minor);
1748 r = init_srcu_struct(&md->io_barrier);
1750 goto bad_io_barrier;
1752 md->numa_node_id = numa_node_id;
1753 md->init_tio_pdu = false;
1754 md->type = DM_TYPE_NONE;
1755 mutex_init(&md->suspend_lock);
1756 mutex_init(&md->type_lock);
1757 mutex_init(&md->table_devices_lock);
1758 spin_lock_init(&md->deferred_lock);
1759 atomic_set(&md->holders, 1);
1760 atomic_set(&md->open_count, 0);
1761 atomic_set(&md->event_nr, 0);
1762 atomic_set(&md->uevent_seq, 0);
1763 INIT_LIST_HEAD(&md->uevent_list);
1764 INIT_LIST_HEAD(&md->table_devices);
1765 spin_lock_init(&md->uevent_lock);
1768 * default to bio-based until DM table is loaded and md->type
1769 * established. If request-based table is loaded: blk-mq will
1770 * override accordingly.
1772 md->disk = blk_alloc_disk(md->numa_node_id);
1775 md->queue = md->disk->queue;
1777 init_waitqueue_head(&md->wait);
1778 INIT_WORK(&md->work, dm_wq_work);
1779 init_waitqueue_head(&md->eventq);
1780 init_completion(&md->kobj_holder.completion);
1782 md->swap_bios = get_swap_bios();
1783 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1784 mutex_init(&md->swap_bios_lock);
1786 md->disk->major = _major;
1787 md->disk->first_minor = minor;
1788 md->disk->minors = 1;
1789 md->disk->fops = &dm_blk_dops;
1790 md->disk->queue = md->queue;
1791 md->disk->private_data = md;
1792 sprintf(md->disk->disk_name, "dm-%d", minor);
1794 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1795 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1797 if (IS_ERR(md->dax_dev))
1801 format_dev_t(md->name, MKDEV(_major, minor));
1803 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1807 dm_stats_init(&md->stats);
1809 /* Populate the mapping, nobody knows we exist yet */
1810 spin_lock(&_minor_lock);
1811 old_md = idr_replace(&_minor_idr, md, minor);
1812 spin_unlock(&_minor_lock);
1814 BUG_ON(old_md != MINOR_ALLOCED);
1819 cleanup_mapped_device(md);
1823 module_put(THIS_MODULE);
1829 static void unlock_fs(struct mapped_device *md);
1831 static void free_dev(struct mapped_device *md)
1833 int minor = MINOR(disk_devt(md->disk));
1837 cleanup_mapped_device(md);
1839 free_table_devices(&md->table_devices);
1840 dm_stats_cleanup(&md->stats);
1843 module_put(THIS_MODULE);
1847 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1849 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1852 if (dm_table_bio_based(t)) {
1854 * The md may already have mempools that need changing.
1855 * If so, reload bioset because front_pad may have changed
1856 * because a different table was loaded.
1858 bioset_exit(&md->bs);
1859 bioset_exit(&md->io_bs);
1861 } else if (bioset_initialized(&md->bs)) {
1863 * There's no need to reload with request-based dm
1864 * because the size of front_pad doesn't change.
1865 * Note for future: If you are to reload bioset,
1866 * prep-ed requests in the queue may refer
1867 * to bio from the old bioset, so you must walk
1868 * through the queue to unprep.
1874 bioset_initialized(&md->bs) ||
1875 bioset_initialized(&md->io_bs));
1877 ret = bioset_init_from_src(&md->bs, &p->bs);
1880 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1882 bioset_exit(&md->bs);
1884 /* mempool bind completed, no longer need any mempools in the table */
1885 dm_table_free_md_mempools(t);
1890 * Bind a table to the device.
1892 static void event_callback(void *context)
1894 unsigned long flags;
1896 struct mapped_device *md = (struct mapped_device *) context;
1898 spin_lock_irqsave(&md->uevent_lock, flags);
1899 list_splice_init(&md->uevent_list, &uevents);
1900 spin_unlock_irqrestore(&md->uevent_lock, flags);
1902 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1904 atomic_inc(&md->event_nr);
1905 wake_up(&md->eventq);
1906 dm_issue_global_event();
1910 * Returns old map, which caller must destroy.
1912 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1913 struct queue_limits *limits)
1915 struct dm_table *old_map;
1916 struct request_queue *q = md->queue;
1917 bool request_based = dm_table_request_based(t);
1921 lockdep_assert_held(&md->suspend_lock);
1923 size = dm_table_get_size(t);
1926 * Wipe any geometry if the size of the table changed.
1928 if (size != dm_get_size(md))
1929 memset(&md->geometry, 0, sizeof(md->geometry));
1931 if (!get_capacity(md->disk))
1932 set_capacity(md->disk, size);
1934 set_capacity_and_notify(md->disk, size);
1936 dm_table_event_callback(t, event_callback, md);
1939 * The queue hasn't been stopped yet, if the old table type wasn't
1940 * for request-based during suspension. So stop it to prevent
1941 * I/O mapping before resume.
1942 * This must be done before setting the queue restrictions,
1943 * because request-based dm may be run just after the setting.
1948 if (request_based) {
1950 * Leverage the fact that request-based DM targets are
1951 * immutable singletons - used to optimize dm_mq_queue_rq.
1953 md->immutable_target = dm_table_get_immutable_target(t);
1956 ret = __bind_mempools(md, t);
1958 old_map = ERR_PTR(ret);
1962 ret = dm_table_set_restrictions(t, q, limits);
1964 old_map = ERR_PTR(ret);
1968 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1969 rcu_assign_pointer(md->map, (void *)t);
1970 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1980 * Returns unbound table for the caller to free.
1982 static struct dm_table *__unbind(struct mapped_device *md)
1984 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1989 dm_table_event_callback(map, NULL, NULL);
1990 RCU_INIT_POINTER(md->map, NULL);
1997 * Constructor for a new device.
1999 int dm_create(int minor, struct mapped_device **result)
2001 struct mapped_device *md;
2003 md = alloc_dev(minor);
2007 dm_ima_reset_data(md);
2014 * Functions to manage md->type.
2015 * All are required to hold md->type_lock.
2017 void dm_lock_md_type(struct mapped_device *md)
2019 mutex_lock(&md->type_lock);
2022 void dm_unlock_md_type(struct mapped_device *md)
2024 mutex_unlock(&md->type_lock);
2027 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2029 BUG_ON(!mutex_is_locked(&md->type_lock));
2033 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2038 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2040 return md->immutable_target_type;
2044 * The queue_limits are only valid as long as you have a reference
2047 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2049 BUG_ON(!atomic_read(&md->holders));
2050 return &md->queue->limits;
2052 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2055 * Setup the DM device's queue based on md's type
2057 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2059 enum dm_queue_mode type = dm_table_get_type(t);
2060 struct queue_limits limits;
2064 case DM_TYPE_REQUEST_BASED:
2065 md->disk->fops = &dm_rq_blk_dops;
2066 r = dm_mq_init_request_queue(md, t);
2068 DMERR("Cannot initialize queue for request-based dm mapped device");
2072 case DM_TYPE_BIO_BASED:
2073 case DM_TYPE_DAX_BIO_BASED:
2080 r = dm_calculate_queue_limits(t, &limits);
2082 DMERR("Cannot calculate initial queue limits");
2085 r = dm_table_set_restrictions(t, md->queue, &limits);
2089 r = add_disk(md->disk);
2093 r = dm_sysfs_init(md);
2095 del_gendisk(md->disk);
2102 struct mapped_device *dm_get_md(dev_t dev)
2104 struct mapped_device *md;
2105 unsigned minor = MINOR(dev);
2107 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2110 spin_lock(&_minor_lock);
2112 md = idr_find(&_minor_idr, minor);
2113 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2114 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2120 spin_unlock(&_minor_lock);
2124 EXPORT_SYMBOL_GPL(dm_get_md);
2126 void *dm_get_mdptr(struct mapped_device *md)
2128 return md->interface_ptr;
2131 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2133 md->interface_ptr = ptr;
2136 void dm_get(struct mapped_device *md)
2138 atomic_inc(&md->holders);
2139 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2142 int dm_hold(struct mapped_device *md)
2144 spin_lock(&_minor_lock);
2145 if (test_bit(DMF_FREEING, &md->flags)) {
2146 spin_unlock(&_minor_lock);
2150 spin_unlock(&_minor_lock);
2153 EXPORT_SYMBOL_GPL(dm_hold);
2155 const char *dm_device_name(struct mapped_device *md)
2159 EXPORT_SYMBOL_GPL(dm_device_name);
2161 static void __dm_destroy(struct mapped_device *md, bool wait)
2163 struct dm_table *map;
2168 spin_lock(&_minor_lock);
2169 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2170 set_bit(DMF_FREEING, &md->flags);
2171 spin_unlock(&_minor_lock);
2173 blk_set_queue_dying(md->queue);
2176 * Take suspend_lock so that presuspend and postsuspend methods
2177 * do not race with internal suspend.
2179 mutex_lock(&md->suspend_lock);
2180 map = dm_get_live_table(md, &srcu_idx);
2181 if (!dm_suspended_md(md)) {
2182 dm_table_presuspend_targets(map);
2183 set_bit(DMF_SUSPENDED, &md->flags);
2184 set_bit(DMF_POST_SUSPENDING, &md->flags);
2185 dm_table_postsuspend_targets(map);
2187 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2188 dm_put_live_table(md, srcu_idx);
2189 mutex_unlock(&md->suspend_lock);
2192 * Rare, but there may be I/O requests still going to complete,
2193 * for example. Wait for all references to disappear.
2194 * No one should increment the reference count of the mapped_device,
2195 * after the mapped_device state becomes DMF_FREEING.
2198 while (atomic_read(&md->holders))
2200 else if (atomic_read(&md->holders))
2201 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2202 dm_device_name(md), atomic_read(&md->holders));
2204 dm_table_destroy(__unbind(md));
2208 void dm_destroy(struct mapped_device *md)
2210 __dm_destroy(md, true);
2213 void dm_destroy_immediate(struct mapped_device *md)
2215 __dm_destroy(md, false);
2218 void dm_put(struct mapped_device *md)
2220 atomic_dec(&md->holders);
2222 EXPORT_SYMBOL_GPL(dm_put);
2224 static bool md_in_flight_bios(struct mapped_device *md)
2227 struct block_device *part = dm_disk(md)->part0;
2230 for_each_possible_cpu(cpu) {
2231 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2232 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2238 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2244 prepare_to_wait(&md->wait, &wait, task_state);
2246 if (!md_in_flight_bios(md))
2249 if (signal_pending_state(task_state, current)) {
2256 finish_wait(&md->wait, &wait);
2261 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2265 if (!queue_is_mq(md->queue))
2266 return dm_wait_for_bios_completion(md, task_state);
2269 if (!blk_mq_queue_inflight(md->queue))
2272 if (signal_pending_state(task_state, current)) {
2284 * Process the deferred bios
2286 static void dm_wq_work(struct work_struct *work)
2288 struct mapped_device *md = container_of(work, struct mapped_device, work);
2291 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2292 spin_lock_irq(&md->deferred_lock);
2293 bio = bio_list_pop(&md->deferred);
2294 spin_unlock_irq(&md->deferred_lock);
2299 submit_bio_noacct(bio);
2303 static void dm_queue_flush(struct mapped_device *md)
2305 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2306 smp_mb__after_atomic();
2307 queue_work(md->wq, &md->work);
2311 * Swap in a new table, returning the old one for the caller to destroy.
2313 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2315 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2316 struct queue_limits limits;
2319 mutex_lock(&md->suspend_lock);
2321 /* device must be suspended */
2322 if (!dm_suspended_md(md))
2326 * If the new table has no data devices, retain the existing limits.
2327 * This helps multipath with queue_if_no_path if all paths disappear,
2328 * then new I/O is queued based on these limits, and then some paths
2331 if (dm_table_has_no_data_devices(table)) {
2332 live_map = dm_get_live_table_fast(md);
2334 limits = md->queue->limits;
2335 dm_put_live_table_fast(md);
2339 r = dm_calculate_queue_limits(table, &limits);
2346 map = __bind(md, table, &limits);
2347 dm_issue_global_event();
2350 mutex_unlock(&md->suspend_lock);
2355 * Functions to lock and unlock any filesystem running on the
2358 static int lock_fs(struct mapped_device *md)
2362 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2364 r = freeze_bdev(md->disk->part0);
2366 set_bit(DMF_FROZEN, &md->flags);
2370 static void unlock_fs(struct mapped_device *md)
2372 if (!test_bit(DMF_FROZEN, &md->flags))
2374 thaw_bdev(md->disk->part0);
2375 clear_bit(DMF_FROZEN, &md->flags);
2379 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2380 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2381 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2383 * If __dm_suspend returns 0, the device is completely quiescent
2384 * now. There is no request-processing activity. All new requests
2385 * are being added to md->deferred list.
2387 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2388 unsigned suspend_flags, unsigned int task_state,
2389 int dmf_suspended_flag)
2391 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2392 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2395 lockdep_assert_held(&md->suspend_lock);
2398 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2399 * This flag is cleared before dm_suspend returns.
2402 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2404 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2407 * This gets reverted if there's an error later and the targets
2408 * provide the .presuspend_undo hook.
2410 dm_table_presuspend_targets(map);
2413 * Flush I/O to the device.
2414 * Any I/O submitted after lock_fs() may not be flushed.
2415 * noflush takes precedence over do_lockfs.
2416 * (lock_fs() flushes I/Os and waits for them to complete.)
2418 if (!noflush && do_lockfs) {
2421 dm_table_presuspend_undo_targets(map);
2427 * Here we must make sure that no processes are submitting requests
2428 * to target drivers i.e. no one may be executing
2429 * __split_and_process_bio from dm_submit_bio.
2431 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2432 * we take the write lock. To prevent any process from reentering
2433 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2434 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2435 * flush_workqueue(md->wq).
2437 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2439 synchronize_srcu(&md->io_barrier);
2442 * Stop md->queue before flushing md->wq in case request-based
2443 * dm defers requests to md->wq from md->queue.
2445 if (dm_request_based(md))
2446 dm_stop_queue(md->queue);
2448 flush_workqueue(md->wq);
2451 * At this point no more requests are entering target request routines.
2452 * We call dm_wait_for_completion to wait for all existing requests
2455 r = dm_wait_for_completion(md, task_state);
2457 set_bit(dmf_suspended_flag, &md->flags);
2460 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2462 synchronize_srcu(&md->io_barrier);
2464 /* were we interrupted ? */
2468 if (dm_request_based(md))
2469 dm_start_queue(md->queue);
2472 dm_table_presuspend_undo_targets(map);
2473 /* pushback list is already flushed, so skip flush */
2480 * We need to be able to change a mapping table under a mounted
2481 * filesystem. For example we might want to move some data in
2482 * the background. Before the table can be swapped with
2483 * dm_bind_table, dm_suspend must be called to flush any in
2484 * flight bios and ensure that any further io gets deferred.
2487 * Suspend mechanism in request-based dm.
2489 * 1. Flush all I/Os by lock_fs() if needed.
2490 * 2. Stop dispatching any I/O by stopping the request_queue.
2491 * 3. Wait for all in-flight I/Os to be completed or requeued.
2493 * To abort suspend, start the request_queue.
2495 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2497 struct dm_table *map = NULL;
2501 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2503 if (dm_suspended_md(md)) {
2508 if (dm_suspended_internally_md(md)) {
2509 /* already internally suspended, wait for internal resume */
2510 mutex_unlock(&md->suspend_lock);
2511 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2517 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2519 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2523 set_bit(DMF_POST_SUSPENDING, &md->flags);
2524 dm_table_postsuspend_targets(map);
2525 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2528 mutex_unlock(&md->suspend_lock);
2532 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2535 int r = dm_table_resume_targets(map);
2543 * Flushing deferred I/Os must be done after targets are resumed
2544 * so that mapping of targets can work correctly.
2545 * Request-based dm is queueing the deferred I/Os in its request_queue.
2547 if (dm_request_based(md))
2548 dm_start_queue(md->queue);
2555 int dm_resume(struct mapped_device *md)
2558 struct dm_table *map = NULL;
2562 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2564 if (!dm_suspended_md(md))
2567 if (dm_suspended_internally_md(md)) {
2568 /* already internally suspended, wait for internal resume */
2569 mutex_unlock(&md->suspend_lock);
2570 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2576 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2577 if (!map || !dm_table_get_size(map))
2580 r = __dm_resume(md, map);
2584 clear_bit(DMF_SUSPENDED, &md->flags);
2586 mutex_unlock(&md->suspend_lock);
2592 * Internal suspend/resume works like userspace-driven suspend. It waits
2593 * until all bios finish and prevents issuing new bios to the target drivers.
2594 * It may be used only from the kernel.
2597 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2599 struct dm_table *map = NULL;
2601 lockdep_assert_held(&md->suspend_lock);
2603 if (md->internal_suspend_count++)
2604 return; /* nested internal suspend */
2606 if (dm_suspended_md(md)) {
2607 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2608 return; /* nest suspend */
2611 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2614 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2615 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2616 * would require changing .presuspend to return an error -- avoid this
2617 * until there is a need for more elaborate variants of internal suspend.
2619 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2620 DMF_SUSPENDED_INTERNALLY);
2622 set_bit(DMF_POST_SUSPENDING, &md->flags);
2623 dm_table_postsuspend_targets(map);
2624 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2627 static void __dm_internal_resume(struct mapped_device *md)
2629 BUG_ON(!md->internal_suspend_count);
2631 if (--md->internal_suspend_count)
2632 return; /* resume from nested internal suspend */
2634 if (dm_suspended_md(md))
2635 goto done; /* resume from nested suspend */
2638 * NOTE: existing callers don't need to call dm_table_resume_targets
2639 * (which may fail -- so best to avoid it for now by passing NULL map)
2641 (void) __dm_resume(md, NULL);
2644 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2645 smp_mb__after_atomic();
2646 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2649 void dm_internal_suspend_noflush(struct mapped_device *md)
2651 mutex_lock(&md->suspend_lock);
2652 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2653 mutex_unlock(&md->suspend_lock);
2655 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2657 void dm_internal_resume(struct mapped_device *md)
2659 mutex_lock(&md->suspend_lock);
2660 __dm_internal_resume(md);
2661 mutex_unlock(&md->suspend_lock);
2663 EXPORT_SYMBOL_GPL(dm_internal_resume);
2666 * Fast variants of internal suspend/resume hold md->suspend_lock,
2667 * which prevents interaction with userspace-driven suspend.
2670 void dm_internal_suspend_fast(struct mapped_device *md)
2672 mutex_lock(&md->suspend_lock);
2673 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2676 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2677 synchronize_srcu(&md->io_barrier);
2678 flush_workqueue(md->wq);
2679 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2681 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2683 void dm_internal_resume_fast(struct mapped_device *md)
2685 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2691 mutex_unlock(&md->suspend_lock);
2693 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2695 /*-----------------------------------------------------------------
2696 * Event notification.
2697 *---------------------------------------------------------------*/
2698 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2703 char udev_cookie[DM_COOKIE_LENGTH];
2704 char *envp[] = { udev_cookie, NULL };
2706 noio_flag = memalloc_noio_save();
2709 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2711 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2712 DM_COOKIE_ENV_VAR_NAME, cookie);
2713 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2717 memalloc_noio_restore(noio_flag);
2722 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2724 return atomic_add_return(1, &md->uevent_seq);
2727 uint32_t dm_get_event_nr(struct mapped_device *md)
2729 return atomic_read(&md->event_nr);
2732 int dm_wait_event(struct mapped_device *md, int event_nr)
2734 return wait_event_interruptible(md->eventq,
2735 (event_nr != atomic_read(&md->event_nr)));
2738 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2740 unsigned long flags;
2742 spin_lock_irqsave(&md->uevent_lock, flags);
2743 list_add(elist, &md->uevent_list);
2744 spin_unlock_irqrestore(&md->uevent_lock, flags);
2748 * The gendisk is only valid as long as you have a reference
2751 struct gendisk *dm_disk(struct mapped_device *md)
2755 EXPORT_SYMBOL_GPL(dm_disk);
2757 struct kobject *dm_kobject(struct mapped_device *md)
2759 return &md->kobj_holder.kobj;
2762 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2764 struct mapped_device *md;
2766 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2768 spin_lock(&_minor_lock);
2769 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2775 spin_unlock(&_minor_lock);
2780 int dm_suspended_md(struct mapped_device *md)
2782 return test_bit(DMF_SUSPENDED, &md->flags);
2785 static int dm_post_suspending_md(struct mapped_device *md)
2787 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2790 int dm_suspended_internally_md(struct mapped_device *md)
2792 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2795 int dm_test_deferred_remove_flag(struct mapped_device *md)
2797 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2800 int dm_suspended(struct dm_target *ti)
2802 return dm_suspended_md(ti->table->md);
2804 EXPORT_SYMBOL_GPL(dm_suspended);
2806 int dm_post_suspending(struct dm_target *ti)
2808 return dm_post_suspending_md(ti->table->md);
2810 EXPORT_SYMBOL_GPL(dm_post_suspending);
2812 int dm_noflush_suspending(struct dm_target *ti)
2814 return __noflush_suspending(ti->table->md);
2816 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2818 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2819 unsigned integrity, unsigned per_io_data_size,
2820 unsigned min_pool_size)
2822 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2823 unsigned int pool_size = 0;
2824 unsigned int front_pad, io_front_pad;
2831 case DM_TYPE_BIO_BASED:
2832 case DM_TYPE_DAX_BIO_BASED:
2833 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2834 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2835 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2836 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2839 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2842 case DM_TYPE_REQUEST_BASED:
2843 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2844 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2845 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2851 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2855 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2861 dm_free_md_mempools(pools);
2866 void dm_free_md_mempools(struct dm_md_mempools *pools)
2871 bioset_exit(&pools->bs);
2872 bioset_exit(&pools->io_bs);
2884 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2887 struct mapped_device *md = bdev->bd_disk->private_data;
2888 struct dm_table *table;
2889 struct dm_target *ti;
2890 int ret = -ENOTTY, srcu_idx;
2892 table = dm_get_live_table(md, &srcu_idx);
2893 if (!table || !dm_table_get_size(table))
2896 /* We only support devices that have a single target */
2897 if (dm_table_get_num_targets(table) != 1)
2899 ti = dm_table_get_target(table, 0);
2902 if (!ti->type->iterate_devices)
2905 ret = ti->type->iterate_devices(ti, fn, data);
2907 dm_put_live_table(md, srcu_idx);
2912 * For register / unregister we need to manually call out to every path.
2914 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2915 sector_t start, sector_t len, void *data)
2917 struct dm_pr *pr = data;
2918 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2920 if (!ops || !ops->pr_register)
2922 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2925 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2936 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2937 if (ret && new_key) {
2938 /* unregister all paths if we failed to register any path */
2939 pr.old_key = new_key;
2942 pr.fail_early = false;
2943 dm_call_pr(bdev, __dm_pr_register, &pr);
2949 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2952 struct mapped_device *md = bdev->bd_disk->private_data;
2953 const struct pr_ops *ops;
2956 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2960 ops = bdev->bd_disk->fops->pr_ops;
2961 if (ops && ops->pr_reserve)
2962 r = ops->pr_reserve(bdev, key, type, flags);
2966 dm_unprepare_ioctl(md, srcu_idx);
2970 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2972 struct mapped_device *md = bdev->bd_disk->private_data;
2973 const struct pr_ops *ops;
2976 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2980 ops = bdev->bd_disk->fops->pr_ops;
2981 if (ops && ops->pr_release)
2982 r = ops->pr_release(bdev, key, type);
2986 dm_unprepare_ioctl(md, srcu_idx);
2990 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2991 enum pr_type type, bool abort)
2993 struct mapped_device *md = bdev->bd_disk->private_data;
2994 const struct pr_ops *ops;
2997 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3001 ops = bdev->bd_disk->fops->pr_ops;
3002 if (ops && ops->pr_preempt)
3003 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3007 dm_unprepare_ioctl(md, srcu_idx);
3011 static int dm_pr_clear(struct block_device *bdev, u64 key)
3013 struct mapped_device *md = bdev->bd_disk->private_data;
3014 const struct pr_ops *ops;
3017 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3021 ops = bdev->bd_disk->fops->pr_ops;
3022 if (ops && ops->pr_clear)
3023 r = ops->pr_clear(bdev, key);
3027 dm_unprepare_ioctl(md, srcu_idx);
3031 static const struct pr_ops dm_pr_ops = {
3032 .pr_register = dm_pr_register,
3033 .pr_reserve = dm_pr_reserve,
3034 .pr_release = dm_pr_release,
3035 .pr_preempt = dm_pr_preempt,
3036 .pr_clear = dm_pr_clear,
3039 static const struct block_device_operations dm_blk_dops = {
3040 .submit_bio = dm_submit_bio,
3041 .open = dm_blk_open,
3042 .release = dm_blk_close,
3043 .ioctl = dm_blk_ioctl,
3044 .getgeo = dm_blk_getgeo,
3045 .report_zones = dm_blk_report_zones,
3046 .pr_ops = &dm_pr_ops,
3047 .owner = THIS_MODULE
3050 static const struct block_device_operations dm_rq_blk_dops = {
3051 .open = dm_blk_open,
3052 .release = dm_blk_close,
3053 .ioctl = dm_blk_ioctl,
3054 .getgeo = dm_blk_getgeo,
3055 .pr_ops = &dm_pr_ops,
3056 .owner = THIS_MODULE
3059 static const struct dax_operations dm_dax_ops = {
3060 .direct_access = dm_dax_direct_access,
3061 .dax_supported = dm_dax_supported,
3062 .copy_from_iter = dm_dax_copy_from_iter,
3063 .copy_to_iter = dm_dax_copy_to_iter,
3064 .zero_page_range = dm_dax_zero_page_range,
3070 module_init(dm_init);
3071 module_exit(dm_exit);
3073 module_param(major, uint, 0);
3074 MODULE_PARM_DESC(major, "The major number of the device mapper");
3076 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3077 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3079 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3080 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3082 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3083 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3085 MODULE_DESCRIPTION(DM_NAME " driver");
3086 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3087 MODULE_LICENSE("GPL");