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 bool bio_is_flush_with_data(struct bio *bio)
489 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
492 static void dm_io_acct(bool end, struct mapped_device *md, struct bio *bio,
493 unsigned long start_time, struct dm_stats_aux *stats_aux)
495 bool is_flush_with_data;
496 unsigned int bi_size;
498 /* If REQ_PREFLUSH set save any payload but do not account it */
499 is_flush_with_data = bio_is_flush_with_data(bio);
500 if (is_flush_with_data) {
501 bi_size = bio->bi_iter.bi_size;
502 bio->bi_iter.bi_size = 0;
506 bio_start_io_acct_time(bio, start_time);
508 bio_end_io_acct(bio, start_time);
510 if (unlikely(dm_stats_used(&md->stats)))
511 dm_stats_account_io(&md->stats, bio_data_dir(bio),
512 bio->bi_iter.bi_sector, bio_sectors(bio),
513 end, start_time, stats_aux);
515 /* Restore bio's payload so it does get accounted upon requeue */
516 if (is_flush_with_data)
517 bio->bi_iter.bi_size = bi_size;
520 static void start_io_acct(struct dm_io *io)
522 dm_io_acct(false, io->md, io->orig_bio, io->start_time, &io->stats_aux);
525 static void end_io_acct(struct mapped_device *md, struct bio *bio,
526 unsigned long start_time, struct dm_stats_aux *stats_aux)
528 dm_io_acct(true, md, bio, start_time, stats_aux);
531 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
534 struct dm_target_io *tio;
537 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
541 tio = container_of(clone, struct dm_target_io, clone);
542 tio->inside_dm_io = true;
545 io = container_of(tio, struct dm_io, tio);
546 io->magic = DM_IO_MAGIC;
548 atomic_set(&io->io_count, 1);
549 this_cpu_inc(*md->pending_io);
552 spin_lock_init(&io->endio_lock);
554 io->start_time = jiffies;
556 dm_stats_record_start(&md->stats, &io->stats_aux);
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);
599 * Add the bio to the list of deferred io.
601 static void queue_io(struct mapped_device *md, struct bio *bio)
605 spin_lock_irqsave(&md->deferred_lock, flags);
606 bio_list_add(&md->deferred, bio);
607 spin_unlock_irqrestore(&md->deferred_lock, flags);
608 queue_work(md->wq, &md->work);
612 * Everyone (including functions in this file), should use this
613 * function to access the md->map field, and make sure they call
614 * dm_put_live_table() when finished.
616 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
618 *srcu_idx = srcu_read_lock(&md->io_barrier);
620 return srcu_dereference(md->map, &md->io_barrier);
623 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
625 srcu_read_unlock(&md->io_barrier, srcu_idx);
628 void dm_sync_table(struct mapped_device *md)
630 synchronize_srcu(&md->io_barrier);
631 synchronize_rcu_expedited();
635 * A fast alternative to dm_get_live_table/dm_put_live_table.
636 * The caller must not block between these two functions.
638 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
641 return rcu_dereference(md->map);
644 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
649 static char *_dm_claim_ptr = "I belong to device-mapper";
652 * Open a table device so we can use it as a map destination.
654 static int open_table_device(struct table_device *td, dev_t dev,
655 struct mapped_device *md)
657 struct block_device *bdev;
661 BUG_ON(td->dm_dev.bdev);
663 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
665 return PTR_ERR(bdev);
667 r = bd_link_disk_holder(bdev, dm_disk(md));
669 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
673 td->dm_dev.bdev = bdev;
674 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev);
679 * Close a table device that we've been using.
681 static void close_table_device(struct table_device *td, struct mapped_device *md)
683 if (!td->dm_dev.bdev)
686 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
687 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
688 put_dax(td->dm_dev.dax_dev);
689 td->dm_dev.bdev = NULL;
690 td->dm_dev.dax_dev = NULL;
693 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
696 struct table_device *td;
698 list_for_each_entry(td, l, list)
699 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
705 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
706 struct dm_dev **result)
709 struct table_device *td;
711 mutex_lock(&md->table_devices_lock);
712 td = find_table_device(&md->table_devices, dev, mode);
714 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
716 mutex_unlock(&md->table_devices_lock);
720 td->dm_dev.mode = mode;
721 td->dm_dev.bdev = NULL;
723 if ((r = open_table_device(td, dev, md))) {
724 mutex_unlock(&md->table_devices_lock);
729 format_dev_t(td->dm_dev.name, dev);
731 refcount_set(&td->count, 1);
732 list_add(&td->list, &md->table_devices);
734 refcount_inc(&td->count);
736 mutex_unlock(&md->table_devices_lock);
738 *result = &td->dm_dev;
742 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
744 struct table_device *td = container_of(d, struct table_device, dm_dev);
746 mutex_lock(&md->table_devices_lock);
747 if (refcount_dec_and_test(&td->count)) {
748 close_table_device(td, md);
752 mutex_unlock(&md->table_devices_lock);
755 static void free_table_devices(struct list_head *devices)
757 struct list_head *tmp, *next;
759 list_for_each_safe(tmp, next, devices) {
760 struct table_device *td = list_entry(tmp, struct table_device, list);
762 DMWARN("dm_destroy: %s still exists with %d references",
763 td->dm_dev.name, refcount_read(&td->count));
769 * Get the geometry associated with a dm device
771 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
779 * Set the geometry of a device.
781 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
783 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
785 if (geo->start > sz) {
786 DMWARN("Start sector is beyond the geometry limits.");
795 static int __noflush_suspending(struct mapped_device *md)
797 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
801 * Decrements the number of outstanding ios that a bio has been
802 * cloned into, completing the original io if necc.
804 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
807 blk_status_t io_error;
809 struct mapped_device *md = io->md;
810 unsigned long start_time = 0;
811 struct dm_stats_aux stats_aux;
813 /* Push-back supersedes any I/O errors */
814 if (unlikely(error)) {
815 spin_lock_irqsave(&io->endio_lock, flags);
816 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
818 spin_unlock_irqrestore(&io->endio_lock, flags);
821 if (atomic_dec_and_test(&io->io_count)) {
823 if (io->status == BLK_STS_DM_REQUEUE) {
825 * Target requested pushing back the I/O.
827 spin_lock_irqsave(&md->deferred_lock, flags);
828 if (__noflush_suspending(md) &&
829 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
830 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
831 bio_list_add_head(&md->deferred, bio);
834 * noflush suspend was interrupted or this is
835 * a write to a zoned target.
837 io->status = BLK_STS_IOERR;
839 spin_unlock_irqrestore(&md->deferred_lock, flags);
842 io_error = io->status;
843 start_time = io->start_time;
844 stats_aux = io->stats_aux;
846 end_io_acct(md, bio, start_time, &stats_aux);
848 this_cpu_dec(*md->pending_io);
850 /* nudge anyone waiting on suspend queue */
851 if (unlikely(wq_has_sleeper(&md->wait)))
854 if (io_error == BLK_STS_DM_REQUEUE)
857 if (bio_is_flush_with_data(bio)) {
859 * Preflush done for flush with data, reissue
860 * without REQ_PREFLUSH.
862 bio->bi_opf &= ~REQ_PREFLUSH;
865 /* done with normal IO or empty flush */
867 bio->bi_status = io_error;
873 void disable_discard(struct mapped_device *md)
875 struct queue_limits *limits = dm_get_queue_limits(md);
877 /* device doesn't really support DISCARD, disable it */
878 limits->max_discard_sectors = 0;
879 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
882 void disable_write_same(struct mapped_device *md)
884 struct queue_limits *limits = dm_get_queue_limits(md);
886 /* device doesn't really support WRITE SAME, disable it */
887 limits->max_write_same_sectors = 0;
890 void disable_write_zeroes(struct mapped_device *md)
892 struct queue_limits *limits = dm_get_queue_limits(md);
894 /* device doesn't really support WRITE ZEROES, disable it */
895 limits->max_write_zeroes_sectors = 0;
898 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
900 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
903 static void clone_endio(struct bio *bio)
905 blk_status_t error = bio->bi_status;
906 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
907 struct dm_io *io = tio->io;
908 struct mapped_device *md = tio->io->md;
909 dm_endio_fn endio = tio->ti->type->end_io;
910 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
912 if (unlikely(error == BLK_STS_TARGET)) {
913 if (bio_op(bio) == REQ_OP_DISCARD &&
914 !q->limits.max_discard_sectors)
916 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
917 !q->limits.max_write_same_sectors)
918 disable_write_same(md);
919 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
920 !q->limits.max_write_zeroes_sectors)
921 disable_write_zeroes(md);
924 if (blk_queue_is_zoned(q))
925 dm_zone_endio(io, bio);
928 int r = endio(tio->ti, bio, &error);
930 case DM_ENDIO_REQUEUE:
932 * Requeuing writes to a sequential zone of a zoned
933 * target will break the sequential write pattern:
936 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
937 error = BLK_STS_IOERR;
939 error = BLK_STS_DM_REQUEUE;
943 case DM_ENDIO_INCOMPLETE:
944 /* The target will handle the io */
947 DMWARN("unimplemented target endio return value: %d", r);
952 if (unlikely(swap_bios_limit(tio->ti, bio))) {
953 struct mapped_device *md = io->md;
954 up(&md->swap_bios_semaphore);
958 dm_io_dec_pending(io, error);
962 * Return maximum size of I/O possible at the supplied sector up to the current
965 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
966 sector_t target_offset)
968 return ti->len - target_offset;
971 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
973 sector_t target_offset = dm_target_offset(ti, sector);
974 sector_t len = max_io_len_target_boundary(ti, target_offset);
978 * Does the target need to split IO even further?
979 * - varied (per target) IO splitting is a tenet of DM; this
980 * explains why stacked chunk_sectors based splitting via
981 * blk_max_size_offset() isn't possible here. So pass in
982 * ti->max_io_len to override stacked chunk_sectors.
984 if (ti->max_io_len) {
985 max_len = blk_max_size_offset(ti->table->md->queue,
986 target_offset, ti->max_io_len);
994 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
996 if (len > UINT_MAX) {
997 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
998 (unsigned long long)len, UINT_MAX);
999 ti->error = "Maximum size of target IO is too large";
1003 ti->max_io_len = (uint32_t) len;
1007 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1009 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1010 sector_t sector, int *srcu_idx)
1011 __acquires(md->io_barrier)
1013 struct dm_table *map;
1014 struct dm_target *ti;
1016 map = dm_get_live_table(md, srcu_idx);
1020 ti = dm_table_find_target(map, sector);
1027 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1028 long nr_pages, void **kaddr, pfn_t *pfn)
1030 struct mapped_device *md = dax_get_private(dax_dev);
1031 sector_t sector = pgoff * PAGE_SECTORS;
1032 struct dm_target *ti;
1033 long len, ret = -EIO;
1036 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1040 if (!ti->type->direct_access)
1042 len = max_io_len(ti, sector) / PAGE_SECTORS;
1045 nr_pages = min(len, nr_pages);
1046 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1049 dm_put_live_table(md, srcu_idx);
1054 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1055 int blocksize, sector_t start, sector_t len)
1057 struct mapped_device *md = dax_get_private(dax_dev);
1058 struct dm_table *map;
1062 map = dm_get_live_table(md, &srcu_idx);
1066 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1069 dm_put_live_table(md, srcu_idx);
1074 static size_t dm_dax_copy_from_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_from_iter) {
1088 ret = copy_from_iter(addr, bytes, i);
1091 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1093 dm_put_live_table(md, srcu_idx);
1098 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1099 void *addr, size_t bytes, struct iov_iter *i)
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 (!ti->type->dax_copy_to_iter) {
1112 ret = copy_to_iter(addr, bytes, i);
1115 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1117 dm_put_live_table(md, srcu_idx);
1122 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1125 struct mapped_device *md = dax_get_private(dax_dev);
1126 sector_t sector = pgoff * PAGE_SECTORS;
1127 struct dm_target *ti;
1131 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1135 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1137 * ->zero_page_range() is mandatory dax operation. If we are
1138 * here, something is wrong.
1142 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1144 dm_put_live_table(md, srcu_idx);
1150 * A target may call dm_accept_partial_bio only from the map routine. It is
1151 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1152 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1154 * dm_accept_partial_bio informs the dm that the target only wants to process
1155 * additional n_sectors sectors of the bio and the rest of the data should be
1156 * sent in a next bio.
1158 * A diagram that explains the arithmetics:
1159 * +--------------------+---------------+-------+
1161 * +--------------------+---------------+-------+
1163 * <-------------- *tio->len_ptr --------------->
1164 * <------- bi_size ------->
1167 * Region 1 was already iterated over with bio_advance or similar function.
1168 * (it may be empty if the target doesn't use bio_advance)
1169 * Region 2 is the remaining bio size that the target wants to process.
1170 * (it may be empty if region 1 is non-empty, although there is no reason
1172 * The target requires that region 3 is to be sent in the next bio.
1174 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1175 * the partially processed part (the sum of regions 1+2) must be the same for all
1176 * copies of the bio.
1178 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1180 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1181 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1183 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1184 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1185 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1186 BUG_ON(bi_size > *tio->len_ptr);
1187 BUG_ON(n_sectors > bi_size);
1189 *tio->len_ptr -= bi_size - n_sectors;
1190 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1192 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1194 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1196 mutex_lock(&md->swap_bios_lock);
1197 while (latch < md->swap_bios) {
1199 down(&md->swap_bios_semaphore);
1202 while (latch > md->swap_bios) {
1204 up(&md->swap_bios_semaphore);
1207 mutex_unlock(&md->swap_bios_lock);
1210 static blk_qc_t __map_bio(struct dm_target_io *tio)
1214 struct bio *clone = &tio->clone;
1215 struct dm_io *io = tio->io;
1216 struct dm_target *ti = tio->ti;
1217 blk_qc_t ret = BLK_QC_T_NONE;
1219 clone->bi_end_io = clone_endio;
1222 * Map the clone. If r == 0 we don't need to do
1223 * anything, the target has assumed ownership of
1226 dm_io_inc_pending(io);
1227 sector = clone->bi_iter.bi_sector;
1229 if (unlikely(swap_bios_limit(ti, clone))) {
1230 struct mapped_device *md = io->md;
1231 int latch = get_swap_bios();
1232 if (unlikely(latch != md->swap_bios))
1233 __set_swap_bios_limit(md, latch);
1234 down(&md->swap_bios_semaphore);
1238 * Check if the IO needs a special mapping due to zone append emulation
1239 * on zoned target. In this case, dm_zone_map_bio() calls the target
1242 if (dm_emulate_zone_append(io->md))
1243 r = dm_zone_map_bio(tio);
1245 r = ti->type->map(ti, clone);
1248 case DM_MAPIO_SUBMITTED:
1250 case DM_MAPIO_REMAPPED:
1251 /* the bio has been remapped so dispatch it */
1252 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1253 ret = submit_bio_noacct(clone);
1256 if (unlikely(swap_bios_limit(ti, clone))) {
1257 struct mapped_device *md = io->md;
1258 up(&md->swap_bios_semaphore);
1261 dm_io_dec_pending(io, BLK_STS_IOERR);
1263 case DM_MAPIO_REQUEUE:
1264 if (unlikely(swap_bios_limit(ti, clone))) {
1265 struct mapped_device *md = io->md;
1266 up(&md->swap_bios_semaphore);
1269 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1272 DMWARN("unimplemented target map return value: %d", r);
1279 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1281 bio->bi_iter.bi_sector = sector;
1282 bio->bi_iter.bi_size = to_bytes(len);
1286 * Creates a bio that consists of range of complete bvecs.
1288 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1289 sector_t sector, unsigned len)
1291 struct bio *clone = &tio->clone;
1294 __bio_clone_fast(clone, bio);
1296 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1300 if (bio_integrity(bio)) {
1301 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1302 !dm_target_passes_integrity(tio->ti->type))) {
1303 DMWARN("%s: the target %s doesn't support integrity data.",
1304 dm_device_name(tio->io->md),
1305 tio->ti->type->name);
1309 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1314 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1315 clone->bi_iter.bi_size = to_bytes(len);
1317 if (bio_integrity(bio))
1318 bio_integrity_trim(clone);
1323 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1324 struct dm_target *ti, unsigned num_bios)
1326 struct dm_target_io *tio;
1332 if (num_bios == 1) {
1333 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1334 bio_list_add(blist, &tio->clone);
1338 for (try = 0; try < 2; try++) {
1343 mutex_lock(&ci->io->md->table_devices_lock);
1344 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1345 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1349 bio_list_add(blist, &tio->clone);
1352 mutex_unlock(&ci->io->md->table_devices_lock);
1353 if (bio_nr == num_bios)
1356 while ((bio = bio_list_pop(blist))) {
1357 tio = container_of(bio, struct dm_target_io, clone);
1363 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1364 struct dm_target_io *tio, unsigned *len)
1366 struct bio *clone = &tio->clone;
1370 __bio_clone_fast(clone, ci->bio);
1372 bio_setup_sector(clone, ci->sector, *len);
1374 return __map_bio(tio);
1377 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1378 unsigned num_bios, unsigned *len)
1380 struct bio_list blist = BIO_EMPTY_LIST;
1382 struct dm_target_io *tio;
1384 alloc_multiple_bios(&blist, ci, ti, num_bios);
1386 while ((bio = bio_list_pop(&blist))) {
1387 tio = container_of(bio, struct dm_target_io, clone);
1388 (void) __clone_and_map_simple_bio(ci, tio, len);
1392 static int __send_empty_flush(struct clone_info *ci)
1394 unsigned target_nr = 0;
1395 struct dm_target *ti;
1396 struct bio flush_bio;
1399 * Use an on-stack bio for this, it's safe since we don't
1400 * need to reference it after submit. It's just used as
1401 * the basis for the clone(s).
1403 bio_init(&flush_bio, NULL, 0);
1404 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1405 bio_set_dev(&flush_bio, ci->io->md->disk->part0);
1407 ci->bio = &flush_bio;
1408 ci->sector_count = 0;
1410 BUG_ON(bio_has_data(ci->bio));
1411 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1412 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1414 bio_uninit(ci->bio);
1418 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1419 sector_t sector, unsigned *len)
1421 struct bio *bio = ci->bio;
1422 struct dm_target_io *tio;
1425 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1427 r = clone_bio(tio, bio, sector, *len);
1432 (void) __map_bio(tio);
1437 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1443 * Even though the device advertised support for this type of
1444 * request, that does not mean every target supports it, and
1445 * reconfiguration might also have changed that since the
1446 * check was performed.
1451 len = min_t(sector_t, ci->sector_count,
1452 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1454 __send_duplicate_bios(ci, ti, num_bios, &len);
1457 ci->sector_count -= len;
1462 static bool is_abnormal_io(struct bio *bio)
1466 switch (bio_op(bio)) {
1467 case REQ_OP_DISCARD:
1468 case REQ_OP_SECURE_ERASE:
1469 case REQ_OP_WRITE_SAME:
1470 case REQ_OP_WRITE_ZEROES:
1478 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1481 struct bio *bio = ci->bio;
1482 unsigned num_bios = 0;
1484 switch (bio_op(bio)) {
1485 case REQ_OP_DISCARD:
1486 num_bios = ti->num_discard_bios;
1488 case REQ_OP_SECURE_ERASE:
1489 num_bios = ti->num_secure_erase_bios;
1491 case REQ_OP_WRITE_SAME:
1492 num_bios = ti->num_write_same_bios;
1494 case REQ_OP_WRITE_ZEROES:
1495 num_bios = ti->num_write_zeroes_bios;
1501 *result = __send_changing_extent_only(ci, ti, num_bios);
1506 * Select the correct strategy for processing a non-flush bio.
1508 static int __split_and_process_non_flush(struct clone_info *ci)
1510 struct dm_target *ti;
1514 ti = dm_table_find_target(ci->map, ci->sector);
1518 if (__process_abnormal_io(ci, ti, &r))
1521 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1523 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1528 ci->sector_count -= len;
1533 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1534 struct dm_table *map, struct bio *bio)
1537 ci->io = alloc_io(md, bio);
1538 ci->sector = bio->bi_iter.bi_sector;
1542 * Entry point to split a bio into clones and submit them to the targets.
1544 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1545 struct dm_table *map, struct bio *bio)
1547 struct clone_info ci;
1548 blk_qc_t ret = BLK_QC_T_NONE;
1551 init_clone_info(&ci, md, map, bio);
1553 if (bio->bi_opf & REQ_PREFLUSH) {
1554 error = __send_empty_flush(&ci);
1555 /* dm_io_dec_pending submits any data associated with flush */
1556 } else if (op_is_zone_mgmt(bio_op(bio))) {
1558 ci.sector_count = 0;
1559 error = __split_and_process_non_flush(&ci);
1562 ci.sector_count = bio_sectors(bio);
1563 error = __split_and_process_non_flush(&ci);
1564 if (ci.sector_count && !error) {
1566 * Remainder must be passed to submit_bio_noacct()
1567 * so that it gets handled *after* bios already submitted
1568 * have been completely processed.
1569 * We take a clone of the original to store in
1570 * ci.io->orig_bio to be used by end_io_acct() and
1571 * for dec_pending to use for completion handling.
1573 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1574 GFP_NOIO, &md->queue->bio_split);
1575 ci.io->orig_bio = b;
1578 trace_block_split(b, bio->bi_iter.bi_sector);
1579 ret = submit_bio_noacct(bio);
1582 start_io_acct(ci.io);
1584 /* drop the extra reference count */
1585 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1589 static blk_qc_t dm_submit_bio(struct bio *bio)
1591 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1592 blk_qc_t ret = BLK_QC_T_NONE;
1594 struct dm_table *map;
1596 map = dm_get_live_table(md, &srcu_idx);
1598 /* If suspended, or map not yet available, queue this IO for later */
1599 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1601 if (bio->bi_opf & REQ_NOWAIT)
1602 bio_wouldblock_error(bio);
1603 else if (bio->bi_opf & REQ_RAHEAD)
1611 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1612 * otherwise associated queue_limits won't be imposed.
1614 if (is_abnormal_io(bio))
1615 blk_queue_split(&bio);
1617 ret = __split_and_process_bio(md, map, bio);
1619 dm_put_live_table(md, srcu_idx);
1623 /*-----------------------------------------------------------------
1624 * An IDR is used to keep track of allocated minor numbers.
1625 *---------------------------------------------------------------*/
1626 static void free_minor(int minor)
1628 spin_lock(&_minor_lock);
1629 idr_remove(&_minor_idr, minor);
1630 spin_unlock(&_minor_lock);
1634 * See if the device with a specific minor # is free.
1636 static int specific_minor(int minor)
1640 if (minor >= (1 << MINORBITS))
1643 idr_preload(GFP_KERNEL);
1644 spin_lock(&_minor_lock);
1646 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1648 spin_unlock(&_minor_lock);
1651 return r == -ENOSPC ? -EBUSY : r;
1655 static int next_free_minor(int *minor)
1659 idr_preload(GFP_KERNEL);
1660 spin_lock(&_minor_lock);
1662 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1664 spin_unlock(&_minor_lock);
1672 static const struct block_device_operations dm_blk_dops;
1673 static const struct block_device_operations dm_rq_blk_dops;
1674 static const struct dax_operations dm_dax_ops;
1676 static void dm_wq_work(struct work_struct *work);
1678 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1679 static void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1681 dm_destroy_keyslot_manager(q->ksm);
1684 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1686 static inline void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1689 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1691 static void cleanup_mapped_device(struct mapped_device *md)
1694 destroy_workqueue(md->wq);
1695 bioset_exit(&md->bs);
1696 bioset_exit(&md->io_bs);
1699 kill_dax(md->dax_dev);
1700 put_dax(md->dax_dev);
1704 dm_cleanup_zoned_dev(md);
1706 spin_lock(&_minor_lock);
1707 md->disk->private_data = NULL;
1708 spin_unlock(&_minor_lock);
1709 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1711 del_gendisk(md->disk);
1713 dm_queue_destroy_keyslot_manager(md->queue);
1714 blk_cleanup_disk(md->disk);
1717 if (md->pending_io) {
1718 free_percpu(md->pending_io);
1719 md->pending_io = NULL;
1722 cleanup_srcu_struct(&md->io_barrier);
1724 mutex_destroy(&md->suspend_lock);
1725 mutex_destroy(&md->type_lock);
1726 mutex_destroy(&md->table_devices_lock);
1727 mutex_destroy(&md->swap_bios_lock);
1729 dm_mq_cleanup_mapped_device(md);
1733 * Allocate and initialise a blank device with a given minor.
1735 static struct mapped_device *alloc_dev(int minor)
1737 int r, numa_node_id = dm_get_numa_node();
1738 struct mapped_device *md;
1741 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1743 DMWARN("unable to allocate device, out of memory.");
1747 if (!try_module_get(THIS_MODULE))
1748 goto bad_module_get;
1750 /* get a minor number for the dev */
1751 if (minor == DM_ANY_MINOR)
1752 r = next_free_minor(&minor);
1754 r = specific_minor(minor);
1758 r = init_srcu_struct(&md->io_barrier);
1760 goto bad_io_barrier;
1762 md->numa_node_id = numa_node_id;
1763 md->init_tio_pdu = false;
1764 md->type = DM_TYPE_NONE;
1765 mutex_init(&md->suspend_lock);
1766 mutex_init(&md->type_lock);
1767 mutex_init(&md->table_devices_lock);
1768 spin_lock_init(&md->deferred_lock);
1769 atomic_set(&md->holders, 1);
1770 atomic_set(&md->open_count, 0);
1771 atomic_set(&md->event_nr, 0);
1772 atomic_set(&md->uevent_seq, 0);
1773 INIT_LIST_HEAD(&md->uevent_list);
1774 INIT_LIST_HEAD(&md->table_devices);
1775 spin_lock_init(&md->uevent_lock);
1778 * default to bio-based until DM table is loaded and md->type
1779 * established. If request-based table is loaded: blk-mq will
1780 * override accordingly.
1782 md->disk = blk_alloc_disk(md->numa_node_id);
1785 md->queue = md->disk->queue;
1787 init_waitqueue_head(&md->wait);
1788 INIT_WORK(&md->work, dm_wq_work);
1789 init_waitqueue_head(&md->eventq);
1790 init_completion(&md->kobj_holder.completion);
1792 md->swap_bios = get_swap_bios();
1793 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1794 mutex_init(&md->swap_bios_lock);
1796 md->disk->major = _major;
1797 md->disk->first_minor = minor;
1798 md->disk->minors = 1;
1799 md->disk->fops = &dm_blk_dops;
1800 md->disk->queue = md->queue;
1801 md->disk->private_data = md;
1802 sprintf(md->disk->disk_name, "dm-%d", minor);
1804 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1805 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1807 if (IS_ERR(md->dax_dev)) {
1813 format_dev_t(md->name, MKDEV(_major, minor));
1815 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1819 md->pending_io = alloc_percpu(unsigned long);
1820 if (!md->pending_io)
1823 dm_stats_init(&md->stats);
1825 /* Populate the mapping, nobody knows we exist yet */
1826 spin_lock(&_minor_lock);
1827 old_md = idr_replace(&_minor_idr, md, minor);
1828 spin_unlock(&_minor_lock);
1830 BUG_ON(old_md != MINOR_ALLOCED);
1835 cleanup_mapped_device(md);
1839 module_put(THIS_MODULE);
1845 static void unlock_fs(struct mapped_device *md);
1847 static void free_dev(struct mapped_device *md)
1849 int minor = MINOR(disk_devt(md->disk));
1853 cleanup_mapped_device(md);
1855 free_table_devices(&md->table_devices);
1856 dm_stats_cleanup(&md->stats);
1859 module_put(THIS_MODULE);
1863 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1865 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1868 if (dm_table_bio_based(t)) {
1870 * The md may already have mempools that need changing.
1871 * If so, reload bioset because front_pad may have changed
1872 * because a different table was loaded.
1874 bioset_exit(&md->bs);
1875 bioset_exit(&md->io_bs);
1877 } else if (bioset_initialized(&md->bs)) {
1879 * There's no need to reload with request-based dm
1880 * because the size of front_pad doesn't change.
1881 * Note for future: If you are to reload bioset,
1882 * prep-ed requests in the queue may refer
1883 * to bio from the old bioset, so you must walk
1884 * through the queue to unprep.
1890 bioset_initialized(&md->bs) ||
1891 bioset_initialized(&md->io_bs));
1893 ret = bioset_init_from_src(&md->bs, &p->bs);
1896 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1898 bioset_exit(&md->bs);
1900 /* mempool bind completed, no longer need any mempools in the table */
1901 dm_table_free_md_mempools(t);
1906 * Bind a table to the device.
1908 static void event_callback(void *context)
1910 unsigned long flags;
1912 struct mapped_device *md = (struct mapped_device *) context;
1914 spin_lock_irqsave(&md->uevent_lock, flags);
1915 list_splice_init(&md->uevent_list, &uevents);
1916 spin_unlock_irqrestore(&md->uevent_lock, flags);
1918 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1920 atomic_inc(&md->event_nr);
1921 wake_up(&md->eventq);
1922 dm_issue_global_event();
1926 * Returns old map, which caller must destroy.
1928 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1929 struct queue_limits *limits)
1931 struct dm_table *old_map;
1932 struct request_queue *q = md->queue;
1933 bool request_based = dm_table_request_based(t);
1937 lockdep_assert_held(&md->suspend_lock);
1939 size = dm_table_get_size(t);
1942 * Wipe any geometry if the size of the table changed.
1944 if (size != dm_get_size(md))
1945 memset(&md->geometry, 0, sizeof(md->geometry));
1947 if (!get_capacity(md->disk))
1948 set_capacity(md->disk, size);
1950 set_capacity_and_notify(md->disk, size);
1952 dm_table_event_callback(t, event_callback, md);
1955 * The queue hasn't been stopped yet, if the old table type wasn't
1956 * for request-based during suspension. So stop it to prevent
1957 * I/O mapping before resume.
1958 * This must be done before setting the queue restrictions,
1959 * because request-based dm may be run just after the setting.
1964 if (request_based) {
1966 * Leverage the fact that request-based DM targets are
1967 * immutable singletons - used to optimize dm_mq_queue_rq.
1969 md->immutable_target = dm_table_get_immutable_target(t);
1972 ret = __bind_mempools(md, t);
1974 old_map = ERR_PTR(ret);
1978 ret = dm_table_set_restrictions(t, q, limits);
1980 old_map = ERR_PTR(ret);
1984 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1985 rcu_assign_pointer(md->map, (void *)t);
1986 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1996 * Returns unbound table for the caller to free.
1998 static struct dm_table *__unbind(struct mapped_device *md)
2000 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2005 dm_table_event_callback(map, NULL, NULL);
2006 RCU_INIT_POINTER(md->map, NULL);
2013 * Constructor for a new device.
2015 int dm_create(int minor, struct mapped_device **result)
2017 struct mapped_device *md;
2019 md = alloc_dev(minor);
2023 dm_ima_reset_data(md);
2030 * Functions to manage md->type.
2031 * All are required to hold md->type_lock.
2033 void dm_lock_md_type(struct mapped_device *md)
2035 mutex_lock(&md->type_lock);
2038 void dm_unlock_md_type(struct mapped_device *md)
2040 mutex_unlock(&md->type_lock);
2043 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2045 BUG_ON(!mutex_is_locked(&md->type_lock));
2049 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2054 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2056 return md->immutable_target_type;
2060 * The queue_limits are only valid as long as you have a reference
2063 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2065 BUG_ON(!atomic_read(&md->holders));
2066 return &md->queue->limits;
2068 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2071 * Setup the DM device's queue based on md's type
2073 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2075 enum dm_queue_mode type = dm_table_get_type(t);
2076 struct queue_limits limits;
2080 case DM_TYPE_REQUEST_BASED:
2081 md->disk->fops = &dm_rq_blk_dops;
2082 r = dm_mq_init_request_queue(md, t);
2084 DMERR("Cannot initialize queue for request-based dm mapped device");
2088 case DM_TYPE_BIO_BASED:
2089 case DM_TYPE_DAX_BIO_BASED:
2096 r = dm_calculate_queue_limits(t, &limits);
2098 DMERR("Cannot calculate initial queue limits");
2101 r = dm_table_set_restrictions(t, md->queue, &limits);
2107 r = dm_sysfs_init(md);
2109 del_gendisk(md->disk);
2116 struct mapped_device *dm_get_md(dev_t dev)
2118 struct mapped_device *md;
2119 unsigned minor = MINOR(dev);
2121 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2124 spin_lock(&_minor_lock);
2126 md = idr_find(&_minor_idr, minor);
2127 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2128 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2134 spin_unlock(&_minor_lock);
2138 EXPORT_SYMBOL_GPL(dm_get_md);
2140 void *dm_get_mdptr(struct mapped_device *md)
2142 return md->interface_ptr;
2145 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2147 md->interface_ptr = ptr;
2150 void dm_get(struct mapped_device *md)
2152 atomic_inc(&md->holders);
2153 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2156 int dm_hold(struct mapped_device *md)
2158 spin_lock(&_minor_lock);
2159 if (test_bit(DMF_FREEING, &md->flags)) {
2160 spin_unlock(&_minor_lock);
2164 spin_unlock(&_minor_lock);
2167 EXPORT_SYMBOL_GPL(dm_hold);
2169 const char *dm_device_name(struct mapped_device *md)
2173 EXPORT_SYMBOL_GPL(dm_device_name);
2175 static void __dm_destroy(struct mapped_device *md, bool wait)
2177 struct dm_table *map;
2182 spin_lock(&_minor_lock);
2183 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2184 set_bit(DMF_FREEING, &md->flags);
2185 spin_unlock(&_minor_lock);
2187 blk_mark_disk_dead(md->disk);
2190 * Take suspend_lock so that presuspend and postsuspend methods
2191 * do not race with internal suspend.
2193 mutex_lock(&md->suspend_lock);
2194 map = dm_get_live_table(md, &srcu_idx);
2195 if (!dm_suspended_md(md)) {
2196 dm_table_presuspend_targets(map);
2197 set_bit(DMF_SUSPENDED, &md->flags);
2198 set_bit(DMF_POST_SUSPENDING, &md->flags);
2199 dm_table_postsuspend_targets(map);
2201 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2202 dm_put_live_table(md, srcu_idx);
2203 mutex_unlock(&md->suspend_lock);
2206 * Rare, but there may be I/O requests still going to complete,
2207 * for example. Wait for all references to disappear.
2208 * No one should increment the reference count of the mapped_device,
2209 * after the mapped_device state becomes DMF_FREEING.
2212 while (atomic_read(&md->holders))
2214 else if (atomic_read(&md->holders))
2215 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2216 dm_device_name(md), atomic_read(&md->holders));
2218 dm_table_destroy(__unbind(md));
2222 void dm_destroy(struct mapped_device *md)
2224 __dm_destroy(md, true);
2227 void dm_destroy_immediate(struct mapped_device *md)
2229 __dm_destroy(md, false);
2232 void dm_put(struct mapped_device *md)
2234 atomic_dec(&md->holders);
2236 EXPORT_SYMBOL_GPL(dm_put);
2238 static bool dm_in_flight_bios(struct mapped_device *md)
2241 unsigned long sum = 0;
2243 for_each_possible_cpu(cpu)
2244 sum += *per_cpu_ptr(md->pending_io, cpu);
2249 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2255 prepare_to_wait(&md->wait, &wait, task_state);
2257 if (!dm_in_flight_bios(md))
2260 if (signal_pending_state(task_state, current)) {
2267 finish_wait(&md->wait, &wait);
2274 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2278 if (!queue_is_mq(md->queue))
2279 return dm_wait_for_bios_completion(md, task_state);
2282 if (!blk_mq_queue_inflight(md->queue))
2285 if (signal_pending_state(task_state, current)) {
2297 * Process the deferred bios
2299 static void dm_wq_work(struct work_struct *work)
2301 struct mapped_device *md = container_of(work, struct mapped_device, work);
2304 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2305 spin_lock_irq(&md->deferred_lock);
2306 bio = bio_list_pop(&md->deferred);
2307 spin_unlock_irq(&md->deferred_lock);
2312 submit_bio_noacct(bio);
2316 static void dm_queue_flush(struct mapped_device *md)
2318 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2319 smp_mb__after_atomic();
2320 queue_work(md->wq, &md->work);
2324 * Swap in a new table, returning the old one for the caller to destroy.
2326 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2328 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2329 struct queue_limits limits;
2332 mutex_lock(&md->suspend_lock);
2334 /* device must be suspended */
2335 if (!dm_suspended_md(md))
2339 * If the new table has no data devices, retain the existing limits.
2340 * This helps multipath with queue_if_no_path if all paths disappear,
2341 * then new I/O is queued based on these limits, and then some paths
2344 if (dm_table_has_no_data_devices(table)) {
2345 live_map = dm_get_live_table_fast(md);
2347 limits = md->queue->limits;
2348 dm_put_live_table_fast(md);
2352 r = dm_calculate_queue_limits(table, &limits);
2359 map = __bind(md, table, &limits);
2360 dm_issue_global_event();
2363 mutex_unlock(&md->suspend_lock);
2368 * Functions to lock and unlock any filesystem running on the
2371 static int lock_fs(struct mapped_device *md)
2375 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2377 r = freeze_bdev(md->disk->part0);
2379 set_bit(DMF_FROZEN, &md->flags);
2383 static void unlock_fs(struct mapped_device *md)
2385 if (!test_bit(DMF_FROZEN, &md->flags))
2387 thaw_bdev(md->disk->part0);
2388 clear_bit(DMF_FROZEN, &md->flags);
2392 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2393 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2394 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2396 * If __dm_suspend returns 0, the device is completely quiescent
2397 * now. There is no request-processing activity. All new requests
2398 * are being added to md->deferred list.
2400 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2401 unsigned suspend_flags, unsigned int task_state,
2402 int dmf_suspended_flag)
2404 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2405 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2408 lockdep_assert_held(&md->suspend_lock);
2411 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2412 * This flag is cleared before dm_suspend returns.
2415 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2417 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2420 * This gets reverted if there's an error later and the targets
2421 * provide the .presuspend_undo hook.
2423 dm_table_presuspend_targets(map);
2426 * Flush I/O to the device.
2427 * Any I/O submitted after lock_fs() may not be flushed.
2428 * noflush takes precedence over do_lockfs.
2429 * (lock_fs() flushes I/Os and waits for them to complete.)
2431 if (!noflush && do_lockfs) {
2434 dm_table_presuspend_undo_targets(map);
2440 * Here we must make sure that no processes are submitting requests
2441 * to target drivers i.e. no one may be executing
2442 * __split_and_process_bio from dm_submit_bio.
2444 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2445 * we take the write lock. To prevent any process from reentering
2446 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2447 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2448 * flush_workqueue(md->wq).
2450 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2452 synchronize_srcu(&md->io_barrier);
2455 * Stop md->queue before flushing md->wq in case request-based
2456 * dm defers requests to md->wq from md->queue.
2458 if (dm_request_based(md))
2459 dm_stop_queue(md->queue);
2461 flush_workqueue(md->wq);
2464 * At this point no more requests are entering target request routines.
2465 * We call dm_wait_for_completion to wait for all existing requests
2468 r = dm_wait_for_completion(md, task_state);
2470 set_bit(dmf_suspended_flag, &md->flags);
2473 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2475 synchronize_srcu(&md->io_barrier);
2477 /* were we interrupted ? */
2481 if (dm_request_based(md))
2482 dm_start_queue(md->queue);
2485 dm_table_presuspend_undo_targets(map);
2486 /* pushback list is already flushed, so skip flush */
2493 * We need to be able to change a mapping table under a mounted
2494 * filesystem. For example we might want to move some data in
2495 * the background. Before the table can be swapped with
2496 * dm_bind_table, dm_suspend must be called to flush any in
2497 * flight bios and ensure that any further io gets deferred.
2500 * Suspend mechanism in request-based dm.
2502 * 1. Flush all I/Os by lock_fs() if needed.
2503 * 2. Stop dispatching any I/O by stopping the request_queue.
2504 * 3. Wait for all in-flight I/Os to be completed or requeued.
2506 * To abort suspend, start the request_queue.
2508 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2510 struct dm_table *map = NULL;
2514 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2516 if (dm_suspended_md(md)) {
2521 if (dm_suspended_internally_md(md)) {
2522 /* already internally suspended, wait for internal resume */
2523 mutex_unlock(&md->suspend_lock);
2524 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2530 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2532 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2536 set_bit(DMF_POST_SUSPENDING, &md->flags);
2537 dm_table_postsuspend_targets(map);
2538 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2541 mutex_unlock(&md->suspend_lock);
2545 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2548 int r = dm_table_resume_targets(map);
2556 * Flushing deferred I/Os must be done after targets are resumed
2557 * so that mapping of targets can work correctly.
2558 * Request-based dm is queueing the deferred I/Os in its request_queue.
2560 if (dm_request_based(md))
2561 dm_start_queue(md->queue);
2568 int dm_resume(struct mapped_device *md)
2571 struct dm_table *map = NULL;
2575 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2577 if (!dm_suspended_md(md))
2580 if (dm_suspended_internally_md(md)) {
2581 /* already internally suspended, wait for internal resume */
2582 mutex_unlock(&md->suspend_lock);
2583 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2589 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2590 if (!map || !dm_table_get_size(map))
2593 r = __dm_resume(md, map);
2597 clear_bit(DMF_SUSPENDED, &md->flags);
2599 mutex_unlock(&md->suspend_lock);
2605 * Internal suspend/resume works like userspace-driven suspend. It waits
2606 * until all bios finish and prevents issuing new bios to the target drivers.
2607 * It may be used only from the kernel.
2610 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2612 struct dm_table *map = NULL;
2614 lockdep_assert_held(&md->suspend_lock);
2616 if (md->internal_suspend_count++)
2617 return; /* nested internal suspend */
2619 if (dm_suspended_md(md)) {
2620 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2621 return; /* nest suspend */
2624 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2627 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2628 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2629 * would require changing .presuspend to return an error -- avoid this
2630 * until there is a need for more elaborate variants of internal suspend.
2632 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2633 DMF_SUSPENDED_INTERNALLY);
2635 set_bit(DMF_POST_SUSPENDING, &md->flags);
2636 dm_table_postsuspend_targets(map);
2637 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2640 static void __dm_internal_resume(struct mapped_device *md)
2642 BUG_ON(!md->internal_suspend_count);
2644 if (--md->internal_suspend_count)
2645 return; /* resume from nested internal suspend */
2647 if (dm_suspended_md(md))
2648 goto done; /* resume from nested suspend */
2651 * NOTE: existing callers don't need to call dm_table_resume_targets
2652 * (which may fail -- so best to avoid it for now by passing NULL map)
2654 (void) __dm_resume(md, NULL);
2657 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2658 smp_mb__after_atomic();
2659 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2662 void dm_internal_suspend_noflush(struct mapped_device *md)
2664 mutex_lock(&md->suspend_lock);
2665 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2666 mutex_unlock(&md->suspend_lock);
2668 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2670 void dm_internal_resume(struct mapped_device *md)
2672 mutex_lock(&md->suspend_lock);
2673 __dm_internal_resume(md);
2674 mutex_unlock(&md->suspend_lock);
2676 EXPORT_SYMBOL_GPL(dm_internal_resume);
2679 * Fast variants of internal suspend/resume hold md->suspend_lock,
2680 * which prevents interaction with userspace-driven suspend.
2683 void dm_internal_suspend_fast(struct mapped_device *md)
2685 mutex_lock(&md->suspend_lock);
2686 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2689 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2690 synchronize_srcu(&md->io_barrier);
2691 flush_workqueue(md->wq);
2692 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2694 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2696 void dm_internal_resume_fast(struct mapped_device *md)
2698 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2704 mutex_unlock(&md->suspend_lock);
2706 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2708 /*-----------------------------------------------------------------
2709 * Event notification.
2710 *---------------------------------------------------------------*/
2711 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2716 char udev_cookie[DM_COOKIE_LENGTH];
2717 char *envp[] = { udev_cookie, NULL };
2719 noio_flag = memalloc_noio_save();
2722 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2724 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2725 DM_COOKIE_ENV_VAR_NAME, cookie);
2726 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2730 memalloc_noio_restore(noio_flag);
2735 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2737 return atomic_add_return(1, &md->uevent_seq);
2740 uint32_t dm_get_event_nr(struct mapped_device *md)
2742 return atomic_read(&md->event_nr);
2745 int dm_wait_event(struct mapped_device *md, int event_nr)
2747 return wait_event_interruptible(md->eventq,
2748 (event_nr != atomic_read(&md->event_nr)));
2751 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2753 unsigned long flags;
2755 spin_lock_irqsave(&md->uevent_lock, flags);
2756 list_add(elist, &md->uevent_list);
2757 spin_unlock_irqrestore(&md->uevent_lock, flags);
2761 * The gendisk is only valid as long as you have a reference
2764 struct gendisk *dm_disk(struct mapped_device *md)
2768 EXPORT_SYMBOL_GPL(dm_disk);
2770 struct kobject *dm_kobject(struct mapped_device *md)
2772 return &md->kobj_holder.kobj;
2775 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2777 struct mapped_device *md;
2779 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2781 spin_lock(&_minor_lock);
2782 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2788 spin_unlock(&_minor_lock);
2793 int dm_suspended_md(struct mapped_device *md)
2795 return test_bit(DMF_SUSPENDED, &md->flags);
2798 static int dm_post_suspending_md(struct mapped_device *md)
2800 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2803 int dm_suspended_internally_md(struct mapped_device *md)
2805 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2808 int dm_test_deferred_remove_flag(struct mapped_device *md)
2810 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2813 int dm_suspended(struct dm_target *ti)
2815 return dm_suspended_md(ti->table->md);
2817 EXPORT_SYMBOL_GPL(dm_suspended);
2819 int dm_post_suspending(struct dm_target *ti)
2821 return dm_post_suspending_md(ti->table->md);
2823 EXPORT_SYMBOL_GPL(dm_post_suspending);
2825 int dm_noflush_suspending(struct dm_target *ti)
2827 return __noflush_suspending(ti->table->md);
2829 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2831 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2832 unsigned integrity, unsigned per_io_data_size,
2833 unsigned min_pool_size)
2835 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2836 unsigned int pool_size = 0;
2837 unsigned int front_pad, io_front_pad;
2844 case DM_TYPE_BIO_BASED:
2845 case DM_TYPE_DAX_BIO_BASED:
2846 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2847 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2848 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2849 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2852 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2855 case DM_TYPE_REQUEST_BASED:
2856 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2857 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2858 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2864 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2868 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2874 dm_free_md_mempools(pools);
2879 void dm_free_md_mempools(struct dm_md_mempools *pools)
2884 bioset_exit(&pools->bs);
2885 bioset_exit(&pools->io_bs);
2897 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2900 struct mapped_device *md = bdev->bd_disk->private_data;
2901 struct dm_table *table;
2902 struct dm_target *ti;
2903 int ret = -ENOTTY, srcu_idx;
2905 table = dm_get_live_table(md, &srcu_idx);
2906 if (!table || !dm_table_get_size(table))
2909 /* We only support devices that have a single target */
2910 if (dm_table_get_num_targets(table) != 1)
2912 ti = dm_table_get_target(table, 0);
2915 if (!ti->type->iterate_devices)
2918 ret = ti->type->iterate_devices(ti, fn, data);
2920 dm_put_live_table(md, srcu_idx);
2925 * For register / unregister we need to manually call out to every path.
2927 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2928 sector_t start, sector_t len, void *data)
2930 struct dm_pr *pr = data;
2931 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2933 if (!ops || !ops->pr_register)
2935 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2938 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2949 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2950 if (ret && new_key) {
2951 /* unregister all paths if we failed to register any path */
2952 pr.old_key = new_key;
2955 pr.fail_early = false;
2956 dm_call_pr(bdev, __dm_pr_register, &pr);
2962 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2965 struct mapped_device *md = bdev->bd_disk->private_data;
2966 const struct pr_ops *ops;
2969 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2973 ops = bdev->bd_disk->fops->pr_ops;
2974 if (ops && ops->pr_reserve)
2975 r = ops->pr_reserve(bdev, key, type, flags);
2979 dm_unprepare_ioctl(md, srcu_idx);
2983 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2985 struct mapped_device *md = bdev->bd_disk->private_data;
2986 const struct pr_ops *ops;
2989 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2993 ops = bdev->bd_disk->fops->pr_ops;
2994 if (ops && ops->pr_release)
2995 r = ops->pr_release(bdev, key, type);
2999 dm_unprepare_ioctl(md, srcu_idx);
3003 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3004 enum pr_type type, bool abort)
3006 struct mapped_device *md = bdev->bd_disk->private_data;
3007 const struct pr_ops *ops;
3010 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3014 ops = bdev->bd_disk->fops->pr_ops;
3015 if (ops && ops->pr_preempt)
3016 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3020 dm_unprepare_ioctl(md, srcu_idx);
3024 static int dm_pr_clear(struct block_device *bdev, u64 key)
3026 struct mapped_device *md = bdev->bd_disk->private_data;
3027 const struct pr_ops *ops;
3030 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3034 ops = bdev->bd_disk->fops->pr_ops;
3035 if (ops && ops->pr_clear)
3036 r = ops->pr_clear(bdev, key);
3040 dm_unprepare_ioctl(md, srcu_idx);
3044 static const struct pr_ops dm_pr_ops = {
3045 .pr_register = dm_pr_register,
3046 .pr_reserve = dm_pr_reserve,
3047 .pr_release = dm_pr_release,
3048 .pr_preempt = dm_pr_preempt,
3049 .pr_clear = dm_pr_clear,
3052 static const struct block_device_operations dm_blk_dops = {
3053 .submit_bio = dm_submit_bio,
3054 .open = dm_blk_open,
3055 .release = dm_blk_close,
3056 .ioctl = dm_blk_ioctl,
3057 .getgeo = dm_blk_getgeo,
3058 .report_zones = dm_blk_report_zones,
3059 .pr_ops = &dm_pr_ops,
3060 .owner = THIS_MODULE
3063 static const struct block_device_operations dm_rq_blk_dops = {
3064 .open = dm_blk_open,
3065 .release = dm_blk_close,
3066 .ioctl = dm_blk_ioctl,
3067 .getgeo = dm_blk_getgeo,
3068 .pr_ops = &dm_pr_ops,
3069 .owner = THIS_MODULE
3072 static const struct dax_operations dm_dax_ops = {
3073 .direct_access = dm_dax_direct_access,
3074 .dax_supported = dm_dax_supported,
3075 .copy_from_iter = dm_dax_copy_from_iter,
3076 .copy_to_iter = dm_dax_copy_to_iter,
3077 .zero_page_range = dm_dax_zero_page_range,
3083 module_init(dm_init);
3084 module_exit(dm_exit);
3086 module_param(major, uint, 0);
3087 MODULE_PARM_DESC(major, "The major number of the device mapper");
3089 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3090 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3092 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3093 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3095 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3096 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3098 MODULE_DESCRIPTION(DM_NAME " driver");
3099 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3100 MODULE_LICENSE("GPL");