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 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
471 dm_unprepare_ioctl(md, srcu_idx);
475 u64 dm_start_time_ns_from_clone(struct bio *bio)
477 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
478 struct dm_io *io = tio->io;
480 return jiffies_to_nsecs(io->start_time);
482 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
484 static bool bio_is_flush_with_data(struct bio *bio)
486 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
489 static void dm_io_acct(bool end, struct mapped_device *md, struct bio *bio,
490 unsigned long start_time, struct dm_stats_aux *stats_aux)
492 bool is_flush_with_data;
493 unsigned int bi_size;
495 /* If REQ_PREFLUSH set save any payload but do not account it */
496 is_flush_with_data = bio_is_flush_with_data(bio);
497 if (is_flush_with_data) {
498 bi_size = bio->bi_iter.bi_size;
499 bio->bi_iter.bi_size = 0;
503 bio_start_io_acct_time(bio, start_time);
505 bio_end_io_acct(bio, start_time);
507 if (unlikely(dm_stats_used(&md->stats)))
508 dm_stats_account_io(&md->stats, bio_data_dir(bio),
509 bio->bi_iter.bi_sector, bio_sectors(bio),
510 end, start_time, stats_aux);
512 /* Restore bio's payload so it does get accounted upon requeue */
513 if (is_flush_with_data)
514 bio->bi_iter.bi_size = bi_size;
517 static void start_io_acct(struct dm_io *io)
519 dm_io_acct(false, io->md, io->orig_bio, io->start_time, &io->stats_aux);
522 static void end_io_acct(struct mapped_device *md, struct bio *bio,
523 unsigned long start_time, struct dm_stats_aux *stats_aux)
525 dm_io_acct(true, md, bio, start_time, stats_aux);
528 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
531 struct dm_target_io *tio;
534 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
538 tio = container_of(clone, struct dm_target_io, clone);
539 tio->inside_dm_io = true;
542 io = container_of(tio, struct dm_io, tio);
543 io->magic = DM_IO_MAGIC;
545 atomic_set(&io->io_count, 1);
546 this_cpu_inc(*md->pending_io);
549 spin_lock_init(&io->endio_lock);
551 io->start_time = jiffies;
553 dm_stats_record_start(&md->stats, &io->stats_aux);
558 static void free_io(struct mapped_device *md, struct dm_io *io)
560 bio_put(&io->tio.clone);
563 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
564 unsigned target_bio_nr, gfp_t gfp_mask)
566 struct dm_target_io *tio;
568 if (!ci->io->tio.io) {
569 /* the dm_target_io embedded in ci->io is available */
572 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
576 tio = container_of(clone, struct dm_target_io, clone);
577 tio->inside_dm_io = false;
580 tio->magic = DM_TIO_MAGIC;
583 tio->target_bio_nr = target_bio_nr;
588 static void free_tio(struct dm_target_io *tio)
590 if (tio->inside_dm_io)
592 bio_put(&tio->clone);
596 * Add the bio to the list of deferred io.
598 static void queue_io(struct mapped_device *md, struct bio *bio)
602 spin_lock_irqsave(&md->deferred_lock, flags);
603 bio_list_add(&md->deferred, bio);
604 spin_unlock_irqrestore(&md->deferred_lock, flags);
605 queue_work(md->wq, &md->work);
609 * Everyone (including functions in this file), should use this
610 * function to access the md->map field, and make sure they call
611 * dm_put_live_table() when finished.
613 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
615 *srcu_idx = srcu_read_lock(&md->io_barrier);
617 return srcu_dereference(md->map, &md->io_barrier);
620 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
622 srcu_read_unlock(&md->io_barrier, srcu_idx);
625 void dm_sync_table(struct mapped_device *md)
627 synchronize_srcu(&md->io_barrier);
628 synchronize_rcu_expedited();
632 * A fast alternative to dm_get_live_table/dm_put_live_table.
633 * The caller must not block between these two functions.
635 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
638 return rcu_dereference(md->map);
641 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
646 static char *_dm_claim_ptr = "I belong to device-mapper";
649 * Open a table device so we can use it as a map destination.
651 static int open_table_device(struct table_device *td, dev_t dev,
652 struct mapped_device *md)
654 struct block_device *bdev;
658 BUG_ON(td->dm_dev.bdev);
660 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
662 return PTR_ERR(bdev);
664 r = bd_link_disk_holder(bdev, dm_disk(md));
666 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
670 td->dm_dev.bdev = bdev;
671 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev);
676 * Close a table device that we've been using.
678 static void close_table_device(struct table_device *td, struct mapped_device *md)
680 if (!td->dm_dev.bdev)
683 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
684 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
685 put_dax(td->dm_dev.dax_dev);
686 td->dm_dev.bdev = NULL;
687 td->dm_dev.dax_dev = NULL;
690 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
693 struct table_device *td;
695 list_for_each_entry(td, l, list)
696 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
702 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
703 struct dm_dev **result)
706 struct table_device *td;
708 mutex_lock(&md->table_devices_lock);
709 td = find_table_device(&md->table_devices, dev, mode);
711 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
713 mutex_unlock(&md->table_devices_lock);
717 td->dm_dev.mode = mode;
718 td->dm_dev.bdev = NULL;
720 if ((r = open_table_device(td, dev, md))) {
721 mutex_unlock(&md->table_devices_lock);
726 format_dev_t(td->dm_dev.name, dev);
728 refcount_set(&td->count, 1);
729 list_add(&td->list, &md->table_devices);
731 refcount_inc(&td->count);
733 mutex_unlock(&md->table_devices_lock);
735 *result = &td->dm_dev;
739 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
741 struct table_device *td = container_of(d, struct table_device, dm_dev);
743 mutex_lock(&md->table_devices_lock);
744 if (refcount_dec_and_test(&td->count)) {
745 close_table_device(td, md);
749 mutex_unlock(&md->table_devices_lock);
752 static void free_table_devices(struct list_head *devices)
754 struct list_head *tmp, *next;
756 list_for_each_safe(tmp, next, devices) {
757 struct table_device *td = list_entry(tmp, struct table_device, list);
759 DMWARN("dm_destroy: %s still exists with %d references",
760 td->dm_dev.name, refcount_read(&td->count));
766 * Get the geometry associated with a dm device
768 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
776 * Set the geometry of a device.
778 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
780 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
782 if (geo->start > sz) {
783 DMWARN("Start sector is beyond the geometry limits.");
792 static int __noflush_suspending(struct mapped_device *md)
794 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
798 * Decrements the number of outstanding ios that a bio has been
799 * cloned into, completing the original io if necc.
801 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
804 blk_status_t io_error;
806 struct mapped_device *md = io->md;
807 unsigned long start_time = 0;
808 struct dm_stats_aux stats_aux;
810 /* Push-back supersedes any I/O errors */
811 if (unlikely(error)) {
812 spin_lock_irqsave(&io->endio_lock, flags);
813 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
815 spin_unlock_irqrestore(&io->endio_lock, flags);
818 if (atomic_dec_and_test(&io->io_count)) {
820 if (io->status == BLK_STS_DM_REQUEUE) {
822 * Target requested pushing back the I/O.
824 spin_lock_irqsave(&md->deferred_lock, flags);
825 if (__noflush_suspending(md) &&
826 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
827 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
828 bio_list_add_head(&md->deferred, bio);
831 * noflush suspend was interrupted or this is
832 * a write to a zoned target.
834 io->status = BLK_STS_IOERR;
836 spin_unlock_irqrestore(&md->deferred_lock, flags);
839 io_error = io->status;
840 start_time = io->start_time;
841 stats_aux = io->stats_aux;
843 end_io_acct(md, bio, start_time, &stats_aux);
845 this_cpu_dec(*md->pending_io);
847 /* nudge anyone waiting on suspend queue */
848 if (unlikely(wq_has_sleeper(&md->wait)))
851 if (io_error == BLK_STS_DM_REQUEUE)
854 if (bio_is_flush_with_data(bio)) {
856 * Preflush done for flush with data, reissue
857 * without REQ_PREFLUSH.
859 bio->bi_opf &= ~REQ_PREFLUSH;
862 /* done with normal IO or empty flush */
864 bio->bi_status = io_error;
870 void disable_discard(struct mapped_device *md)
872 struct queue_limits *limits = dm_get_queue_limits(md);
874 /* device doesn't really support DISCARD, disable it */
875 limits->max_discard_sectors = 0;
876 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
879 void disable_write_same(struct mapped_device *md)
881 struct queue_limits *limits = dm_get_queue_limits(md);
883 /* device doesn't really support WRITE SAME, disable it */
884 limits->max_write_same_sectors = 0;
887 void disable_write_zeroes(struct mapped_device *md)
889 struct queue_limits *limits = dm_get_queue_limits(md);
891 /* device doesn't really support WRITE ZEROES, disable it */
892 limits->max_write_zeroes_sectors = 0;
895 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
897 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
900 static void clone_endio(struct bio *bio)
902 blk_status_t error = bio->bi_status;
903 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
904 struct dm_io *io = tio->io;
905 struct mapped_device *md = tio->io->md;
906 dm_endio_fn endio = tio->ti->type->end_io;
907 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
909 if (unlikely(error == BLK_STS_TARGET)) {
910 if (bio_op(bio) == REQ_OP_DISCARD &&
911 !q->limits.max_discard_sectors)
913 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
914 !q->limits.max_write_same_sectors)
915 disable_write_same(md);
916 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
917 !q->limits.max_write_zeroes_sectors)
918 disable_write_zeroes(md);
921 if (blk_queue_is_zoned(q))
922 dm_zone_endio(io, bio);
925 int r = endio(tio->ti, bio, &error);
927 case DM_ENDIO_REQUEUE:
929 * Requeuing writes to a sequential zone of a zoned
930 * target will break the sequential write pattern:
933 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
934 error = BLK_STS_IOERR;
936 error = BLK_STS_DM_REQUEUE;
940 case DM_ENDIO_INCOMPLETE:
941 /* The target will handle the io */
944 DMWARN("unimplemented target endio return value: %d", r);
949 if (unlikely(swap_bios_limit(tio->ti, bio))) {
950 struct mapped_device *md = io->md;
951 up(&md->swap_bios_semaphore);
955 dm_io_dec_pending(io, error);
959 * Return maximum size of I/O possible at the supplied sector up to the current
962 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
963 sector_t target_offset)
965 return ti->len - target_offset;
968 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
970 sector_t target_offset = dm_target_offset(ti, sector);
971 sector_t len = max_io_len_target_boundary(ti, target_offset);
975 * Does the target need to split IO even further?
976 * - varied (per target) IO splitting is a tenet of DM; this
977 * explains why stacked chunk_sectors based splitting via
978 * blk_max_size_offset() isn't possible here. So pass in
979 * ti->max_io_len to override stacked chunk_sectors.
981 if (ti->max_io_len) {
982 max_len = blk_max_size_offset(ti->table->md->queue,
983 target_offset, ti->max_io_len);
991 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
993 if (len > UINT_MAX) {
994 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
995 (unsigned long long)len, UINT_MAX);
996 ti->error = "Maximum size of target IO is too large";
1000 ti->max_io_len = (uint32_t) len;
1004 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1006 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1007 sector_t sector, int *srcu_idx)
1008 __acquires(md->io_barrier)
1010 struct dm_table *map;
1011 struct dm_target *ti;
1013 map = dm_get_live_table(md, srcu_idx);
1017 ti = dm_table_find_target(map, sector);
1024 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1025 long nr_pages, void **kaddr, pfn_t *pfn)
1027 struct mapped_device *md = dax_get_private(dax_dev);
1028 sector_t sector = pgoff * PAGE_SECTORS;
1029 struct dm_target *ti;
1030 long len, ret = -EIO;
1033 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1037 if (!ti->type->direct_access)
1039 len = max_io_len(ti, sector) / PAGE_SECTORS;
1042 nr_pages = min(len, nr_pages);
1043 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1046 dm_put_live_table(md, srcu_idx);
1051 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1052 int blocksize, sector_t start, sector_t len)
1054 struct mapped_device *md = dax_get_private(dax_dev);
1055 struct dm_table *map;
1059 map = dm_get_live_table(md, &srcu_idx);
1063 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1066 dm_put_live_table(md, srcu_idx);
1071 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1072 void *addr, size_t bytes, struct iov_iter *i)
1074 struct mapped_device *md = dax_get_private(dax_dev);
1075 sector_t sector = pgoff * PAGE_SECTORS;
1076 struct dm_target *ti;
1080 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1084 if (!ti->type->dax_copy_from_iter) {
1085 ret = copy_from_iter(addr, bytes, i);
1088 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1090 dm_put_live_table(md, srcu_idx);
1095 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1096 void *addr, size_t bytes, struct iov_iter *i)
1098 struct mapped_device *md = dax_get_private(dax_dev);
1099 sector_t sector = pgoff * PAGE_SECTORS;
1100 struct dm_target *ti;
1104 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1108 if (!ti->type->dax_copy_to_iter) {
1109 ret = copy_to_iter(addr, bytes, i);
1112 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1114 dm_put_live_table(md, srcu_idx);
1119 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1122 struct mapped_device *md = dax_get_private(dax_dev);
1123 sector_t sector = pgoff * PAGE_SECTORS;
1124 struct dm_target *ti;
1128 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1132 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1134 * ->zero_page_range() is mandatory dax operation. If we are
1135 * here, something is wrong.
1139 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1141 dm_put_live_table(md, srcu_idx);
1147 * A target may call dm_accept_partial_bio only from the map routine. It is
1148 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1149 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1151 * dm_accept_partial_bio informs the dm that the target only wants to process
1152 * additional n_sectors sectors of the bio and the rest of the data should be
1153 * sent in a next bio.
1155 * A diagram that explains the arithmetics:
1156 * +--------------------+---------------+-------+
1158 * +--------------------+---------------+-------+
1160 * <-------------- *tio->len_ptr --------------->
1161 * <------- bi_size ------->
1164 * Region 1 was already iterated over with bio_advance or similar function.
1165 * (it may be empty if the target doesn't use bio_advance)
1166 * Region 2 is the remaining bio size that the target wants to process.
1167 * (it may be empty if region 1 is non-empty, although there is no reason
1169 * The target requires that region 3 is to be sent in the next bio.
1171 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1172 * the partially processed part (the sum of regions 1+2) must be the same for all
1173 * copies of the bio.
1175 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1177 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1178 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1180 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1181 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1182 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1183 BUG_ON(bi_size > *tio->len_ptr);
1184 BUG_ON(n_sectors > bi_size);
1186 *tio->len_ptr -= bi_size - n_sectors;
1187 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1189 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1191 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1193 mutex_lock(&md->swap_bios_lock);
1194 while (latch < md->swap_bios) {
1196 down(&md->swap_bios_semaphore);
1199 while (latch > md->swap_bios) {
1201 up(&md->swap_bios_semaphore);
1204 mutex_unlock(&md->swap_bios_lock);
1207 static blk_qc_t __map_bio(struct dm_target_io *tio)
1211 struct bio *clone = &tio->clone;
1212 struct dm_io *io = tio->io;
1213 struct dm_target *ti = tio->ti;
1214 blk_qc_t ret = BLK_QC_T_NONE;
1216 clone->bi_end_io = clone_endio;
1219 * Map the clone. If r == 0 we don't need to do
1220 * anything, the target has assumed ownership of
1223 dm_io_inc_pending(io);
1224 sector = clone->bi_iter.bi_sector;
1226 if (unlikely(swap_bios_limit(ti, clone))) {
1227 struct mapped_device *md = io->md;
1228 int latch = get_swap_bios();
1229 if (unlikely(latch != md->swap_bios))
1230 __set_swap_bios_limit(md, latch);
1231 down(&md->swap_bios_semaphore);
1235 * Check if the IO needs a special mapping due to zone append emulation
1236 * on zoned target. In this case, dm_zone_map_bio() calls the target
1239 if (dm_emulate_zone_append(io->md))
1240 r = dm_zone_map_bio(tio);
1242 r = ti->type->map(ti, clone);
1245 case DM_MAPIO_SUBMITTED:
1247 case DM_MAPIO_REMAPPED:
1248 /* the bio has been remapped so dispatch it */
1249 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1250 ret = submit_bio_noacct(clone);
1253 if (unlikely(swap_bios_limit(ti, clone))) {
1254 struct mapped_device *md = io->md;
1255 up(&md->swap_bios_semaphore);
1258 dm_io_dec_pending(io, BLK_STS_IOERR);
1260 case DM_MAPIO_REQUEUE:
1261 if (unlikely(swap_bios_limit(ti, clone))) {
1262 struct mapped_device *md = io->md;
1263 up(&md->swap_bios_semaphore);
1266 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1269 DMWARN("unimplemented target map return value: %d", r);
1276 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1278 bio->bi_iter.bi_sector = sector;
1279 bio->bi_iter.bi_size = to_bytes(len);
1283 * Creates a bio that consists of range of complete bvecs.
1285 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1286 sector_t sector, unsigned len)
1288 struct bio *clone = &tio->clone;
1291 __bio_clone_fast(clone, bio);
1293 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1297 if (bio_integrity(bio)) {
1298 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1299 !dm_target_passes_integrity(tio->ti->type))) {
1300 DMWARN("%s: the target %s doesn't support integrity data.",
1301 dm_device_name(tio->io->md),
1302 tio->ti->type->name);
1306 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1311 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1312 clone->bi_iter.bi_size = to_bytes(len);
1314 if (bio_integrity(bio))
1315 bio_integrity_trim(clone);
1320 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1321 struct dm_target *ti, unsigned num_bios)
1323 struct dm_target_io *tio;
1329 if (num_bios == 1) {
1330 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1331 bio_list_add(blist, &tio->clone);
1335 for (try = 0; try < 2; try++) {
1340 mutex_lock(&ci->io->md->table_devices_lock);
1341 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1342 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1346 bio_list_add(blist, &tio->clone);
1349 mutex_unlock(&ci->io->md->table_devices_lock);
1350 if (bio_nr == num_bios)
1353 while ((bio = bio_list_pop(blist))) {
1354 tio = container_of(bio, struct dm_target_io, clone);
1360 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1361 struct dm_target_io *tio, unsigned *len)
1363 struct bio *clone = &tio->clone;
1367 __bio_clone_fast(clone, ci->bio);
1369 bio_setup_sector(clone, ci->sector, *len);
1371 return __map_bio(tio);
1374 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1375 unsigned num_bios, unsigned *len)
1377 struct bio_list blist = BIO_EMPTY_LIST;
1379 struct dm_target_io *tio;
1381 alloc_multiple_bios(&blist, ci, ti, num_bios);
1383 while ((bio = bio_list_pop(&blist))) {
1384 tio = container_of(bio, struct dm_target_io, clone);
1385 (void) __clone_and_map_simple_bio(ci, tio, len);
1389 static int __send_empty_flush(struct clone_info *ci)
1391 unsigned target_nr = 0;
1392 struct dm_target *ti;
1393 struct bio flush_bio;
1396 * Use an on-stack bio for this, it's safe since we don't
1397 * need to reference it after submit. It's just used as
1398 * the basis for the clone(s).
1400 bio_init(&flush_bio, NULL, 0);
1401 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1402 bio_set_dev(&flush_bio, ci->io->md->disk->part0);
1404 ci->bio = &flush_bio;
1405 ci->sector_count = 0;
1407 BUG_ON(bio_has_data(ci->bio));
1408 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1409 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1411 bio_uninit(ci->bio);
1415 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1416 sector_t sector, unsigned *len)
1418 struct bio *bio = ci->bio;
1419 struct dm_target_io *tio;
1422 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1424 r = clone_bio(tio, bio, sector, *len);
1429 (void) __map_bio(tio);
1434 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1440 * Even though the device advertised support for this type of
1441 * request, that does not mean every target supports it, and
1442 * reconfiguration might also have changed that since the
1443 * check was performed.
1448 len = min_t(sector_t, ci->sector_count,
1449 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1451 __send_duplicate_bios(ci, ti, num_bios, &len);
1454 ci->sector_count -= len;
1459 static bool is_abnormal_io(struct bio *bio)
1463 switch (bio_op(bio)) {
1464 case REQ_OP_DISCARD:
1465 case REQ_OP_SECURE_ERASE:
1466 case REQ_OP_WRITE_SAME:
1467 case REQ_OP_WRITE_ZEROES:
1475 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1478 struct bio *bio = ci->bio;
1479 unsigned num_bios = 0;
1481 switch (bio_op(bio)) {
1482 case REQ_OP_DISCARD:
1483 num_bios = ti->num_discard_bios;
1485 case REQ_OP_SECURE_ERASE:
1486 num_bios = ti->num_secure_erase_bios;
1488 case REQ_OP_WRITE_SAME:
1489 num_bios = ti->num_write_same_bios;
1491 case REQ_OP_WRITE_ZEROES:
1492 num_bios = ti->num_write_zeroes_bios;
1498 *result = __send_changing_extent_only(ci, ti, num_bios);
1503 * Select the correct strategy for processing a non-flush bio.
1505 static int __split_and_process_non_flush(struct clone_info *ci)
1507 struct dm_target *ti;
1511 ti = dm_table_find_target(ci->map, ci->sector);
1515 if (__process_abnormal_io(ci, ti, &r))
1518 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1520 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1525 ci->sector_count -= len;
1530 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1531 struct dm_table *map, struct bio *bio)
1534 ci->io = alloc_io(md, bio);
1535 ci->sector = bio->bi_iter.bi_sector;
1539 * Entry point to split a bio into clones and submit them to the targets.
1541 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1542 struct dm_table *map, struct bio *bio)
1544 struct clone_info ci;
1545 blk_qc_t ret = BLK_QC_T_NONE;
1548 init_clone_info(&ci, md, map, bio);
1550 if (bio->bi_opf & REQ_PREFLUSH) {
1551 error = __send_empty_flush(&ci);
1552 /* dm_io_dec_pending submits any data associated with flush */
1553 } else if (op_is_zone_mgmt(bio_op(bio))) {
1555 ci.sector_count = 0;
1556 error = __split_and_process_non_flush(&ci);
1559 ci.sector_count = bio_sectors(bio);
1560 error = __split_and_process_non_flush(&ci);
1561 if (ci.sector_count && !error) {
1563 * Remainder must be passed to submit_bio_noacct()
1564 * so that it gets handled *after* bios already submitted
1565 * have been completely processed.
1566 * We take a clone of the original to store in
1567 * ci.io->orig_bio to be used by end_io_acct() and
1568 * for dec_pending to use for completion handling.
1570 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1571 GFP_NOIO, &md->queue->bio_split);
1572 ci.io->orig_bio = b;
1575 trace_block_split(b, bio->bi_iter.bi_sector);
1576 ret = submit_bio_noacct(bio);
1579 start_io_acct(ci.io);
1581 /* drop the extra reference count */
1582 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1586 static blk_qc_t dm_submit_bio(struct bio *bio)
1588 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1589 blk_qc_t ret = BLK_QC_T_NONE;
1591 struct dm_table *map;
1593 map = dm_get_live_table(md, &srcu_idx);
1595 /* If suspended, or map not yet available, queue this IO for later */
1596 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1598 if (bio->bi_opf & REQ_NOWAIT)
1599 bio_wouldblock_error(bio);
1600 else if (bio->bi_opf & REQ_RAHEAD)
1608 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1609 * otherwise associated queue_limits won't be imposed.
1611 if (is_abnormal_io(bio))
1612 blk_queue_split(&bio);
1614 ret = __split_and_process_bio(md, map, bio);
1616 dm_put_live_table(md, srcu_idx);
1620 /*-----------------------------------------------------------------
1621 * An IDR is used to keep track of allocated minor numbers.
1622 *---------------------------------------------------------------*/
1623 static void free_minor(int minor)
1625 spin_lock(&_minor_lock);
1626 idr_remove(&_minor_idr, minor);
1627 spin_unlock(&_minor_lock);
1631 * See if the device with a specific minor # is free.
1633 static int specific_minor(int minor)
1637 if (minor >= (1 << MINORBITS))
1640 idr_preload(GFP_KERNEL);
1641 spin_lock(&_minor_lock);
1643 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1645 spin_unlock(&_minor_lock);
1648 return r == -ENOSPC ? -EBUSY : r;
1652 static int next_free_minor(int *minor)
1656 idr_preload(GFP_KERNEL);
1657 spin_lock(&_minor_lock);
1659 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1661 spin_unlock(&_minor_lock);
1669 static const struct block_device_operations dm_blk_dops;
1670 static const struct block_device_operations dm_rq_blk_dops;
1671 static const struct dax_operations dm_dax_ops;
1673 static void dm_wq_work(struct work_struct *work);
1675 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1676 static void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1678 dm_destroy_keyslot_manager(q->ksm);
1681 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1683 static inline void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1686 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1688 static void cleanup_mapped_device(struct mapped_device *md)
1691 destroy_workqueue(md->wq);
1692 bioset_exit(&md->bs);
1693 bioset_exit(&md->io_bs);
1696 kill_dax(md->dax_dev);
1697 put_dax(md->dax_dev);
1701 dm_cleanup_zoned_dev(md);
1703 spin_lock(&_minor_lock);
1704 md->disk->private_data = NULL;
1705 spin_unlock(&_minor_lock);
1706 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1708 del_gendisk(md->disk);
1710 dm_queue_destroy_keyslot_manager(md->queue);
1711 blk_cleanup_disk(md->disk);
1714 if (md->pending_io) {
1715 free_percpu(md->pending_io);
1716 md->pending_io = NULL;
1719 cleanup_srcu_struct(&md->io_barrier);
1721 mutex_destroy(&md->suspend_lock);
1722 mutex_destroy(&md->type_lock);
1723 mutex_destroy(&md->table_devices_lock);
1724 mutex_destroy(&md->swap_bios_lock);
1726 dm_mq_cleanup_mapped_device(md);
1730 * Allocate and initialise a blank device with a given minor.
1732 static struct mapped_device *alloc_dev(int minor)
1734 int r, numa_node_id = dm_get_numa_node();
1735 struct mapped_device *md;
1738 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1740 DMWARN("unable to allocate device, out of memory.");
1744 if (!try_module_get(THIS_MODULE))
1745 goto bad_module_get;
1747 /* get a minor number for the dev */
1748 if (minor == DM_ANY_MINOR)
1749 r = next_free_minor(&minor);
1751 r = specific_minor(minor);
1755 r = init_srcu_struct(&md->io_barrier);
1757 goto bad_io_barrier;
1759 md->numa_node_id = numa_node_id;
1760 md->init_tio_pdu = false;
1761 md->type = DM_TYPE_NONE;
1762 mutex_init(&md->suspend_lock);
1763 mutex_init(&md->type_lock);
1764 mutex_init(&md->table_devices_lock);
1765 spin_lock_init(&md->deferred_lock);
1766 atomic_set(&md->holders, 1);
1767 atomic_set(&md->open_count, 0);
1768 atomic_set(&md->event_nr, 0);
1769 atomic_set(&md->uevent_seq, 0);
1770 INIT_LIST_HEAD(&md->uevent_list);
1771 INIT_LIST_HEAD(&md->table_devices);
1772 spin_lock_init(&md->uevent_lock);
1775 * default to bio-based until DM table is loaded and md->type
1776 * established. If request-based table is loaded: blk-mq will
1777 * override accordingly.
1779 md->disk = blk_alloc_disk(md->numa_node_id);
1782 md->queue = md->disk->queue;
1784 init_waitqueue_head(&md->wait);
1785 INIT_WORK(&md->work, dm_wq_work);
1786 init_waitqueue_head(&md->eventq);
1787 init_completion(&md->kobj_holder.completion);
1789 md->swap_bios = get_swap_bios();
1790 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1791 mutex_init(&md->swap_bios_lock);
1793 md->disk->major = _major;
1794 md->disk->first_minor = minor;
1795 md->disk->minors = 1;
1796 md->disk->fops = &dm_blk_dops;
1797 md->disk->private_data = md;
1798 sprintf(md->disk->disk_name, "dm-%d", minor);
1800 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1801 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1803 if (IS_ERR(md->dax_dev)) {
1809 format_dev_t(md->name, MKDEV(_major, minor));
1811 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1815 md->pending_io = alloc_percpu(unsigned long);
1816 if (!md->pending_io)
1819 dm_stats_init(&md->stats);
1821 /* Populate the mapping, nobody knows we exist yet */
1822 spin_lock(&_minor_lock);
1823 old_md = idr_replace(&_minor_idr, md, minor);
1824 spin_unlock(&_minor_lock);
1826 BUG_ON(old_md != MINOR_ALLOCED);
1831 cleanup_mapped_device(md);
1835 module_put(THIS_MODULE);
1841 static void unlock_fs(struct mapped_device *md);
1843 static void free_dev(struct mapped_device *md)
1845 int minor = MINOR(disk_devt(md->disk));
1849 cleanup_mapped_device(md);
1851 free_table_devices(&md->table_devices);
1852 dm_stats_cleanup(&md->stats);
1855 module_put(THIS_MODULE);
1859 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1861 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1864 if (dm_table_bio_based(t)) {
1866 * The md may already have mempools that need changing.
1867 * If so, reload bioset because front_pad may have changed
1868 * because a different table was loaded.
1870 bioset_exit(&md->bs);
1871 bioset_exit(&md->io_bs);
1873 } else if (bioset_initialized(&md->bs)) {
1875 * There's no need to reload with request-based dm
1876 * because the size of front_pad doesn't change.
1877 * Note for future: If you are to reload bioset,
1878 * prep-ed requests in the queue may refer
1879 * to bio from the old bioset, so you must walk
1880 * through the queue to unprep.
1886 bioset_initialized(&md->bs) ||
1887 bioset_initialized(&md->io_bs));
1889 ret = bioset_init_from_src(&md->bs, &p->bs);
1892 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1894 bioset_exit(&md->bs);
1896 /* mempool bind completed, no longer need any mempools in the table */
1897 dm_table_free_md_mempools(t);
1902 * Bind a table to the device.
1904 static void event_callback(void *context)
1906 unsigned long flags;
1908 struct mapped_device *md = (struct mapped_device *) context;
1910 spin_lock_irqsave(&md->uevent_lock, flags);
1911 list_splice_init(&md->uevent_list, &uevents);
1912 spin_unlock_irqrestore(&md->uevent_lock, flags);
1914 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1916 atomic_inc(&md->event_nr);
1917 wake_up(&md->eventq);
1918 dm_issue_global_event();
1922 * Returns old map, which caller must destroy.
1924 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1925 struct queue_limits *limits)
1927 struct dm_table *old_map;
1928 struct request_queue *q = md->queue;
1929 bool request_based = dm_table_request_based(t);
1933 lockdep_assert_held(&md->suspend_lock);
1935 size = dm_table_get_size(t);
1938 * Wipe any geometry if the size of the table changed.
1940 if (size != dm_get_size(md))
1941 memset(&md->geometry, 0, sizeof(md->geometry));
1943 if (!get_capacity(md->disk))
1944 set_capacity(md->disk, size);
1946 set_capacity_and_notify(md->disk, size);
1948 dm_table_event_callback(t, event_callback, md);
1951 * The queue hasn't been stopped yet, if the old table type wasn't
1952 * for request-based during suspension. So stop it to prevent
1953 * I/O mapping before resume.
1954 * This must be done before setting the queue restrictions,
1955 * because request-based dm may be run just after the setting.
1960 if (request_based) {
1962 * Leverage the fact that request-based DM targets are
1963 * immutable singletons - used to optimize dm_mq_queue_rq.
1965 md->immutable_target = dm_table_get_immutable_target(t);
1968 ret = __bind_mempools(md, t);
1970 old_map = ERR_PTR(ret);
1974 ret = dm_table_set_restrictions(t, q, limits);
1976 old_map = ERR_PTR(ret);
1980 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1981 rcu_assign_pointer(md->map, (void *)t);
1982 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1992 * Returns unbound table for the caller to free.
1994 static struct dm_table *__unbind(struct mapped_device *md)
1996 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2001 dm_table_event_callback(map, NULL, NULL);
2002 RCU_INIT_POINTER(md->map, NULL);
2009 * Constructor for a new device.
2011 int dm_create(int minor, struct mapped_device **result)
2013 struct mapped_device *md;
2015 md = alloc_dev(minor);
2019 dm_ima_reset_data(md);
2026 * Functions to manage md->type.
2027 * All are required to hold md->type_lock.
2029 void dm_lock_md_type(struct mapped_device *md)
2031 mutex_lock(&md->type_lock);
2034 void dm_unlock_md_type(struct mapped_device *md)
2036 mutex_unlock(&md->type_lock);
2039 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2041 BUG_ON(!mutex_is_locked(&md->type_lock));
2045 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2050 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2052 return md->immutable_target_type;
2056 * The queue_limits are only valid as long as you have a reference
2059 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2061 BUG_ON(!atomic_read(&md->holders));
2062 return &md->queue->limits;
2064 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2067 * Setup the DM device's queue based on md's type
2069 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2071 enum dm_queue_mode type = dm_table_get_type(t);
2072 struct queue_limits limits;
2076 case DM_TYPE_REQUEST_BASED:
2077 md->disk->fops = &dm_rq_blk_dops;
2078 r = dm_mq_init_request_queue(md, t);
2080 DMERR("Cannot initialize queue for request-based dm mapped device");
2084 case DM_TYPE_BIO_BASED:
2085 case DM_TYPE_DAX_BIO_BASED:
2092 r = dm_calculate_queue_limits(t, &limits);
2094 DMERR("Cannot calculate initial queue limits");
2097 r = dm_table_set_restrictions(t, md->queue, &limits);
2103 r = dm_sysfs_init(md);
2105 del_gendisk(md->disk);
2112 struct mapped_device *dm_get_md(dev_t dev)
2114 struct mapped_device *md;
2115 unsigned minor = MINOR(dev);
2117 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2120 spin_lock(&_minor_lock);
2122 md = idr_find(&_minor_idr, minor);
2123 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2124 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2130 spin_unlock(&_minor_lock);
2134 EXPORT_SYMBOL_GPL(dm_get_md);
2136 void *dm_get_mdptr(struct mapped_device *md)
2138 return md->interface_ptr;
2141 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2143 md->interface_ptr = ptr;
2146 void dm_get(struct mapped_device *md)
2148 atomic_inc(&md->holders);
2149 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2152 int dm_hold(struct mapped_device *md)
2154 spin_lock(&_minor_lock);
2155 if (test_bit(DMF_FREEING, &md->flags)) {
2156 spin_unlock(&_minor_lock);
2160 spin_unlock(&_minor_lock);
2163 EXPORT_SYMBOL_GPL(dm_hold);
2165 const char *dm_device_name(struct mapped_device *md)
2169 EXPORT_SYMBOL_GPL(dm_device_name);
2171 static void __dm_destroy(struct mapped_device *md, bool wait)
2173 struct dm_table *map;
2178 spin_lock(&_minor_lock);
2179 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2180 set_bit(DMF_FREEING, &md->flags);
2181 spin_unlock(&_minor_lock);
2183 blk_mark_disk_dead(md->disk);
2186 * Take suspend_lock so that presuspend and postsuspend methods
2187 * do not race with internal suspend.
2189 mutex_lock(&md->suspend_lock);
2190 map = dm_get_live_table(md, &srcu_idx);
2191 if (!dm_suspended_md(md)) {
2192 dm_table_presuspend_targets(map);
2193 set_bit(DMF_SUSPENDED, &md->flags);
2194 set_bit(DMF_POST_SUSPENDING, &md->flags);
2195 dm_table_postsuspend_targets(map);
2197 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2198 dm_put_live_table(md, srcu_idx);
2199 mutex_unlock(&md->suspend_lock);
2202 * Rare, but there may be I/O requests still going to complete,
2203 * for example. Wait for all references to disappear.
2204 * No one should increment the reference count of the mapped_device,
2205 * after the mapped_device state becomes DMF_FREEING.
2208 while (atomic_read(&md->holders))
2210 else if (atomic_read(&md->holders))
2211 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2212 dm_device_name(md), atomic_read(&md->holders));
2214 dm_table_destroy(__unbind(md));
2218 void dm_destroy(struct mapped_device *md)
2220 __dm_destroy(md, true);
2223 void dm_destroy_immediate(struct mapped_device *md)
2225 __dm_destroy(md, false);
2228 void dm_put(struct mapped_device *md)
2230 atomic_dec(&md->holders);
2232 EXPORT_SYMBOL_GPL(dm_put);
2234 static bool dm_in_flight_bios(struct mapped_device *md)
2237 unsigned long sum = 0;
2239 for_each_possible_cpu(cpu)
2240 sum += *per_cpu_ptr(md->pending_io, cpu);
2245 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2251 prepare_to_wait(&md->wait, &wait, task_state);
2253 if (!dm_in_flight_bios(md))
2256 if (signal_pending_state(task_state, current)) {
2263 finish_wait(&md->wait, &wait);
2270 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2274 if (!queue_is_mq(md->queue))
2275 return dm_wait_for_bios_completion(md, task_state);
2278 if (!blk_mq_queue_inflight(md->queue))
2281 if (signal_pending_state(task_state, current)) {
2293 * Process the deferred bios
2295 static void dm_wq_work(struct work_struct *work)
2297 struct mapped_device *md = container_of(work, struct mapped_device, work);
2300 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2301 spin_lock_irq(&md->deferred_lock);
2302 bio = bio_list_pop(&md->deferred);
2303 spin_unlock_irq(&md->deferred_lock);
2308 submit_bio_noacct(bio);
2312 static void dm_queue_flush(struct mapped_device *md)
2314 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2315 smp_mb__after_atomic();
2316 queue_work(md->wq, &md->work);
2320 * Swap in a new table, returning the old one for the caller to destroy.
2322 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2324 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2325 struct queue_limits limits;
2328 mutex_lock(&md->suspend_lock);
2330 /* device must be suspended */
2331 if (!dm_suspended_md(md))
2335 * If the new table has no data devices, retain the existing limits.
2336 * This helps multipath with queue_if_no_path if all paths disappear,
2337 * then new I/O is queued based on these limits, and then some paths
2340 if (dm_table_has_no_data_devices(table)) {
2341 live_map = dm_get_live_table_fast(md);
2343 limits = md->queue->limits;
2344 dm_put_live_table_fast(md);
2348 r = dm_calculate_queue_limits(table, &limits);
2355 map = __bind(md, table, &limits);
2356 dm_issue_global_event();
2359 mutex_unlock(&md->suspend_lock);
2364 * Functions to lock and unlock any filesystem running on the
2367 static int lock_fs(struct mapped_device *md)
2371 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2373 r = freeze_bdev(md->disk->part0);
2375 set_bit(DMF_FROZEN, &md->flags);
2379 static void unlock_fs(struct mapped_device *md)
2381 if (!test_bit(DMF_FROZEN, &md->flags))
2383 thaw_bdev(md->disk->part0);
2384 clear_bit(DMF_FROZEN, &md->flags);
2388 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2389 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2390 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2392 * If __dm_suspend returns 0, the device is completely quiescent
2393 * now. There is no request-processing activity. All new requests
2394 * are being added to md->deferred list.
2396 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2397 unsigned suspend_flags, unsigned int task_state,
2398 int dmf_suspended_flag)
2400 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2401 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2404 lockdep_assert_held(&md->suspend_lock);
2407 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2408 * This flag is cleared before dm_suspend returns.
2411 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2413 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2416 * This gets reverted if there's an error later and the targets
2417 * provide the .presuspend_undo hook.
2419 dm_table_presuspend_targets(map);
2422 * Flush I/O to the device.
2423 * Any I/O submitted after lock_fs() may not be flushed.
2424 * noflush takes precedence over do_lockfs.
2425 * (lock_fs() flushes I/Os and waits for them to complete.)
2427 if (!noflush && do_lockfs) {
2430 dm_table_presuspend_undo_targets(map);
2436 * Here we must make sure that no processes are submitting requests
2437 * to target drivers i.e. no one may be executing
2438 * __split_and_process_bio from dm_submit_bio.
2440 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2441 * we take the write lock. To prevent any process from reentering
2442 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2443 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2444 * flush_workqueue(md->wq).
2446 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2448 synchronize_srcu(&md->io_barrier);
2451 * Stop md->queue before flushing md->wq in case request-based
2452 * dm defers requests to md->wq from md->queue.
2454 if (dm_request_based(md))
2455 dm_stop_queue(md->queue);
2457 flush_workqueue(md->wq);
2460 * At this point no more requests are entering target request routines.
2461 * We call dm_wait_for_completion to wait for all existing requests
2464 r = dm_wait_for_completion(md, task_state);
2466 set_bit(dmf_suspended_flag, &md->flags);
2469 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2471 synchronize_srcu(&md->io_barrier);
2473 /* were we interrupted ? */
2477 if (dm_request_based(md))
2478 dm_start_queue(md->queue);
2481 dm_table_presuspend_undo_targets(map);
2482 /* pushback list is already flushed, so skip flush */
2489 * We need to be able to change a mapping table under a mounted
2490 * filesystem. For example we might want to move some data in
2491 * the background. Before the table can be swapped with
2492 * dm_bind_table, dm_suspend must be called to flush any in
2493 * flight bios and ensure that any further io gets deferred.
2496 * Suspend mechanism in request-based dm.
2498 * 1. Flush all I/Os by lock_fs() if needed.
2499 * 2. Stop dispatching any I/O by stopping the request_queue.
2500 * 3. Wait for all in-flight I/Os to be completed or requeued.
2502 * To abort suspend, start the request_queue.
2504 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2506 struct dm_table *map = NULL;
2510 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2512 if (dm_suspended_md(md)) {
2517 if (dm_suspended_internally_md(md)) {
2518 /* already internally suspended, wait for internal resume */
2519 mutex_unlock(&md->suspend_lock);
2520 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2526 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2528 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2532 set_bit(DMF_POST_SUSPENDING, &md->flags);
2533 dm_table_postsuspend_targets(map);
2534 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2537 mutex_unlock(&md->suspend_lock);
2541 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2544 int r = dm_table_resume_targets(map);
2552 * Flushing deferred I/Os must be done after targets are resumed
2553 * so that mapping of targets can work correctly.
2554 * Request-based dm is queueing the deferred I/Os in its request_queue.
2556 if (dm_request_based(md))
2557 dm_start_queue(md->queue);
2564 int dm_resume(struct mapped_device *md)
2567 struct dm_table *map = NULL;
2571 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2573 if (!dm_suspended_md(md))
2576 if (dm_suspended_internally_md(md)) {
2577 /* already internally suspended, wait for internal resume */
2578 mutex_unlock(&md->suspend_lock);
2579 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2585 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2586 if (!map || !dm_table_get_size(map))
2589 r = __dm_resume(md, map);
2593 clear_bit(DMF_SUSPENDED, &md->flags);
2595 mutex_unlock(&md->suspend_lock);
2601 * Internal suspend/resume works like userspace-driven suspend. It waits
2602 * until all bios finish and prevents issuing new bios to the target drivers.
2603 * It may be used only from the kernel.
2606 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2608 struct dm_table *map = NULL;
2610 lockdep_assert_held(&md->suspend_lock);
2612 if (md->internal_suspend_count++)
2613 return; /* nested internal suspend */
2615 if (dm_suspended_md(md)) {
2616 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2617 return; /* nest suspend */
2620 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2623 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2624 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2625 * would require changing .presuspend to return an error -- avoid this
2626 * until there is a need for more elaborate variants of internal suspend.
2628 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2629 DMF_SUSPENDED_INTERNALLY);
2631 set_bit(DMF_POST_SUSPENDING, &md->flags);
2632 dm_table_postsuspend_targets(map);
2633 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2636 static void __dm_internal_resume(struct mapped_device *md)
2638 BUG_ON(!md->internal_suspend_count);
2640 if (--md->internal_suspend_count)
2641 return; /* resume from nested internal suspend */
2643 if (dm_suspended_md(md))
2644 goto done; /* resume from nested suspend */
2647 * NOTE: existing callers don't need to call dm_table_resume_targets
2648 * (which may fail -- so best to avoid it for now by passing NULL map)
2650 (void) __dm_resume(md, NULL);
2653 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2654 smp_mb__after_atomic();
2655 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2658 void dm_internal_suspend_noflush(struct mapped_device *md)
2660 mutex_lock(&md->suspend_lock);
2661 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2662 mutex_unlock(&md->suspend_lock);
2664 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2666 void dm_internal_resume(struct mapped_device *md)
2668 mutex_lock(&md->suspend_lock);
2669 __dm_internal_resume(md);
2670 mutex_unlock(&md->suspend_lock);
2672 EXPORT_SYMBOL_GPL(dm_internal_resume);
2675 * Fast variants of internal suspend/resume hold md->suspend_lock,
2676 * which prevents interaction with userspace-driven suspend.
2679 void dm_internal_suspend_fast(struct mapped_device *md)
2681 mutex_lock(&md->suspend_lock);
2682 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2685 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2686 synchronize_srcu(&md->io_barrier);
2687 flush_workqueue(md->wq);
2688 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2690 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2692 void dm_internal_resume_fast(struct mapped_device *md)
2694 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2700 mutex_unlock(&md->suspend_lock);
2702 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2704 /*-----------------------------------------------------------------
2705 * Event notification.
2706 *---------------------------------------------------------------*/
2707 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2712 char udev_cookie[DM_COOKIE_LENGTH];
2713 char *envp[] = { udev_cookie, NULL };
2715 noio_flag = memalloc_noio_save();
2718 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2720 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2721 DM_COOKIE_ENV_VAR_NAME, cookie);
2722 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2726 memalloc_noio_restore(noio_flag);
2731 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2733 return atomic_add_return(1, &md->uevent_seq);
2736 uint32_t dm_get_event_nr(struct mapped_device *md)
2738 return atomic_read(&md->event_nr);
2741 int dm_wait_event(struct mapped_device *md, int event_nr)
2743 return wait_event_interruptible(md->eventq,
2744 (event_nr != atomic_read(&md->event_nr)));
2747 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2749 unsigned long flags;
2751 spin_lock_irqsave(&md->uevent_lock, flags);
2752 list_add(elist, &md->uevent_list);
2753 spin_unlock_irqrestore(&md->uevent_lock, flags);
2757 * The gendisk is only valid as long as you have a reference
2760 struct gendisk *dm_disk(struct mapped_device *md)
2764 EXPORT_SYMBOL_GPL(dm_disk);
2766 struct kobject *dm_kobject(struct mapped_device *md)
2768 return &md->kobj_holder.kobj;
2771 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2773 struct mapped_device *md;
2775 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2777 spin_lock(&_minor_lock);
2778 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2784 spin_unlock(&_minor_lock);
2789 int dm_suspended_md(struct mapped_device *md)
2791 return test_bit(DMF_SUSPENDED, &md->flags);
2794 static int dm_post_suspending_md(struct mapped_device *md)
2796 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2799 int dm_suspended_internally_md(struct mapped_device *md)
2801 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2804 int dm_test_deferred_remove_flag(struct mapped_device *md)
2806 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2809 int dm_suspended(struct dm_target *ti)
2811 return dm_suspended_md(ti->table->md);
2813 EXPORT_SYMBOL_GPL(dm_suspended);
2815 int dm_post_suspending(struct dm_target *ti)
2817 return dm_post_suspending_md(ti->table->md);
2819 EXPORT_SYMBOL_GPL(dm_post_suspending);
2821 int dm_noflush_suspending(struct dm_target *ti)
2823 return __noflush_suspending(ti->table->md);
2825 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2827 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2828 unsigned integrity, unsigned per_io_data_size,
2829 unsigned min_pool_size)
2831 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2832 unsigned int pool_size = 0;
2833 unsigned int front_pad, io_front_pad;
2840 case DM_TYPE_BIO_BASED:
2841 case DM_TYPE_DAX_BIO_BASED:
2842 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2843 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2844 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2845 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2848 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2851 case DM_TYPE_REQUEST_BASED:
2852 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2853 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2854 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2860 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2864 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2870 dm_free_md_mempools(pools);
2875 void dm_free_md_mempools(struct dm_md_mempools *pools)
2880 bioset_exit(&pools->bs);
2881 bioset_exit(&pools->io_bs);
2893 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2896 struct mapped_device *md = bdev->bd_disk->private_data;
2897 struct dm_table *table;
2898 struct dm_target *ti;
2899 int ret = -ENOTTY, srcu_idx;
2901 table = dm_get_live_table(md, &srcu_idx);
2902 if (!table || !dm_table_get_size(table))
2905 /* We only support devices that have a single target */
2906 if (dm_table_get_num_targets(table) != 1)
2908 ti = dm_table_get_target(table, 0);
2910 if (dm_suspended_md(md)) {
2916 if (!ti->type->iterate_devices)
2919 ret = ti->type->iterate_devices(ti, fn, data);
2921 dm_put_live_table(md, srcu_idx);
2926 * For register / unregister we need to manually call out to every path.
2928 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2929 sector_t start, sector_t len, void *data)
2931 struct dm_pr *pr = data;
2932 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2934 if (!ops || !ops->pr_register)
2936 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2939 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2950 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2951 if (ret && new_key) {
2952 /* unregister all paths if we failed to register any path */
2953 pr.old_key = new_key;
2956 pr.fail_early = false;
2957 dm_call_pr(bdev, __dm_pr_register, &pr);
2963 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2966 struct mapped_device *md = bdev->bd_disk->private_data;
2967 const struct pr_ops *ops;
2970 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2974 ops = bdev->bd_disk->fops->pr_ops;
2975 if (ops && ops->pr_reserve)
2976 r = ops->pr_reserve(bdev, key, type, flags);
2980 dm_unprepare_ioctl(md, srcu_idx);
2984 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2986 struct mapped_device *md = bdev->bd_disk->private_data;
2987 const struct pr_ops *ops;
2990 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2994 ops = bdev->bd_disk->fops->pr_ops;
2995 if (ops && ops->pr_release)
2996 r = ops->pr_release(bdev, key, type);
3000 dm_unprepare_ioctl(md, srcu_idx);
3004 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3005 enum pr_type type, bool abort)
3007 struct mapped_device *md = bdev->bd_disk->private_data;
3008 const struct pr_ops *ops;
3011 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3015 ops = bdev->bd_disk->fops->pr_ops;
3016 if (ops && ops->pr_preempt)
3017 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3021 dm_unprepare_ioctl(md, srcu_idx);
3025 static int dm_pr_clear(struct block_device *bdev, u64 key)
3027 struct mapped_device *md = bdev->bd_disk->private_data;
3028 const struct pr_ops *ops;
3031 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3035 ops = bdev->bd_disk->fops->pr_ops;
3036 if (ops && ops->pr_clear)
3037 r = ops->pr_clear(bdev, key);
3041 dm_unprepare_ioctl(md, srcu_idx);
3045 static const struct pr_ops dm_pr_ops = {
3046 .pr_register = dm_pr_register,
3047 .pr_reserve = dm_pr_reserve,
3048 .pr_release = dm_pr_release,
3049 .pr_preempt = dm_pr_preempt,
3050 .pr_clear = dm_pr_clear,
3053 static const struct block_device_operations dm_blk_dops = {
3054 .submit_bio = dm_submit_bio,
3055 .open = dm_blk_open,
3056 .release = dm_blk_close,
3057 .ioctl = dm_blk_ioctl,
3058 .getgeo = dm_blk_getgeo,
3059 .report_zones = dm_blk_report_zones,
3060 .pr_ops = &dm_pr_ops,
3061 .owner = THIS_MODULE
3064 static const struct block_device_operations dm_rq_blk_dops = {
3065 .open = dm_blk_open,
3066 .release = dm_blk_close,
3067 .ioctl = dm_blk_ioctl,
3068 .getgeo = dm_blk_getgeo,
3069 .pr_ops = &dm_pr_ops,
3070 .owner = THIS_MODULE
3073 static const struct dax_operations dm_dax_ops = {
3074 .direct_access = dm_dax_direct_access,
3075 .dax_supported = dm_dax_supported,
3076 .copy_from_iter = dm_dax_copy_from_iter,
3077 .copy_to_iter = dm_dax_copy_to_iter,
3078 .zero_page_range = dm_dax_zero_page_range,
3084 module_init(dm_init);
3085 module_exit(dm_exit);
3087 module_param(major, uint, 0);
3088 MODULE_PARM_DESC(major, "The major number of the device mapper");
3090 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3091 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3093 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3094 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3096 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3097 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3099 MODULE_DESCRIPTION(DM_NAME " driver");
3100 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3101 MODULE_LICENSE("GPL");