2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
23 #include <linux/wait.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
33 DEFAULT_RATELIMIT_INTERVAL,
34 DEFAULT_RATELIMIT_BURST);
35 EXPORT_SYMBOL(dm_ratelimit_state);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name = DM_NAME;
47 static unsigned int major = 0;
48 static unsigned int _major = 0;
50 static DEFINE_IDR(_minor_idr);
52 static DEFINE_SPINLOCK(_minor_lock);
54 static void do_deferred_remove(struct work_struct *w);
56 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
58 static struct workqueue_struct *deferred_remove_workqueue;
61 * One of these is allocated per bio.
64 struct mapped_device *md;
68 unsigned long start_time;
69 spinlock_t endio_lock;
70 struct dm_stats_aux stats_aux;
73 #define MINOR_ALLOCED ((void *)-1)
76 * Bits for the md->flags field.
78 #define DMF_BLOCK_IO_FOR_SUSPEND 0
79 #define DMF_SUSPENDED 1
82 #define DMF_DELETING 4
83 #define DMF_NOFLUSH_SUSPENDING 5
84 #define DMF_DEFERRED_REMOVE 6
85 #define DMF_SUSPENDED_INTERNALLY 7
87 #define DM_NUMA_NODE NUMA_NO_NODE
88 static int dm_numa_node = DM_NUMA_NODE;
91 * For mempools pre-allocation at the table loading time.
93 struct dm_md_mempools {
99 struct list_head list;
101 struct dm_dev dm_dev;
104 static struct kmem_cache *_io_cache;
105 static struct kmem_cache *_rq_tio_cache;
106 static struct kmem_cache *_rq_cache;
109 * Bio-based DM's mempools' reserved IOs set by the user.
111 #define RESERVED_BIO_BASED_IOS 16
112 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
114 static int __dm_get_module_param_int(int *module_param, int min, int max)
116 int param = ACCESS_ONCE(*module_param);
117 int modified_param = 0;
118 bool modified = true;
121 modified_param = min;
122 else if (param > max)
123 modified_param = max;
128 (void)cmpxchg(module_param, param, modified_param);
129 param = modified_param;
135 unsigned __dm_get_module_param(unsigned *module_param,
136 unsigned def, unsigned max)
138 unsigned param = ACCESS_ONCE(*module_param);
139 unsigned modified_param = 0;
142 modified_param = def;
143 else if (param > max)
144 modified_param = max;
146 if (modified_param) {
147 (void)cmpxchg(module_param, param, modified_param);
148 param = modified_param;
154 unsigned dm_get_reserved_bio_based_ios(void)
156 return __dm_get_module_param(&reserved_bio_based_ios,
157 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
159 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
161 static unsigned dm_get_numa_node(void)
163 return __dm_get_module_param_int(&dm_numa_node,
164 DM_NUMA_NODE, num_online_nodes() - 1);
167 static int __init local_init(void)
171 /* allocate a slab for the dm_ios */
172 _io_cache = KMEM_CACHE(dm_io, 0);
176 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
178 goto out_free_io_cache;
180 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
181 __alignof__(struct request), 0, NULL);
183 goto out_free_rq_tio_cache;
185 r = dm_uevent_init();
187 goto out_free_rq_cache;
189 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
190 if (!deferred_remove_workqueue) {
192 goto out_uevent_exit;
196 r = register_blkdev(_major, _name);
198 goto out_free_workqueue;
206 destroy_workqueue(deferred_remove_workqueue);
210 kmem_cache_destroy(_rq_cache);
211 out_free_rq_tio_cache:
212 kmem_cache_destroy(_rq_tio_cache);
214 kmem_cache_destroy(_io_cache);
219 static void local_exit(void)
221 flush_scheduled_work();
222 destroy_workqueue(deferred_remove_workqueue);
224 kmem_cache_destroy(_rq_cache);
225 kmem_cache_destroy(_rq_tio_cache);
226 kmem_cache_destroy(_io_cache);
227 unregister_blkdev(_major, _name);
232 DMINFO("cleaned up");
235 static int (*_inits[])(void) __initdata = {
246 static void (*_exits[])(void) = {
257 static int __init dm_init(void)
259 const int count = ARRAY_SIZE(_inits);
263 for (i = 0; i < count; i++) {
278 static void __exit dm_exit(void)
280 int i = ARRAY_SIZE(_exits);
286 * Should be empty by this point.
288 idr_destroy(&_minor_idr);
292 * Block device functions
294 int dm_deleting_md(struct mapped_device *md)
296 return test_bit(DMF_DELETING, &md->flags);
299 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
301 struct mapped_device *md;
303 spin_lock(&_minor_lock);
305 md = bdev->bd_disk->private_data;
309 if (test_bit(DMF_FREEING, &md->flags) ||
310 dm_deleting_md(md)) {
316 atomic_inc(&md->open_count);
318 spin_unlock(&_minor_lock);
320 return md ? 0 : -ENXIO;
323 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
325 struct mapped_device *md;
327 spin_lock(&_minor_lock);
329 md = disk->private_data;
333 if (atomic_dec_and_test(&md->open_count) &&
334 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
335 queue_work(deferred_remove_workqueue, &deferred_remove_work);
339 spin_unlock(&_minor_lock);
342 int dm_open_count(struct mapped_device *md)
344 return atomic_read(&md->open_count);
348 * Guarantees nothing is using the device before it's deleted.
350 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
354 spin_lock(&_minor_lock);
356 if (dm_open_count(md)) {
359 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
360 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
363 set_bit(DMF_DELETING, &md->flags);
365 spin_unlock(&_minor_lock);
370 int dm_cancel_deferred_remove(struct mapped_device *md)
374 spin_lock(&_minor_lock);
376 if (test_bit(DMF_DELETING, &md->flags))
379 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
381 spin_unlock(&_minor_lock);
386 static void do_deferred_remove(struct work_struct *w)
388 dm_deferred_remove();
391 sector_t dm_get_size(struct mapped_device *md)
393 return get_capacity(md->disk);
396 struct request_queue *dm_get_md_queue(struct mapped_device *md)
401 struct dm_stats *dm_get_stats(struct mapped_device *md)
406 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
408 struct mapped_device *md = bdev->bd_disk->private_data;
410 return dm_get_geometry(md, geo);
413 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
414 struct block_device **bdev,
417 struct dm_target *tgt;
418 struct dm_table *map;
423 map = dm_get_live_table(md, &srcu_idx);
424 if (!map || !dm_table_get_size(map))
427 /* We only support devices that have a single target */
428 if (dm_table_get_num_targets(map) != 1)
431 tgt = dm_table_get_target(map, 0);
432 if (!tgt->type->prepare_ioctl)
435 if (dm_suspended_md(md)) {
440 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
445 dm_put_live_table(md, srcu_idx);
449 dm_put_live_table(md, srcu_idx);
450 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
457 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
458 unsigned int cmd, unsigned long arg)
460 struct mapped_device *md = bdev->bd_disk->private_data;
463 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
469 * Target determined this ioctl is being issued against a
470 * subset of the parent bdev; require extra privileges.
472 if (!capable(CAP_SYS_RAWIO)) {
474 "%s: sending ioctl %x to DM device without required privilege.",
481 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
487 static struct dm_io *alloc_io(struct mapped_device *md)
489 return mempool_alloc(md->io_pool, GFP_NOIO);
492 static void free_io(struct mapped_device *md, struct dm_io *io)
494 mempool_free(io, md->io_pool);
497 static void free_tio(struct dm_target_io *tio)
499 bio_put(&tio->clone);
502 int md_in_flight(struct mapped_device *md)
504 return atomic_read(&md->pending[READ]) +
505 atomic_read(&md->pending[WRITE]);
508 static void start_io_acct(struct dm_io *io)
510 struct mapped_device *md = io->md;
511 struct bio *bio = io->bio;
513 int rw = bio_data_dir(bio);
515 io->start_time = jiffies;
517 cpu = part_stat_lock();
518 part_round_stats(cpu, &dm_disk(md)->part0);
520 atomic_set(&dm_disk(md)->part0.in_flight[rw],
521 atomic_inc_return(&md->pending[rw]));
523 if (unlikely(dm_stats_used(&md->stats)))
524 dm_stats_account_io(&md->stats, bio_data_dir(bio),
525 bio->bi_iter.bi_sector, bio_sectors(bio),
526 false, 0, &io->stats_aux);
529 static void end_io_acct(struct dm_io *io)
531 struct mapped_device *md = io->md;
532 struct bio *bio = io->bio;
533 unsigned long duration = jiffies - io->start_time;
535 int rw = bio_data_dir(bio);
537 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
539 if (unlikely(dm_stats_used(&md->stats)))
540 dm_stats_account_io(&md->stats, bio_data_dir(bio),
541 bio->bi_iter.bi_sector, bio_sectors(bio),
542 true, duration, &io->stats_aux);
545 * After this is decremented the bio must not be touched if it is
548 pending = atomic_dec_return(&md->pending[rw]);
549 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
550 pending += atomic_read(&md->pending[rw^0x1]);
552 /* nudge anyone waiting on suspend queue */
558 * Add the bio to the list of deferred io.
560 static void queue_io(struct mapped_device *md, struct bio *bio)
564 spin_lock_irqsave(&md->deferred_lock, flags);
565 bio_list_add(&md->deferred, bio);
566 spin_unlock_irqrestore(&md->deferred_lock, flags);
567 queue_work(md->wq, &md->work);
571 * Everyone (including functions in this file), should use this
572 * function to access the md->map field, and make sure they call
573 * dm_put_live_table() when finished.
575 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
577 *srcu_idx = srcu_read_lock(&md->io_barrier);
579 return srcu_dereference(md->map, &md->io_barrier);
582 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
584 srcu_read_unlock(&md->io_barrier, srcu_idx);
587 void dm_sync_table(struct mapped_device *md)
589 synchronize_srcu(&md->io_barrier);
590 synchronize_rcu_expedited();
594 * A fast alternative to dm_get_live_table/dm_put_live_table.
595 * The caller must not block between these two functions.
597 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
600 return rcu_dereference(md->map);
603 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
609 * Open a table device so we can use it as a map destination.
611 static int open_table_device(struct table_device *td, dev_t dev,
612 struct mapped_device *md)
614 static char *_claim_ptr = "I belong to device-mapper";
615 struct block_device *bdev;
619 BUG_ON(td->dm_dev.bdev);
621 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
623 return PTR_ERR(bdev);
625 r = bd_link_disk_holder(bdev, dm_disk(md));
627 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
631 td->dm_dev.bdev = bdev;
636 * Close a table device that we've been using.
638 static void close_table_device(struct table_device *td, struct mapped_device *md)
640 if (!td->dm_dev.bdev)
643 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
644 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
645 td->dm_dev.bdev = NULL;
648 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
650 struct table_device *td;
652 list_for_each_entry(td, l, list)
653 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
659 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
660 struct dm_dev **result) {
662 struct table_device *td;
664 mutex_lock(&md->table_devices_lock);
665 td = find_table_device(&md->table_devices, dev, mode);
667 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
669 mutex_unlock(&md->table_devices_lock);
673 td->dm_dev.mode = mode;
674 td->dm_dev.bdev = NULL;
676 if ((r = open_table_device(td, dev, md))) {
677 mutex_unlock(&md->table_devices_lock);
682 format_dev_t(td->dm_dev.name, dev);
684 atomic_set(&td->count, 0);
685 list_add(&td->list, &md->table_devices);
687 atomic_inc(&td->count);
688 mutex_unlock(&md->table_devices_lock);
690 *result = &td->dm_dev;
693 EXPORT_SYMBOL_GPL(dm_get_table_device);
695 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
697 struct table_device *td = container_of(d, struct table_device, dm_dev);
699 mutex_lock(&md->table_devices_lock);
700 if (atomic_dec_and_test(&td->count)) {
701 close_table_device(td, md);
705 mutex_unlock(&md->table_devices_lock);
707 EXPORT_SYMBOL(dm_put_table_device);
709 static void free_table_devices(struct list_head *devices)
711 struct list_head *tmp, *next;
713 list_for_each_safe(tmp, next, devices) {
714 struct table_device *td = list_entry(tmp, struct table_device, list);
716 DMWARN("dm_destroy: %s still exists with %d references",
717 td->dm_dev.name, atomic_read(&td->count));
723 * Get the geometry associated with a dm device
725 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
733 * Set the geometry of a device.
735 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
737 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
739 if (geo->start > sz) {
740 DMWARN("Start sector is beyond the geometry limits.");
749 /*-----------------------------------------------------------------
751 * A more elegant soln is in the works that uses the queue
752 * merge fn, unfortunately there are a couple of changes to
753 * the block layer that I want to make for this. So in the
754 * interests of getting something for people to use I give
755 * you this clearly demarcated crap.
756 *---------------------------------------------------------------*/
758 static int __noflush_suspending(struct mapped_device *md)
760 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
764 * Decrements the number of outstanding ios that a bio has been
765 * cloned into, completing the original io if necc.
767 static void dec_pending(struct dm_io *io, int error)
772 struct mapped_device *md = io->md;
774 /* Push-back supersedes any I/O errors */
775 if (unlikely(error)) {
776 spin_lock_irqsave(&io->endio_lock, flags);
777 if (!(io->error > 0 && __noflush_suspending(md)))
779 spin_unlock_irqrestore(&io->endio_lock, flags);
782 if (atomic_dec_and_test(&io->io_count)) {
783 if (io->error == DM_ENDIO_REQUEUE) {
785 * Target requested pushing back the I/O.
787 spin_lock_irqsave(&md->deferred_lock, flags);
788 if (__noflush_suspending(md))
789 bio_list_add_head(&md->deferred, io->bio);
791 /* noflush suspend was interrupted. */
793 spin_unlock_irqrestore(&md->deferred_lock, flags);
796 io_error = io->error;
801 if (io_error == DM_ENDIO_REQUEUE)
804 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
806 * Preflush done for flush with data, reissue
807 * without REQ_PREFLUSH.
809 bio->bi_opf &= ~REQ_PREFLUSH;
812 /* done with normal IO or empty flush */
813 bio->bi_error = io_error;
819 void disable_write_same(struct mapped_device *md)
821 struct queue_limits *limits = dm_get_queue_limits(md);
823 /* device doesn't really support WRITE SAME, disable it */
824 limits->max_write_same_sectors = 0;
827 void disable_write_zeroes(struct mapped_device *md)
829 struct queue_limits *limits = dm_get_queue_limits(md);
831 /* device doesn't really support WRITE ZEROES, disable it */
832 limits->max_write_zeroes_sectors = 0;
835 static void clone_endio(struct bio *bio)
837 int error = bio->bi_error;
839 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
840 struct dm_io *io = tio->io;
841 struct mapped_device *md = tio->io->md;
842 dm_endio_fn endio = tio->ti->type->end_io;
845 r = endio(tio->ti, bio, error);
846 if (r < 0 || r == DM_ENDIO_REQUEUE)
848 * error and requeue request are handled
852 else if (r == DM_ENDIO_INCOMPLETE)
853 /* The target will handle the io */
856 DMWARN("unimplemented target endio return value: %d", r);
861 if (unlikely(r == -EREMOTEIO)) {
862 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
863 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)
864 disable_write_same(md);
865 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
866 !bdev_get_queue(bio->bi_bdev)->limits.max_write_zeroes_sectors)
867 disable_write_zeroes(md);
871 dec_pending(io, error);
875 * Return maximum size of I/O possible at the supplied sector up to the current
878 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
880 sector_t target_offset = dm_target_offset(ti, sector);
882 return ti->len - target_offset;
885 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
887 sector_t len = max_io_len_target_boundary(sector, ti);
888 sector_t offset, max_len;
891 * Does the target need to split even further?
893 if (ti->max_io_len) {
894 offset = dm_target_offset(ti, sector);
895 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
896 max_len = sector_div(offset, ti->max_io_len);
898 max_len = offset & (ti->max_io_len - 1);
899 max_len = ti->max_io_len - max_len;
908 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
910 if (len > UINT_MAX) {
911 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
912 (unsigned long long)len, UINT_MAX);
913 ti->error = "Maximum size of target IO is too large";
917 ti->max_io_len = (uint32_t) len;
921 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
923 static long dm_blk_direct_access(struct block_device *bdev, sector_t sector,
924 void **kaddr, pfn_t *pfn, long size)
926 struct mapped_device *md = bdev->bd_disk->private_data;
927 struct dm_table *map;
928 struct dm_target *ti;
930 long len, ret = -EIO;
932 map = dm_get_live_table(md, &srcu_idx);
936 ti = dm_table_find_target(map, sector);
937 if (!dm_target_is_valid(ti))
940 len = max_io_len(sector, ti) << SECTOR_SHIFT;
941 size = min(len, size);
943 if (ti->type->direct_access)
944 ret = ti->type->direct_access(ti, sector, kaddr, pfn, size);
946 dm_put_live_table(md, srcu_idx);
947 return min(ret, size);
951 * A target may call dm_accept_partial_bio only from the map routine. It is
952 * allowed for all bio types except REQ_PREFLUSH.
954 * dm_accept_partial_bio informs the dm that the target only wants to process
955 * additional n_sectors sectors of the bio and the rest of the data should be
956 * sent in a next bio.
958 * A diagram that explains the arithmetics:
959 * +--------------------+---------------+-------+
961 * +--------------------+---------------+-------+
963 * <-------------- *tio->len_ptr --------------->
964 * <------- bi_size ------->
967 * Region 1 was already iterated over with bio_advance or similar function.
968 * (it may be empty if the target doesn't use bio_advance)
969 * Region 2 is the remaining bio size that the target wants to process.
970 * (it may be empty if region 1 is non-empty, although there is no reason
972 * The target requires that region 3 is to be sent in the next bio.
974 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
975 * the partially processed part (the sum of regions 1+2) must be the same for all
978 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
980 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
981 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
982 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
983 BUG_ON(bi_size > *tio->len_ptr);
984 BUG_ON(n_sectors > bi_size);
985 *tio->len_ptr -= bi_size - n_sectors;
986 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
988 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
991 * Flush current->bio_list when the target map method blocks.
992 * This fixes deadlocks in snapshot and possibly in other targets.
995 struct blk_plug plug;
996 struct blk_plug_cb cb;
999 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1001 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1002 struct bio_list list;
1006 INIT_LIST_HEAD(&o->cb.list);
1008 if (unlikely(!current->bio_list))
1011 for (i = 0; i < 2; i++) {
1012 list = current->bio_list[i];
1013 bio_list_init(¤t->bio_list[i]);
1015 while ((bio = bio_list_pop(&list))) {
1016 struct bio_set *bs = bio->bi_pool;
1017 if (unlikely(!bs) || bs == fs_bio_set) {
1018 bio_list_add(¤t->bio_list[i], bio);
1022 spin_lock(&bs->rescue_lock);
1023 bio_list_add(&bs->rescue_list, bio);
1024 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1025 spin_unlock(&bs->rescue_lock);
1030 static void dm_offload_start(struct dm_offload *o)
1032 blk_start_plug(&o->plug);
1033 o->cb.callback = flush_current_bio_list;
1034 list_add(&o->cb.list, ¤t->plug->cb_list);
1037 static void dm_offload_end(struct dm_offload *o)
1039 list_del(&o->cb.list);
1040 blk_finish_plug(&o->plug);
1043 static void __map_bio(struct dm_target_io *tio)
1047 struct dm_offload o;
1048 struct bio *clone = &tio->clone;
1049 struct dm_target *ti = tio->ti;
1051 clone->bi_end_io = clone_endio;
1054 * Map the clone. If r == 0 we don't need to do
1055 * anything, the target has assumed ownership of
1058 atomic_inc(&tio->io->io_count);
1059 sector = clone->bi_iter.bi_sector;
1061 dm_offload_start(&o);
1062 r = ti->type->map(ti, clone);
1065 if (r == DM_MAPIO_REMAPPED) {
1066 /* the bio has been remapped so dispatch it */
1068 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1069 tio->io->bio->bi_bdev->bd_dev, sector);
1071 generic_make_request(clone);
1072 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1073 /* error the io and bail out, or requeue it if needed */
1074 dec_pending(tio->io, r);
1076 } else if (r != DM_MAPIO_SUBMITTED) {
1077 DMWARN("unimplemented target map return value: %d", r);
1083 struct mapped_device *md;
1084 struct dm_table *map;
1088 unsigned sector_count;
1091 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1093 bio->bi_iter.bi_sector = sector;
1094 bio->bi_iter.bi_size = to_bytes(len);
1098 * Creates a bio that consists of range of complete bvecs.
1100 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1101 sector_t sector, unsigned len)
1103 struct bio *clone = &tio->clone;
1105 __bio_clone_fast(clone, bio);
1107 if (unlikely(bio_integrity(bio) != NULL)) {
1110 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1111 !dm_target_passes_integrity(tio->ti->type))) {
1112 DMWARN("%s: the target %s doesn't support integrity data.",
1113 dm_device_name(tio->io->md),
1114 tio->ti->type->name);
1118 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1123 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1124 clone->bi_iter.bi_size = to_bytes(len);
1126 if (unlikely(bio_integrity(bio) != NULL))
1127 bio_integrity_trim(clone, 0, len);
1132 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1133 struct dm_target *ti,
1134 unsigned target_bio_nr)
1136 struct dm_target_io *tio;
1139 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1140 tio = container_of(clone, struct dm_target_io, clone);
1144 tio->target_bio_nr = target_bio_nr;
1149 static void __clone_and_map_simple_bio(struct clone_info *ci,
1150 struct dm_target *ti,
1151 unsigned target_bio_nr, unsigned *len)
1153 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1154 struct bio *clone = &tio->clone;
1158 __bio_clone_fast(clone, ci->bio);
1160 bio_setup_sector(clone, ci->sector, *len);
1165 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1166 unsigned num_bios, unsigned *len)
1168 unsigned target_bio_nr;
1170 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1171 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1174 static int __send_empty_flush(struct clone_info *ci)
1176 unsigned target_nr = 0;
1177 struct dm_target *ti;
1179 BUG_ON(bio_has_data(ci->bio));
1180 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1181 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1186 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1187 sector_t sector, unsigned *len)
1189 struct bio *bio = ci->bio;
1190 struct dm_target_io *tio;
1191 unsigned target_bio_nr;
1192 unsigned num_target_bios = 1;
1196 * Does the target want to receive duplicate copies of the bio?
1198 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1199 num_target_bios = ti->num_write_bios(ti, bio);
1201 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1202 tio = alloc_tio(ci, ti, target_bio_nr);
1204 r = clone_bio(tio, bio, sector, *len);
1215 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1217 static unsigned get_num_discard_bios(struct dm_target *ti)
1219 return ti->num_discard_bios;
1222 static unsigned get_num_write_same_bios(struct dm_target *ti)
1224 return ti->num_write_same_bios;
1227 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1229 return ti->num_write_zeroes_bios;
1232 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1234 static bool is_split_required_for_discard(struct dm_target *ti)
1236 return ti->split_discard_bios;
1239 static int __send_changing_extent_only(struct clone_info *ci,
1240 get_num_bios_fn get_num_bios,
1241 is_split_required_fn is_split_required)
1243 struct dm_target *ti;
1248 ti = dm_table_find_target(ci->map, ci->sector);
1249 if (!dm_target_is_valid(ti))
1253 * Even though the device advertised support for this type of
1254 * request, that does not mean every target supports it, and
1255 * reconfiguration might also have changed that since the
1256 * check was performed.
1258 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1262 if (is_split_required && !is_split_required(ti))
1263 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1265 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1267 __send_duplicate_bios(ci, ti, num_bios, &len);
1270 } while (ci->sector_count -= len);
1275 static int __send_discard(struct clone_info *ci)
1277 return __send_changing_extent_only(ci, get_num_discard_bios,
1278 is_split_required_for_discard);
1281 static int __send_write_same(struct clone_info *ci)
1283 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1286 static int __send_write_zeroes(struct clone_info *ci)
1288 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1292 * Select the correct strategy for processing a non-flush bio.
1294 static int __split_and_process_non_flush(struct clone_info *ci)
1296 struct bio *bio = ci->bio;
1297 struct dm_target *ti;
1301 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1302 return __send_discard(ci);
1303 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1304 return __send_write_same(ci);
1305 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1306 return __send_write_zeroes(ci);
1308 ti = dm_table_find_target(ci->map, ci->sector);
1309 if (!dm_target_is_valid(ti))
1312 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1314 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1319 ci->sector_count -= len;
1325 * Entry point to split a bio into clones and submit them to the targets.
1327 static void __split_and_process_bio(struct mapped_device *md,
1328 struct dm_table *map, struct bio *bio)
1330 struct clone_info ci;
1333 if (unlikely(!map)) {
1340 ci.io = alloc_io(md);
1342 atomic_set(&ci.io->io_count, 1);
1345 spin_lock_init(&ci.io->endio_lock);
1346 ci.sector = bio->bi_iter.bi_sector;
1348 start_io_acct(ci.io);
1350 if (bio->bi_opf & REQ_PREFLUSH) {
1351 ci.bio = &ci.md->flush_bio;
1352 ci.sector_count = 0;
1353 error = __send_empty_flush(&ci);
1354 /* dec_pending submits any data associated with flush */
1357 ci.sector_count = bio_sectors(bio);
1358 while (ci.sector_count && !error)
1359 error = __split_and_process_non_flush(&ci);
1362 /* drop the extra reference count */
1363 dec_pending(ci.io, error);
1365 /*-----------------------------------------------------------------
1367 *---------------------------------------------------------------*/
1370 * The request function that just remaps the bio built up by
1373 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1375 int rw = bio_data_dir(bio);
1376 struct mapped_device *md = q->queuedata;
1378 struct dm_table *map;
1380 map = dm_get_live_table(md, &srcu_idx);
1382 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1384 /* if we're suspended, we have to queue this io for later */
1385 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1386 dm_put_live_table(md, srcu_idx);
1388 if (!(bio->bi_opf & REQ_RAHEAD))
1392 return BLK_QC_T_NONE;
1395 __split_and_process_bio(md, map, bio);
1396 dm_put_live_table(md, srcu_idx);
1397 return BLK_QC_T_NONE;
1400 static int dm_any_congested(void *congested_data, int bdi_bits)
1403 struct mapped_device *md = congested_data;
1404 struct dm_table *map;
1406 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1407 if (dm_request_based(md)) {
1409 * With request-based DM we only need to check the
1410 * top-level queue for congestion.
1412 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1414 map = dm_get_live_table_fast(md);
1416 r = dm_table_any_congested(map, bdi_bits);
1417 dm_put_live_table_fast(md);
1424 /*-----------------------------------------------------------------
1425 * An IDR is used to keep track of allocated minor numbers.
1426 *---------------------------------------------------------------*/
1427 static void free_minor(int minor)
1429 spin_lock(&_minor_lock);
1430 idr_remove(&_minor_idr, minor);
1431 spin_unlock(&_minor_lock);
1435 * See if the device with a specific minor # is free.
1437 static int specific_minor(int minor)
1441 if (minor >= (1 << MINORBITS))
1444 idr_preload(GFP_KERNEL);
1445 spin_lock(&_minor_lock);
1447 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1449 spin_unlock(&_minor_lock);
1452 return r == -ENOSPC ? -EBUSY : r;
1456 static int next_free_minor(int *minor)
1460 idr_preload(GFP_KERNEL);
1461 spin_lock(&_minor_lock);
1463 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1465 spin_unlock(&_minor_lock);
1473 static const struct block_device_operations dm_blk_dops;
1475 static void dm_wq_work(struct work_struct *work);
1477 void dm_init_md_queue(struct mapped_device *md)
1480 * Request-based dm devices cannot be stacked on top of bio-based dm
1481 * devices. The type of this dm device may not have been decided yet.
1482 * The type is decided at the first table loading time.
1483 * To prevent problematic device stacking, clear the queue flag
1484 * for request stacking support until then.
1486 * This queue is new, so no concurrency on the queue_flags.
1488 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1491 * Initialize data that will only be used by a non-blk-mq DM queue
1492 * - must do so here (in alloc_dev callchain) before queue is used
1494 md->queue->queuedata = md;
1495 md->queue->backing_dev_info->congested_data = md;
1498 void dm_init_normal_md_queue(struct mapped_device *md)
1500 md->use_blk_mq = false;
1501 dm_init_md_queue(md);
1504 * Initialize aspects of queue that aren't relevant for blk-mq
1506 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1507 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1510 static void cleanup_mapped_device(struct mapped_device *md)
1513 destroy_workqueue(md->wq);
1514 if (md->kworker_task)
1515 kthread_stop(md->kworker_task);
1516 mempool_destroy(md->io_pool);
1518 bioset_free(md->bs);
1521 spin_lock(&_minor_lock);
1522 md->disk->private_data = NULL;
1523 spin_unlock(&_minor_lock);
1524 del_gendisk(md->disk);
1529 blk_cleanup_queue(md->queue);
1531 cleanup_srcu_struct(&md->io_barrier);
1538 dm_mq_cleanup_mapped_device(md);
1542 * Allocate and initialise a blank device with a given minor.
1544 static struct mapped_device *alloc_dev(int minor)
1546 int r, numa_node_id = dm_get_numa_node();
1547 struct mapped_device *md;
1550 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1552 DMWARN("unable to allocate device, out of memory.");
1556 if (!try_module_get(THIS_MODULE))
1557 goto bad_module_get;
1559 /* get a minor number for the dev */
1560 if (minor == DM_ANY_MINOR)
1561 r = next_free_minor(&minor);
1563 r = specific_minor(minor);
1567 r = init_srcu_struct(&md->io_barrier);
1569 goto bad_io_barrier;
1571 md->numa_node_id = numa_node_id;
1572 md->use_blk_mq = dm_use_blk_mq_default();
1573 md->init_tio_pdu = false;
1574 md->type = DM_TYPE_NONE;
1575 mutex_init(&md->suspend_lock);
1576 mutex_init(&md->type_lock);
1577 mutex_init(&md->table_devices_lock);
1578 spin_lock_init(&md->deferred_lock);
1579 atomic_set(&md->holders, 1);
1580 atomic_set(&md->open_count, 0);
1581 atomic_set(&md->event_nr, 0);
1582 atomic_set(&md->uevent_seq, 0);
1583 INIT_LIST_HEAD(&md->uevent_list);
1584 INIT_LIST_HEAD(&md->table_devices);
1585 spin_lock_init(&md->uevent_lock);
1587 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1591 dm_init_md_queue(md);
1593 md->disk = alloc_disk_node(1, numa_node_id);
1597 atomic_set(&md->pending[0], 0);
1598 atomic_set(&md->pending[1], 0);
1599 init_waitqueue_head(&md->wait);
1600 INIT_WORK(&md->work, dm_wq_work);
1601 init_waitqueue_head(&md->eventq);
1602 init_completion(&md->kobj_holder.completion);
1603 md->kworker_task = NULL;
1605 md->disk->major = _major;
1606 md->disk->first_minor = minor;
1607 md->disk->fops = &dm_blk_dops;
1608 md->disk->queue = md->queue;
1609 md->disk->private_data = md;
1610 sprintf(md->disk->disk_name, "dm-%d", minor);
1612 format_dev_t(md->name, MKDEV(_major, minor));
1614 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1618 md->bdev = bdget_disk(md->disk, 0);
1622 bio_init(&md->flush_bio, NULL, 0);
1623 md->flush_bio.bi_bdev = md->bdev;
1624 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
1626 dm_stats_init(&md->stats);
1628 /* Populate the mapping, nobody knows we exist yet */
1629 spin_lock(&_minor_lock);
1630 old_md = idr_replace(&_minor_idr, md, minor);
1631 spin_unlock(&_minor_lock);
1633 BUG_ON(old_md != MINOR_ALLOCED);
1638 cleanup_mapped_device(md);
1642 module_put(THIS_MODULE);
1648 static void unlock_fs(struct mapped_device *md);
1650 static void free_dev(struct mapped_device *md)
1652 int minor = MINOR(disk_devt(md->disk));
1656 cleanup_mapped_device(md);
1658 free_table_devices(&md->table_devices);
1659 dm_stats_cleanup(&md->stats);
1662 module_put(THIS_MODULE);
1666 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1668 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1671 /* The md already has necessary mempools. */
1672 if (dm_table_bio_based(t)) {
1674 * Reload bioset because front_pad may have changed
1675 * because a different table was loaded.
1677 bioset_free(md->bs);
1682 * There's no need to reload with request-based dm
1683 * because the size of front_pad doesn't change.
1684 * Note for future: If you are to reload bioset,
1685 * prep-ed requests in the queue may refer
1686 * to bio from the old bioset, so you must walk
1687 * through the queue to unprep.
1692 BUG_ON(!p || md->io_pool || md->bs);
1694 md->io_pool = p->io_pool;
1700 /* mempool bind completed, no longer need any mempools in the table */
1701 dm_table_free_md_mempools(t);
1705 * Bind a table to the device.
1707 static void event_callback(void *context)
1709 unsigned long flags;
1711 struct mapped_device *md = (struct mapped_device *) context;
1713 spin_lock_irqsave(&md->uevent_lock, flags);
1714 list_splice_init(&md->uevent_list, &uevents);
1715 spin_unlock_irqrestore(&md->uevent_lock, flags);
1717 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1719 atomic_inc(&md->event_nr);
1720 wake_up(&md->eventq);
1724 * Protected by md->suspend_lock obtained by dm_swap_table().
1726 static void __set_size(struct mapped_device *md, sector_t size)
1728 lockdep_assert_held(&md->suspend_lock);
1730 set_capacity(md->disk, size);
1732 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1736 * Returns old map, which caller must destroy.
1738 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1739 struct queue_limits *limits)
1741 struct dm_table *old_map;
1742 struct request_queue *q = md->queue;
1745 lockdep_assert_held(&md->suspend_lock);
1747 size = dm_table_get_size(t);
1750 * Wipe any geometry if the size of the table changed.
1752 if (size != dm_get_size(md))
1753 memset(&md->geometry, 0, sizeof(md->geometry));
1755 __set_size(md, size);
1757 dm_table_event_callback(t, event_callback, md);
1760 * The queue hasn't been stopped yet, if the old table type wasn't
1761 * for request-based during suspension. So stop it to prevent
1762 * I/O mapping before resume.
1763 * This must be done before setting the queue restrictions,
1764 * because request-based dm may be run just after the setting.
1766 if (dm_table_request_based(t)) {
1769 * Leverage the fact that request-based DM targets are
1770 * immutable singletons and establish md->immutable_target
1771 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1773 md->immutable_target = dm_table_get_immutable_target(t);
1776 __bind_mempools(md, t);
1778 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1779 rcu_assign_pointer(md->map, (void *)t);
1780 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1782 dm_table_set_restrictions(t, q, limits);
1790 * Returns unbound table for the caller to free.
1792 static struct dm_table *__unbind(struct mapped_device *md)
1794 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1799 dm_table_event_callback(map, NULL, NULL);
1800 RCU_INIT_POINTER(md->map, NULL);
1807 * Constructor for a new device.
1809 int dm_create(int minor, struct mapped_device **result)
1811 struct mapped_device *md;
1813 md = alloc_dev(minor);
1824 * Functions to manage md->type.
1825 * All are required to hold md->type_lock.
1827 void dm_lock_md_type(struct mapped_device *md)
1829 mutex_lock(&md->type_lock);
1832 void dm_unlock_md_type(struct mapped_device *md)
1834 mutex_unlock(&md->type_lock);
1837 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1839 BUG_ON(!mutex_is_locked(&md->type_lock));
1843 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1848 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1850 return md->immutable_target_type;
1854 * The queue_limits are only valid as long as you have a reference
1857 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1859 BUG_ON(!atomic_read(&md->holders));
1860 return &md->queue->limits;
1862 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1865 * Setup the DM device's queue based on md's type
1867 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1870 enum dm_queue_mode type = dm_get_md_type(md);
1873 case DM_TYPE_REQUEST_BASED:
1874 r = dm_old_init_request_queue(md, t);
1876 DMERR("Cannot initialize queue for request-based mapped device");
1880 case DM_TYPE_MQ_REQUEST_BASED:
1881 r = dm_mq_init_request_queue(md, t);
1883 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1887 case DM_TYPE_BIO_BASED:
1888 case DM_TYPE_DAX_BIO_BASED:
1889 dm_init_normal_md_queue(md);
1890 blk_queue_make_request(md->queue, dm_make_request);
1892 * DM handles splitting bios as needed. Free the bio_split bioset
1893 * since it won't be used (saves 1 process per bio-based DM device).
1895 bioset_free(md->queue->bio_split);
1896 md->queue->bio_split = NULL;
1898 if (type == DM_TYPE_DAX_BIO_BASED)
1899 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1909 struct mapped_device *dm_get_md(dev_t dev)
1911 struct mapped_device *md;
1912 unsigned minor = MINOR(dev);
1914 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1917 spin_lock(&_minor_lock);
1919 md = idr_find(&_minor_idr, minor);
1921 if ((md == MINOR_ALLOCED ||
1922 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1923 dm_deleting_md(md) ||
1924 test_bit(DMF_FREEING, &md->flags))) {
1932 spin_unlock(&_minor_lock);
1936 EXPORT_SYMBOL_GPL(dm_get_md);
1938 void *dm_get_mdptr(struct mapped_device *md)
1940 return md->interface_ptr;
1943 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1945 md->interface_ptr = ptr;
1948 void dm_get(struct mapped_device *md)
1950 atomic_inc(&md->holders);
1951 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1954 int dm_hold(struct mapped_device *md)
1956 spin_lock(&_minor_lock);
1957 if (test_bit(DMF_FREEING, &md->flags)) {
1958 spin_unlock(&_minor_lock);
1962 spin_unlock(&_minor_lock);
1965 EXPORT_SYMBOL_GPL(dm_hold);
1967 const char *dm_device_name(struct mapped_device *md)
1971 EXPORT_SYMBOL_GPL(dm_device_name);
1973 static void __dm_destroy(struct mapped_device *md, bool wait)
1975 struct request_queue *q = dm_get_md_queue(md);
1976 struct dm_table *map;
1981 spin_lock(&_minor_lock);
1982 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
1983 set_bit(DMF_FREEING, &md->flags);
1984 spin_unlock(&_minor_lock);
1986 blk_set_queue_dying(q);
1988 if (dm_request_based(md) && md->kworker_task)
1989 kthread_flush_worker(&md->kworker);
1992 * Take suspend_lock so that presuspend and postsuspend methods
1993 * do not race with internal suspend.
1995 mutex_lock(&md->suspend_lock);
1996 map = dm_get_live_table(md, &srcu_idx);
1997 if (!dm_suspended_md(md)) {
1998 dm_table_presuspend_targets(map);
1999 dm_table_postsuspend_targets(map);
2001 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2002 dm_put_live_table(md, srcu_idx);
2003 mutex_unlock(&md->suspend_lock);
2006 * Rare, but there may be I/O requests still going to complete,
2007 * for example. Wait for all references to disappear.
2008 * No one should increment the reference count of the mapped_device,
2009 * after the mapped_device state becomes DMF_FREEING.
2012 while (atomic_read(&md->holders))
2014 else if (atomic_read(&md->holders))
2015 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2016 dm_device_name(md), atomic_read(&md->holders));
2019 dm_table_destroy(__unbind(md));
2023 void dm_destroy(struct mapped_device *md)
2025 __dm_destroy(md, true);
2028 void dm_destroy_immediate(struct mapped_device *md)
2030 __dm_destroy(md, false);
2033 void dm_put(struct mapped_device *md)
2035 atomic_dec(&md->holders);
2037 EXPORT_SYMBOL_GPL(dm_put);
2039 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2045 prepare_to_wait(&md->wait, &wait, task_state);
2047 if (!md_in_flight(md))
2050 if (signal_pending_state(task_state, current)) {
2057 finish_wait(&md->wait, &wait);
2063 * Process the deferred bios
2065 static void dm_wq_work(struct work_struct *work)
2067 struct mapped_device *md = container_of(work, struct mapped_device,
2071 struct dm_table *map;
2073 map = dm_get_live_table(md, &srcu_idx);
2075 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2076 spin_lock_irq(&md->deferred_lock);
2077 c = bio_list_pop(&md->deferred);
2078 spin_unlock_irq(&md->deferred_lock);
2083 if (dm_request_based(md))
2084 generic_make_request(c);
2086 __split_and_process_bio(md, map, c);
2089 dm_put_live_table(md, srcu_idx);
2092 static void dm_queue_flush(struct mapped_device *md)
2094 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2095 smp_mb__after_atomic();
2096 queue_work(md->wq, &md->work);
2100 * Swap in a new table, returning the old one for the caller to destroy.
2102 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2104 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2105 struct queue_limits limits;
2108 mutex_lock(&md->suspend_lock);
2110 /* device must be suspended */
2111 if (!dm_suspended_md(md))
2115 * If the new table has no data devices, retain the existing limits.
2116 * This helps multipath with queue_if_no_path if all paths disappear,
2117 * then new I/O is queued based on these limits, and then some paths
2120 if (dm_table_has_no_data_devices(table)) {
2121 live_map = dm_get_live_table_fast(md);
2123 limits = md->queue->limits;
2124 dm_put_live_table_fast(md);
2128 r = dm_calculate_queue_limits(table, &limits);
2135 map = __bind(md, table, &limits);
2138 mutex_unlock(&md->suspend_lock);
2143 * Functions to lock and unlock any filesystem running on the
2146 static int lock_fs(struct mapped_device *md)
2150 WARN_ON(md->frozen_sb);
2152 md->frozen_sb = freeze_bdev(md->bdev);
2153 if (IS_ERR(md->frozen_sb)) {
2154 r = PTR_ERR(md->frozen_sb);
2155 md->frozen_sb = NULL;
2159 set_bit(DMF_FROZEN, &md->flags);
2164 static void unlock_fs(struct mapped_device *md)
2166 if (!test_bit(DMF_FROZEN, &md->flags))
2169 thaw_bdev(md->bdev, md->frozen_sb);
2170 md->frozen_sb = NULL;
2171 clear_bit(DMF_FROZEN, &md->flags);
2175 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2176 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2177 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2179 * If __dm_suspend returns 0, the device is completely quiescent
2180 * now. There is no request-processing activity. All new requests
2181 * are being added to md->deferred list.
2183 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2184 unsigned suspend_flags, long task_state,
2185 int dmf_suspended_flag)
2187 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2188 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2191 lockdep_assert_held(&md->suspend_lock);
2194 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2195 * This flag is cleared before dm_suspend returns.
2198 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2200 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2203 * This gets reverted if there's an error later and the targets
2204 * provide the .presuspend_undo hook.
2206 dm_table_presuspend_targets(map);
2209 * Flush I/O to the device.
2210 * Any I/O submitted after lock_fs() may not be flushed.
2211 * noflush takes precedence over do_lockfs.
2212 * (lock_fs() flushes I/Os and waits for them to complete.)
2214 if (!noflush && do_lockfs) {
2217 dm_table_presuspend_undo_targets(map);
2223 * Here we must make sure that no processes are submitting requests
2224 * to target drivers i.e. no one may be executing
2225 * __split_and_process_bio. This is called from dm_request and
2228 * To get all processes out of __split_and_process_bio in dm_request,
2229 * we take the write lock. To prevent any process from reentering
2230 * __split_and_process_bio from dm_request and quiesce the thread
2231 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2232 * flush_workqueue(md->wq).
2234 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2236 synchronize_srcu(&md->io_barrier);
2239 * Stop md->queue before flushing md->wq in case request-based
2240 * dm defers requests to md->wq from md->queue.
2242 if (dm_request_based(md)) {
2243 dm_stop_queue(md->queue);
2244 if (md->kworker_task)
2245 kthread_flush_worker(&md->kworker);
2248 flush_workqueue(md->wq);
2251 * At this point no more requests are entering target request routines.
2252 * We call dm_wait_for_completion to wait for all existing requests
2255 r = dm_wait_for_completion(md, task_state);
2257 set_bit(dmf_suspended_flag, &md->flags);
2260 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2262 synchronize_srcu(&md->io_barrier);
2264 /* were we interrupted ? */
2268 if (dm_request_based(md))
2269 dm_start_queue(md->queue);
2272 dm_table_presuspend_undo_targets(map);
2273 /* pushback list is already flushed, so skip flush */
2280 * We need to be able to change a mapping table under a mounted
2281 * filesystem. For example we might want to move some data in
2282 * the background. Before the table can be swapped with
2283 * dm_bind_table, dm_suspend must be called to flush any in
2284 * flight bios and ensure that any further io gets deferred.
2287 * Suspend mechanism in request-based dm.
2289 * 1. Flush all I/Os by lock_fs() if needed.
2290 * 2. Stop dispatching any I/O by stopping the request_queue.
2291 * 3. Wait for all in-flight I/Os to be completed or requeued.
2293 * To abort suspend, start the request_queue.
2295 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2297 struct dm_table *map = NULL;
2301 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2303 if (dm_suspended_md(md)) {
2308 if (dm_suspended_internally_md(md)) {
2309 /* already internally suspended, wait for internal resume */
2310 mutex_unlock(&md->suspend_lock);
2311 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2317 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2319 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2323 dm_table_postsuspend_targets(map);
2326 mutex_unlock(&md->suspend_lock);
2330 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2333 int r = dm_table_resume_targets(map);
2341 * Flushing deferred I/Os must be done after targets are resumed
2342 * so that mapping of targets can work correctly.
2343 * Request-based dm is queueing the deferred I/Os in its request_queue.
2345 if (dm_request_based(md))
2346 dm_start_queue(md->queue);
2353 int dm_resume(struct mapped_device *md)
2356 struct dm_table *map = NULL;
2360 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2362 if (!dm_suspended_md(md))
2365 if (dm_suspended_internally_md(md)) {
2366 /* already internally suspended, wait for internal resume */
2367 mutex_unlock(&md->suspend_lock);
2368 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2374 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2375 if (!map || !dm_table_get_size(map))
2378 r = __dm_resume(md, map);
2382 clear_bit(DMF_SUSPENDED, &md->flags);
2384 mutex_unlock(&md->suspend_lock);
2390 * Internal suspend/resume works like userspace-driven suspend. It waits
2391 * until all bios finish and prevents issuing new bios to the target drivers.
2392 * It may be used only from the kernel.
2395 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2397 struct dm_table *map = NULL;
2399 lockdep_assert_held(&md->suspend_lock);
2401 if (md->internal_suspend_count++)
2402 return; /* nested internal suspend */
2404 if (dm_suspended_md(md)) {
2405 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2406 return; /* nest suspend */
2409 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2412 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2413 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2414 * would require changing .presuspend to return an error -- avoid this
2415 * until there is a need for more elaborate variants of internal suspend.
2417 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2418 DMF_SUSPENDED_INTERNALLY);
2420 dm_table_postsuspend_targets(map);
2423 static void __dm_internal_resume(struct mapped_device *md)
2425 BUG_ON(!md->internal_suspend_count);
2427 if (--md->internal_suspend_count)
2428 return; /* resume from nested internal suspend */
2430 if (dm_suspended_md(md))
2431 goto done; /* resume from nested suspend */
2434 * NOTE: existing callers don't need to call dm_table_resume_targets
2435 * (which may fail -- so best to avoid it for now by passing NULL map)
2437 (void) __dm_resume(md, NULL);
2440 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2441 smp_mb__after_atomic();
2442 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2445 void dm_internal_suspend_noflush(struct mapped_device *md)
2447 mutex_lock(&md->suspend_lock);
2448 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2449 mutex_unlock(&md->suspend_lock);
2451 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2453 void dm_internal_resume(struct mapped_device *md)
2455 mutex_lock(&md->suspend_lock);
2456 __dm_internal_resume(md);
2457 mutex_unlock(&md->suspend_lock);
2459 EXPORT_SYMBOL_GPL(dm_internal_resume);
2462 * Fast variants of internal suspend/resume hold md->suspend_lock,
2463 * which prevents interaction with userspace-driven suspend.
2466 void dm_internal_suspend_fast(struct mapped_device *md)
2468 mutex_lock(&md->suspend_lock);
2469 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2472 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2473 synchronize_srcu(&md->io_barrier);
2474 flush_workqueue(md->wq);
2475 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2477 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2479 void dm_internal_resume_fast(struct mapped_device *md)
2481 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2487 mutex_unlock(&md->suspend_lock);
2489 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2491 /*-----------------------------------------------------------------
2492 * Event notification.
2493 *---------------------------------------------------------------*/
2494 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2497 char udev_cookie[DM_COOKIE_LENGTH];
2498 char *envp[] = { udev_cookie, NULL };
2501 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2503 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2504 DM_COOKIE_ENV_VAR_NAME, cookie);
2505 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2510 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2512 return atomic_add_return(1, &md->uevent_seq);
2515 uint32_t dm_get_event_nr(struct mapped_device *md)
2517 return atomic_read(&md->event_nr);
2520 int dm_wait_event(struct mapped_device *md, int event_nr)
2522 return wait_event_interruptible(md->eventq,
2523 (event_nr != atomic_read(&md->event_nr)));
2526 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2528 unsigned long flags;
2530 spin_lock_irqsave(&md->uevent_lock, flags);
2531 list_add(elist, &md->uevent_list);
2532 spin_unlock_irqrestore(&md->uevent_lock, flags);
2536 * The gendisk is only valid as long as you have a reference
2539 struct gendisk *dm_disk(struct mapped_device *md)
2543 EXPORT_SYMBOL_GPL(dm_disk);
2545 struct kobject *dm_kobject(struct mapped_device *md)
2547 return &md->kobj_holder.kobj;
2550 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2552 struct mapped_device *md;
2554 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2556 if (test_bit(DMF_FREEING, &md->flags) ||
2564 int dm_suspended_md(struct mapped_device *md)
2566 return test_bit(DMF_SUSPENDED, &md->flags);
2569 int dm_suspended_internally_md(struct mapped_device *md)
2571 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2574 int dm_test_deferred_remove_flag(struct mapped_device *md)
2576 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2579 int dm_suspended(struct dm_target *ti)
2581 return dm_suspended_md(dm_table_get_md(ti->table));
2583 EXPORT_SYMBOL_GPL(dm_suspended);
2585 int dm_noflush_suspending(struct dm_target *ti)
2587 return __noflush_suspending(dm_table_get_md(ti->table));
2589 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2591 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2592 unsigned integrity, unsigned per_io_data_size)
2594 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2595 unsigned int pool_size = 0;
2596 unsigned int front_pad;
2602 case DM_TYPE_BIO_BASED:
2603 case DM_TYPE_DAX_BIO_BASED:
2604 pool_size = dm_get_reserved_bio_based_ios();
2605 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2607 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2608 if (!pools->io_pool)
2611 case DM_TYPE_REQUEST_BASED:
2612 case DM_TYPE_MQ_REQUEST_BASED:
2613 pool_size = dm_get_reserved_rq_based_ios();
2614 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2615 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2621 pools->bs = bioset_create_nobvec(pool_size, front_pad);
2625 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2631 dm_free_md_mempools(pools);
2636 void dm_free_md_mempools(struct dm_md_mempools *pools)
2641 mempool_destroy(pools->io_pool);
2644 bioset_free(pools->bs);
2656 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2659 struct mapped_device *md = bdev->bd_disk->private_data;
2660 struct dm_table *table;
2661 struct dm_target *ti;
2662 int ret = -ENOTTY, srcu_idx;
2664 table = dm_get_live_table(md, &srcu_idx);
2665 if (!table || !dm_table_get_size(table))
2668 /* We only support devices that have a single target */
2669 if (dm_table_get_num_targets(table) != 1)
2671 ti = dm_table_get_target(table, 0);
2674 if (!ti->type->iterate_devices)
2677 ret = ti->type->iterate_devices(ti, fn, data);
2679 dm_put_live_table(md, srcu_idx);
2684 * For register / unregister we need to manually call out to every path.
2686 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2687 sector_t start, sector_t len, void *data)
2689 struct dm_pr *pr = data;
2690 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2692 if (!ops || !ops->pr_register)
2694 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2697 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2708 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2709 if (ret && new_key) {
2710 /* unregister all paths if we failed to register any path */
2711 pr.old_key = new_key;
2714 pr.fail_early = false;
2715 dm_call_pr(bdev, __dm_pr_register, &pr);
2721 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2724 struct mapped_device *md = bdev->bd_disk->private_data;
2725 const struct pr_ops *ops;
2729 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2733 ops = bdev->bd_disk->fops->pr_ops;
2734 if (ops && ops->pr_reserve)
2735 r = ops->pr_reserve(bdev, key, type, flags);
2743 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2745 struct mapped_device *md = bdev->bd_disk->private_data;
2746 const struct pr_ops *ops;
2750 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2754 ops = bdev->bd_disk->fops->pr_ops;
2755 if (ops && ops->pr_release)
2756 r = ops->pr_release(bdev, key, type);
2764 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2765 enum pr_type type, bool abort)
2767 struct mapped_device *md = bdev->bd_disk->private_data;
2768 const struct pr_ops *ops;
2772 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2776 ops = bdev->bd_disk->fops->pr_ops;
2777 if (ops && ops->pr_preempt)
2778 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2786 static int dm_pr_clear(struct block_device *bdev, u64 key)
2788 struct mapped_device *md = bdev->bd_disk->private_data;
2789 const struct pr_ops *ops;
2793 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2797 ops = bdev->bd_disk->fops->pr_ops;
2798 if (ops && ops->pr_clear)
2799 r = ops->pr_clear(bdev, key);
2807 static const struct pr_ops dm_pr_ops = {
2808 .pr_register = dm_pr_register,
2809 .pr_reserve = dm_pr_reserve,
2810 .pr_release = dm_pr_release,
2811 .pr_preempt = dm_pr_preempt,
2812 .pr_clear = dm_pr_clear,
2815 static const struct block_device_operations dm_blk_dops = {
2816 .open = dm_blk_open,
2817 .release = dm_blk_close,
2818 .ioctl = dm_blk_ioctl,
2819 .direct_access = dm_blk_direct_access,
2820 .getgeo = dm_blk_getgeo,
2821 .pr_ops = &dm_pr_ops,
2822 .owner = THIS_MODULE
2828 module_init(dm_init);
2829 module_exit(dm_exit);
2831 module_param(major, uint, 0);
2832 MODULE_PARM_DESC(major, "The major number of the device mapper");
2834 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2835 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2837 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2838 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2840 MODULE_DESCRIPTION(DM_NAME " driver");
2842 MODULE_LICENSE("GPL");