2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/delay.h>
24 #include <linux/wait.h>
27 #define DM_MSG_PREFIX "core"
31 * ratelimit state to be used in DMXXX_LIMIT().
33 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
34 DEFAULT_RATELIMIT_INTERVAL,
35 DEFAULT_RATELIMIT_BURST);
36 EXPORT_SYMBOL(dm_ratelimit_state);
40 * Cookies are numeric values sent with CHANGE and REMOVE
41 * uevents while resuming, removing or renaming the device.
43 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
44 #define DM_COOKIE_LENGTH 24
46 static const char *_name = DM_NAME;
48 static unsigned int major = 0;
49 static unsigned int _major = 0;
51 static DEFINE_IDR(_minor_idr);
53 static DEFINE_SPINLOCK(_minor_lock);
55 static void do_deferred_remove(struct work_struct *w);
57 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
59 static struct workqueue_struct *deferred_remove_workqueue;
62 * One of these is allocated per bio.
65 struct mapped_device *md;
69 unsigned long start_time;
70 spinlock_t endio_lock;
71 struct dm_stats_aux stats_aux;
74 #define MINOR_ALLOCED ((void *)-1)
77 * Bits for the md->flags field.
79 #define DMF_BLOCK_IO_FOR_SUSPEND 0
80 #define DMF_SUSPENDED 1
83 #define DMF_DELETING 4
84 #define DMF_NOFLUSH_SUSPENDING 5
85 #define DMF_DEFERRED_REMOVE 6
86 #define DMF_SUSPENDED_INTERNALLY 7
88 #define DM_NUMA_NODE NUMA_NO_NODE
89 static int dm_numa_node = DM_NUMA_NODE;
92 * For mempools pre-allocation at the table loading time.
94 struct dm_md_mempools {
100 struct list_head list;
102 struct dm_dev dm_dev;
105 static struct kmem_cache *_io_cache;
106 static struct kmem_cache *_rq_tio_cache;
107 static struct kmem_cache *_rq_cache;
110 * Bio-based DM's mempools' reserved IOs set by the user.
112 #define RESERVED_BIO_BASED_IOS 16
113 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
115 static int __dm_get_module_param_int(int *module_param, int min, int max)
117 int param = ACCESS_ONCE(*module_param);
118 int modified_param = 0;
119 bool modified = true;
122 modified_param = min;
123 else if (param > max)
124 modified_param = max;
129 (void)cmpxchg(module_param, param, modified_param);
130 param = modified_param;
136 unsigned __dm_get_module_param(unsigned *module_param,
137 unsigned def, unsigned max)
139 unsigned param = ACCESS_ONCE(*module_param);
140 unsigned modified_param = 0;
143 modified_param = def;
144 else if (param > max)
145 modified_param = max;
147 if (modified_param) {
148 (void)cmpxchg(module_param, param, modified_param);
149 param = modified_param;
155 unsigned dm_get_reserved_bio_based_ios(void)
157 return __dm_get_module_param(&reserved_bio_based_ios,
158 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
160 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
162 static unsigned dm_get_numa_node(void)
164 return __dm_get_module_param_int(&dm_numa_node,
165 DM_NUMA_NODE, num_online_nodes() - 1);
168 static int __init local_init(void)
172 /* allocate a slab for the dm_ios */
173 _io_cache = KMEM_CACHE(dm_io, 0);
177 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
179 goto out_free_io_cache;
181 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
182 __alignof__(struct request), 0, NULL);
184 goto out_free_rq_tio_cache;
186 r = dm_uevent_init();
188 goto out_free_rq_cache;
190 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
191 if (!deferred_remove_workqueue) {
193 goto out_uevent_exit;
197 r = register_blkdev(_major, _name);
199 goto out_free_workqueue;
207 destroy_workqueue(deferred_remove_workqueue);
211 kmem_cache_destroy(_rq_cache);
212 out_free_rq_tio_cache:
213 kmem_cache_destroy(_rq_tio_cache);
215 kmem_cache_destroy(_io_cache);
220 static void local_exit(void)
222 flush_scheduled_work();
223 destroy_workqueue(deferred_remove_workqueue);
225 kmem_cache_destroy(_rq_cache);
226 kmem_cache_destroy(_rq_tio_cache);
227 kmem_cache_destroy(_io_cache);
228 unregister_blkdev(_major, _name);
233 DMINFO("cleaned up");
236 static int (*_inits[])(void) __initdata = {
247 static void (*_exits[])(void) = {
258 static int __init dm_init(void)
260 const int count = ARRAY_SIZE(_inits);
264 for (i = 0; i < count; i++) {
279 static void __exit dm_exit(void)
281 int i = ARRAY_SIZE(_exits);
287 * Should be empty by this point.
289 idr_destroy(&_minor_idr);
293 * Block device functions
295 int dm_deleting_md(struct mapped_device *md)
297 return test_bit(DMF_DELETING, &md->flags);
300 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
302 struct mapped_device *md;
304 spin_lock(&_minor_lock);
306 md = bdev->bd_disk->private_data;
310 if (test_bit(DMF_FREEING, &md->flags) ||
311 dm_deleting_md(md)) {
317 atomic_inc(&md->open_count);
319 spin_unlock(&_minor_lock);
321 return md ? 0 : -ENXIO;
324 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
326 struct mapped_device *md;
328 spin_lock(&_minor_lock);
330 md = disk->private_data;
334 if (atomic_dec_and_test(&md->open_count) &&
335 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
336 queue_work(deferred_remove_workqueue, &deferred_remove_work);
340 spin_unlock(&_minor_lock);
343 int dm_open_count(struct mapped_device *md)
345 return atomic_read(&md->open_count);
349 * Guarantees nothing is using the device before it's deleted.
351 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
355 spin_lock(&_minor_lock);
357 if (dm_open_count(md)) {
360 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
361 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
364 set_bit(DMF_DELETING, &md->flags);
366 spin_unlock(&_minor_lock);
371 int dm_cancel_deferred_remove(struct mapped_device *md)
375 spin_lock(&_minor_lock);
377 if (test_bit(DMF_DELETING, &md->flags))
380 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
382 spin_unlock(&_minor_lock);
387 static void do_deferred_remove(struct work_struct *w)
389 dm_deferred_remove();
392 sector_t dm_get_size(struct mapped_device *md)
394 return get_capacity(md->disk);
397 struct request_queue *dm_get_md_queue(struct mapped_device *md)
402 struct dm_stats *dm_get_stats(struct mapped_device *md)
407 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
409 struct mapped_device *md = bdev->bd_disk->private_data;
411 return dm_get_geometry(md, geo);
414 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
415 struct block_device **bdev,
418 struct dm_target *tgt;
419 struct dm_table *map;
424 map = dm_get_live_table(md, &srcu_idx);
425 if (!map || !dm_table_get_size(map))
428 /* We only support devices that have a single target */
429 if (dm_table_get_num_targets(map) != 1)
432 tgt = dm_table_get_target(map, 0);
433 if (!tgt->type->prepare_ioctl)
436 if (dm_suspended_md(md)) {
441 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
446 dm_put_live_table(md, srcu_idx);
450 dm_put_live_table(md, srcu_idx);
451 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
458 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
459 unsigned int cmd, unsigned long arg)
461 struct mapped_device *md = bdev->bd_disk->private_data;
464 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
470 * Target determined this ioctl is being issued against a
471 * subset of the parent bdev; require extra privileges.
473 if (!capable(CAP_SYS_RAWIO)) {
475 "%s: sending ioctl %x to DM device without required privilege.",
482 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
488 static struct dm_io *alloc_io(struct mapped_device *md)
490 return mempool_alloc(md->io_pool, GFP_NOIO);
493 static void free_io(struct mapped_device *md, struct dm_io *io)
495 mempool_free(io, md->io_pool);
498 static void free_tio(struct dm_target_io *tio)
500 bio_put(&tio->clone);
503 int md_in_flight(struct mapped_device *md)
505 return atomic_read(&md->pending[READ]) +
506 atomic_read(&md->pending[WRITE]);
509 static void start_io_acct(struct dm_io *io)
511 struct mapped_device *md = io->md;
512 struct bio *bio = io->bio;
514 int rw = bio_data_dir(bio);
516 io->start_time = jiffies;
518 cpu = part_stat_lock();
519 part_round_stats(cpu, &dm_disk(md)->part0);
521 atomic_set(&dm_disk(md)->part0.in_flight[rw],
522 atomic_inc_return(&md->pending[rw]));
524 if (unlikely(dm_stats_used(&md->stats)))
525 dm_stats_account_io(&md->stats, bio_data_dir(bio),
526 bio->bi_iter.bi_sector, bio_sectors(bio),
527 false, 0, &io->stats_aux);
530 static void end_io_acct(struct dm_io *io)
532 struct mapped_device *md = io->md;
533 struct bio *bio = io->bio;
534 unsigned long duration = jiffies - io->start_time;
536 int rw = bio_data_dir(bio);
538 generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time);
540 if (unlikely(dm_stats_used(&md->stats)))
541 dm_stats_account_io(&md->stats, bio_data_dir(bio),
542 bio->bi_iter.bi_sector, bio_sectors(bio),
543 true, duration, &io->stats_aux);
546 * After this is decremented the bio must not be touched if it is
549 pending = atomic_dec_return(&md->pending[rw]);
550 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
551 pending += atomic_read(&md->pending[rw^0x1]);
553 /* nudge anyone waiting on suspend queue */
559 * Add the bio to the list of deferred io.
561 static void queue_io(struct mapped_device *md, struct bio *bio)
565 spin_lock_irqsave(&md->deferred_lock, flags);
566 bio_list_add(&md->deferred, bio);
567 spin_unlock_irqrestore(&md->deferred_lock, flags);
568 queue_work(md->wq, &md->work);
572 * Everyone (including functions in this file), should use this
573 * function to access the md->map field, and make sure they call
574 * dm_put_live_table() when finished.
576 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
578 *srcu_idx = srcu_read_lock(&md->io_barrier);
580 return srcu_dereference(md->map, &md->io_barrier);
583 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
585 srcu_read_unlock(&md->io_barrier, srcu_idx);
588 void dm_sync_table(struct mapped_device *md)
590 synchronize_srcu(&md->io_barrier);
591 synchronize_rcu_expedited();
595 * A fast alternative to dm_get_live_table/dm_put_live_table.
596 * The caller must not block between these two functions.
598 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
601 return rcu_dereference(md->map);
604 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
610 * Open a table device so we can use it as a map destination.
612 static int open_table_device(struct table_device *td, dev_t dev,
613 struct mapped_device *md)
615 static char *_claim_ptr = "I belong to device-mapper";
616 struct block_device *bdev;
620 BUG_ON(td->dm_dev.bdev);
622 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
624 return PTR_ERR(bdev);
626 r = bd_link_disk_holder(bdev, dm_disk(md));
628 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
632 td->dm_dev.bdev = bdev;
633 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
638 * Close a table device that we've been using.
640 static void close_table_device(struct table_device *td, struct mapped_device *md)
642 if (!td->dm_dev.bdev)
645 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
646 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
647 put_dax(td->dm_dev.dax_dev);
648 td->dm_dev.bdev = NULL;
649 td->dm_dev.dax_dev = NULL;
652 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
654 struct table_device *td;
656 list_for_each_entry(td, l, list)
657 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
663 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
664 struct dm_dev **result) {
666 struct table_device *td;
668 mutex_lock(&md->table_devices_lock);
669 td = find_table_device(&md->table_devices, dev, mode);
671 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
673 mutex_unlock(&md->table_devices_lock);
677 td->dm_dev.mode = mode;
678 td->dm_dev.bdev = NULL;
680 if ((r = open_table_device(td, dev, md))) {
681 mutex_unlock(&md->table_devices_lock);
686 format_dev_t(td->dm_dev.name, dev);
688 atomic_set(&td->count, 0);
689 list_add(&td->list, &md->table_devices);
691 atomic_inc(&td->count);
692 mutex_unlock(&md->table_devices_lock);
694 *result = &td->dm_dev;
697 EXPORT_SYMBOL_GPL(dm_get_table_device);
699 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
701 struct table_device *td = container_of(d, struct table_device, dm_dev);
703 mutex_lock(&md->table_devices_lock);
704 if (atomic_dec_and_test(&td->count)) {
705 close_table_device(td, md);
709 mutex_unlock(&md->table_devices_lock);
711 EXPORT_SYMBOL(dm_put_table_device);
713 static void free_table_devices(struct list_head *devices)
715 struct list_head *tmp, *next;
717 list_for_each_safe(tmp, next, devices) {
718 struct table_device *td = list_entry(tmp, struct table_device, list);
720 DMWARN("dm_destroy: %s still exists with %d references",
721 td->dm_dev.name, atomic_read(&td->count));
727 * Get the geometry associated with a dm device
729 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
737 * Set the geometry of a device.
739 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
741 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
743 if (geo->start > sz) {
744 DMWARN("Start sector is beyond the geometry limits.");
753 /*-----------------------------------------------------------------
755 * A more elegant soln is in the works that uses the queue
756 * merge fn, unfortunately there are a couple of changes to
757 * the block layer that I want to make for this. So in the
758 * interests of getting something for people to use I give
759 * you this clearly demarcated crap.
760 *---------------------------------------------------------------*/
762 static int __noflush_suspending(struct mapped_device *md)
764 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
768 * Decrements the number of outstanding ios that a bio has been
769 * cloned into, completing the original io if necc.
771 static void dec_pending(struct dm_io *io, int error)
776 struct mapped_device *md = io->md;
778 /* Push-back supersedes any I/O errors */
779 if (unlikely(error)) {
780 spin_lock_irqsave(&io->endio_lock, flags);
781 if (!(io->error > 0 && __noflush_suspending(md)))
783 spin_unlock_irqrestore(&io->endio_lock, flags);
786 if (atomic_dec_and_test(&io->io_count)) {
787 if (io->error == DM_ENDIO_REQUEUE) {
789 * Target requested pushing back the I/O.
791 spin_lock_irqsave(&md->deferred_lock, flags);
792 if (__noflush_suspending(md))
793 bio_list_add_head(&md->deferred, io->bio);
795 /* noflush suspend was interrupted. */
797 spin_unlock_irqrestore(&md->deferred_lock, flags);
800 io_error = io->error;
805 if (io_error == DM_ENDIO_REQUEUE)
808 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
810 * Preflush done for flush with data, reissue
811 * without REQ_PREFLUSH.
813 bio->bi_opf &= ~REQ_PREFLUSH;
816 /* done with normal IO or empty flush */
817 bio->bi_error = io_error;
823 void disable_write_same(struct mapped_device *md)
825 struct queue_limits *limits = dm_get_queue_limits(md);
827 /* device doesn't really support WRITE SAME, disable it */
828 limits->max_write_same_sectors = 0;
831 void disable_write_zeroes(struct mapped_device *md)
833 struct queue_limits *limits = dm_get_queue_limits(md);
835 /* device doesn't really support WRITE ZEROES, disable it */
836 limits->max_write_zeroes_sectors = 0;
839 static void clone_endio(struct bio *bio)
841 int error = bio->bi_error;
843 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
844 struct dm_io *io = tio->io;
845 struct mapped_device *md = tio->io->md;
846 dm_endio_fn endio = tio->ti->type->end_io;
849 r = endio(tio->ti, bio, error);
850 if (r < 0 || r == DM_ENDIO_REQUEUE)
852 * error and requeue request are handled
856 else if (r == DM_ENDIO_INCOMPLETE)
857 /* The target will handle the io */
860 DMWARN("unimplemented target endio return value: %d", r);
865 if (unlikely(r == -EREMOTEIO)) {
866 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
867 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)
868 disable_write_same(md);
869 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
870 !bdev_get_queue(bio->bi_bdev)->limits.max_write_zeroes_sectors)
871 disable_write_zeroes(md);
875 dec_pending(io, error);
879 * Return maximum size of I/O possible at the supplied sector up to the current
882 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
884 sector_t target_offset = dm_target_offset(ti, sector);
886 return ti->len - target_offset;
889 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
891 sector_t len = max_io_len_target_boundary(sector, ti);
892 sector_t offset, max_len;
895 * Does the target need to split even further?
897 if (ti->max_io_len) {
898 offset = dm_target_offset(ti, sector);
899 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
900 max_len = sector_div(offset, ti->max_io_len);
902 max_len = offset & (ti->max_io_len - 1);
903 max_len = ti->max_io_len - max_len;
912 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
914 if (len > UINT_MAX) {
915 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
916 (unsigned long long)len, UINT_MAX);
917 ti->error = "Maximum size of target IO is too large";
921 ti->max_io_len = (uint32_t) len;
925 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
927 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
928 sector_t sector, int *srcu_idx)
930 struct dm_table *map;
931 struct dm_target *ti;
933 map = dm_get_live_table(md, srcu_idx);
937 ti = dm_table_find_target(map, sector);
938 if (!dm_target_is_valid(ti))
944 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
945 long nr_pages, void **kaddr, pfn_t *pfn)
947 struct mapped_device *md = dax_get_private(dax_dev);
948 sector_t sector = pgoff * PAGE_SECTORS;
949 struct dm_target *ti;
950 long len, ret = -EIO;
953 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
957 if (!ti->type->direct_access)
959 len = max_io_len(sector, ti) / PAGE_SECTORS;
962 nr_pages = min(len, nr_pages);
963 if (ti->type->direct_access)
964 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
967 dm_put_live_table(md, srcu_idx);
973 * A target may call dm_accept_partial_bio only from the map routine. It is
974 * allowed for all bio types except REQ_PREFLUSH.
976 * dm_accept_partial_bio informs the dm that the target only wants to process
977 * additional n_sectors sectors of the bio and the rest of the data should be
978 * sent in a next bio.
980 * A diagram that explains the arithmetics:
981 * +--------------------+---------------+-------+
983 * +--------------------+---------------+-------+
985 * <-------------- *tio->len_ptr --------------->
986 * <------- bi_size ------->
989 * Region 1 was already iterated over with bio_advance or similar function.
990 * (it may be empty if the target doesn't use bio_advance)
991 * Region 2 is the remaining bio size that the target wants to process.
992 * (it may be empty if region 1 is non-empty, although there is no reason
994 * The target requires that region 3 is to be sent in the next bio.
996 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
997 * the partially processed part (the sum of regions 1+2) must be the same for all
1000 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1002 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1003 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1004 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1005 BUG_ON(bi_size > *tio->len_ptr);
1006 BUG_ON(n_sectors > bi_size);
1007 *tio->len_ptr -= bi_size - n_sectors;
1008 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1010 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1013 * Flush current->bio_list when the target map method blocks.
1014 * This fixes deadlocks in snapshot and possibly in other targets.
1017 struct blk_plug plug;
1018 struct blk_plug_cb cb;
1021 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1023 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1024 struct bio_list list;
1028 INIT_LIST_HEAD(&o->cb.list);
1030 if (unlikely(!current->bio_list))
1033 for (i = 0; i < 2; i++) {
1034 list = current->bio_list[i];
1035 bio_list_init(¤t->bio_list[i]);
1037 while ((bio = bio_list_pop(&list))) {
1038 struct bio_set *bs = bio->bi_pool;
1039 if (unlikely(!bs) || bs == fs_bio_set) {
1040 bio_list_add(¤t->bio_list[i], bio);
1044 spin_lock(&bs->rescue_lock);
1045 bio_list_add(&bs->rescue_list, bio);
1046 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1047 spin_unlock(&bs->rescue_lock);
1052 static void dm_offload_start(struct dm_offload *o)
1054 blk_start_plug(&o->plug);
1055 o->cb.callback = flush_current_bio_list;
1056 list_add(&o->cb.list, ¤t->plug->cb_list);
1059 static void dm_offload_end(struct dm_offload *o)
1061 list_del(&o->cb.list);
1062 blk_finish_plug(&o->plug);
1065 static void __map_bio(struct dm_target_io *tio)
1069 struct dm_offload o;
1070 struct bio *clone = &tio->clone;
1071 struct dm_target *ti = tio->ti;
1073 clone->bi_end_io = clone_endio;
1076 * Map the clone. If r == 0 we don't need to do
1077 * anything, the target has assumed ownership of
1080 atomic_inc(&tio->io->io_count);
1081 sector = clone->bi_iter.bi_sector;
1083 dm_offload_start(&o);
1084 r = ti->type->map(ti, clone);
1087 if (r == DM_MAPIO_REMAPPED) {
1088 /* the bio has been remapped so dispatch it */
1090 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1091 tio->io->bio->bi_bdev->bd_dev, sector);
1093 generic_make_request(clone);
1094 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1095 /* error the io and bail out, or requeue it if needed */
1096 dec_pending(tio->io, r);
1098 } else if (r != DM_MAPIO_SUBMITTED) {
1099 DMWARN("unimplemented target map return value: %d", r);
1105 struct mapped_device *md;
1106 struct dm_table *map;
1110 unsigned sector_count;
1113 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1115 bio->bi_iter.bi_sector = sector;
1116 bio->bi_iter.bi_size = to_bytes(len);
1120 * Creates a bio that consists of range of complete bvecs.
1122 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1123 sector_t sector, unsigned len)
1125 struct bio *clone = &tio->clone;
1127 __bio_clone_fast(clone, bio);
1129 if (unlikely(bio_integrity(bio) != NULL)) {
1132 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1133 !dm_target_passes_integrity(tio->ti->type))) {
1134 DMWARN("%s: the target %s doesn't support integrity data.",
1135 dm_device_name(tio->io->md),
1136 tio->ti->type->name);
1140 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1145 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1146 clone->bi_iter.bi_size = to_bytes(len);
1148 if (unlikely(bio_integrity(bio) != NULL))
1149 bio_integrity_trim(clone, 0, len);
1154 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1155 struct dm_target *ti,
1156 unsigned target_bio_nr)
1158 struct dm_target_io *tio;
1161 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1162 tio = container_of(clone, struct dm_target_io, clone);
1166 tio->target_bio_nr = target_bio_nr;
1171 static void __clone_and_map_simple_bio(struct clone_info *ci,
1172 struct dm_target *ti,
1173 unsigned target_bio_nr, unsigned *len)
1175 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1176 struct bio *clone = &tio->clone;
1180 __bio_clone_fast(clone, ci->bio);
1182 bio_setup_sector(clone, ci->sector, *len);
1187 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1188 unsigned num_bios, unsigned *len)
1190 unsigned target_bio_nr;
1192 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1193 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1196 static int __send_empty_flush(struct clone_info *ci)
1198 unsigned target_nr = 0;
1199 struct dm_target *ti;
1201 BUG_ON(bio_has_data(ci->bio));
1202 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1203 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1208 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1209 sector_t sector, unsigned *len)
1211 struct bio *bio = ci->bio;
1212 struct dm_target_io *tio;
1213 unsigned target_bio_nr;
1214 unsigned num_target_bios = 1;
1218 * Does the target want to receive duplicate copies of the bio?
1220 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1221 num_target_bios = ti->num_write_bios(ti, bio);
1223 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1224 tio = alloc_tio(ci, ti, target_bio_nr);
1226 r = clone_bio(tio, bio, sector, *len);
1237 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1239 static unsigned get_num_discard_bios(struct dm_target *ti)
1241 return ti->num_discard_bios;
1244 static unsigned get_num_write_same_bios(struct dm_target *ti)
1246 return ti->num_write_same_bios;
1249 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1251 return ti->num_write_zeroes_bios;
1254 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1256 static bool is_split_required_for_discard(struct dm_target *ti)
1258 return ti->split_discard_bios;
1261 static int __send_changing_extent_only(struct clone_info *ci,
1262 get_num_bios_fn get_num_bios,
1263 is_split_required_fn is_split_required)
1265 struct dm_target *ti;
1270 ti = dm_table_find_target(ci->map, ci->sector);
1271 if (!dm_target_is_valid(ti))
1275 * Even though the device advertised support for this type of
1276 * request, that does not mean every target supports it, and
1277 * reconfiguration might also have changed that since the
1278 * check was performed.
1280 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1284 if (is_split_required && !is_split_required(ti))
1285 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1287 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1289 __send_duplicate_bios(ci, ti, num_bios, &len);
1292 } while (ci->sector_count -= len);
1297 static int __send_discard(struct clone_info *ci)
1299 return __send_changing_extent_only(ci, get_num_discard_bios,
1300 is_split_required_for_discard);
1303 static int __send_write_same(struct clone_info *ci)
1305 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1308 static int __send_write_zeroes(struct clone_info *ci)
1310 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1314 * Select the correct strategy for processing a non-flush bio.
1316 static int __split_and_process_non_flush(struct clone_info *ci)
1318 struct bio *bio = ci->bio;
1319 struct dm_target *ti;
1323 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1324 return __send_discard(ci);
1325 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1326 return __send_write_same(ci);
1327 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1328 return __send_write_zeroes(ci);
1330 ti = dm_table_find_target(ci->map, ci->sector);
1331 if (!dm_target_is_valid(ti))
1334 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1336 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1341 ci->sector_count -= len;
1347 * Entry point to split a bio into clones and submit them to the targets.
1349 static void __split_and_process_bio(struct mapped_device *md,
1350 struct dm_table *map, struct bio *bio)
1352 struct clone_info ci;
1355 if (unlikely(!map)) {
1362 ci.io = alloc_io(md);
1364 atomic_set(&ci.io->io_count, 1);
1367 spin_lock_init(&ci.io->endio_lock);
1368 ci.sector = bio->bi_iter.bi_sector;
1370 start_io_acct(ci.io);
1372 if (bio->bi_opf & REQ_PREFLUSH) {
1373 ci.bio = &ci.md->flush_bio;
1374 ci.sector_count = 0;
1375 error = __send_empty_flush(&ci);
1376 /* dec_pending submits any data associated with flush */
1379 ci.sector_count = bio_sectors(bio);
1380 while (ci.sector_count && !error)
1381 error = __split_and_process_non_flush(&ci);
1384 /* drop the extra reference count */
1385 dec_pending(ci.io, error);
1387 /*-----------------------------------------------------------------
1389 *---------------------------------------------------------------*/
1392 * The request function that just remaps the bio built up by
1395 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1397 int rw = bio_data_dir(bio);
1398 struct mapped_device *md = q->queuedata;
1400 struct dm_table *map;
1402 map = dm_get_live_table(md, &srcu_idx);
1404 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1406 /* if we're suspended, we have to queue this io for later */
1407 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1408 dm_put_live_table(md, srcu_idx);
1410 if (!(bio->bi_opf & REQ_RAHEAD))
1414 return BLK_QC_T_NONE;
1417 __split_and_process_bio(md, map, bio);
1418 dm_put_live_table(md, srcu_idx);
1419 return BLK_QC_T_NONE;
1422 static int dm_any_congested(void *congested_data, int bdi_bits)
1425 struct mapped_device *md = congested_data;
1426 struct dm_table *map;
1428 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1429 if (dm_request_based(md)) {
1431 * With request-based DM we only need to check the
1432 * top-level queue for congestion.
1434 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1436 map = dm_get_live_table_fast(md);
1438 r = dm_table_any_congested(map, bdi_bits);
1439 dm_put_live_table_fast(md);
1446 /*-----------------------------------------------------------------
1447 * An IDR is used to keep track of allocated minor numbers.
1448 *---------------------------------------------------------------*/
1449 static void free_minor(int minor)
1451 spin_lock(&_minor_lock);
1452 idr_remove(&_minor_idr, minor);
1453 spin_unlock(&_minor_lock);
1457 * See if the device with a specific minor # is free.
1459 static int specific_minor(int minor)
1463 if (minor >= (1 << MINORBITS))
1466 idr_preload(GFP_KERNEL);
1467 spin_lock(&_minor_lock);
1469 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1471 spin_unlock(&_minor_lock);
1474 return r == -ENOSPC ? -EBUSY : r;
1478 static int next_free_minor(int *minor)
1482 idr_preload(GFP_KERNEL);
1483 spin_lock(&_minor_lock);
1485 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1487 spin_unlock(&_minor_lock);
1495 static const struct block_device_operations dm_blk_dops;
1496 static const struct dax_operations dm_dax_ops;
1498 static void dm_wq_work(struct work_struct *work);
1500 void dm_init_md_queue(struct mapped_device *md)
1503 * Request-based dm devices cannot be stacked on top of bio-based dm
1504 * devices. The type of this dm device may not have been decided yet.
1505 * The type is decided at the first table loading time.
1506 * To prevent problematic device stacking, clear the queue flag
1507 * for request stacking support until then.
1509 * This queue is new, so no concurrency on the queue_flags.
1511 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1514 * Initialize data that will only be used by a non-blk-mq DM queue
1515 * - must do so here (in alloc_dev callchain) before queue is used
1517 md->queue->queuedata = md;
1518 md->queue->backing_dev_info->congested_data = md;
1521 void dm_init_normal_md_queue(struct mapped_device *md)
1523 md->use_blk_mq = false;
1524 dm_init_md_queue(md);
1527 * Initialize aspects of queue that aren't relevant for blk-mq
1529 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1530 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1533 static void cleanup_mapped_device(struct mapped_device *md)
1536 destroy_workqueue(md->wq);
1537 if (md->kworker_task)
1538 kthread_stop(md->kworker_task);
1539 mempool_destroy(md->io_pool);
1541 bioset_free(md->bs);
1544 kill_dax(md->dax_dev);
1545 put_dax(md->dax_dev);
1550 spin_lock(&_minor_lock);
1551 md->disk->private_data = NULL;
1552 spin_unlock(&_minor_lock);
1553 del_gendisk(md->disk);
1558 blk_cleanup_queue(md->queue);
1560 cleanup_srcu_struct(&md->io_barrier);
1567 dm_mq_cleanup_mapped_device(md);
1571 * Allocate and initialise a blank device with a given minor.
1573 static struct mapped_device *alloc_dev(int minor)
1575 int r, numa_node_id = dm_get_numa_node();
1576 struct dax_device *dax_dev;
1577 struct mapped_device *md;
1580 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1582 DMWARN("unable to allocate device, out of memory.");
1586 if (!try_module_get(THIS_MODULE))
1587 goto bad_module_get;
1589 /* get a minor number for the dev */
1590 if (minor == DM_ANY_MINOR)
1591 r = next_free_minor(&minor);
1593 r = specific_minor(minor);
1597 r = init_srcu_struct(&md->io_barrier);
1599 goto bad_io_barrier;
1601 md->numa_node_id = numa_node_id;
1602 md->use_blk_mq = dm_use_blk_mq_default();
1603 md->init_tio_pdu = false;
1604 md->type = DM_TYPE_NONE;
1605 mutex_init(&md->suspend_lock);
1606 mutex_init(&md->type_lock);
1607 mutex_init(&md->table_devices_lock);
1608 spin_lock_init(&md->deferred_lock);
1609 atomic_set(&md->holders, 1);
1610 atomic_set(&md->open_count, 0);
1611 atomic_set(&md->event_nr, 0);
1612 atomic_set(&md->uevent_seq, 0);
1613 INIT_LIST_HEAD(&md->uevent_list);
1614 INIT_LIST_HEAD(&md->table_devices);
1615 spin_lock_init(&md->uevent_lock);
1617 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1621 dm_init_md_queue(md);
1623 md->disk = alloc_disk_node(1, numa_node_id);
1627 atomic_set(&md->pending[0], 0);
1628 atomic_set(&md->pending[1], 0);
1629 init_waitqueue_head(&md->wait);
1630 INIT_WORK(&md->work, dm_wq_work);
1631 init_waitqueue_head(&md->eventq);
1632 init_completion(&md->kobj_holder.completion);
1633 md->kworker_task = NULL;
1635 md->disk->major = _major;
1636 md->disk->first_minor = minor;
1637 md->disk->fops = &dm_blk_dops;
1638 md->disk->queue = md->queue;
1639 md->disk->private_data = md;
1640 sprintf(md->disk->disk_name, "dm-%d", minor);
1642 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1645 md->dax_dev = dax_dev;
1648 format_dev_t(md->name, MKDEV(_major, minor));
1650 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1654 md->bdev = bdget_disk(md->disk, 0);
1658 bio_init(&md->flush_bio, NULL, 0);
1659 md->flush_bio.bi_bdev = md->bdev;
1660 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1662 dm_stats_init(&md->stats);
1664 /* Populate the mapping, nobody knows we exist yet */
1665 spin_lock(&_minor_lock);
1666 old_md = idr_replace(&_minor_idr, md, minor);
1667 spin_unlock(&_minor_lock);
1669 BUG_ON(old_md != MINOR_ALLOCED);
1674 cleanup_mapped_device(md);
1678 module_put(THIS_MODULE);
1684 static void unlock_fs(struct mapped_device *md);
1686 static void free_dev(struct mapped_device *md)
1688 int minor = MINOR(disk_devt(md->disk));
1692 cleanup_mapped_device(md);
1694 free_table_devices(&md->table_devices);
1695 dm_stats_cleanup(&md->stats);
1698 module_put(THIS_MODULE);
1702 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1704 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1707 /* The md already has necessary mempools. */
1708 if (dm_table_bio_based(t)) {
1710 * Reload bioset because front_pad may have changed
1711 * because a different table was loaded.
1713 bioset_free(md->bs);
1718 * There's no need to reload with request-based dm
1719 * because the size of front_pad doesn't change.
1720 * Note for future: If you are to reload bioset,
1721 * prep-ed requests in the queue may refer
1722 * to bio from the old bioset, so you must walk
1723 * through the queue to unprep.
1728 BUG_ON(!p || md->io_pool || md->bs);
1730 md->io_pool = p->io_pool;
1736 /* mempool bind completed, no longer need any mempools in the table */
1737 dm_table_free_md_mempools(t);
1741 * Bind a table to the device.
1743 static void event_callback(void *context)
1745 unsigned long flags;
1747 struct mapped_device *md = (struct mapped_device *) context;
1749 spin_lock_irqsave(&md->uevent_lock, flags);
1750 list_splice_init(&md->uevent_list, &uevents);
1751 spin_unlock_irqrestore(&md->uevent_lock, flags);
1753 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1755 atomic_inc(&md->event_nr);
1756 wake_up(&md->eventq);
1760 * Protected by md->suspend_lock obtained by dm_swap_table().
1762 static void __set_size(struct mapped_device *md, sector_t size)
1764 lockdep_assert_held(&md->suspend_lock);
1766 set_capacity(md->disk, size);
1768 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1772 * Returns old map, which caller must destroy.
1774 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1775 struct queue_limits *limits)
1777 struct dm_table *old_map;
1778 struct request_queue *q = md->queue;
1781 lockdep_assert_held(&md->suspend_lock);
1783 size = dm_table_get_size(t);
1786 * Wipe any geometry if the size of the table changed.
1788 if (size != dm_get_size(md))
1789 memset(&md->geometry, 0, sizeof(md->geometry));
1791 __set_size(md, size);
1793 dm_table_event_callback(t, event_callback, md);
1796 * The queue hasn't been stopped yet, if the old table type wasn't
1797 * for request-based during suspension. So stop it to prevent
1798 * I/O mapping before resume.
1799 * This must be done before setting the queue restrictions,
1800 * because request-based dm may be run just after the setting.
1802 if (dm_table_request_based(t)) {
1805 * Leverage the fact that request-based DM targets are
1806 * immutable singletons and establish md->immutable_target
1807 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1809 md->immutable_target = dm_table_get_immutable_target(t);
1812 __bind_mempools(md, t);
1814 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1815 rcu_assign_pointer(md->map, (void *)t);
1816 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1818 dm_table_set_restrictions(t, q, limits);
1826 * Returns unbound table for the caller to free.
1828 static struct dm_table *__unbind(struct mapped_device *md)
1830 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1835 dm_table_event_callback(map, NULL, NULL);
1836 RCU_INIT_POINTER(md->map, NULL);
1843 * Constructor for a new device.
1845 int dm_create(int minor, struct mapped_device **result)
1847 struct mapped_device *md;
1849 md = alloc_dev(minor);
1860 * Functions to manage md->type.
1861 * All are required to hold md->type_lock.
1863 void dm_lock_md_type(struct mapped_device *md)
1865 mutex_lock(&md->type_lock);
1868 void dm_unlock_md_type(struct mapped_device *md)
1870 mutex_unlock(&md->type_lock);
1873 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1875 BUG_ON(!mutex_is_locked(&md->type_lock));
1879 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1884 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1886 return md->immutable_target_type;
1890 * The queue_limits are only valid as long as you have a reference
1893 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1895 BUG_ON(!atomic_read(&md->holders));
1896 return &md->queue->limits;
1898 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1901 * Setup the DM device's queue based on md's type
1903 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1906 enum dm_queue_mode type = dm_get_md_type(md);
1909 case DM_TYPE_REQUEST_BASED:
1910 r = dm_old_init_request_queue(md, t);
1912 DMERR("Cannot initialize queue for request-based mapped device");
1916 case DM_TYPE_MQ_REQUEST_BASED:
1917 r = dm_mq_init_request_queue(md, t);
1919 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1923 case DM_TYPE_BIO_BASED:
1924 case DM_TYPE_DAX_BIO_BASED:
1925 dm_init_normal_md_queue(md);
1926 blk_queue_make_request(md->queue, dm_make_request);
1928 * DM handles splitting bios as needed. Free the bio_split bioset
1929 * since it won't be used (saves 1 process per bio-based DM device).
1931 bioset_free(md->queue->bio_split);
1932 md->queue->bio_split = NULL;
1934 if (type == DM_TYPE_DAX_BIO_BASED)
1935 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1945 struct mapped_device *dm_get_md(dev_t dev)
1947 struct mapped_device *md;
1948 unsigned minor = MINOR(dev);
1950 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1953 spin_lock(&_minor_lock);
1955 md = idr_find(&_minor_idr, minor);
1957 if ((md == MINOR_ALLOCED ||
1958 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1959 dm_deleting_md(md) ||
1960 test_bit(DMF_FREEING, &md->flags))) {
1968 spin_unlock(&_minor_lock);
1972 EXPORT_SYMBOL_GPL(dm_get_md);
1974 void *dm_get_mdptr(struct mapped_device *md)
1976 return md->interface_ptr;
1979 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1981 md->interface_ptr = ptr;
1984 void dm_get(struct mapped_device *md)
1986 atomic_inc(&md->holders);
1987 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1990 int dm_hold(struct mapped_device *md)
1992 spin_lock(&_minor_lock);
1993 if (test_bit(DMF_FREEING, &md->flags)) {
1994 spin_unlock(&_minor_lock);
1998 spin_unlock(&_minor_lock);
2001 EXPORT_SYMBOL_GPL(dm_hold);
2003 const char *dm_device_name(struct mapped_device *md)
2007 EXPORT_SYMBOL_GPL(dm_device_name);
2009 static void __dm_destroy(struct mapped_device *md, bool wait)
2011 struct request_queue *q = dm_get_md_queue(md);
2012 struct dm_table *map;
2017 spin_lock(&_minor_lock);
2018 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2019 set_bit(DMF_FREEING, &md->flags);
2020 spin_unlock(&_minor_lock);
2022 blk_set_queue_dying(q);
2024 if (dm_request_based(md) && md->kworker_task)
2025 kthread_flush_worker(&md->kworker);
2028 * Take suspend_lock so that presuspend and postsuspend methods
2029 * do not race with internal suspend.
2031 mutex_lock(&md->suspend_lock);
2032 map = dm_get_live_table(md, &srcu_idx);
2033 if (!dm_suspended_md(md)) {
2034 dm_table_presuspend_targets(map);
2035 dm_table_postsuspend_targets(map);
2037 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2038 dm_put_live_table(md, srcu_idx);
2039 mutex_unlock(&md->suspend_lock);
2042 * Rare, but there may be I/O requests still going to complete,
2043 * for example. Wait for all references to disappear.
2044 * No one should increment the reference count of the mapped_device,
2045 * after the mapped_device state becomes DMF_FREEING.
2048 while (atomic_read(&md->holders))
2050 else if (atomic_read(&md->holders))
2051 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2052 dm_device_name(md), atomic_read(&md->holders));
2055 dm_table_destroy(__unbind(md));
2059 void dm_destroy(struct mapped_device *md)
2061 __dm_destroy(md, true);
2064 void dm_destroy_immediate(struct mapped_device *md)
2066 __dm_destroy(md, false);
2069 void dm_put(struct mapped_device *md)
2071 atomic_dec(&md->holders);
2073 EXPORT_SYMBOL_GPL(dm_put);
2075 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2081 prepare_to_wait(&md->wait, &wait, task_state);
2083 if (!md_in_flight(md))
2086 if (signal_pending_state(task_state, current)) {
2093 finish_wait(&md->wait, &wait);
2099 * Process the deferred bios
2101 static void dm_wq_work(struct work_struct *work)
2103 struct mapped_device *md = container_of(work, struct mapped_device,
2107 struct dm_table *map;
2109 map = dm_get_live_table(md, &srcu_idx);
2111 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2112 spin_lock_irq(&md->deferred_lock);
2113 c = bio_list_pop(&md->deferred);
2114 spin_unlock_irq(&md->deferred_lock);
2119 if (dm_request_based(md))
2120 generic_make_request(c);
2122 __split_and_process_bio(md, map, c);
2125 dm_put_live_table(md, srcu_idx);
2128 static void dm_queue_flush(struct mapped_device *md)
2130 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2131 smp_mb__after_atomic();
2132 queue_work(md->wq, &md->work);
2136 * Swap in a new table, returning the old one for the caller to destroy.
2138 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2140 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2141 struct queue_limits limits;
2144 mutex_lock(&md->suspend_lock);
2146 /* device must be suspended */
2147 if (!dm_suspended_md(md))
2151 * If the new table has no data devices, retain the existing limits.
2152 * This helps multipath with queue_if_no_path if all paths disappear,
2153 * then new I/O is queued based on these limits, and then some paths
2156 if (dm_table_has_no_data_devices(table)) {
2157 live_map = dm_get_live_table_fast(md);
2159 limits = md->queue->limits;
2160 dm_put_live_table_fast(md);
2164 r = dm_calculate_queue_limits(table, &limits);
2171 map = __bind(md, table, &limits);
2174 mutex_unlock(&md->suspend_lock);
2179 * Functions to lock and unlock any filesystem running on the
2182 static int lock_fs(struct mapped_device *md)
2186 WARN_ON(md->frozen_sb);
2188 md->frozen_sb = freeze_bdev(md->bdev);
2189 if (IS_ERR(md->frozen_sb)) {
2190 r = PTR_ERR(md->frozen_sb);
2191 md->frozen_sb = NULL;
2195 set_bit(DMF_FROZEN, &md->flags);
2200 static void unlock_fs(struct mapped_device *md)
2202 if (!test_bit(DMF_FROZEN, &md->flags))
2205 thaw_bdev(md->bdev, md->frozen_sb);
2206 md->frozen_sb = NULL;
2207 clear_bit(DMF_FROZEN, &md->flags);
2211 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2212 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2213 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2215 * If __dm_suspend returns 0, the device is completely quiescent
2216 * now. There is no request-processing activity. All new requests
2217 * are being added to md->deferred list.
2219 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2220 unsigned suspend_flags, long task_state,
2221 int dmf_suspended_flag)
2223 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2224 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2227 lockdep_assert_held(&md->suspend_lock);
2230 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2231 * This flag is cleared before dm_suspend returns.
2234 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2236 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2239 * This gets reverted if there's an error later and the targets
2240 * provide the .presuspend_undo hook.
2242 dm_table_presuspend_targets(map);
2245 * Flush I/O to the device.
2246 * Any I/O submitted after lock_fs() may not be flushed.
2247 * noflush takes precedence over do_lockfs.
2248 * (lock_fs() flushes I/Os and waits for them to complete.)
2250 if (!noflush && do_lockfs) {
2253 dm_table_presuspend_undo_targets(map);
2259 * Here we must make sure that no processes are submitting requests
2260 * to target drivers i.e. no one may be executing
2261 * __split_and_process_bio. This is called from dm_request and
2264 * To get all processes out of __split_and_process_bio in dm_request,
2265 * we take the write lock. To prevent any process from reentering
2266 * __split_and_process_bio from dm_request and quiesce the thread
2267 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2268 * flush_workqueue(md->wq).
2270 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2272 synchronize_srcu(&md->io_barrier);
2275 * Stop md->queue before flushing md->wq in case request-based
2276 * dm defers requests to md->wq from md->queue.
2278 if (dm_request_based(md)) {
2279 dm_stop_queue(md->queue);
2280 if (md->kworker_task)
2281 kthread_flush_worker(&md->kworker);
2284 flush_workqueue(md->wq);
2287 * At this point no more requests are entering target request routines.
2288 * We call dm_wait_for_completion to wait for all existing requests
2291 r = dm_wait_for_completion(md, task_state);
2293 set_bit(dmf_suspended_flag, &md->flags);
2296 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2298 synchronize_srcu(&md->io_barrier);
2300 /* were we interrupted ? */
2304 if (dm_request_based(md))
2305 dm_start_queue(md->queue);
2308 dm_table_presuspend_undo_targets(map);
2309 /* pushback list is already flushed, so skip flush */
2316 * We need to be able to change a mapping table under a mounted
2317 * filesystem. For example we might want to move some data in
2318 * the background. Before the table can be swapped with
2319 * dm_bind_table, dm_suspend must be called to flush any in
2320 * flight bios and ensure that any further io gets deferred.
2323 * Suspend mechanism in request-based dm.
2325 * 1. Flush all I/Os by lock_fs() if needed.
2326 * 2. Stop dispatching any I/O by stopping the request_queue.
2327 * 3. Wait for all in-flight I/Os to be completed or requeued.
2329 * To abort suspend, start the request_queue.
2331 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2333 struct dm_table *map = NULL;
2337 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2339 if (dm_suspended_md(md)) {
2344 if (dm_suspended_internally_md(md)) {
2345 /* already internally suspended, wait for internal resume */
2346 mutex_unlock(&md->suspend_lock);
2347 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2353 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2355 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2359 dm_table_postsuspend_targets(map);
2362 mutex_unlock(&md->suspend_lock);
2366 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2369 int r = dm_table_resume_targets(map);
2377 * Flushing deferred I/Os must be done after targets are resumed
2378 * so that mapping of targets can work correctly.
2379 * Request-based dm is queueing the deferred I/Os in its request_queue.
2381 if (dm_request_based(md))
2382 dm_start_queue(md->queue);
2389 int dm_resume(struct mapped_device *md)
2392 struct dm_table *map = NULL;
2396 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2398 if (!dm_suspended_md(md))
2401 if (dm_suspended_internally_md(md)) {
2402 /* already internally suspended, wait for internal resume */
2403 mutex_unlock(&md->suspend_lock);
2404 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2410 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2411 if (!map || !dm_table_get_size(map))
2414 r = __dm_resume(md, map);
2418 clear_bit(DMF_SUSPENDED, &md->flags);
2420 mutex_unlock(&md->suspend_lock);
2426 * Internal suspend/resume works like userspace-driven suspend. It waits
2427 * until all bios finish and prevents issuing new bios to the target drivers.
2428 * It may be used only from the kernel.
2431 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2433 struct dm_table *map = NULL;
2435 lockdep_assert_held(&md->suspend_lock);
2437 if (md->internal_suspend_count++)
2438 return; /* nested internal suspend */
2440 if (dm_suspended_md(md)) {
2441 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2442 return; /* nest suspend */
2445 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2448 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2449 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2450 * would require changing .presuspend to return an error -- avoid this
2451 * until there is a need for more elaborate variants of internal suspend.
2453 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2454 DMF_SUSPENDED_INTERNALLY);
2456 dm_table_postsuspend_targets(map);
2459 static void __dm_internal_resume(struct mapped_device *md)
2461 BUG_ON(!md->internal_suspend_count);
2463 if (--md->internal_suspend_count)
2464 return; /* resume from nested internal suspend */
2466 if (dm_suspended_md(md))
2467 goto done; /* resume from nested suspend */
2470 * NOTE: existing callers don't need to call dm_table_resume_targets
2471 * (which may fail -- so best to avoid it for now by passing NULL map)
2473 (void) __dm_resume(md, NULL);
2476 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2477 smp_mb__after_atomic();
2478 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2481 void dm_internal_suspend_noflush(struct mapped_device *md)
2483 mutex_lock(&md->suspend_lock);
2484 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2485 mutex_unlock(&md->suspend_lock);
2487 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2489 void dm_internal_resume(struct mapped_device *md)
2491 mutex_lock(&md->suspend_lock);
2492 __dm_internal_resume(md);
2493 mutex_unlock(&md->suspend_lock);
2495 EXPORT_SYMBOL_GPL(dm_internal_resume);
2498 * Fast variants of internal suspend/resume hold md->suspend_lock,
2499 * which prevents interaction with userspace-driven suspend.
2502 void dm_internal_suspend_fast(struct mapped_device *md)
2504 mutex_lock(&md->suspend_lock);
2505 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2508 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2509 synchronize_srcu(&md->io_barrier);
2510 flush_workqueue(md->wq);
2511 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2513 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2515 void dm_internal_resume_fast(struct mapped_device *md)
2517 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2523 mutex_unlock(&md->suspend_lock);
2525 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2527 /*-----------------------------------------------------------------
2528 * Event notification.
2529 *---------------------------------------------------------------*/
2530 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2533 char udev_cookie[DM_COOKIE_LENGTH];
2534 char *envp[] = { udev_cookie, NULL };
2537 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2539 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2540 DM_COOKIE_ENV_VAR_NAME, cookie);
2541 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2546 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2548 return atomic_add_return(1, &md->uevent_seq);
2551 uint32_t dm_get_event_nr(struct mapped_device *md)
2553 return atomic_read(&md->event_nr);
2556 int dm_wait_event(struct mapped_device *md, int event_nr)
2558 return wait_event_interruptible(md->eventq,
2559 (event_nr != atomic_read(&md->event_nr)));
2562 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2564 unsigned long flags;
2566 spin_lock_irqsave(&md->uevent_lock, flags);
2567 list_add(elist, &md->uevent_list);
2568 spin_unlock_irqrestore(&md->uevent_lock, flags);
2572 * The gendisk is only valid as long as you have a reference
2575 struct gendisk *dm_disk(struct mapped_device *md)
2579 EXPORT_SYMBOL_GPL(dm_disk);
2581 struct kobject *dm_kobject(struct mapped_device *md)
2583 return &md->kobj_holder.kobj;
2586 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2588 struct mapped_device *md;
2590 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2592 if (test_bit(DMF_FREEING, &md->flags) ||
2600 int dm_suspended_md(struct mapped_device *md)
2602 return test_bit(DMF_SUSPENDED, &md->flags);
2605 int dm_suspended_internally_md(struct mapped_device *md)
2607 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2610 int dm_test_deferred_remove_flag(struct mapped_device *md)
2612 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2615 int dm_suspended(struct dm_target *ti)
2617 return dm_suspended_md(dm_table_get_md(ti->table));
2619 EXPORT_SYMBOL_GPL(dm_suspended);
2621 int dm_noflush_suspending(struct dm_target *ti)
2623 return __noflush_suspending(dm_table_get_md(ti->table));
2625 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2627 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2628 unsigned integrity, unsigned per_io_data_size)
2630 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2631 unsigned int pool_size = 0;
2632 unsigned int front_pad;
2638 case DM_TYPE_BIO_BASED:
2639 case DM_TYPE_DAX_BIO_BASED:
2640 pool_size = dm_get_reserved_bio_based_ios();
2641 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2643 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2644 if (!pools->io_pool)
2647 case DM_TYPE_REQUEST_BASED:
2648 case DM_TYPE_MQ_REQUEST_BASED:
2649 pool_size = dm_get_reserved_rq_based_ios();
2650 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2651 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2657 pools->bs = bioset_create_nobvec(pool_size, front_pad);
2661 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2667 dm_free_md_mempools(pools);
2672 void dm_free_md_mempools(struct dm_md_mempools *pools)
2677 mempool_destroy(pools->io_pool);
2680 bioset_free(pools->bs);
2692 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2695 struct mapped_device *md = bdev->bd_disk->private_data;
2696 struct dm_table *table;
2697 struct dm_target *ti;
2698 int ret = -ENOTTY, srcu_idx;
2700 table = dm_get_live_table(md, &srcu_idx);
2701 if (!table || !dm_table_get_size(table))
2704 /* We only support devices that have a single target */
2705 if (dm_table_get_num_targets(table) != 1)
2707 ti = dm_table_get_target(table, 0);
2710 if (!ti->type->iterate_devices)
2713 ret = ti->type->iterate_devices(ti, fn, data);
2715 dm_put_live_table(md, srcu_idx);
2720 * For register / unregister we need to manually call out to every path.
2722 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2723 sector_t start, sector_t len, void *data)
2725 struct dm_pr *pr = data;
2726 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2728 if (!ops || !ops->pr_register)
2730 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2733 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2744 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2745 if (ret && new_key) {
2746 /* unregister all paths if we failed to register any path */
2747 pr.old_key = new_key;
2750 pr.fail_early = false;
2751 dm_call_pr(bdev, __dm_pr_register, &pr);
2757 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2760 struct mapped_device *md = bdev->bd_disk->private_data;
2761 const struct pr_ops *ops;
2765 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2769 ops = bdev->bd_disk->fops->pr_ops;
2770 if (ops && ops->pr_reserve)
2771 r = ops->pr_reserve(bdev, key, type, flags);
2779 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2781 struct mapped_device *md = bdev->bd_disk->private_data;
2782 const struct pr_ops *ops;
2786 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2790 ops = bdev->bd_disk->fops->pr_ops;
2791 if (ops && ops->pr_release)
2792 r = ops->pr_release(bdev, key, type);
2800 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2801 enum pr_type type, bool abort)
2803 struct mapped_device *md = bdev->bd_disk->private_data;
2804 const struct pr_ops *ops;
2808 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2812 ops = bdev->bd_disk->fops->pr_ops;
2813 if (ops && ops->pr_preempt)
2814 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2822 static int dm_pr_clear(struct block_device *bdev, u64 key)
2824 struct mapped_device *md = bdev->bd_disk->private_data;
2825 const struct pr_ops *ops;
2829 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2833 ops = bdev->bd_disk->fops->pr_ops;
2834 if (ops && ops->pr_clear)
2835 r = ops->pr_clear(bdev, key);
2843 static const struct pr_ops dm_pr_ops = {
2844 .pr_register = dm_pr_register,
2845 .pr_reserve = dm_pr_reserve,
2846 .pr_release = dm_pr_release,
2847 .pr_preempt = dm_pr_preempt,
2848 .pr_clear = dm_pr_clear,
2851 static const struct block_device_operations dm_blk_dops = {
2852 .open = dm_blk_open,
2853 .release = dm_blk_close,
2854 .ioctl = dm_blk_ioctl,
2855 .getgeo = dm_blk_getgeo,
2856 .pr_ops = &dm_pr_ops,
2857 .owner = THIS_MODULE
2860 static const struct dax_operations dm_dax_ops = {
2861 .direct_access = dm_dax_direct_access,
2867 module_init(dm_init);
2868 module_exit(dm_exit);
2870 module_param(major, uint, 0);
2871 MODULE_PARM_DESC(major, "The major number of the device mapper");
2873 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2874 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2876 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2877 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2879 MODULE_DESCRIPTION(DM_NAME " driver");
2880 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2881 MODULE_LICENSE("GPL");