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, blk_status_t error)
774 blk_status_t io_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->status == BLK_STS_DM_REQUEUE &&
782 __noflush_suspending(md)))
784 spin_unlock_irqrestore(&io->endio_lock, flags);
787 if (atomic_dec_and_test(&io->io_count)) {
788 if (io->status == BLK_STS_DM_REQUEUE) {
790 * Target requested pushing back the I/O.
792 spin_lock_irqsave(&md->deferred_lock, flags);
793 if (__noflush_suspending(md))
794 bio_list_add_head(&md->deferred, io->bio);
796 /* noflush suspend was interrupted. */
797 io->status = BLK_STS_IOERR;
798 spin_unlock_irqrestore(&md->deferred_lock, flags);
801 io_error = io->status;
806 if (io_error == BLK_STS_DM_REQUEUE)
809 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
811 * Preflush done for flush with data, reissue
812 * without REQ_PREFLUSH.
814 bio->bi_opf &= ~REQ_PREFLUSH;
817 /* done with normal IO or empty flush */
818 bio->bi_status = io_error;
824 void disable_write_same(struct mapped_device *md)
826 struct queue_limits *limits = dm_get_queue_limits(md);
828 /* device doesn't really support WRITE SAME, disable it */
829 limits->max_write_same_sectors = 0;
832 void disable_write_zeroes(struct mapped_device *md)
834 struct queue_limits *limits = dm_get_queue_limits(md);
836 /* device doesn't really support WRITE ZEROES, disable it */
837 limits->max_write_zeroes_sectors = 0;
840 static void clone_endio(struct bio *bio)
842 blk_status_t error = bio->bi_status;
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;
848 if (unlikely(error == BLK_STS_TARGET)) {
849 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
850 !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)
851 disable_write_same(md);
852 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
853 !bdev_get_queue(bio->bi_bdev)->limits.max_write_zeroes_sectors)
854 disable_write_zeroes(md);
858 int r = endio(tio->ti, bio, &error);
860 case DM_ENDIO_REQUEUE:
861 error = BLK_STS_DM_REQUEUE;
865 case DM_ENDIO_INCOMPLETE:
866 /* The target will handle the io */
869 DMWARN("unimplemented target endio return value: %d", r);
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);
1088 case DM_MAPIO_SUBMITTED:
1090 case DM_MAPIO_REMAPPED:
1091 /* the bio has been remapped so dispatch it */
1092 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1093 tio->io->bio->bi_bdev->bd_dev, sector);
1094 generic_make_request(clone);
1097 dec_pending(tio->io, BLK_STS_IOERR);
1100 case DM_MAPIO_REQUEUE:
1101 dec_pending(tio->io, BLK_STS_DM_REQUEUE);
1105 DMWARN("unimplemented target map return value: %d", r);
1111 struct mapped_device *md;
1112 struct dm_table *map;
1116 unsigned sector_count;
1119 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1121 bio->bi_iter.bi_sector = sector;
1122 bio->bi_iter.bi_size = to_bytes(len);
1126 * Creates a bio that consists of range of complete bvecs.
1128 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1129 sector_t sector, unsigned len)
1131 struct bio *clone = &tio->clone;
1133 __bio_clone_fast(clone, bio);
1135 if (unlikely(bio_integrity(bio) != NULL)) {
1138 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1139 !dm_target_passes_integrity(tio->ti->type))) {
1140 DMWARN("%s: the target %s doesn't support integrity data.",
1141 dm_device_name(tio->io->md),
1142 tio->ti->type->name);
1146 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1151 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1152 clone->bi_iter.bi_size = to_bytes(len);
1154 if (unlikely(bio_integrity(bio) != NULL))
1155 bio_integrity_trim(clone, 0, len);
1160 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1161 struct dm_target *ti,
1162 unsigned target_bio_nr)
1164 struct dm_target_io *tio;
1167 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1168 tio = container_of(clone, struct dm_target_io, clone);
1172 tio->target_bio_nr = target_bio_nr;
1177 static void __clone_and_map_simple_bio(struct clone_info *ci,
1178 struct dm_target *ti,
1179 unsigned target_bio_nr, unsigned *len)
1181 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1182 struct bio *clone = &tio->clone;
1186 __bio_clone_fast(clone, ci->bio);
1188 bio_setup_sector(clone, ci->sector, *len);
1193 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1194 unsigned num_bios, unsigned *len)
1196 unsigned target_bio_nr;
1198 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1199 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1202 static int __send_empty_flush(struct clone_info *ci)
1204 unsigned target_nr = 0;
1205 struct dm_target *ti;
1207 BUG_ON(bio_has_data(ci->bio));
1208 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1209 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1214 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1215 sector_t sector, unsigned *len)
1217 struct bio *bio = ci->bio;
1218 struct dm_target_io *tio;
1219 unsigned target_bio_nr;
1220 unsigned num_target_bios = 1;
1224 * Does the target want to receive duplicate copies of the bio?
1226 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1227 num_target_bios = ti->num_write_bios(ti, bio);
1229 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1230 tio = alloc_tio(ci, ti, target_bio_nr);
1232 r = clone_bio(tio, bio, sector, *len);
1243 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1245 static unsigned get_num_discard_bios(struct dm_target *ti)
1247 return ti->num_discard_bios;
1250 static unsigned get_num_write_same_bios(struct dm_target *ti)
1252 return ti->num_write_same_bios;
1255 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1257 return ti->num_write_zeroes_bios;
1260 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1262 static bool is_split_required_for_discard(struct dm_target *ti)
1264 return ti->split_discard_bios;
1267 static int __send_changing_extent_only(struct clone_info *ci,
1268 get_num_bios_fn get_num_bios,
1269 is_split_required_fn is_split_required)
1271 struct dm_target *ti;
1276 ti = dm_table_find_target(ci->map, ci->sector);
1277 if (!dm_target_is_valid(ti))
1281 * Even though the device advertised support for this type of
1282 * request, that does not mean every target supports it, and
1283 * reconfiguration might also have changed that since the
1284 * check was performed.
1286 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1290 if (is_split_required && !is_split_required(ti))
1291 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1293 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1295 __send_duplicate_bios(ci, ti, num_bios, &len);
1298 } while (ci->sector_count -= len);
1303 static int __send_discard(struct clone_info *ci)
1305 return __send_changing_extent_only(ci, get_num_discard_bios,
1306 is_split_required_for_discard);
1309 static int __send_write_same(struct clone_info *ci)
1311 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1314 static int __send_write_zeroes(struct clone_info *ci)
1316 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1320 * Select the correct strategy for processing a non-flush bio.
1322 static int __split_and_process_non_flush(struct clone_info *ci)
1324 struct bio *bio = ci->bio;
1325 struct dm_target *ti;
1329 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1330 return __send_discard(ci);
1331 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1332 return __send_write_same(ci);
1333 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1334 return __send_write_zeroes(ci);
1336 ti = dm_table_find_target(ci->map, ci->sector);
1337 if (!dm_target_is_valid(ti))
1340 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1342 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1347 ci->sector_count -= len;
1353 * Entry point to split a bio into clones and submit them to the targets.
1355 static void __split_and_process_bio(struct mapped_device *md,
1356 struct dm_table *map, struct bio *bio)
1358 struct clone_info ci;
1361 if (unlikely(!map)) {
1368 ci.io = alloc_io(md);
1370 atomic_set(&ci.io->io_count, 1);
1373 spin_lock_init(&ci.io->endio_lock);
1374 ci.sector = bio->bi_iter.bi_sector;
1376 start_io_acct(ci.io);
1378 if (bio->bi_opf & REQ_PREFLUSH) {
1379 ci.bio = &ci.md->flush_bio;
1380 ci.sector_count = 0;
1381 error = __send_empty_flush(&ci);
1382 /* dec_pending submits any data associated with flush */
1385 ci.sector_count = bio_sectors(bio);
1386 while (ci.sector_count && !error)
1387 error = __split_and_process_non_flush(&ci);
1390 /* drop the extra reference count */
1391 dec_pending(ci.io, error);
1393 /*-----------------------------------------------------------------
1395 *---------------------------------------------------------------*/
1398 * The request function that just remaps the bio built up by
1401 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1403 int rw = bio_data_dir(bio);
1404 struct mapped_device *md = q->queuedata;
1406 struct dm_table *map;
1408 map = dm_get_live_table(md, &srcu_idx);
1410 generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0);
1412 /* if we're suspended, we have to queue this io for later */
1413 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1414 dm_put_live_table(md, srcu_idx);
1416 if (!(bio->bi_opf & REQ_RAHEAD))
1420 return BLK_QC_T_NONE;
1423 __split_and_process_bio(md, map, bio);
1424 dm_put_live_table(md, srcu_idx);
1425 return BLK_QC_T_NONE;
1428 static int dm_any_congested(void *congested_data, int bdi_bits)
1431 struct mapped_device *md = congested_data;
1432 struct dm_table *map;
1434 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1435 if (dm_request_based(md)) {
1437 * With request-based DM we only need to check the
1438 * top-level queue for congestion.
1440 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1442 map = dm_get_live_table_fast(md);
1444 r = dm_table_any_congested(map, bdi_bits);
1445 dm_put_live_table_fast(md);
1452 /*-----------------------------------------------------------------
1453 * An IDR is used to keep track of allocated minor numbers.
1454 *---------------------------------------------------------------*/
1455 static void free_minor(int minor)
1457 spin_lock(&_minor_lock);
1458 idr_remove(&_minor_idr, minor);
1459 spin_unlock(&_minor_lock);
1463 * See if the device with a specific minor # is free.
1465 static int specific_minor(int minor)
1469 if (minor >= (1 << MINORBITS))
1472 idr_preload(GFP_KERNEL);
1473 spin_lock(&_minor_lock);
1475 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1477 spin_unlock(&_minor_lock);
1480 return r == -ENOSPC ? -EBUSY : r;
1484 static int next_free_minor(int *minor)
1488 idr_preload(GFP_KERNEL);
1489 spin_lock(&_minor_lock);
1491 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1493 spin_unlock(&_minor_lock);
1501 static const struct block_device_operations dm_blk_dops;
1502 static const struct dax_operations dm_dax_ops;
1504 static void dm_wq_work(struct work_struct *work);
1506 void dm_init_md_queue(struct mapped_device *md)
1509 * Request-based dm devices cannot be stacked on top of bio-based dm
1510 * devices. The type of this dm device may not have been decided yet.
1511 * The type is decided at the first table loading time.
1512 * To prevent problematic device stacking, clear the queue flag
1513 * for request stacking support until then.
1515 * This queue is new, so no concurrency on the queue_flags.
1517 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1520 * Initialize data that will only be used by a non-blk-mq DM queue
1521 * - must do so here (in alloc_dev callchain) before queue is used
1523 md->queue->queuedata = md;
1524 md->queue->backing_dev_info->congested_data = md;
1527 void dm_init_normal_md_queue(struct mapped_device *md)
1529 md->use_blk_mq = false;
1530 dm_init_md_queue(md);
1533 * Initialize aspects of queue that aren't relevant for blk-mq
1535 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1536 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1539 static void cleanup_mapped_device(struct mapped_device *md)
1542 destroy_workqueue(md->wq);
1543 if (md->kworker_task)
1544 kthread_stop(md->kworker_task);
1545 mempool_destroy(md->io_pool);
1547 bioset_free(md->bs);
1550 kill_dax(md->dax_dev);
1551 put_dax(md->dax_dev);
1556 spin_lock(&_minor_lock);
1557 md->disk->private_data = NULL;
1558 spin_unlock(&_minor_lock);
1559 del_gendisk(md->disk);
1564 blk_cleanup_queue(md->queue);
1566 cleanup_srcu_struct(&md->io_barrier);
1573 dm_mq_cleanup_mapped_device(md);
1577 * Allocate and initialise a blank device with a given minor.
1579 static struct mapped_device *alloc_dev(int minor)
1581 int r, numa_node_id = dm_get_numa_node();
1582 struct dax_device *dax_dev;
1583 struct mapped_device *md;
1586 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1588 DMWARN("unable to allocate device, out of memory.");
1592 if (!try_module_get(THIS_MODULE))
1593 goto bad_module_get;
1595 /* get a minor number for the dev */
1596 if (minor == DM_ANY_MINOR)
1597 r = next_free_minor(&minor);
1599 r = specific_minor(minor);
1603 r = init_srcu_struct(&md->io_barrier);
1605 goto bad_io_barrier;
1607 md->numa_node_id = numa_node_id;
1608 md->use_blk_mq = dm_use_blk_mq_default();
1609 md->init_tio_pdu = false;
1610 md->type = DM_TYPE_NONE;
1611 mutex_init(&md->suspend_lock);
1612 mutex_init(&md->type_lock);
1613 mutex_init(&md->table_devices_lock);
1614 spin_lock_init(&md->deferred_lock);
1615 atomic_set(&md->holders, 1);
1616 atomic_set(&md->open_count, 0);
1617 atomic_set(&md->event_nr, 0);
1618 atomic_set(&md->uevent_seq, 0);
1619 INIT_LIST_HEAD(&md->uevent_list);
1620 INIT_LIST_HEAD(&md->table_devices);
1621 spin_lock_init(&md->uevent_lock);
1623 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1627 dm_init_md_queue(md);
1629 md->disk = alloc_disk_node(1, numa_node_id);
1633 atomic_set(&md->pending[0], 0);
1634 atomic_set(&md->pending[1], 0);
1635 init_waitqueue_head(&md->wait);
1636 INIT_WORK(&md->work, dm_wq_work);
1637 init_waitqueue_head(&md->eventq);
1638 init_completion(&md->kobj_holder.completion);
1639 md->kworker_task = NULL;
1641 md->disk->major = _major;
1642 md->disk->first_minor = minor;
1643 md->disk->fops = &dm_blk_dops;
1644 md->disk->queue = md->queue;
1645 md->disk->private_data = md;
1646 sprintf(md->disk->disk_name, "dm-%d", minor);
1648 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1651 md->dax_dev = dax_dev;
1654 format_dev_t(md->name, MKDEV(_major, minor));
1656 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1660 md->bdev = bdget_disk(md->disk, 0);
1664 bio_init(&md->flush_bio, NULL, 0);
1665 md->flush_bio.bi_bdev = md->bdev;
1666 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1668 dm_stats_init(&md->stats);
1670 /* Populate the mapping, nobody knows we exist yet */
1671 spin_lock(&_minor_lock);
1672 old_md = idr_replace(&_minor_idr, md, minor);
1673 spin_unlock(&_minor_lock);
1675 BUG_ON(old_md != MINOR_ALLOCED);
1680 cleanup_mapped_device(md);
1684 module_put(THIS_MODULE);
1690 static void unlock_fs(struct mapped_device *md);
1692 static void free_dev(struct mapped_device *md)
1694 int minor = MINOR(disk_devt(md->disk));
1698 cleanup_mapped_device(md);
1700 free_table_devices(&md->table_devices);
1701 dm_stats_cleanup(&md->stats);
1704 module_put(THIS_MODULE);
1708 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1710 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1713 /* The md already has necessary mempools. */
1714 if (dm_table_bio_based(t)) {
1716 * Reload bioset because front_pad may have changed
1717 * because a different table was loaded.
1719 bioset_free(md->bs);
1724 * There's no need to reload with request-based dm
1725 * because the size of front_pad doesn't change.
1726 * Note for future: If you are to reload bioset,
1727 * prep-ed requests in the queue may refer
1728 * to bio from the old bioset, so you must walk
1729 * through the queue to unprep.
1734 BUG_ON(!p || md->io_pool || md->bs);
1736 md->io_pool = p->io_pool;
1742 /* mempool bind completed, no longer need any mempools in the table */
1743 dm_table_free_md_mempools(t);
1747 * Bind a table to the device.
1749 static void event_callback(void *context)
1751 unsigned long flags;
1753 struct mapped_device *md = (struct mapped_device *) context;
1755 spin_lock_irqsave(&md->uevent_lock, flags);
1756 list_splice_init(&md->uevent_list, &uevents);
1757 spin_unlock_irqrestore(&md->uevent_lock, flags);
1759 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1761 atomic_inc(&md->event_nr);
1762 wake_up(&md->eventq);
1766 * Protected by md->suspend_lock obtained by dm_swap_table().
1768 static void __set_size(struct mapped_device *md, sector_t size)
1770 lockdep_assert_held(&md->suspend_lock);
1772 set_capacity(md->disk, size);
1774 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1778 * Returns old map, which caller must destroy.
1780 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1781 struct queue_limits *limits)
1783 struct dm_table *old_map;
1784 struct request_queue *q = md->queue;
1787 lockdep_assert_held(&md->suspend_lock);
1789 size = dm_table_get_size(t);
1792 * Wipe any geometry if the size of the table changed.
1794 if (size != dm_get_size(md))
1795 memset(&md->geometry, 0, sizeof(md->geometry));
1797 __set_size(md, size);
1799 dm_table_event_callback(t, event_callback, md);
1802 * The queue hasn't been stopped yet, if the old table type wasn't
1803 * for request-based during suspension. So stop it to prevent
1804 * I/O mapping before resume.
1805 * This must be done before setting the queue restrictions,
1806 * because request-based dm may be run just after the setting.
1808 if (dm_table_request_based(t)) {
1811 * Leverage the fact that request-based DM targets are
1812 * immutable singletons and establish md->immutable_target
1813 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1815 md->immutable_target = dm_table_get_immutable_target(t);
1818 __bind_mempools(md, t);
1820 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1821 rcu_assign_pointer(md->map, (void *)t);
1822 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1824 dm_table_set_restrictions(t, q, limits);
1832 * Returns unbound table for the caller to free.
1834 static struct dm_table *__unbind(struct mapped_device *md)
1836 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1841 dm_table_event_callback(map, NULL, NULL);
1842 RCU_INIT_POINTER(md->map, NULL);
1849 * Constructor for a new device.
1851 int dm_create(int minor, struct mapped_device **result)
1853 struct mapped_device *md;
1855 md = alloc_dev(minor);
1866 * Functions to manage md->type.
1867 * All are required to hold md->type_lock.
1869 void dm_lock_md_type(struct mapped_device *md)
1871 mutex_lock(&md->type_lock);
1874 void dm_unlock_md_type(struct mapped_device *md)
1876 mutex_unlock(&md->type_lock);
1879 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1881 BUG_ON(!mutex_is_locked(&md->type_lock));
1885 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1890 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
1892 return md->immutable_target_type;
1896 * The queue_limits are only valid as long as you have a reference
1899 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
1901 BUG_ON(!atomic_read(&md->holders));
1902 return &md->queue->limits;
1904 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
1907 * Setup the DM device's queue based on md's type
1909 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
1912 enum dm_queue_mode type = dm_get_md_type(md);
1915 case DM_TYPE_REQUEST_BASED:
1916 r = dm_old_init_request_queue(md, t);
1918 DMERR("Cannot initialize queue for request-based mapped device");
1922 case DM_TYPE_MQ_REQUEST_BASED:
1923 r = dm_mq_init_request_queue(md, t);
1925 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1929 case DM_TYPE_BIO_BASED:
1930 case DM_TYPE_DAX_BIO_BASED:
1931 dm_init_normal_md_queue(md);
1932 blk_queue_make_request(md->queue, dm_make_request);
1934 * DM handles splitting bios as needed. Free the bio_split bioset
1935 * since it won't be used (saves 1 process per bio-based DM device).
1937 bioset_free(md->queue->bio_split);
1938 md->queue->bio_split = NULL;
1940 if (type == DM_TYPE_DAX_BIO_BASED)
1941 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
1951 struct mapped_device *dm_get_md(dev_t dev)
1953 struct mapped_device *md;
1954 unsigned minor = MINOR(dev);
1956 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1959 spin_lock(&_minor_lock);
1961 md = idr_find(&_minor_idr, minor);
1963 if ((md == MINOR_ALLOCED ||
1964 (MINOR(disk_devt(dm_disk(md))) != minor) ||
1965 dm_deleting_md(md) ||
1966 test_bit(DMF_FREEING, &md->flags))) {
1974 spin_unlock(&_minor_lock);
1978 EXPORT_SYMBOL_GPL(dm_get_md);
1980 void *dm_get_mdptr(struct mapped_device *md)
1982 return md->interface_ptr;
1985 void dm_set_mdptr(struct mapped_device *md, void *ptr)
1987 md->interface_ptr = ptr;
1990 void dm_get(struct mapped_device *md)
1992 atomic_inc(&md->holders);
1993 BUG_ON(test_bit(DMF_FREEING, &md->flags));
1996 int dm_hold(struct mapped_device *md)
1998 spin_lock(&_minor_lock);
1999 if (test_bit(DMF_FREEING, &md->flags)) {
2000 spin_unlock(&_minor_lock);
2004 spin_unlock(&_minor_lock);
2007 EXPORT_SYMBOL_GPL(dm_hold);
2009 const char *dm_device_name(struct mapped_device *md)
2013 EXPORT_SYMBOL_GPL(dm_device_name);
2015 static void __dm_destroy(struct mapped_device *md, bool wait)
2017 struct request_queue *q = dm_get_md_queue(md);
2018 struct dm_table *map;
2023 spin_lock(&_minor_lock);
2024 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2025 set_bit(DMF_FREEING, &md->flags);
2026 spin_unlock(&_minor_lock);
2028 blk_set_queue_dying(q);
2030 if (dm_request_based(md) && md->kworker_task)
2031 kthread_flush_worker(&md->kworker);
2034 * Take suspend_lock so that presuspend and postsuspend methods
2035 * do not race with internal suspend.
2037 mutex_lock(&md->suspend_lock);
2038 map = dm_get_live_table(md, &srcu_idx);
2039 if (!dm_suspended_md(md)) {
2040 dm_table_presuspend_targets(map);
2041 dm_table_postsuspend_targets(map);
2043 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2044 dm_put_live_table(md, srcu_idx);
2045 mutex_unlock(&md->suspend_lock);
2048 * Rare, but there may be I/O requests still going to complete,
2049 * for example. Wait for all references to disappear.
2050 * No one should increment the reference count of the mapped_device,
2051 * after the mapped_device state becomes DMF_FREEING.
2054 while (atomic_read(&md->holders))
2056 else if (atomic_read(&md->holders))
2057 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2058 dm_device_name(md), atomic_read(&md->holders));
2061 dm_table_destroy(__unbind(md));
2065 void dm_destroy(struct mapped_device *md)
2067 __dm_destroy(md, true);
2070 void dm_destroy_immediate(struct mapped_device *md)
2072 __dm_destroy(md, false);
2075 void dm_put(struct mapped_device *md)
2077 atomic_dec(&md->holders);
2079 EXPORT_SYMBOL_GPL(dm_put);
2081 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2087 prepare_to_wait(&md->wait, &wait, task_state);
2089 if (!md_in_flight(md))
2092 if (signal_pending_state(task_state, current)) {
2099 finish_wait(&md->wait, &wait);
2105 * Process the deferred bios
2107 static void dm_wq_work(struct work_struct *work)
2109 struct mapped_device *md = container_of(work, struct mapped_device,
2113 struct dm_table *map;
2115 map = dm_get_live_table(md, &srcu_idx);
2117 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2118 spin_lock_irq(&md->deferred_lock);
2119 c = bio_list_pop(&md->deferred);
2120 spin_unlock_irq(&md->deferred_lock);
2125 if (dm_request_based(md))
2126 generic_make_request(c);
2128 __split_and_process_bio(md, map, c);
2131 dm_put_live_table(md, srcu_idx);
2134 static void dm_queue_flush(struct mapped_device *md)
2136 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2137 smp_mb__after_atomic();
2138 queue_work(md->wq, &md->work);
2142 * Swap in a new table, returning the old one for the caller to destroy.
2144 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2146 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2147 struct queue_limits limits;
2150 mutex_lock(&md->suspend_lock);
2152 /* device must be suspended */
2153 if (!dm_suspended_md(md))
2157 * If the new table has no data devices, retain the existing limits.
2158 * This helps multipath with queue_if_no_path if all paths disappear,
2159 * then new I/O is queued based on these limits, and then some paths
2162 if (dm_table_has_no_data_devices(table)) {
2163 live_map = dm_get_live_table_fast(md);
2165 limits = md->queue->limits;
2166 dm_put_live_table_fast(md);
2170 r = dm_calculate_queue_limits(table, &limits);
2177 map = __bind(md, table, &limits);
2180 mutex_unlock(&md->suspend_lock);
2185 * Functions to lock and unlock any filesystem running on the
2188 static int lock_fs(struct mapped_device *md)
2192 WARN_ON(md->frozen_sb);
2194 md->frozen_sb = freeze_bdev(md->bdev);
2195 if (IS_ERR(md->frozen_sb)) {
2196 r = PTR_ERR(md->frozen_sb);
2197 md->frozen_sb = NULL;
2201 set_bit(DMF_FROZEN, &md->flags);
2206 static void unlock_fs(struct mapped_device *md)
2208 if (!test_bit(DMF_FROZEN, &md->flags))
2211 thaw_bdev(md->bdev, md->frozen_sb);
2212 md->frozen_sb = NULL;
2213 clear_bit(DMF_FROZEN, &md->flags);
2217 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2218 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2219 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2221 * If __dm_suspend returns 0, the device is completely quiescent
2222 * now. There is no request-processing activity. All new requests
2223 * are being added to md->deferred list.
2225 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2226 unsigned suspend_flags, long task_state,
2227 int dmf_suspended_flag)
2229 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2230 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2233 lockdep_assert_held(&md->suspend_lock);
2236 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2237 * This flag is cleared before dm_suspend returns.
2240 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2242 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2245 * This gets reverted if there's an error later and the targets
2246 * provide the .presuspend_undo hook.
2248 dm_table_presuspend_targets(map);
2251 * Flush I/O to the device.
2252 * Any I/O submitted after lock_fs() may not be flushed.
2253 * noflush takes precedence over do_lockfs.
2254 * (lock_fs() flushes I/Os and waits for them to complete.)
2256 if (!noflush && do_lockfs) {
2259 dm_table_presuspend_undo_targets(map);
2265 * Here we must make sure that no processes are submitting requests
2266 * to target drivers i.e. no one may be executing
2267 * __split_and_process_bio. This is called from dm_request and
2270 * To get all processes out of __split_and_process_bio in dm_request,
2271 * we take the write lock. To prevent any process from reentering
2272 * __split_and_process_bio from dm_request and quiesce the thread
2273 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2274 * flush_workqueue(md->wq).
2276 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2278 synchronize_srcu(&md->io_barrier);
2281 * Stop md->queue before flushing md->wq in case request-based
2282 * dm defers requests to md->wq from md->queue.
2284 if (dm_request_based(md)) {
2285 dm_stop_queue(md->queue);
2286 if (md->kworker_task)
2287 kthread_flush_worker(&md->kworker);
2290 flush_workqueue(md->wq);
2293 * At this point no more requests are entering target request routines.
2294 * We call dm_wait_for_completion to wait for all existing requests
2297 r = dm_wait_for_completion(md, task_state);
2299 set_bit(dmf_suspended_flag, &md->flags);
2302 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2304 synchronize_srcu(&md->io_barrier);
2306 /* were we interrupted ? */
2310 if (dm_request_based(md))
2311 dm_start_queue(md->queue);
2314 dm_table_presuspend_undo_targets(map);
2315 /* pushback list is already flushed, so skip flush */
2322 * We need to be able to change a mapping table under a mounted
2323 * filesystem. For example we might want to move some data in
2324 * the background. Before the table can be swapped with
2325 * dm_bind_table, dm_suspend must be called to flush any in
2326 * flight bios and ensure that any further io gets deferred.
2329 * Suspend mechanism in request-based dm.
2331 * 1. Flush all I/Os by lock_fs() if needed.
2332 * 2. Stop dispatching any I/O by stopping the request_queue.
2333 * 3. Wait for all in-flight I/Os to be completed or requeued.
2335 * To abort suspend, start the request_queue.
2337 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2339 struct dm_table *map = NULL;
2343 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2345 if (dm_suspended_md(md)) {
2350 if (dm_suspended_internally_md(md)) {
2351 /* already internally suspended, wait for internal resume */
2352 mutex_unlock(&md->suspend_lock);
2353 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2359 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2361 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2365 dm_table_postsuspend_targets(map);
2368 mutex_unlock(&md->suspend_lock);
2372 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2375 int r = dm_table_resume_targets(map);
2383 * Flushing deferred I/Os must be done after targets are resumed
2384 * so that mapping of targets can work correctly.
2385 * Request-based dm is queueing the deferred I/Os in its request_queue.
2387 if (dm_request_based(md))
2388 dm_start_queue(md->queue);
2395 int dm_resume(struct mapped_device *md)
2398 struct dm_table *map = NULL;
2402 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2404 if (!dm_suspended_md(md))
2407 if (dm_suspended_internally_md(md)) {
2408 /* already internally suspended, wait for internal resume */
2409 mutex_unlock(&md->suspend_lock);
2410 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2416 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2417 if (!map || !dm_table_get_size(map))
2420 r = __dm_resume(md, map);
2424 clear_bit(DMF_SUSPENDED, &md->flags);
2426 mutex_unlock(&md->suspend_lock);
2432 * Internal suspend/resume works like userspace-driven suspend. It waits
2433 * until all bios finish and prevents issuing new bios to the target drivers.
2434 * It may be used only from the kernel.
2437 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2439 struct dm_table *map = NULL;
2441 lockdep_assert_held(&md->suspend_lock);
2443 if (md->internal_suspend_count++)
2444 return; /* nested internal suspend */
2446 if (dm_suspended_md(md)) {
2447 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2448 return; /* nest suspend */
2451 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2454 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2455 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2456 * would require changing .presuspend to return an error -- avoid this
2457 * until there is a need for more elaborate variants of internal suspend.
2459 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2460 DMF_SUSPENDED_INTERNALLY);
2462 dm_table_postsuspend_targets(map);
2465 static void __dm_internal_resume(struct mapped_device *md)
2467 BUG_ON(!md->internal_suspend_count);
2469 if (--md->internal_suspend_count)
2470 return; /* resume from nested internal suspend */
2472 if (dm_suspended_md(md))
2473 goto done; /* resume from nested suspend */
2476 * NOTE: existing callers don't need to call dm_table_resume_targets
2477 * (which may fail -- so best to avoid it for now by passing NULL map)
2479 (void) __dm_resume(md, NULL);
2482 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2483 smp_mb__after_atomic();
2484 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2487 void dm_internal_suspend_noflush(struct mapped_device *md)
2489 mutex_lock(&md->suspend_lock);
2490 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2491 mutex_unlock(&md->suspend_lock);
2493 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2495 void dm_internal_resume(struct mapped_device *md)
2497 mutex_lock(&md->suspend_lock);
2498 __dm_internal_resume(md);
2499 mutex_unlock(&md->suspend_lock);
2501 EXPORT_SYMBOL_GPL(dm_internal_resume);
2504 * Fast variants of internal suspend/resume hold md->suspend_lock,
2505 * which prevents interaction with userspace-driven suspend.
2508 void dm_internal_suspend_fast(struct mapped_device *md)
2510 mutex_lock(&md->suspend_lock);
2511 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2514 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2515 synchronize_srcu(&md->io_barrier);
2516 flush_workqueue(md->wq);
2517 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2519 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2521 void dm_internal_resume_fast(struct mapped_device *md)
2523 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2529 mutex_unlock(&md->suspend_lock);
2531 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2533 /*-----------------------------------------------------------------
2534 * Event notification.
2535 *---------------------------------------------------------------*/
2536 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2539 char udev_cookie[DM_COOKIE_LENGTH];
2540 char *envp[] = { udev_cookie, NULL };
2543 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2545 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2546 DM_COOKIE_ENV_VAR_NAME, cookie);
2547 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2552 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2554 return atomic_add_return(1, &md->uevent_seq);
2557 uint32_t dm_get_event_nr(struct mapped_device *md)
2559 return atomic_read(&md->event_nr);
2562 int dm_wait_event(struct mapped_device *md, int event_nr)
2564 return wait_event_interruptible(md->eventq,
2565 (event_nr != atomic_read(&md->event_nr)));
2568 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2570 unsigned long flags;
2572 spin_lock_irqsave(&md->uevent_lock, flags);
2573 list_add(elist, &md->uevent_list);
2574 spin_unlock_irqrestore(&md->uevent_lock, flags);
2578 * The gendisk is only valid as long as you have a reference
2581 struct gendisk *dm_disk(struct mapped_device *md)
2585 EXPORT_SYMBOL_GPL(dm_disk);
2587 struct kobject *dm_kobject(struct mapped_device *md)
2589 return &md->kobj_holder.kobj;
2592 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2594 struct mapped_device *md;
2596 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2598 if (test_bit(DMF_FREEING, &md->flags) ||
2606 int dm_suspended_md(struct mapped_device *md)
2608 return test_bit(DMF_SUSPENDED, &md->flags);
2611 int dm_suspended_internally_md(struct mapped_device *md)
2613 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2616 int dm_test_deferred_remove_flag(struct mapped_device *md)
2618 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2621 int dm_suspended(struct dm_target *ti)
2623 return dm_suspended_md(dm_table_get_md(ti->table));
2625 EXPORT_SYMBOL_GPL(dm_suspended);
2627 int dm_noflush_suspending(struct dm_target *ti)
2629 return __noflush_suspending(dm_table_get_md(ti->table));
2631 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2633 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2634 unsigned integrity, unsigned per_io_data_size)
2636 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2637 unsigned int pool_size = 0;
2638 unsigned int front_pad;
2644 case DM_TYPE_BIO_BASED:
2645 case DM_TYPE_DAX_BIO_BASED:
2646 pool_size = dm_get_reserved_bio_based_ios();
2647 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2649 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2650 if (!pools->io_pool)
2653 case DM_TYPE_REQUEST_BASED:
2654 case DM_TYPE_MQ_REQUEST_BASED:
2655 pool_size = dm_get_reserved_rq_based_ios();
2656 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2657 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2663 pools->bs = bioset_create(pool_size, front_pad, 0);
2667 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2673 dm_free_md_mempools(pools);
2678 void dm_free_md_mempools(struct dm_md_mempools *pools)
2683 mempool_destroy(pools->io_pool);
2686 bioset_free(pools->bs);
2698 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2701 struct mapped_device *md = bdev->bd_disk->private_data;
2702 struct dm_table *table;
2703 struct dm_target *ti;
2704 int ret = -ENOTTY, srcu_idx;
2706 table = dm_get_live_table(md, &srcu_idx);
2707 if (!table || !dm_table_get_size(table))
2710 /* We only support devices that have a single target */
2711 if (dm_table_get_num_targets(table) != 1)
2713 ti = dm_table_get_target(table, 0);
2716 if (!ti->type->iterate_devices)
2719 ret = ti->type->iterate_devices(ti, fn, data);
2721 dm_put_live_table(md, srcu_idx);
2726 * For register / unregister we need to manually call out to every path.
2728 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2729 sector_t start, sector_t len, void *data)
2731 struct dm_pr *pr = data;
2732 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2734 if (!ops || !ops->pr_register)
2736 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2739 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2750 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2751 if (ret && new_key) {
2752 /* unregister all paths if we failed to register any path */
2753 pr.old_key = new_key;
2756 pr.fail_early = false;
2757 dm_call_pr(bdev, __dm_pr_register, &pr);
2763 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2766 struct mapped_device *md = bdev->bd_disk->private_data;
2767 const struct pr_ops *ops;
2771 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2775 ops = bdev->bd_disk->fops->pr_ops;
2776 if (ops && ops->pr_reserve)
2777 r = ops->pr_reserve(bdev, key, type, flags);
2785 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2787 struct mapped_device *md = bdev->bd_disk->private_data;
2788 const struct pr_ops *ops;
2792 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2796 ops = bdev->bd_disk->fops->pr_ops;
2797 if (ops && ops->pr_release)
2798 r = ops->pr_release(bdev, key, type);
2806 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2807 enum pr_type type, bool abort)
2809 struct mapped_device *md = bdev->bd_disk->private_data;
2810 const struct pr_ops *ops;
2814 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2818 ops = bdev->bd_disk->fops->pr_ops;
2819 if (ops && ops->pr_preempt)
2820 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2828 static int dm_pr_clear(struct block_device *bdev, u64 key)
2830 struct mapped_device *md = bdev->bd_disk->private_data;
2831 const struct pr_ops *ops;
2835 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2839 ops = bdev->bd_disk->fops->pr_ops;
2840 if (ops && ops->pr_clear)
2841 r = ops->pr_clear(bdev, key);
2849 static const struct pr_ops dm_pr_ops = {
2850 .pr_register = dm_pr_register,
2851 .pr_reserve = dm_pr_reserve,
2852 .pr_release = dm_pr_release,
2853 .pr_preempt = dm_pr_preempt,
2854 .pr_clear = dm_pr_clear,
2857 static const struct block_device_operations dm_blk_dops = {
2858 .open = dm_blk_open,
2859 .release = dm_blk_close,
2860 .ioctl = dm_blk_ioctl,
2861 .getgeo = dm_blk_getgeo,
2862 .pr_ops = &dm_pr_ops,
2863 .owner = THIS_MODULE
2866 static const struct dax_operations dm_dax_ops = {
2867 .direct_access = dm_dax_direct_access,
2873 module_init(dm_init);
2874 module_exit(dm_exit);
2876 module_param(major, uint, 0);
2877 MODULE_PARM_DESC(major, "The major number of the device mapper");
2879 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2880 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2882 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2883 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2885 MODULE_DESCRIPTION(DM_NAME " driver");
2886 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2887 MODULE_LICENSE("GPL");