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/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
28 #define DM_MSG_PREFIX "core"
31 * Cookies are numeric values sent with CHANGE and REMOVE
32 * uevents while resuming, removing or renaming the device.
34 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
35 #define DM_COOKIE_LENGTH 24
37 static const char *_name = DM_NAME;
39 static unsigned int major = 0;
40 static unsigned int _major = 0;
42 static DEFINE_IDR(_minor_idr);
44 static DEFINE_SPINLOCK(_minor_lock);
46 static void do_deferred_remove(struct work_struct *w);
48 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
50 static struct workqueue_struct *deferred_remove_workqueue;
52 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
53 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
56 * One of these is allocated per bio.
59 struct mapped_device *md;
63 unsigned long start_time;
64 spinlock_t endio_lock;
65 struct dm_stats_aux stats_aux;
68 #define MINOR_ALLOCED ((void *)-1)
71 * Bits for the md->flags field.
73 #define DMF_BLOCK_IO_FOR_SUSPEND 0
74 #define DMF_SUSPENDED 1
77 #define DMF_DELETING 4
78 #define DMF_NOFLUSH_SUSPENDING 5
79 #define DMF_DEFERRED_REMOVE 6
80 #define DMF_SUSPENDED_INTERNALLY 7
82 #define DM_NUMA_NODE NUMA_NO_NODE
83 static int dm_numa_node = DM_NUMA_NODE;
86 * For mempools pre-allocation at the table loading time.
88 struct dm_md_mempools {
94 struct list_head list;
99 static struct kmem_cache *_io_cache;
100 static struct kmem_cache *_rq_tio_cache;
101 static struct kmem_cache *_rq_cache;
104 * Bio-based DM's mempools' reserved IOs set by the user.
106 #define RESERVED_BIO_BASED_IOS 16
107 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
109 static int __dm_get_module_param_int(int *module_param, int min, int max)
111 int param = ACCESS_ONCE(*module_param);
112 int modified_param = 0;
113 bool modified = true;
116 modified_param = min;
117 else if (param > max)
118 modified_param = max;
123 (void)cmpxchg(module_param, param, modified_param);
124 param = modified_param;
130 unsigned __dm_get_module_param(unsigned *module_param,
131 unsigned def, unsigned max)
133 unsigned param = ACCESS_ONCE(*module_param);
134 unsigned modified_param = 0;
137 modified_param = def;
138 else if (param > max)
139 modified_param = max;
141 if (modified_param) {
142 (void)cmpxchg(module_param, param, modified_param);
143 param = modified_param;
149 unsigned dm_get_reserved_bio_based_ios(void)
151 return __dm_get_module_param(&reserved_bio_based_ios,
152 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
154 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
156 static unsigned dm_get_numa_node(void)
158 return __dm_get_module_param_int(&dm_numa_node,
159 DM_NUMA_NODE, num_online_nodes() - 1);
162 static int __init local_init(void)
166 /* allocate a slab for the dm_ios */
167 _io_cache = KMEM_CACHE(dm_io, 0);
171 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
173 goto out_free_io_cache;
175 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
176 __alignof__(struct request), 0, NULL);
178 goto out_free_rq_tio_cache;
180 r = dm_uevent_init();
182 goto out_free_rq_cache;
184 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
185 if (!deferred_remove_workqueue) {
187 goto out_uevent_exit;
191 r = register_blkdev(_major, _name);
193 goto out_free_workqueue;
201 destroy_workqueue(deferred_remove_workqueue);
205 kmem_cache_destroy(_rq_cache);
206 out_free_rq_tio_cache:
207 kmem_cache_destroy(_rq_tio_cache);
209 kmem_cache_destroy(_io_cache);
214 static void local_exit(void)
216 flush_scheduled_work();
217 destroy_workqueue(deferred_remove_workqueue);
219 kmem_cache_destroy(_rq_cache);
220 kmem_cache_destroy(_rq_tio_cache);
221 kmem_cache_destroy(_io_cache);
222 unregister_blkdev(_major, _name);
227 DMINFO("cleaned up");
230 static int (*_inits[])(void) __initdata = {
241 static void (*_exits[])(void) = {
252 static int __init dm_init(void)
254 const int count = ARRAY_SIZE(_inits);
258 for (i = 0; i < count; i++) {
273 static void __exit dm_exit(void)
275 int i = ARRAY_SIZE(_exits);
281 * Should be empty by this point.
283 idr_destroy(&_minor_idr);
287 * Block device functions
289 int dm_deleting_md(struct mapped_device *md)
291 return test_bit(DMF_DELETING, &md->flags);
294 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
296 struct mapped_device *md;
298 spin_lock(&_minor_lock);
300 md = bdev->bd_disk->private_data;
304 if (test_bit(DMF_FREEING, &md->flags) ||
305 dm_deleting_md(md)) {
311 atomic_inc(&md->open_count);
313 spin_unlock(&_minor_lock);
315 return md ? 0 : -ENXIO;
318 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
320 struct mapped_device *md;
322 spin_lock(&_minor_lock);
324 md = disk->private_data;
328 if (atomic_dec_and_test(&md->open_count) &&
329 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
330 queue_work(deferred_remove_workqueue, &deferred_remove_work);
334 spin_unlock(&_minor_lock);
337 int dm_open_count(struct mapped_device *md)
339 return atomic_read(&md->open_count);
343 * Guarantees nothing is using the device before it's deleted.
345 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
349 spin_lock(&_minor_lock);
351 if (dm_open_count(md)) {
354 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
355 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
358 set_bit(DMF_DELETING, &md->flags);
360 spin_unlock(&_minor_lock);
365 int dm_cancel_deferred_remove(struct mapped_device *md)
369 spin_lock(&_minor_lock);
371 if (test_bit(DMF_DELETING, &md->flags))
374 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
376 spin_unlock(&_minor_lock);
381 static void do_deferred_remove(struct work_struct *w)
383 dm_deferred_remove();
386 sector_t dm_get_size(struct mapped_device *md)
388 return get_capacity(md->disk);
391 struct request_queue *dm_get_md_queue(struct mapped_device *md)
396 struct dm_stats *dm_get_stats(struct mapped_device *md)
401 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
403 struct mapped_device *md = bdev->bd_disk->private_data;
405 return dm_get_geometry(md, geo);
408 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
409 struct block_device **bdev,
412 struct dm_target *tgt;
413 struct dm_table *map;
418 map = dm_get_live_table(md, &srcu_idx);
419 if (!map || !dm_table_get_size(map))
422 /* We only support devices that have a single target */
423 if (dm_table_get_num_targets(map) != 1)
426 tgt = dm_table_get_target(map, 0);
427 if (!tgt->type->prepare_ioctl)
430 if (dm_suspended_md(md)) {
435 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
440 dm_put_live_table(md, srcu_idx);
444 dm_put_live_table(md, srcu_idx);
445 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
452 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
453 unsigned int cmd, unsigned long arg)
455 struct mapped_device *md = bdev->bd_disk->private_data;
458 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
464 * Target determined this ioctl is being issued against a
465 * subset of the parent bdev; require extra privileges.
467 if (!capable(CAP_SYS_RAWIO)) {
469 "%s: sending ioctl %x to DM device without required privilege.",
476 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
482 static struct dm_io *alloc_io(struct mapped_device *md)
484 return mempool_alloc(md->io_pool, GFP_NOIO);
487 static void free_io(struct mapped_device *md, struct dm_io *io)
489 mempool_free(io, md->io_pool);
492 static void free_tio(struct dm_target_io *tio)
494 bio_put(&tio->clone);
497 int md_in_flight(struct mapped_device *md)
499 return atomic_read(&md->pending[READ]) +
500 atomic_read(&md->pending[WRITE]);
503 static void start_io_acct(struct dm_io *io)
505 struct mapped_device *md = io->md;
506 struct bio *bio = io->bio;
508 int rw = bio_data_dir(bio);
510 io->start_time = jiffies;
512 cpu = part_stat_lock();
513 part_round_stats(md->queue, cpu, &dm_disk(md)->part0);
515 atomic_set(&dm_disk(md)->part0.in_flight[rw],
516 atomic_inc_return(&md->pending[rw]));
518 if (unlikely(dm_stats_used(&md->stats)))
519 dm_stats_account_io(&md->stats, bio_data_dir(bio),
520 bio->bi_iter.bi_sector, bio_sectors(bio),
521 false, 0, &io->stats_aux);
524 static void end_io_acct(struct dm_io *io)
526 struct mapped_device *md = io->md;
527 struct bio *bio = io->bio;
528 unsigned long duration = jiffies - io->start_time;
530 int rw = bio_data_dir(bio);
532 generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time);
534 if (unlikely(dm_stats_used(&md->stats)))
535 dm_stats_account_io(&md->stats, bio_data_dir(bio),
536 bio->bi_iter.bi_sector, bio_sectors(bio),
537 true, duration, &io->stats_aux);
540 * After this is decremented the bio must not be touched if it is
543 pending = atomic_dec_return(&md->pending[rw]);
544 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
545 pending += atomic_read(&md->pending[rw^0x1]);
547 /* nudge anyone waiting on suspend queue */
553 * Add the bio to the list of deferred io.
555 static void queue_io(struct mapped_device *md, struct bio *bio)
559 spin_lock_irqsave(&md->deferred_lock, flags);
560 bio_list_add(&md->deferred, bio);
561 spin_unlock_irqrestore(&md->deferred_lock, flags);
562 queue_work(md->wq, &md->work);
566 * Everyone (including functions in this file), should use this
567 * function to access the md->map field, and make sure they call
568 * dm_put_live_table() when finished.
570 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
572 *srcu_idx = srcu_read_lock(&md->io_barrier);
574 return srcu_dereference(md->map, &md->io_barrier);
577 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
579 srcu_read_unlock(&md->io_barrier, srcu_idx);
582 void dm_sync_table(struct mapped_device *md)
584 synchronize_srcu(&md->io_barrier);
585 synchronize_rcu_expedited();
589 * A fast alternative to dm_get_live_table/dm_put_live_table.
590 * The caller must not block between these two functions.
592 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
595 return rcu_dereference(md->map);
598 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
604 * Open a table device so we can use it as a map destination.
606 static int open_table_device(struct table_device *td, dev_t dev,
607 struct mapped_device *md)
609 static char *_claim_ptr = "I belong to device-mapper";
610 struct block_device *bdev;
614 BUG_ON(td->dm_dev.bdev);
616 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
618 return PTR_ERR(bdev);
620 r = bd_link_disk_holder(bdev, dm_disk(md));
622 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
626 td->dm_dev.bdev = bdev;
627 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
632 * Close a table device that we've been using.
634 static void close_table_device(struct table_device *td, struct mapped_device *md)
636 if (!td->dm_dev.bdev)
639 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
640 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
641 put_dax(td->dm_dev.dax_dev);
642 td->dm_dev.bdev = NULL;
643 td->dm_dev.dax_dev = NULL;
646 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
648 struct table_device *td;
650 list_for_each_entry(td, l, list)
651 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
657 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
658 struct dm_dev **result) {
660 struct table_device *td;
662 mutex_lock(&md->table_devices_lock);
663 td = find_table_device(&md->table_devices, dev, mode);
665 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
667 mutex_unlock(&md->table_devices_lock);
671 td->dm_dev.mode = mode;
672 td->dm_dev.bdev = NULL;
674 if ((r = open_table_device(td, dev, md))) {
675 mutex_unlock(&md->table_devices_lock);
680 format_dev_t(td->dm_dev.name, dev);
682 atomic_set(&td->count, 0);
683 list_add(&td->list, &md->table_devices);
685 atomic_inc(&td->count);
686 mutex_unlock(&md->table_devices_lock);
688 *result = &td->dm_dev;
691 EXPORT_SYMBOL_GPL(dm_get_table_device);
693 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
695 struct table_device *td = container_of(d, struct table_device, dm_dev);
697 mutex_lock(&md->table_devices_lock);
698 if (atomic_dec_and_test(&td->count)) {
699 close_table_device(td, md);
703 mutex_unlock(&md->table_devices_lock);
705 EXPORT_SYMBOL(dm_put_table_device);
707 static void free_table_devices(struct list_head *devices)
709 struct list_head *tmp, *next;
711 list_for_each_safe(tmp, next, devices) {
712 struct table_device *td = list_entry(tmp, struct table_device, list);
714 DMWARN("dm_destroy: %s still exists with %d references",
715 td->dm_dev.name, atomic_read(&td->count));
721 * Get the geometry associated with a dm device
723 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
731 * Set the geometry of a device.
733 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
735 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
737 if (geo->start > sz) {
738 DMWARN("Start sector is beyond the geometry limits.");
747 /*-----------------------------------------------------------------
749 * A more elegant soln is in the works that uses the queue
750 * merge fn, unfortunately there are a couple of changes to
751 * the block layer that I want to make for this. So in the
752 * interests of getting something for people to use I give
753 * you this clearly demarcated crap.
754 *---------------------------------------------------------------*/
756 static int __noflush_suspending(struct mapped_device *md)
758 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
762 * Decrements the number of outstanding ios that a bio has been
763 * cloned into, completing the original io if necc.
765 static void dec_pending(struct dm_io *io, blk_status_t error)
768 blk_status_t io_error;
770 struct mapped_device *md = io->md;
772 /* Push-back supersedes any I/O errors */
773 if (unlikely(error)) {
774 spin_lock_irqsave(&io->endio_lock, flags);
775 if (!(io->status == BLK_STS_DM_REQUEUE &&
776 __noflush_suspending(md)))
778 spin_unlock_irqrestore(&io->endio_lock, flags);
781 if (atomic_dec_and_test(&io->io_count)) {
782 if (io->status == BLK_STS_DM_REQUEUE) {
784 * Target requested pushing back the I/O.
786 spin_lock_irqsave(&md->deferred_lock, flags);
787 if (__noflush_suspending(md))
788 bio_list_add_head(&md->deferred, io->bio);
790 /* noflush suspend was interrupted. */
791 io->status = BLK_STS_IOERR;
792 spin_unlock_irqrestore(&md->deferred_lock, flags);
795 io_error = io->status;
800 if (io_error == BLK_STS_DM_REQUEUE)
803 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
805 * Preflush done for flush with data, reissue
806 * without REQ_PREFLUSH.
808 bio->bi_opf &= ~REQ_PREFLUSH;
811 /* done with normal IO or empty flush */
812 bio->bi_status = io_error;
818 void disable_write_same(struct mapped_device *md)
820 struct queue_limits *limits = dm_get_queue_limits(md);
822 /* device doesn't really support WRITE SAME, disable it */
823 limits->max_write_same_sectors = 0;
826 void disable_write_zeroes(struct mapped_device *md)
828 struct queue_limits *limits = dm_get_queue_limits(md);
830 /* device doesn't really support WRITE ZEROES, disable it */
831 limits->max_write_zeroes_sectors = 0;
834 static void clone_endio(struct bio *bio)
836 blk_status_t error = bio->bi_status;
837 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
838 struct dm_io *io = tio->io;
839 struct mapped_device *md = tio->io->md;
840 dm_endio_fn endio = tio->ti->type->end_io;
842 if (unlikely(error == BLK_STS_TARGET)) {
843 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
844 !bio->bi_disk->queue->limits.max_write_same_sectors)
845 disable_write_same(md);
846 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
847 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
848 disable_write_zeroes(md);
852 int r = endio(tio->ti, bio, &error);
854 case DM_ENDIO_REQUEUE:
855 error = BLK_STS_DM_REQUEUE;
859 case DM_ENDIO_INCOMPLETE:
860 /* The target will handle the io */
863 DMWARN("unimplemented target endio return value: %d", r);
869 dec_pending(io, error);
873 * Return maximum size of I/O possible at the supplied sector up to the current
876 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
878 sector_t target_offset = dm_target_offset(ti, sector);
880 return ti->len - target_offset;
883 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
885 sector_t len = max_io_len_target_boundary(sector, ti);
886 sector_t offset, max_len;
889 * Does the target need to split even further?
891 if (ti->max_io_len) {
892 offset = dm_target_offset(ti, sector);
893 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
894 max_len = sector_div(offset, ti->max_io_len);
896 max_len = offset & (ti->max_io_len - 1);
897 max_len = ti->max_io_len - max_len;
906 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
908 if (len > UINT_MAX) {
909 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
910 (unsigned long long)len, UINT_MAX);
911 ti->error = "Maximum size of target IO is too large";
915 ti->max_io_len = (uint32_t) len;
919 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
921 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
922 sector_t sector, int *srcu_idx)
924 struct dm_table *map;
925 struct dm_target *ti;
927 map = dm_get_live_table(md, srcu_idx);
931 ti = dm_table_find_target(map, sector);
932 if (!dm_target_is_valid(ti))
938 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
939 long nr_pages, void **kaddr, pfn_t *pfn)
941 struct mapped_device *md = dax_get_private(dax_dev);
942 sector_t sector = pgoff * PAGE_SECTORS;
943 struct dm_target *ti;
944 long len, ret = -EIO;
947 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
951 if (!ti->type->direct_access)
953 len = max_io_len(sector, ti) / PAGE_SECTORS;
956 nr_pages = min(len, nr_pages);
957 if (ti->type->direct_access)
958 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
961 dm_put_live_table(md, srcu_idx);
966 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
967 void *addr, size_t bytes, struct iov_iter *i)
969 struct mapped_device *md = dax_get_private(dax_dev);
970 sector_t sector = pgoff * PAGE_SECTORS;
971 struct dm_target *ti;
975 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
979 if (!ti->type->dax_copy_from_iter) {
980 ret = copy_from_iter(addr, bytes, i);
983 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
985 dm_put_live_table(md, srcu_idx);
991 * A target may call dm_accept_partial_bio only from the map routine. It is
992 * allowed for all bio types except REQ_PREFLUSH.
994 * dm_accept_partial_bio informs the dm that the target only wants to process
995 * additional n_sectors sectors of the bio and the rest of the data should be
996 * sent in a next bio.
998 * A diagram that explains the arithmetics:
999 * +--------------------+---------------+-------+
1001 * +--------------------+---------------+-------+
1003 * <-------------- *tio->len_ptr --------------->
1004 * <------- bi_size ------->
1007 * Region 1 was already iterated over with bio_advance or similar function.
1008 * (it may be empty if the target doesn't use bio_advance)
1009 * Region 2 is the remaining bio size that the target wants to process.
1010 * (it may be empty if region 1 is non-empty, although there is no reason
1012 * The target requires that region 3 is to be sent in the next bio.
1014 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1015 * the partially processed part (the sum of regions 1+2) must be the same for all
1016 * copies of the bio.
1018 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1020 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1021 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1022 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1023 BUG_ON(bi_size > *tio->len_ptr);
1024 BUG_ON(n_sectors > bi_size);
1025 *tio->len_ptr -= bi_size - n_sectors;
1026 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1028 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1031 * The zone descriptors obtained with a zone report indicate
1032 * zone positions within the target device. The zone descriptors
1033 * must be remapped to match their position within the dm device.
1034 * A target may call dm_remap_zone_report after completion of a
1035 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained
1036 * from the target device mapping to the dm device.
1038 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1040 #ifdef CONFIG_BLK_DEV_ZONED
1041 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1042 struct bio *report_bio = tio->io->bio;
1043 struct blk_zone_report_hdr *hdr = NULL;
1044 struct blk_zone *zone;
1045 unsigned int nr_rep = 0;
1047 struct bio_vec bvec;
1048 struct bvec_iter iter;
1055 * Remap the start sector of the reported zones. For sequential zones,
1056 * also remap the write pointer position.
1058 bio_for_each_segment(bvec, report_bio, iter) {
1059 addr = kmap_atomic(bvec.bv_page);
1061 /* Remember the report header in the first page */
1064 ofst = sizeof(struct blk_zone_report_hdr);
1068 /* Set zones start sector */
1069 while (hdr->nr_zones && ofst < bvec.bv_len) {
1071 if (zone->start >= start + ti->len) {
1075 zone->start = zone->start + ti->begin - start;
1076 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1077 if (zone->cond == BLK_ZONE_COND_FULL)
1078 zone->wp = zone->start + zone->len;
1079 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1080 zone->wp = zone->start;
1082 zone->wp = zone->wp + ti->begin - start;
1084 ofst += sizeof(struct blk_zone);
1090 kunmap_atomic(addr);
1097 hdr->nr_zones = nr_rep;
1101 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1103 #else /* !CONFIG_BLK_DEV_ZONED */
1104 bio->bi_status = BLK_STS_NOTSUPP;
1107 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1110 * Flush current->bio_list when the target map method blocks.
1111 * This fixes deadlocks in snapshot and possibly in other targets.
1114 struct blk_plug plug;
1115 struct blk_plug_cb cb;
1118 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1120 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1121 struct bio_list list;
1125 INIT_LIST_HEAD(&o->cb.list);
1127 if (unlikely(!current->bio_list))
1130 for (i = 0; i < 2; i++) {
1131 list = current->bio_list[i];
1132 bio_list_init(¤t->bio_list[i]);
1134 while ((bio = bio_list_pop(&list))) {
1135 struct bio_set *bs = bio->bi_pool;
1136 if (unlikely(!bs) || bs == fs_bio_set ||
1137 !bs->rescue_workqueue) {
1138 bio_list_add(¤t->bio_list[i], bio);
1142 spin_lock(&bs->rescue_lock);
1143 bio_list_add(&bs->rescue_list, bio);
1144 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1145 spin_unlock(&bs->rescue_lock);
1150 static void dm_offload_start(struct dm_offload *o)
1152 blk_start_plug(&o->plug);
1153 o->cb.callback = flush_current_bio_list;
1154 list_add(&o->cb.list, ¤t->plug->cb_list);
1157 static void dm_offload_end(struct dm_offload *o)
1159 list_del(&o->cb.list);
1160 blk_finish_plug(&o->plug);
1163 static void __map_bio(struct dm_target_io *tio)
1167 struct dm_offload o;
1168 struct bio *clone = &tio->clone;
1169 struct dm_target *ti = tio->ti;
1171 clone->bi_end_io = clone_endio;
1174 * Map the clone. If r == 0 we don't need to do
1175 * anything, the target has assumed ownership of
1178 atomic_inc(&tio->io->io_count);
1179 sector = clone->bi_iter.bi_sector;
1181 dm_offload_start(&o);
1182 r = ti->type->map(ti, clone);
1186 case DM_MAPIO_SUBMITTED:
1188 case DM_MAPIO_REMAPPED:
1189 /* the bio has been remapped so dispatch it */
1190 trace_block_bio_remap(clone->bi_disk->queue, clone,
1191 bio_dev(tio->io->bio), sector);
1192 generic_make_request(clone);
1195 dec_pending(tio->io, BLK_STS_IOERR);
1198 case DM_MAPIO_REQUEUE:
1199 dec_pending(tio->io, BLK_STS_DM_REQUEUE);
1203 DMWARN("unimplemented target map return value: %d", r);
1209 struct mapped_device *md;
1210 struct dm_table *map;
1214 unsigned sector_count;
1217 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1219 bio->bi_iter.bi_sector = sector;
1220 bio->bi_iter.bi_size = to_bytes(len);
1224 * Creates a bio that consists of range of complete bvecs.
1226 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1227 sector_t sector, unsigned len)
1229 struct bio *clone = &tio->clone;
1231 __bio_clone_fast(clone, bio);
1233 if (unlikely(bio_integrity(bio) != NULL)) {
1236 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1237 !dm_target_passes_integrity(tio->ti->type))) {
1238 DMWARN("%s: the target %s doesn't support integrity data.",
1239 dm_device_name(tio->io->md),
1240 tio->ti->type->name);
1244 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1249 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1250 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1251 clone->bi_iter.bi_size = to_bytes(len);
1253 if (unlikely(bio_integrity(bio) != NULL))
1254 bio_integrity_trim(clone);
1259 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1260 struct dm_target *ti,
1261 unsigned target_bio_nr)
1263 struct dm_target_io *tio;
1266 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1267 tio = container_of(clone, struct dm_target_io, clone);
1271 tio->target_bio_nr = target_bio_nr;
1276 static void __clone_and_map_simple_bio(struct clone_info *ci,
1277 struct dm_target *ti,
1278 unsigned target_bio_nr, unsigned *len)
1280 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1281 struct bio *clone = &tio->clone;
1285 __bio_clone_fast(clone, ci->bio);
1287 bio_setup_sector(clone, ci->sector, *len);
1292 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1293 unsigned num_bios, unsigned *len)
1295 unsigned target_bio_nr;
1297 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1298 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1301 static int __send_empty_flush(struct clone_info *ci)
1303 unsigned target_nr = 0;
1304 struct dm_target *ti;
1306 BUG_ON(bio_has_data(ci->bio));
1307 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1308 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1313 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1314 sector_t sector, unsigned *len)
1316 struct bio *bio = ci->bio;
1317 struct dm_target_io *tio;
1318 unsigned target_bio_nr;
1319 unsigned num_target_bios = 1;
1323 * Does the target want to receive duplicate copies of the bio?
1325 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1326 num_target_bios = ti->num_write_bios(ti, bio);
1328 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1329 tio = alloc_tio(ci, ti, target_bio_nr);
1331 r = clone_bio(tio, bio, sector, *len);
1342 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1344 static unsigned get_num_discard_bios(struct dm_target *ti)
1346 return ti->num_discard_bios;
1349 static unsigned get_num_write_same_bios(struct dm_target *ti)
1351 return ti->num_write_same_bios;
1354 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1356 return ti->num_write_zeroes_bios;
1359 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1361 static bool is_split_required_for_discard(struct dm_target *ti)
1363 return ti->split_discard_bios;
1366 static int __send_changing_extent_only(struct clone_info *ci,
1367 get_num_bios_fn get_num_bios,
1368 is_split_required_fn is_split_required)
1370 struct dm_target *ti;
1375 ti = dm_table_find_target(ci->map, ci->sector);
1376 if (!dm_target_is_valid(ti))
1380 * Even though the device advertised support for this type of
1381 * request, that does not mean every target supports it, and
1382 * reconfiguration might also have changed that since the
1383 * check was performed.
1385 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1389 if (is_split_required && !is_split_required(ti))
1390 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1392 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1394 __send_duplicate_bios(ci, ti, num_bios, &len);
1397 } while (ci->sector_count -= len);
1402 static int __send_discard(struct clone_info *ci)
1404 return __send_changing_extent_only(ci, get_num_discard_bios,
1405 is_split_required_for_discard);
1408 static int __send_write_same(struct clone_info *ci)
1410 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1413 static int __send_write_zeroes(struct clone_info *ci)
1415 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1419 * Select the correct strategy for processing a non-flush bio.
1421 static int __split_and_process_non_flush(struct clone_info *ci)
1423 struct bio *bio = ci->bio;
1424 struct dm_target *ti;
1428 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1429 return __send_discard(ci);
1430 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1431 return __send_write_same(ci);
1432 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1433 return __send_write_zeroes(ci);
1435 ti = dm_table_find_target(ci->map, ci->sector);
1436 if (!dm_target_is_valid(ti))
1439 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1440 len = ci->sector_count;
1442 len = min_t(sector_t, max_io_len(ci->sector, ti),
1445 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1450 ci->sector_count -= len;
1456 * Entry point to split a bio into clones and submit them to the targets.
1458 static void __split_and_process_bio(struct mapped_device *md,
1459 struct dm_table *map, struct bio *bio)
1461 struct clone_info ci;
1464 if (unlikely(!map)) {
1471 ci.io = alloc_io(md);
1473 atomic_set(&ci.io->io_count, 1);
1476 spin_lock_init(&ci.io->endio_lock);
1477 ci.sector = bio->bi_iter.bi_sector;
1479 start_io_acct(ci.io);
1481 if (bio->bi_opf & REQ_PREFLUSH) {
1482 ci.bio = &ci.md->flush_bio;
1483 ci.sector_count = 0;
1484 error = __send_empty_flush(&ci);
1485 /* dec_pending submits any data associated with flush */
1486 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1488 ci.sector_count = 0;
1489 error = __split_and_process_non_flush(&ci);
1492 ci.sector_count = bio_sectors(bio);
1493 while (ci.sector_count && !error)
1494 error = __split_and_process_non_flush(&ci);
1497 /* drop the extra reference count */
1498 dec_pending(ci.io, errno_to_blk_status(error));
1500 /*-----------------------------------------------------------------
1502 *---------------------------------------------------------------*/
1505 * The request function that just remaps the bio built up by
1508 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1510 int rw = bio_data_dir(bio);
1511 struct mapped_device *md = q->queuedata;
1513 struct dm_table *map;
1515 map = dm_get_live_table(md, &srcu_idx);
1517 generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0);
1519 /* if we're suspended, we have to queue this io for later */
1520 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1521 dm_put_live_table(md, srcu_idx);
1523 if (!(bio->bi_opf & REQ_RAHEAD))
1527 return BLK_QC_T_NONE;
1530 __split_and_process_bio(md, map, bio);
1531 dm_put_live_table(md, srcu_idx);
1532 return BLK_QC_T_NONE;
1535 static int dm_any_congested(void *congested_data, int bdi_bits)
1538 struct mapped_device *md = congested_data;
1539 struct dm_table *map;
1541 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1542 if (dm_request_based(md)) {
1544 * With request-based DM we only need to check the
1545 * top-level queue for congestion.
1547 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1549 map = dm_get_live_table_fast(md);
1551 r = dm_table_any_congested(map, bdi_bits);
1552 dm_put_live_table_fast(md);
1559 /*-----------------------------------------------------------------
1560 * An IDR is used to keep track of allocated minor numbers.
1561 *---------------------------------------------------------------*/
1562 static void free_minor(int minor)
1564 spin_lock(&_minor_lock);
1565 idr_remove(&_minor_idr, minor);
1566 spin_unlock(&_minor_lock);
1570 * See if the device with a specific minor # is free.
1572 static int specific_minor(int minor)
1576 if (minor >= (1 << MINORBITS))
1579 idr_preload(GFP_KERNEL);
1580 spin_lock(&_minor_lock);
1582 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1584 spin_unlock(&_minor_lock);
1587 return r == -ENOSPC ? -EBUSY : r;
1591 static int next_free_minor(int *minor)
1595 idr_preload(GFP_KERNEL);
1596 spin_lock(&_minor_lock);
1598 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1600 spin_unlock(&_minor_lock);
1608 static const struct block_device_operations dm_blk_dops;
1609 static const struct dax_operations dm_dax_ops;
1611 static void dm_wq_work(struct work_struct *work);
1613 void dm_init_md_queue(struct mapped_device *md)
1616 * Request-based dm devices cannot be stacked on top of bio-based dm
1617 * devices. The type of this dm device may not have been decided yet.
1618 * The type is decided at the first table loading time.
1619 * To prevent problematic device stacking, clear the queue flag
1620 * for request stacking support until then.
1622 * This queue is new, so no concurrency on the queue_flags.
1624 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1627 * Initialize data that will only be used by a non-blk-mq DM queue
1628 * - must do so here (in alloc_dev callchain) before queue is used
1630 md->queue->queuedata = md;
1631 md->queue->backing_dev_info->congested_data = md;
1634 void dm_init_normal_md_queue(struct mapped_device *md)
1636 md->use_blk_mq = false;
1637 dm_init_md_queue(md);
1640 * Initialize aspects of queue that aren't relevant for blk-mq
1642 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1645 static void cleanup_mapped_device(struct mapped_device *md)
1648 destroy_workqueue(md->wq);
1649 if (md->kworker_task)
1650 kthread_stop(md->kworker_task);
1651 mempool_destroy(md->io_pool);
1653 bioset_free(md->bs);
1656 kill_dax(md->dax_dev);
1657 put_dax(md->dax_dev);
1662 spin_lock(&_minor_lock);
1663 md->disk->private_data = NULL;
1664 spin_unlock(&_minor_lock);
1665 del_gendisk(md->disk);
1670 blk_cleanup_queue(md->queue);
1672 cleanup_srcu_struct(&md->io_barrier);
1679 dm_mq_cleanup_mapped_device(md);
1683 * Allocate and initialise a blank device with a given minor.
1685 static struct mapped_device *alloc_dev(int minor)
1687 int r, numa_node_id = dm_get_numa_node();
1688 struct dax_device *dax_dev;
1689 struct mapped_device *md;
1692 md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1694 DMWARN("unable to allocate device, out of memory.");
1698 if (!try_module_get(THIS_MODULE))
1699 goto bad_module_get;
1701 /* get a minor number for the dev */
1702 if (minor == DM_ANY_MINOR)
1703 r = next_free_minor(&minor);
1705 r = specific_minor(minor);
1709 r = init_srcu_struct(&md->io_barrier);
1711 goto bad_io_barrier;
1713 md->numa_node_id = numa_node_id;
1714 md->use_blk_mq = dm_use_blk_mq_default();
1715 md->init_tio_pdu = false;
1716 md->type = DM_TYPE_NONE;
1717 mutex_init(&md->suspend_lock);
1718 mutex_init(&md->type_lock);
1719 mutex_init(&md->table_devices_lock);
1720 spin_lock_init(&md->deferred_lock);
1721 atomic_set(&md->holders, 1);
1722 atomic_set(&md->open_count, 0);
1723 atomic_set(&md->event_nr, 0);
1724 atomic_set(&md->uevent_seq, 0);
1725 INIT_LIST_HEAD(&md->uevent_list);
1726 INIT_LIST_HEAD(&md->table_devices);
1727 spin_lock_init(&md->uevent_lock);
1729 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1733 dm_init_md_queue(md);
1735 md->disk = alloc_disk_node(1, numa_node_id);
1739 atomic_set(&md->pending[0], 0);
1740 atomic_set(&md->pending[1], 0);
1741 init_waitqueue_head(&md->wait);
1742 INIT_WORK(&md->work, dm_wq_work);
1743 init_waitqueue_head(&md->eventq);
1744 init_completion(&md->kobj_holder.completion);
1745 md->kworker_task = NULL;
1747 md->disk->major = _major;
1748 md->disk->first_minor = minor;
1749 md->disk->fops = &dm_blk_dops;
1750 md->disk->queue = md->queue;
1751 md->disk->private_data = md;
1752 sprintf(md->disk->disk_name, "dm-%d", minor);
1754 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1757 md->dax_dev = dax_dev;
1760 format_dev_t(md->name, MKDEV(_major, minor));
1762 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1766 md->bdev = bdget_disk(md->disk, 0);
1770 bio_init(&md->flush_bio, NULL, 0);
1771 bio_set_dev(&md->flush_bio, md->bdev);
1772 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1774 dm_stats_init(&md->stats);
1776 /* Populate the mapping, nobody knows we exist yet */
1777 spin_lock(&_minor_lock);
1778 old_md = idr_replace(&_minor_idr, md, minor);
1779 spin_unlock(&_minor_lock);
1781 BUG_ON(old_md != MINOR_ALLOCED);
1786 cleanup_mapped_device(md);
1790 module_put(THIS_MODULE);
1796 static void unlock_fs(struct mapped_device *md);
1798 static void free_dev(struct mapped_device *md)
1800 int minor = MINOR(disk_devt(md->disk));
1804 cleanup_mapped_device(md);
1806 free_table_devices(&md->table_devices);
1807 dm_stats_cleanup(&md->stats);
1810 module_put(THIS_MODULE);
1814 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1816 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1819 /* The md already has necessary mempools. */
1820 if (dm_table_bio_based(t)) {
1822 * Reload bioset because front_pad may have changed
1823 * because a different table was loaded.
1825 bioset_free(md->bs);
1830 * There's no need to reload with request-based dm
1831 * because the size of front_pad doesn't change.
1832 * Note for future: If you are to reload bioset,
1833 * prep-ed requests in the queue may refer
1834 * to bio from the old bioset, so you must walk
1835 * through the queue to unprep.
1840 BUG_ON(!p || md->io_pool || md->bs);
1842 md->io_pool = p->io_pool;
1848 /* mempool bind completed, no longer need any mempools in the table */
1849 dm_table_free_md_mempools(t);
1853 * Bind a table to the device.
1855 static void event_callback(void *context)
1857 unsigned long flags;
1859 struct mapped_device *md = (struct mapped_device *) context;
1861 spin_lock_irqsave(&md->uevent_lock, flags);
1862 list_splice_init(&md->uevent_list, &uevents);
1863 spin_unlock_irqrestore(&md->uevent_lock, flags);
1865 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1867 atomic_inc(&md->event_nr);
1868 atomic_inc(&dm_global_event_nr);
1869 wake_up(&md->eventq);
1870 wake_up(&dm_global_eventq);
1874 * Protected by md->suspend_lock obtained by dm_swap_table().
1876 static void __set_size(struct mapped_device *md, sector_t size)
1878 lockdep_assert_held(&md->suspend_lock);
1880 set_capacity(md->disk, size);
1882 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1886 * Returns old map, which caller must destroy.
1888 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1889 struct queue_limits *limits)
1891 struct dm_table *old_map;
1892 struct request_queue *q = md->queue;
1895 lockdep_assert_held(&md->suspend_lock);
1897 size = dm_table_get_size(t);
1900 * Wipe any geometry if the size of the table changed.
1902 if (size != dm_get_size(md))
1903 memset(&md->geometry, 0, sizeof(md->geometry));
1905 __set_size(md, size);
1907 dm_table_event_callback(t, event_callback, md);
1910 * The queue hasn't been stopped yet, if the old table type wasn't
1911 * for request-based during suspension. So stop it to prevent
1912 * I/O mapping before resume.
1913 * This must be done before setting the queue restrictions,
1914 * because request-based dm may be run just after the setting.
1916 if (dm_table_request_based(t)) {
1919 * Leverage the fact that request-based DM targets are
1920 * immutable singletons and establish md->immutable_target
1921 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1923 md->immutable_target = dm_table_get_immutable_target(t);
1926 __bind_mempools(md, t);
1928 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1929 rcu_assign_pointer(md->map, (void *)t);
1930 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1932 dm_table_set_restrictions(t, q, limits);
1940 * Returns unbound table for the caller to free.
1942 static struct dm_table *__unbind(struct mapped_device *md)
1944 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1949 dm_table_event_callback(map, NULL, NULL);
1950 RCU_INIT_POINTER(md->map, NULL);
1957 * Constructor for a new device.
1959 int dm_create(int minor, struct mapped_device **result)
1961 struct mapped_device *md;
1963 md = alloc_dev(minor);
1974 * Functions to manage md->type.
1975 * All are required to hold md->type_lock.
1977 void dm_lock_md_type(struct mapped_device *md)
1979 mutex_lock(&md->type_lock);
1982 void dm_unlock_md_type(struct mapped_device *md)
1984 mutex_unlock(&md->type_lock);
1987 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
1989 BUG_ON(!mutex_is_locked(&md->type_lock));
1993 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
1998 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2000 return md->immutable_target_type;
2004 * The queue_limits are only valid as long as you have a reference
2007 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2009 BUG_ON(!atomic_read(&md->holders));
2010 return &md->queue->limits;
2012 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2015 * Setup the DM device's queue based on md's type
2017 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2020 enum dm_queue_mode type = dm_get_md_type(md);
2023 case DM_TYPE_REQUEST_BASED:
2024 r = dm_old_init_request_queue(md, t);
2026 DMERR("Cannot initialize queue for request-based mapped device");
2030 case DM_TYPE_MQ_REQUEST_BASED:
2031 r = dm_mq_init_request_queue(md, t);
2033 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2037 case DM_TYPE_BIO_BASED:
2038 case DM_TYPE_DAX_BIO_BASED:
2039 dm_init_normal_md_queue(md);
2040 blk_queue_make_request(md->queue, dm_make_request);
2042 * DM handles splitting bios as needed. Free the bio_split bioset
2043 * since it won't be used (saves 1 process per bio-based DM device).
2045 bioset_free(md->queue->bio_split);
2046 md->queue->bio_split = NULL;
2048 if (type == DM_TYPE_DAX_BIO_BASED)
2049 queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue);
2059 struct mapped_device *dm_get_md(dev_t dev)
2061 struct mapped_device *md;
2062 unsigned minor = MINOR(dev);
2064 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2067 spin_lock(&_minor_lock);
2069 md = idr_find(&_minor_idr, minor);
2071 if ((md == MINOR_ALLOCED ||
2072 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2073 dm_deleting_md(md) ||
2074 test_bit(DMF_FREEING, &md->flags))) {
2082 spin_unlock(&_minor_lock);
2086 EXPORT_SYMBOL_GPL(dm_get_md);
2088 void *dm_get_mdptr(struct mapped_device *md)
2090 return md->interface_ptr;
2093 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2095 md->interface_ptr = ptr;
2098 void dm_get(struct mapped_device *md)
2100 atomic_inc(&md->holders);
2101 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2104 int dm_hold(struct mapped_device *md)
2106 spin_lock(&_minor_lock);
2107 if (test_bit(DMF_FREEING, &md->flags)) {
2108 spin_unlock(&_minor_lock);
2112 spin_unlock(&_minor_lock);
2115 EXPORT_SYMBOL_GPL(dm_hold);
2117 const char *dm_device_name(struct mapped_device *md)
2121 EXPORT_SYMBOL_GPL(dm_device_name);
2123 static void __dm_destroy(struct mapped_device *md, bool wait)
2125 struct request_queue *q = dm_get_md_queue(md);
2126 struct dm_table *map;
2131 spin_lock(&_minor_lock);
2132 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2133 set_bit(DMF_FREEING, &md->flags);
2134 spin_unlock(&_minor_lock);
2136 blk_set_queue_dying(q);
2138 if (dm_request_based(md) && md->kworker_task)
2139 kthread_flush_worker(&md->kworker);
2142 * Take suspend_lock so that presuspend and postsuspend methods
2143 * do not race with internal suspend.
2145 mutex_lock(&md->suspend_lock);
2146 map = dm_get_live_table(md, &srcu_idx);
2147 if (!dm_suspended_md(md)) {
2148 dm_table_presuspend_targets(map);
2149 dm_table_postsuspend_targets(map);
2151 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2152 dm_put_live_table(md, srcu_idx);
2153 mutex_unlock(&md->suspend_lock);
2156 * Rare, but there may be I/O requests still going to complete,
2157 * for example. Wait for all references to disappear.
2158 * No one should increment the reference count of the mapped_device,
2159 * after the mapped_device state becomes DMF_FREEING.
2162 while (atomic_read(&md->holders))
2164 else if (atomic_read(&md->holders))
2165 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2166 dm_device_name(md), atomic_read(&md->holders));
2169 dm_table_destroy(__unbind(md));
2173 void dm_destroy(struct mapped_device *md)
2175 __dm_destroy(md, true);
2178 void dm_destroy_immediate(struct mapped_device *md)
2180 __dm_destroy(md, false);
2183 void dm_put(struct mapped_device *md)
2185 atomic_dec(&md->holders);
2187 EXPORT_SYMBOL_GPL(dm_put);
2189 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2195 prepare_to_wait(&md->wait, &wait, task_state);
2197 if (!md_in_flight(md))
2200 if (signal_pending_state(task_state, current)) {
2207 finish_wait(&md->wait, &wait);
2213 * Process the deferred bios
2215 static void dm_wq_work(struct work_struct *work)
2217 struct mapped_device *md = container_of(work, struct mapped_device,
2221 struct dm_table *map;
2223 map = dm_get_live_table(md, &srcu_idx);
2225 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2226 spin_lock_irq(&md->deferred_lock);
2227 c = bio_list_pop(&md->deferred);
2228 spin_unlock_irq(&md->deferred_lock);
2233 if (dm_request_based(md))
2234 generic_make_request(c);
2236 __split_and_process_bio(md, map, c);
2239 dm_put_live_table(md, srcu_idx);
2242 static void dm_queue_flush(struct mapped_device *md)
2244 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2245 smp_mb__after_atomic();
2246 queue_work(md->wq, &md->work);
2250 * Swap in a new table, returning the old one for the caller to destroy.
2252 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2254 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2255 struct queue_limits limits;
2258 mutex_lock(&md->suspend_lock);
2260 /* device must be suspended */
2261 if (!dm_suspended_md(md))
2265 * If the new table has no data devices, retain the existing limits.
2266 * This helps multipath with queue_if_no_path if all paths disappear,
2267 * then new I/O is queued based on these limits, and then some paths
2270 if (dm_table_has_no_data_devices(table)) {
2271 live_map = dm_get_live_table_fast(md);
2273 limits = md->queue->limits;
2274 dm_put_live_table_fast(md);
2278 r = dm_calculate_queue_limits(table, &limits);
2285 map = __bind(md, table, &limits);
2288 mutex_unlock(&md->suspend_lock);
2293 * Functions to lock and unlock any filesystem running on the
2296 static int lock_fs(struct mapped_device *md)
2300 WARN_ON(md->frozen_sb);
2302 md->frozen_sb = freeze_bdev(md->bdev);
2303 if (IS_ERR(md->frozen_sb)) {
2304 r = PTR_ERR(md->frozen_sb);
2305 md->frozen_sb = NULL;
2309 set_bit(DMF_FROZEN, &md->flags);
2314 static void unlock_fs(struct mapped_device *md)
2316 if (!test_bit(DMF_FROZEN, &md->flags))
2319 thaw_bdev(md->bdev, md->frozen_sb);
2320 md->frozen_sb = NULL;
2321 clear_bit(DMF_FROZEN, &md->flags);
2325 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2326 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2327 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2329 * If __dm_suspend returns 0, the device is completely quiescent
2330 * now. There is no request-processing activity. All new requests
2331 * are being added to md->deferred list.
2333 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2334 unsigned suspend_flags, long task_state,
2335 int dmf_suspended_flag)
2337 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2338 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2341 lockdep_assert_held(&md->suspend_lock);
2344 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2345 * This flag is cleared before dm_suspend returns.
2348 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2350 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2353 * This gets reverted if there's an error later and the targets
2354 * provide the .presuspend_undo hook.
2356 dm_table_presuspend_targets(map);
2359 * Flush I/O to the device.
2360 * Any I/O submitted after lock_fs() may not be flushed.
2361 * noflush takes precedence over do_lockfs.
2362 * (lock_fs() flushes I/Os and waits for them to complete.)
2364 if (!noflush && do_lockfs) {
2367 dm_table_presuspend_undo_targets(map);
2373 * Here we must make sure that no processes are submitting requests
2374 * to target drivers i.e. no one may be executing
2375 * __split_and_process_bio. This is called from dm_request and
2378 * To get all processes out of __split_and_process_bio in dm_request,
2379 * we take the write lock. To prevent any process from reentering
2380 * __split_and_process_bio from dm_request and quiesce the thread
2381 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2382 * flush_workqueue(md->wq).
2384 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2386 synchronize_srcu(&md->io_barrier);
2389 * Stop md->queue before flushing md->wq in case request-based
2390 * dm defers requests to md->wq from md->queue.
2392 if (dm_request_based(md)) {
2393 dm_stop_queue(md->queue);
2394 if (md->kworker_task)
2395 kthread_flush_worker(&md->kworker);
2398 flush_workqueue(md->wq);
2401 * At this point no more requests are entering target request routines.
2402 * We call dm_wait_for_completion to wait for all existing requests
2405 r = dm_wait_for_completion(md, task_state);
2407 set_bit(dmf_suspended_flag, &md->flags);
2410 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2412 synchronize_srcu(&md->io_barrier);
2414 /* were we interrupted ? */
2418 if (dm_request_based(md))
2419 dm_start_queue(md->queue);
2422 dm_table_presuspend_undo_targets(map);
2423 /* pushback list is already flushed, so skip flush */
2430 * We need to be able to change a mapping table under a mounted
2431 * filesystem. For example we might want to move some data in
2432 * the background. Before the table can be swapped with
2433 * dm_bind_table, dm_suspend must be called to flush any in
2434 * flight bios and ensure that any further io gets deferred.
2437 * Suspend mechanism in request-based dm.
2439 * 1. Flush all I/Os by lock_fs() if needed.
2440 * 2. Stop dispatching any I/O by stopping the request_queue.
2441 * 3. Wait for all in-flight I/Os to be completed or requeued.
2443 * To abort suspend, start the request_queue.
2445 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2447 struct dm_table *map = NULL;
2451 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2453 if (dm_suspended_md(md)) {
2458 if (dm_suspended_internally_md(md)) {
2459 /* already internally suspended, wait for internal resume */
2460 mutex_unlock(&md->suspend_lock);
2461 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2467 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2469 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2473 dm_table_postsuspend_targets(map);
2476 mutex_unlock(&md->suspend_lock);
2480 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2483 int r = dm_table_resume_targets(map);
2491 * Flushing deferred I/Os must be done after targets are resumed
2492 * so that mapping of targets can work correctly.
2493 * Request-based dm is queueing the deferred I/Os in its request_queue.
2495 if (dm_request_based(md))
2496 dm_start_queue(md->queue);
2503 int dm_resume(struct mapped_device *md)
2506 struct dm_table *map = NULL;
2510 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2512 if (!dm_suspended_md(md))
2515 if (dm_suspended_internally_md(md)) {
2516 /* already internally suspended, wait for internal resume */
2517 mutex_unlock(&md->suspend_lock);
2518 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2524 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2525 if (!map || !dm_table_get_size(map))
2528 r = __dm_resume(md, map);
2532 clear_bit(DMF_SUSPENDED, &md->flags);
2534 mutex_unlock(&md->suspend_lock);
2540 * Internal suspend/resume works like userspace-driven suspend. It waits
2541 * until all bios finish and prevents issuing new bios to the target drivers.
2542 * It may be used only from the kernel.
2545 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2547 struct dm_table *map = NULL;
2549 lockdep_assert_held(&md->suspend_lock);
2551 if (md->internal_suspend_count++)
2552 return; /* nested internal suspend */
2554 if (dm_suspended_md(md)) {
2555 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2556 return; /* nest suspend */
2559 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2562 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2563 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2564 * would require changing .presuspend to return an error -- avoid this
2565 * until there is a need for more elaborate variants of internal suspend.
2567 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2568 DMF_SUSPENDED_INTERNALLY);
2570 dm_table_postsuspend_targets(map);
2573 static void __dm_internal_resume(struct mapped_device *md)
2575 BUG_ON(!md->internal_suspend_count);
2577 if (--md->internal_suspend_count)
2578 return; /* resume from nested internal suspend */
2580 if (dm_suspended_md(md))
2581 goto done; /* resume from nested suspend */
2584 * NOTE: existing callers don't need to call dm_table_resume_targets
2585 * (which may fail -- so best to avoid it for now by passing NULL map)
2587 (void) __dm_resume(md, NULL);
2590 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2591 smp_mb__after_atomic();
2592 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2595 void dm_internal_suspend_noflush(struct mapped_device *md)
2597 mutex_lock(&md->suspend_lock);
2598 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2599 mutex_unlock(&md->suspend_lock);
2601 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2603 void dm_internal_resume(struct mapped_device *md)
2605 mutex_lock(&md->suspend_lock);
2606 __dm_internal_resume(md);
2607 mutex_unlock(&md->suspend_lock);
2609 EXPORT_SYMBOL_GPL(dm_internal_resume);
2612 * Fast variants of internal suspend/resume hold md->suspend_lock,
2613 * which prevents interaction with userspace-driven suspend.
2616 void dm_internal_suspend_fast(struct mapped_device *md)
2618 mutex_lock(&md->suspend_lock);
2619 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2622 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2623 synchronize_srcu(&md->io_barrier);
2624 flush_workqueue(md->wq);
2625 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2627 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2629 void dm_internal_resume_fast(struct mapped_device *md)
2631 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2637 mutex_unlock(&md->suspend_lock);
2639 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2641 /*-----------------------------------------------------------------
2642 * Event notification.
2643 *---------------------------------------------------------------*/
2644 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2647 char udev_cookie[DM_COOKIE_LENGTH];
2648 char *envp[] = { udev_cookie, NULL };
2651 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2653 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2654 DM_COOKIE_ENV_VAR_NAME, cookie);
2655 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2660 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2662 return atomic_add_return(1, &md->uevent_seq);
2665 uint32_t dm_get_event_nr(struct mapped_device *md)
2667 return atomic_read(&md->event_nr);
2670 int dm_wait_event(struct mapped_device *md, int event_nr)
2672 return wait_event_interruptible(md->eventq,
2673 (event_nr != atomic_read(&md->event_nr)));
2676 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2678 unsigned long flags;
2680 spin_lock_irqsave(&md->uevent_lock, flags);
2681 list_add(elist, &md->uevent_list);
2682 spin_unlock_irqrestore(&md->uevent_lock, flags);
2686 * The gendisk is only valid as long as you have a reference
2689 struct gendisk *dm_disk(struct mapped_device *md)
2693 EXPORT_SYMBOL_GPL(dm_disk);
2695 struct kobject *dm_kobject(struct mapped_device *md)
2697 return &md->kobj_holder.kobj;
2700 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2702 struct mapped_device *md;
2704 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2706 if (test_bit(DMF_FREEING, &md->flags) ||
2714 int dm_suspended_md(struct mapped_device *md)
2716 return test_bit(DMF_SUSPENDED, &md->flags);
2719 int dm_suspended_internally_md(struct mapped_device *md)
2721 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2724 int dm_test_deferred_remove_flag(struct mapped_device *md)
2726 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2729 int dm_suspended(struct dm_target *ti)
2731 return dm_suspended_md(dm_table_get_md(ti->table));
2733 EXPORT_SYMBOL_GPL(dm_suspended);
2735 int dm_noflush_suspending(struct dm_target *ti)
2737 return __noflush_suspending(dm_table_get_md(ti->table));
2739 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2741 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2742 unsigned integrity, unsigned per_io_data_size)
2744 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2745 unsigned int pool_size = 0;
2746 unsigned int front_pad;
2752 case DM_TYPE_BIO_BASED:
2753 case DM_TYPE_DAX_BIO_BASED:
2754 pool_size = dm_get_reserved_bio_based_ios();
2755 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2757 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2758 if (!pools->io_pool)
2761 case DM_TYPE_REQUEST_BASED:
2762 case DM_TYPE_MQ_REQUEST_BASED:
2763 pool_size = dm_get_reserved_rq_based_ios();
2764 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2765 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2771 pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER);
2775 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2781 dm_free_md_mempools(pools);
2786 void dm_free_md_mempools(struct dm_md_mempools *pools)
2791 mempool_destroy(pools->io_pool);
2794 bioset_free(pools->bs);
2806 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2809 struct mapped_device *md = bdev->bd_disk->private_data;
2810 struct dm_table *table;
2811 struct dm_target *ti;
2812 int ret = -ENOTTY, srcu_idx;
2814 table = dm_get_live_table(md, &srcu_idx);
2815 if (!table || !dm_table_get_size(table))
2818 /* We only support devices that have a single target */
2819 if (dm_table_get_num_targets(table) != 1)
2821 ti = dm_table_get_target(table, 0);
2824 if (!ti->type->iterate_devices)
2827 ret = ti->type->iterate_devices(ti, fn, data);
2829 dm_put_live_table(md, srcu_idx);
2834 * For register / unregister we need to manually call out to every path.
2836 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2837 sector_t start, sector_t len, void *data)
2839 struct dm_pr *pr = data;
2840 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2842 if (!ops || !ops->pr_register)
2844 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2847 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2858 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2859 if (ret && new_key) {
2860 /* unregister all paths if we failed to register any path */
2861 pr.old_key = new_key;
2864 pr.fail_early = false;
2865 dm_call_pr(bdev, __dm_pr_register, &pr);
2871 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2874 struct mapped_device *md = bdev->bd_disk->private_data;
2875 const struct pr_ops *ops;
2879 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2883 ops = bdev->bd_disk->fops->pr_ops;
2884 if (ops && ops->pr_reserve)
2885 r = ops->pr_reserve(bdev, key, type, flags);
2893 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2895 struct mapped_device *md = bdev->bd_disk->private_data;
2896 const struct pr_ops *ops;
2900 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2904 ops = bdev->bd_disk->fops->pr_ops;
2905 if (ops && ops->pr_release)
2906 r = ops->pr_release(bdev, key, type);
2914 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2915 enum pr_type type, bool abort)
2917 struct mapped_device *md = bdev->bd_disk->private_data;
2918 const struct pr_ops *ops;
2922 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2926 ops = bdev->bd_disk->fops->pr_ops;
2927 if (ops && ops->pr_preempt)
2928 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2936 static int dm_pr_clear(struct block_device *bdev, u64 key)
2938 struct mapped_device *md = bdev->bd_disk->private_data;
2939 const struct pr_ops *ops;
2943 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2947 ops = bdev->bd_disk->fops->pr_ops;
2948 if (ops && ops->pr_clear)
2949 r = ops->pr_clear(bdev, key);
2957 static const struct pr_ops dm_pr_ops = {
2958 .pr_register = dm_pr_register,
2959 .pr_reserve = dm_pr_reserve,
2960 .pr_release = dm_pr_release,
2961 .pr_preempt = dm_pr_preempt,
2962 .pr_clear = dm_pr_clear,
2965 static const struct block_device_operations dm_blk_dops = {
2966 .open = dm_blk_open,
2967 .release = dm_blk_close,
2968 .ioctl = dm_blk_ioctl,
2969 .getgeo = dm_blk_getgeo,
2970 .pr_ops = &dm_pr_ops,
2971 .owner = THIS_MODULE
2974 static const struct dax_operations dm_dax_ops = {
2975 .direct_access = dm_dax_direct_access,
2976 .copy_from_iter = dm_dax_copy_from_iter,
2982 module_init(dm_init);
2983 module_exit(dm_exit);
2985 module_param(major, uint, 0);
2986 MODULE_PARM_DESC(major, "The major number of the device mapper");
2988 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2989 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2991 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
2992 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
2994 MODULE_DESCRIPTION(DM_NAME " driver");
2996 MODULE_LICENSE("GPL");