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/mm.h>
16 #include <linux/sched/signal.h>
17 #include <linux/blkpg.h>
18 #include <linux/bio.h>
19 #include <linux/mempool.h>
20 #include <linux/dax.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/uio.h>
24 #include <linux/hdreg.h>
25 #include <linux/delay.h>
26 #include <linux/wait.h>
29 #define DM_MSG_PREFIX "core"
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name = DM_NAME;
40 static unsigned int major = 0;
41 static unsigned int _major = 0;
43 static DEFINE_IDR(_minor_idr);
45 static DEFINE_SPINLOCK(_minor_lock);
47 static void do_deferred_remove(struct work_struct *w);
49 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
51 static struct workqueue_struct *deferred_remove_workqueue;
53 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
56 void dm_issue_global_event(void)
58 atomic_inc(&dm_global_event_nr);
59 wake_up(&dm_global_eventq);
63 * One of these is allocated per bio.
66 struct mapped_device *md;
70 unsigned long start_time;
71 spinlock_t endio_lock;
72 struct dm_stats_aux stats_aux;
75 #define MINOR_ALLOCED ((void *)-1)
78 * Bits for the md->flags field.
80 #define DMF_BLOCK_IO_FOR_SUSPEND 0
81 #define DMF_SUSPENDED 1
84 #define DMF_DELETING 4
85 #define DMF_NOFLUSH_SUSPENDING 5
86 #define DMF_DEFERRED_REMOVE 6
87 #define DMF_SUSPENDED_INTERNALLY 7
89 #define DM_NUMA_NODE NUMA_NO_NODE
90 static int dm_numa_node = DM_NUMA_NODE;
93 * For mempools pre-allocation at the table loading time.
95 struct dm_md_mempools {
100 struct table_device {
101 struct list_head list;
103 struct dm_dev dm_dev;
106 static struct kmem_cache *_io_cache;
107 static struct kmem_cache *_rq_tio_cache;
108 static struct kmem_cache *_rq_cache;
111 * Bio-based DM's mempools' reserved IOs set by the user.
113 #define RESERVED_BIO_BASED_IOS 16
114 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
116 static int __dm_get_module_param_int(int *module_param, int min, int max)
118 int param = ACCESS_ONCE(*module_param);
119 int modified_param = 0;
120 bool modified = true;
123 modified_param = min;
124 else if (param > max)
125 modified_param = max;
130 (void)cmpxchg(module_param, param, modified_param);
131 param = modified_param;
137 unsigned __dm_get_module_param(unsigned *module_param,
138 unsigned def, unsigned max)
140 unsigned param = ACCESS_ONCE(*module_param);
141 unsigned modified_param = 0;
144 modified_param = def;
145 else if (param > max)
146 modified_param = max;
148 if (modified_param) {
149 (void)cmpxchg(module_param, param, modified_param);
150 param = modified_param;
156 unsigned dm_get_reserved_bio_based_ios(void)
158 return __dm_get_module_param(&reserved_bio_based_ios,
159 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
161 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
163 static unsigned dm_get_numa_node(void)
165 return __dm_get_module_param_int(&dm_numa_node,
166 DM_NUMA_NODE, num_online_nodes() - 1);
169 static int __init local_init(void)
173 /* allocate a slab for the dm_ios */
174 _io_cache = KMEM_CACHE(dm_io, 0);
178 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
180 goto out_free_io_cache;
182 _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request),
183 __alignof__(struct request), 0, NULL);
185 goto out_free_rq_tio_cache;
187 r = dm_uevent_init();
189 goto out_free_rq_cache;
191 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
192 if (!deferred_remove_workqueue) {
194 goto out_uevent_exit;
198 r = register_blkdev(_major, _name);
200 goto out_free_workqueue;
208 destroy_workqueue(deferred_remove_workqueue);
212 kmem_cache_destroy(_rq_cache);
213 out_free_rq_tio_cache:
214 kmem_cache_destroy(_rq_tio_cache);
216 kmem_cache_destroy(_io_cache);
221 static void local_exit(void)
223 flush_scheduled_work();
224 destroy_workqueue(deferred_remove_workqueue);
226 kmem_cache_destroy(_rq_cache);
227 kmem_cache_destroy(_rq_tio_cache);
228 kmem_cache_destroy(_io_cache);
229 unregister_blkdev(_major, _name);
234 DMINFO("cleaned up");
237 static int (*_inits[])(void) __initdata = {
248 static void (*_exits[])(void) = {
259 static int __init dm_init(void)
261 const int count = ARRAY_SIZE(_inits);
265 for (i = 0; i < count; i++) {
280 static void __exit dm_exit(void)
282 int i = ARRAY_SIZE(_exits);
288 * Should be empty by this point.
290 idr_destroy(&_minor_idr);
294 * Block device functions
296 int dm_deleting_md(struct mapped_device *md)
298 return test_bit(DMF_DELETING, &md->flags);
301 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
303 struct mapped_device *md;
305 spin_lock(&_minor_lock);
307 md = bdev->bd_disk->private_data;
311 if (test_bit(DMF_FREEING, &md->flags) ||
312 dm_deleting_md(md)) {
318 atomic_inc(&md->open_count);
320 spin_unlock(&_minor_lock);
322 return md ? 0 : -ENXIO;
325 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
327 struct mapped_device *md;
329 spin_lock(&_minor_lock);
331 md = disk->private_data;
335 if (atomic_dec_and_test(&md->open_count) &&
336 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
337 queue_work(deferred_remove_workqueue, &deferred_remove_work);
341 spin_unlock(&_minor_lock);
344 int dm_open_count(struct mapped_device *md)
346 return atomic_read(&md->open_count);
350 * Guarantees nothing is using the device before it's deleted.
352 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
356 spin_lock(&_minor_lock);
358 if (dm_open_count(md)) {
361 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
362 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
365 set_bit(DMF_DELETING, &md->flags);
367 spin_unlock(&_minor_lock);
372 int dm_cancel_deferred_remove(struct mapped_device *md)
376 spin_lock(&_minor_lock);
378 if (test_bit(DMF_DELETING, &md->flags))
381 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
383 spin_unlock(&_minor_lock);
388 static void do_deferred_remove(struct work_struct *w)
390 dm_deferred_remove();
393 sector_t dm_get_size(struct mapped_device *md)
395 return get_capacity(md->disk);
398 struct request_queue *dm_get_md_queue(struct mapped_device *md)
403 struct dm_stats *dm_get_stats(struct mapped_device *md)
408 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
410 struct mapped_device *md = bdev->bd_disk->private_data;
412 return dm_get_geometry(md, geo);
415 static int dm_grab_bdev_for_ioctl(struct mapped_device *md,
416 struct block_device **bdev,
419 struct dm_target *tgt;
420 struct dm_table *map;
425 map = dm_get_live_table(md, &srcu_idx);
426 if (!map || !dm_table_get_size(map))
429 /* We only support devices that have a single target */
430 if (dm_table_get_num_targets(map) != 1)
433 tgt = dm_table_get_target(map, 0);
434 if (!tgt->type->prepare_ioctl)
437 if (dm_suspended_md(md)) {
442 r = tgt->type->prepare_ioctl(tgt, bdev, mode);
447 dm_put_live_table(md, srcu_idx);
451 dm_put_live_table(md, srcu_idx);
452 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
459 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
460 unsigned int cmd, unsigned long arg)
462 struct mapped_device *md = bdev->bd_disk->private_data;
465 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
471 * Target determined this ioctl is being issued against a
472 * subset of the parent bdev; require extra privileges.
474 if (!capable(CAP_SYS_RAWIO)) {
476 "%s: sending ioctl %x to DM device without required privilege.",
483 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
489 static struct dm_io *alloc_io(struct mapped_device *md)
491 return mempool_alloc(md->io_pool, GFP_NOIO);
494 static void free_io(struct mapped_device *md, struct dm_io *io)
496 mempool_free(io, md->io_pool);
499 static void free_tio(struct dm_target_io *tio)
501 bio_put(&tio->clone);
504 int md_in_flight(struct mapped_device *md)
506 return atomic_read(&md->pending[READ]) +
507 atomic_read(&md->pending[WRITE]);
510 static void start_io_acct(struct dm_io *io)
512 struct mapped_device *md = io->md;
513 struct bio *bio = io->bio;
515 int rw = bio_data_dir(bio);
517 io->start_time = jiffies;
519 cpu = part_stat_lock();
520 part_round_stats(md->queue, cpu, &dm_disk(md)->part0);
522 atomic_set(&dm_disk(md)->part0.in_flight[rw],
523 atomic_inc_return(&md->pending[rw]));
525 if (unlikely(dm_stats_used(&md->stats)))
526 dm_stats_account_io(&md->stats, bio_data_dir(bio),
527 bio->bi_iter.bi_sector, bio_sectors(bio),
528 false, 0, &io->stats_aux);
531 static void end_io_acct(struct dm_io *io)
533 struct mapped_device *md = io->md;
534 struct bio *bio = io->bio;
535 unsigned long duration = jiffies - io->start_time;
537 int rw = bio_data_dir(bio);
539 generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time);
541 if (unlikely(dm_stats_used(&md->stats)))
542 dm_stats_account_io(&md->stats, bio_data_dir(bio),
543 bio->bi_iter.bi_sector, bio_sectors(bio),
544 true, duration, &io->stats_aux);
547 * After this is decremented the bio must not be touched if it is
550 pending = atomic_dec_return(&md->pending[rw]);
551 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
552 pending += atomic_read(&md->pending[rw^0x1]);
554 /* nudge anyone waiting on suspend queue */
560 * Add the bio to the list of deferred io.
562 static void queue_io(struct mapped_device *md, struct bio *bio)
566 spin_lock_irqsave(&md->deferred_lock, flags);
567 bio_list_add(&md->deferred, bio);
568 spin_unlock_irqrestore(&md->deferred_lock, flags);
569 queue_work(md->wq, &md->work);
573 * Everyone (including functions in this file), should use this
574 * function to access the md->map field, and make sure they call
575 * dm_put_live_table() when finished.
577 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
579 *srcu_idx = srcu_read_lock(&md->io_barrier);
581 return srcu_dereference(md->map, &md->io_barrier);
584 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
586 srcu_read_unlock(&md->io_barrier, srcu_idx);
589 void dm_sync_table(struct mapped_device *md)
591 synchronize_srcu(&md->io_barrier);
592 synchronize_rcu_expedited();
596 * A fast alternative to dm_get_live_table/dm_put_live_table.
597 * The caller must not block between these two functions.
599 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
602 return rcu_dereference(md->map);
605 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
611 * Open a table device so we can use it as a map destination.
613 static int open_table_device(struct table_device *td, dev_t dev,
614 struct mapped_device *md)
616 static char *_claim_ptr = "I belong to device-mapper";
617 struct block_device *bdev;
621 BUG_ON(td->dm_dev.bdev);
623 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
625 return PTR_ERR(bdev);
627 r = bd_link_disk_holder(bdev, dm_disk(md));
629 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
633 td->dm_dev.bdev = bdev;
634 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
639 * Close a table device that we've been using.
641 static void close_table_device(struct table_device *td, struct mapped_device *md)
643 if (!td->dm_dev.bdev)
646 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
647 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
648 put_dax(td->dm_dev.dax_dev);
649 td->dm_dev.bdev = NULL;
650 td->dm_dev.dax_dev = NULL;
653 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
655 struct table_device *td;
657 list_for_each_entry(td, l, list)
658 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
664 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
665 struct dm_dev **result) {
667 struct table_device *td;
669 mutex_lock(&md->table_devices_lock);
670 td = find_table_device(&md->table_devices, dev, mode);
672 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
674 mutex_unlock(&md->table_devices_lock);
678 td->dm_dev.mode = mode;
679 td->dm_dev.bdev = NULL;
681 if ((r = open_table_device(td, dev, md))) {
682 mutex_unlock(&md->table_devices_lock);
687 format_dev_t(td->dm_dev.name, dev);
689 atomic_set(&td->count, 0);
690 list_add(&td->list, &md->table_devices);
692 atomic_inc(&td->count);
693 mutex_unlock(&md->table_devices_lock);
695 *result = &td->dm_dev;
698 EXPORT_SYMBOL_GPL(dm_get_table_device);
700 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
702 struct table_device *td = container_of(d, struct table_device, dm_dev);
704 mutex_lock(&md->table_devices_lock);
705 if (atomic_dec_and_test(&td->count)) {
706 close_table_device(td, md);
710 mutex_unlock(&md->table_devices_lock);
712 EXPORT_SYMBOL(dm_put_table_device);
714 static void free_table_devices(struct list_head *devices)
716 struct list_head *tmp, *next;
718 list_for_each_safe(tmp, next, devices) {
719 struct table_device *td = list_entry(tmp, struct table_device, list);
721 DMWARN("dm_destroy: %s still exists with %d references",
722 td->dm_dev.name, atomic_read(&td->count));
728 * Get the geometry associated with a dm device
730 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
738 * Set the geometry of a device.
740 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
742 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
744 if (geo->start > sz) {
745 DMWARN("Start sector is beyond the geometry limits.");
754 /*-----------------------------------------------------------------
756 * A more elegant soln is in the works that uses the queue
757 * merge fn, unfortunately there are a couple of changes to
758 * the block layer that I want to make for this. So in the
759 * interests of getting something for people to use I give
760 * you this clearly demarcated crap.
761 *---------------------------------------------------------------*/
763 static int __noflush_suspending(struct mapped_device *md)
765 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
769 * Decrements the number of outstanding ios that a bio has been
770 * cloned into, completing the original io if necc.
772 static void dec_pending(struct dm_io *io, blk_status_t error)
775 blk_status_t io_error;
777 struct mapped_device *md = io->md;
779 /* Push-back supersedes any I/O errors */
780 if (unlikely(error)) {
781 spin_lock_irqsave(&io->endio_lock, flags);
782 if (!(io->status == BLK_STS_DM_REQUEUE &&
783 __noflush_suspending(md)))
785 spin_unlock_irqrestore(&io->endio_lock, flags);
788 if (atomic_dec_and_test(&io->io_count)) {
789 if (io->status == BLK_STS_DM_REQUEUE) {
791 * Target requested pushing back the I/O.
793 spin_lock_irqsave(&md->deferred_lock, flags);
794 if (__noflush_suspending(md))
795 bio_list_add_head(&md->deferred, io->bio);
797 /* noflush suspend was interrupted. */
798 io->status = BLK_STS_IOERR;
799 spin_unlock_irqrestore(&md->deferred_lock, flags);
802 io_error = io->status;
807 if (io_error == BLK_STS_DM_REQUEUE)
810 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
812 * Preflush done for flush with data, reissue
813 * without REQ_PREFLUSH.
815 bio->bi_opf &= ~REQ_PREFLUSH;
818 /* done with normal IO or empty flush */
820 bio->bi_status = io_error;
826 void disable_write_same(struct mapped_device *md)
828 struct queue_limits *limits = dm_get_queue_limits(md);
830 /* device doesn't really support WRITE SAME, disable it */
831 limits->max_write_same_sectors = 0;
834 void disable_write_zeroes(struct mapped_device *md)
836 struct queue_limits *limits = dm_get_queue_limits(md);
838 /* device doesn't really support WRITE ZEROES, disable it */
839 limits->max_write_zeroes_sectors = 0;
842 static void clone_endio(struct bio *bio)
844 blk_status_t error = bio->bi_status;
845 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
846 struct dm_io *io = tio->io;
847 struct mapped_device *md = tio->io->md;
848 dm_endio_fn endio = tio->ti->type->end_io;
850 if (unlikely(error == BLK_STS_TARGET)) {
851 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
852 !bio->bi_disk->queue->limits.max_write_same_sectors)
853 disable_write_same(md);
854 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
855 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
856 disable_write_zeroes(md);
860 int r = endio(tio->ti, bio, &error);
862 case DM_ENDIO_REQUEUE:
863 error = BLK_STS_DM_REQUEUE;
867 case DM_ENDIO_INCOMPLETE:
868 /* The target will handle the io */
871 DMWARN("unimplemented target endio return value: %d", r);
877 dec_pending(io, error);
881 * Return maximum size of I/O possible at the supplied sector up to the current
884 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
886 sector_t target_offset = dm_target_offset(ti, sector);
888 return ti->len - target_offset;
891 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
893 sector_t len = max_io_len_target_boundary(sector, ti);
894 sector_t offset, max_len;
897 * Does the target need to split even further?
899 if (ti->max_io_len) {
900 offset = dm_target_offset(ti, sector);
901 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
902 max_len = sector_div(offset, ti->max_io_len);
904 max_len = offset & (ti->max_io_len - 1);
905 max_len = ti->max_io_len - max_len;
914 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
916 if (len > UINT_MAX) {
917 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
918 (unsigned long long)len, UINT_MAX);
919 ti->error = "Maximum size of target IO is too large";
923 ti->max_io_len = (uint32_t) len;
927 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
929 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
930 sector_t sector, int *srcu_idx)
932 struct dm_table *map;
933 struct dm_target *ti;
935 map = dm_get_live_table(md, srcu_idx);
939 ti = dm_table_find_target(map, sector);
940 if (!dm_target_is_valid(ti))
946 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
947 long nr_pages, void **kaddr, pfn_t *pfn)
949 struct mapped_device *md = dax_get_private(dax_dev);
950 sector_t sector = pgoff * PAGE_SECTORS;
951 struct dm_target *ti;
952 long len, ret = -EIO;
955 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
959 if (!ti->type->direct_access)
961 len = max_io_len(sector, ti) / PAGE_SECTORS;
964 nr_pages = min(len, nr_pages);
965 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
968 dm_put_live_table(md, srcu_idx);
973 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
974 void *addr, size_t bytes, struct iov_iter *i)
976 struct mapped_device *md = dax_get_private(dax_dev);
977 sector_t sector = pgoff * PAGE_SECTORS;
978 struct dm_target *ti;
982 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
986 if (!ti->type->dax_copy_from_iter) {
987 ret = copy_from_iter(addr, bytes, i);
990 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
992 dm_put_live_table(md, srcu_idx);
998 * A target may call dm_accept_partial_bio only from the map routine. It is
999 * allowed for all bio types except REQ_PREFLUSH.
1001 * dm_accept_partial_bio informs the dm that the target only wants to process
1002 * additional n_sectors sectors of the bio and the rest of the data should be
1003 * sent in a next bio.
1005 * A diagram that explains the arithmetics:
1006 * +--------------------+---------------+-------+
1008 * +--------------------+---------------+-------+
1010 * <-------------- *tio->len_ptr --------------->
1011 * <------- bi_size ------->
1014 * Region 1 was already iterated over with bio_advance or similar function.
1015 * (it may be empty if the target doesn't use bio_advance)
1016 * Region 2 is the remaining bio size that the target wants to process.
1017 * (it may be empty if region 1 is non-empty, although there is no reason
1019 * The target requires that region 3 is to be sent in the next bio.
1021 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1022 * the partially processed part (the sum of regions 1+2) must be the same for all
1023 * copies of the bio.
1025 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1027 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1028 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1029 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1030 BUG_ON(bi_size > *tio->len_ptr);
1031 BUG_ON(n_sectors > bi_size);
1032 *tio->len_ptr -= bi_size - n_sectors;
1033 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1035 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1038 * The zone descriptors obtained with a zone report indicate zone positions
1039 * within the target backing device, regardless of that device is a partition
1040 * and regardless of the target mapping start sector on the device or partition.
1041 * The zone descriptors start sector and write pointer position must be adjusted
1042 * to match their relative position within the dm device.
1043 * A target may call dm_remap_zone_report() after completion of a
1044 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1047 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1049 #ifdef CONFIG_BLK_DEV_ZONED
1050 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1051 struct bio *report_bio = tio->io->bio;
1052 struct blk_zone_report_hdr *hdr = NULL;
1053 struct blk_zone *zone;
1054 unsigned int nr_rep = 0;
1056 sector_t part_offset;
1057 struct bio_vec bvec;
1058 struct bvec_iter iter;
1065 * bio sector was incremented by the request size on completion. Taking
1066 * into account the original request sector, the target start offset on
1067 * the backing device and the target mapping offset (ti->begin), the
1068 * start sector of the backing device. The partition offset is always 0
1069 * if the target uses a whole device.
1071 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1074 * Remap the start sector of the reported zones. For sequential zones,
1075 * also remap the write pointer position.
1077 bio_for_each_segment(bvec, report_bio, iter) {
1078 addr = kmap_atomic(bvec.bv_page);
1080 /* Remember the report header in the first page */
1083 ofst = sizeof(struct blk_zone_report_hdr);
1087 /* Set zones start sector */
1088 while (hdr->nr_zones && ofst < bvec.bv_len) {
1090 zone->start -= part_offset;
1091 if (zone->start >= start + ti->len) {
1095 zone->start = zone->start + ti->begin - start;
1096 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1097 if (zone->cond == BLK_ZONE_COND_FULL)
1098 zone->wp = zone->start + zone->len;
1099 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1100 zone->wp = zone->start;
1102 zone->wp = zone->wp + ti->begin - start - part_offset;
1104 ofst += sizeof(struct blk_zone);
1110 kunmap_atomic(addr);
1117 hdr->nr_zones = nr_rep;
1121 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1123 #else /* !CONFIG_BLK_DEV_ZONED */
1124 bio->bi_status = BLK_STS_NOTSUPP;
1127 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1130 * Flush current->bio_list when the target map method blocks.
1131 * This fixes deadlocks in snapshot and possibly in other targets.
1134 struct blk_plug plug;
1135 struct blk_plug_cb cb;
1138 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1140 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1141 struct bio_list list;
1145 INIT_LIST_HEAD(&o->cb.list);
1147 if (unlikely(!current->bio_list))
1150 for (i = 0; i < 2; i++) {
1151 list = current->bio_list[i];
1152 bio_list_init(¤t->bio_list[i]);
1154 while ((bio = bio_list_pop(&list))) {
1155 struct bio_set *bs = bio->bi_pool;
1156 if (unlikely(!bs) || bs == fs_bio_set ||
1157 !bs->rescue_workqueue) {
1158 bio_list_add(¤t->bio_list[i], bio);
1162 spin_lock(&bs->rescue_lock);
1163 bio_list_add(&bs->rescue_list, bio);
1164 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1165 spin_unlock(&bs->rescue_lock);
1170 static void dm_offload_start(struct dm_offload *o)
1172 blk_start_plug(&o->plug);
1173 o->cb.callback = flush_current_bio_list;
1174 list_add(&o->cb.list, ¤t->plug->cb_list);
1177 static void dm_offload_end(struct dm_offload *o)
1179 list_del(&o->cb.list);
1180 blk_finish_plug(&o->plug);
1183 static void __map_bio(struct dm_target_io *tio)
1187 struct dm_offload o;
1188 struct bio *clone = &tio->clone;
1189 struct dm_target *ti = tio->ti;
1191 clone->bi_end_io = clone_endio;
1194 * Map the clone. If r == 0 we don't need to do
1195 * anything, the target has assumed ownership of
1198 atomic_inc(&tio->io->io_count);
1199 sector = clone->bi_iter.bi_sector;
1201 dm_offload_start(&o);
1202 r = ti->type->map(ti, clone);
1206 case DM_MAPIO_SUBMITTED:
1208 case DM_MAPIO_REMAPPED:
1209 /* the bio has been remapped so dispatch it */
1210 trace_block_bio_remap(clone->bi_disk->queue, clone,
1211 bio_dev(tio->io->bio), sector);
1212 generic_make_request(clone);
1215 dec_pending(tio->io, BLK_STS_IOERR);
1218 case DM_MAPIO_REQUEUE:
1219 dec_pending(tio->io, BLK_STS_DM_REQUEUE);
1223 DMWARN("unimplemented target map return value: %d", r);
1229 struct mapped_device *md;
1230 struct dm_table *map;
1234 unsigned sector_count;
1237 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1239 bio->bi_iter.bi_sector = sector;
1240 bio->bi_iter.bi_size = to_bytes(len);
1244 * Creates a bio that consists of range of complete bvecs.
1246 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1247 sector_t sector, unsigned len)
1249 struct bio *clone = &tio->clone;
1251 __bio_clone_fast(clone, bio);
1253 if (unlikely(bio_integrity(bio) != NULL)) {
1256 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1257 !dm_target_passes_integrity(tio->ti->type))) {
1258 DMWARN("%s: the target %s doesn't support integrity data.",
1259 dm_device_name(tio->io->md),
1260 tio->ti->type->name);
1264 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1269 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1270 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1271 clone->bi_iter.bi_size = to_bytes(len);
1273 if (unlikely(bio_integrity(bio) != NULL))
1274 bio_integrity_trim(clone);
1279 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1280 struct dm_target *ti,
1281 unsigned target_bio_nr)
1283 struct dm_target_io *tio;
1286 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1287 tio = container_of(clone, struct dm_target_io, clone);
1291 tio->target_bio_nr = target_bio_nr;
1296 static void __clone_and_map_simple_bio(struct clone_info *ci,
1297 struct dm_target *ti,
1298 unsigned target_bio_nr, unsigned *len)
1300 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1301 struct bio *clone = &tio->clone;
1305 __bio_clone_fast(clone, ci->bio);
1307 bio_setup_sector(clone, ci->sector, *len);
1312 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1313 unsigned num_bios, unsigned *len)
1315 unsigned target_bio_nr;
1317 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1318 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1321 static int __send_empty_flush(struct clone_info *ci)
1323 unsigned target_nr = 0;
1324 struct dm_target *ti;
1326 BUG_ON(bio_has_data(ci->bio));
1327 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1328 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1333 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1334 sector_t sector, unsigned *len)
1336 struct bio *bio = ci->bio;
1337 struct dm_target_io *tio;
1338 unsigned target_bio_nr;
1339 unsigned num_target_bios = 1;
1343 * Does the target want to receive duplicate copies of the bio?
1345 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1346 num_target_bios = ti->num_write_bios(ti, bio);
1348 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1349 tio = alloc_tio(ci, ti, target_bio_nr);
1351 r = clone_bio(tio, bio, sector, *len);
1362 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1364 static unsigned get_num_discard_bios(struct dm_target *ti)
1366 return ti->num_discard_bios;
1369 static unsigned get_num_write_same_bios(struct dm_target *ti)
1371 return ti->num_write_same_bios;
1374 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1376 return ti->num_write_zeroes_bios;
1379 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1381 static bool is_split_required_for_discard(struct dm_target *ti)
1383 return ti->split_discard_bios;
1386 static int __send_changing_extent_only(struct clone_info *ci,
1387 get_num_bios_fn get_num_bios,
1388 is_split_required_fn is_split_required)
1390 struct dm_target *ti;
1395 ti = dm_table_find_target(ci->map, ci->sector);
1396 if (!dm_target_is_valid(ti))
1400 * Even though the device advertised support for this type of
1401 * request, that does not mean every target supports it, and
1402 * reconfiguration might also have changed that since the
1403 * check was performed.
1405 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1409 if (is_split_required && !is_split_required(ti))
1410 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1412 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1414 __send_duplicate_bios(ci, ti, num_bios, &len);
1417 } while (ci->sector_count -= len);
1422 static int __send_discard(struct clone_info *ci)
1424 return __send_changing_extent_only(ci, get_num_discard_bios,
1425 is_split_required_for_discard);
1428 static int __send_write_same(struct clone_info *ci)
1430 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1433 static int __send_write_zeroes(struct clone_info *ci)
1435 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1439 * Select the correct strategy for processing a non-flush bio.
1441 static int __split_and_process_non_flush(struct clone_info *ci)
1443 struct bio *bio = ci->bio;
1444 struct dm_target *ti;
1448 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1449 return __send_discard(ci);
1450 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1451 return __send_write_same(ci);
1452 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1453 return __send_write_zeroes(ci);
1455 ti = dm_table_find_target(ci->map, ci->sector);
1456 if (!dm_target_is_valid(ti))
1459 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1460 len = ci->sector_count;
1462 len = min_t(sector_t, max_io_len(ci->sector, ti),
1465 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1470 ci->sector_count -= len;
1476 * Entry point to split a bio into clones and submit them to the targets.
1478 static void __split_and_process_bio(struct mapped_device *md,
1479 struct dm_table *map, struct bio *bio)
1481 struct clone_info ci;
1484 if (unlikely(!map)) {
1491 ci.io = alloc_io(md);
1493 atomic_set(&ci.io->io_count, 1);
1496 spin_lock_init(&ci.io->endio_lock);
1497 ci.sector = bio->bi_iter.bi_sector;
1499 start_io_acct(ci.io);
1501 if (bio->bi_opf & REQ_PREFLUSH) {
1502 ci.bio = &ci.md->flush_bio;
1503 ci.sector_count = 0;
1504 error = __send_empty_flush(&ci);
1505 /* dec_pending submits any data associated with flush */
1506 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1508 ci.sector_count = 0;
1509 error = __split_and_process_non_flush(&ci);
1512 ci.sector_count = bio_sectors(bio);
1513 while (ci.sector_count && !error)
1514 error = __split_and_process_non_flush(&ci);
1517 /* drop the extra reference count */
1518 dec_pending(ci.io, errno_to_blk_status(error));
1520 /*-----------------------------------------------------------------
1522 *---------------------------------------------------------------*/
1525 * The request function that just remaps the bio built up by
1528 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1530 int rw = bio_data_dir(bio);
1531 struct mapped_device *md = q->queuedata;
1533 struct dm_table *map;
1535 map = dm_get_live_table(md, &srcu_idx);
1537 generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0);
1539 /* if we're suspended, we have to queue this io for later */
1540 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1541 dm_put_live_table(md, srcu_idx);
1543 if (!(bio->bi_opf & REQ_RAHEAD))
1547 return BLK_QC_T_NONE;
1550 __split_and_process_bio(md, map, bio);
1551 dm_put_live_table(md, srcu_idx);
1552 return BLK_QC_T_NONE;
1555 static int dm_any_congested(void *congested_data, int bdi_bits)
1558 struct mapped_device *md = congested_data;
1559 struct dm_table *map;
1561 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1562 if (dm_request_based(md)) {
1564 * With request-based DM we only need to check the
1565 * top-level queue for congestion.
1567 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1569 map = dm_get_live_table_fast(md);
1571 r = dm_table_any_congested(map, bdi_bits);
1572 dm_put_live_table_fast(md);
1579 /*-----------------------------------------------------------------
1580 * An IDR is used to keep track of allocated minor numbers.
1581 *---------------------------------------------------------------*/
1582 static void free_minor(int minor)
1584 spin_lock(&_minor_lock);
1585 idr_remove(&_minor_idr, minor);
1586 spin_unlock(&_minor_lock);
1590 * See if the device with a specific minor # is free.
1592 static int specific_minor(int minor)
1596 if (minor >= (1 << MINORBITS))
1599 idr_preload(GFP_KERNEL);
1600 spin_lock(&_minor_lock);
1602 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1604 spin_unlock(&_minor_lock);
1607 return r == -ENOSPC ? -EBUSY : r;
1611 static int next_free_minor(int *minor)
1615 idr_preload(GFP_KERNEL);
1616 spin_lock(&_minor_lock);
1618 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1620 spin_unlock(&_minor_lock);
1628 static const struct block_device_operations dm_blk_dops;
1629 static const struct dax_operations dm_dax_ops;
1631 static void dm_wq_work(struct work_struct *work);
1633 void dm_init_md_queue(struct mapped_device *md)
1636 * Request-based dm devices cannot be stacked on top of bio-based dm
1637 * devices. The type of this dm device may not have been decided yet.
1638 * The type is decided at the first table loading time.
1639 * To prevent problematic device stacking, clear the queue flag
1640 * for request stacking support until then.
1642 * This queue is new, so no concurrency on the queue_flags.
1644 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1647 * Initialize data that will only be used by a non-blk-mq DM queue
1648 * - must do so here (in alloc_dev callchain) before queue is used
1650 md->queue->queuedata = md;
1653 void dm_init_normal_md_queue(struct mapped_device *md)
1655 md->use_blk_mq = false;
1656 dm_init_md_queue(md);
1659 * Initialize aspects of queue that aren't relevant for blk-mq
1661 md->queue->backing_dev_info->congested_data = md;
1662 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1665 static void cleanup_mapped_device(struct mapped_device *md)
1668 destroy_workqueue(md->wq);
1669 if (md->kworker_task)
1670 kthread_stop(md->kworker_task);
1671 mempool_destroy(md->io_pool);
1673 bioset_free(md->bs);
1676 kill_dax(md->dax_dev);
1677 put_dax(md->dax_dev);
1682 spin_lock(&_minor_lock);
1683 md->disk->private_data = NULL;
1684 spin_unlock(&_minor_lock);
1685 del_gendisk(md->disk);
1690 blk_cleanup_queue(md->queue);
1692 cleanup_srcu_struct(&md->io_barrier);
1699 dm_mq_cleanup_mapped_device(md);
1703 * Allocate and initialise a blank device with a given minor.
1705 static struct mapped_device *alloc_dev(int minor)
1707 int r, numa_node_id = dm_get_numa_node();
1708 struct dax_device *dax_dev;
1709 struct mapped_device *md;
1712 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1714 DMWARN("unable to allocate device, out of memory.");
1718 if (!try_module_get(THIS_MODULE))
1719 goto bad_module_get;
1721 /* get a minor number for the dev */
1722 if (minor == DM_ANY_MINOR)
1723 r = next_free_minor(&minor);
1725 r = specific_minor(minor);
1729 r = init_srcu_struct(&md->io_barrier);
1731 goto bad_io_barrier;
1733 md->numa_node_id = numa_node_id;
1734 md->use_blk_mq = dm_use_blk_mq_default();
1735 md->init_tio_pdu = false;
1736 md->type = DM_TYPE_NONE;
1737 mutex_init(&md->suspend_lock);
1738 mutex_init(&md->type_lock);
1739 mutex_init(&md->table_devices_lock);
1740 spin_lock_init(&md->deferred_lock);
1741 atomic_set(&md->holders, 1);
1742 atomic_set(&md->open_count, 0);
1743 atomic_set(&md->event_nr, 0);
1744 atomic_set(&md->uevent_seq, 0);
1745 INIT_LIST_HEAD(&md->uevent_list);
1746 INIT_LIST_HEAD(&md->table_devices);
1747 spin_lock_init(&md->uevent_lock);
1749 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1753 dm_init_md_queue(md);
1755 * default to bio-based required ->make_request_fn until DM
1756 * table is loaded and md->type established. If request-based
1757 * table is loaded: blk-mq will override accordingly.
1759 blk_queue_make_request(md->queue, dm_make_request);
1761 md->disk = alloc_disk_node(1, numa_node_id);
1765 atomic_set(&md->pending[0], 0);
1766 atomic_set(&md->pending[1], 0);
1767 init_waitqueue_head(&md->wait);
1768 INIT_WORK(&md->work, dm_wq_work);
1769 init_waitqueue_head(&md->eventq);
1770 init_completion(&md->kobj_holder.completion);
1771 md->kworker_task = NULL;
1773 md->disk->major = _major;
1774 md->disk->first_minor = minor;
1775 md->disk->fops = &dm_blk_dops;
1776 md->disk->queue = md->queue;
1777 md->disk->private_data = md;
1778 sprintf(md->disk->disk_name, "dm-%d", minor);
1780 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1783 md->dax_dev = dax_dev;
1786 format_dev_t(md->name, MKDEV(_major, minor));
1788 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1792 md->bdev = bdget_disk(md->disk, 0);
1796 bio_init(&md->flush_bio, NULL, 0);
1797 bio_set_dev(&md->flush_bio, md->bdev);
1798 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1800 dm_stats_init(&md->stats);
1802 /* Populate the mapping, nobody knows we exist yet */
1803 spin_lock(&_minor_lock);
1804 old_md = idr_replace(&_minor_idr, md, minor);
1805 spin_unlock(&_minor_lock);
1807 BUG_ON(old_md != MINOR_ALLOCED);
1812 cleanup_mapped_device(md);
1816 module_put(THIS_MODULE);
1822 static void unlock_fs(struct mapped_device *md);
1824 static void free_dev(struct mapped_device *md)
1826 int minor = MINOR(disk_devt(md->disk));
1830 cleanup_mapped_device(md);
1832 free_table_devices(&md->table_devices);
1833 dm_stats_cleanup(&md->stats);
1836 module_put(THIS_MODULE);
1840 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1842 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1845 /* The md already has necessary mempools. */
1846 if (dm_table_bio_based(t)) {
1848 * Reload bioset because front_pad may have changed
1849 * because a different table was loaded.
1851 bioset_free(md->bs);
1856 * There's no need to reload with request-based dm
1857 * because the size of front_pad doesn't change.
1858 * Note for future: If you are to reload bioset,
1859 * prep-ed requests in the queue may refer
1860 * to bio from the old bioset, so you must walk
1861 * through the queue to unprep.
1866 BUG_ON(!p || md->io_pool || md->bs);
1868 md->io_pool = p->io_pool;
1874 /* mempool bind completed, no longer need any mempools in the table */
1875 dm_table_free_md_mempools(t);
1879 * Bind a table to the device.
1881 static void event_callback(void *context)
1883 unsigned long flags;
1885 struct mapped_device *md = (struct mapped_device *) context;
1887 spin_lock_irqsave(&md->uevent_lock, flags);
1888 list_splice_init(&md->uevent_list, &uevents);
1889 spin_unlock_irqrestore(&md->uevent_lock, flags);
1891 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1893 atomic_inc(&md->event_nr);
1894 wake_up(&md->eventq);
1895 dm_issue_global_event();
1899 * Protected by md->suspend_lock obtained by dm_swap_table().
1901 static void __set_size(struct mapped_device *md, sector_t size)
1903 lockdep_assert_held(&md->suspend_lock);
1905 set_capacity(md->disk, size);
1907 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1911 * Returns old map, which caller must destroy.
1913 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1914 struct queue_limits *limits)
1916 struct dm_table *old_map;
1917 struct request_queue *q = md->queue;
1920 lockdep_assert_held(&md->suspend_lock);
1922 size = dm_table_get_size(t);
1925 * Wipe any geometry if the size of the table changed.
1927 if (size != dm_get_size(md))
1928 memset(&md->geometry, 0, sizeof(md->geometry));
1930 __set_size(md, size);
1932 dm_table_event_callback(t, event_callback, md);
1935 * The queue hasn't been stopped yet, if the old table type wasn't
1936 * for request-based during suspension. So stop it to prevent
1937 * I/O mapping before resume.
1938 * This must be done before setting the queue restrictions,
1939 * because request-based dm may be run just after the setting.
1941 if (dm_table_request_based(t)) {
1944 * Leverage the fact that request-based DM targets are
1945 * immutable singletons and establish md->immutable_target
1946 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1948 md->immutable_target = dm_table_get_immutable_target(t);
1951 __bind_mempools(md, t);
1953 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1954 rcu_assign_pointer(md->map, (void *)t);
1955 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1957 dm_table_set_restrictions(t, q, limits);
1965 * Returns unbound table for the caller to free.
1967 static struct dm_table *__unbind(struct mapped_device *md)
1969 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1974 dm_table_event_callback(map, NULL, NULL);
1975 RCU_INIT_POINTER(md->map, NULL);
1982 * Constructor for a new device.
1984 int dm_create(int minor, struct mapped_device **result)
1986 struct mapped_device *md;
1988 md = alloc_dev(minor);
1999 * Functions to manage md->type.
2000 * All are required to hold md->type_lock.
2002 void dm_lock_md_type(struct mapped_device *md)
2004 mutex_lock(&md->type_lock);
2007 void dm_unlock_md_type(struct mapped_device *md)
2009 mutex_unlock(&md->type_lock);
2012 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2014 BUG_ON(!mutex_is_locked(&md->type_lock));
2018 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2023 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2025 return md->immutable_target_type;
2029 * The queue_limits are only valid as long as you have a reference
2032 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2034 BUG_ON(!atomic_read(&md->holders));
2035 return &md->queue->limits;
2037 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2040 * Setup the DM device's queue based on md's type
2042 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2045 enum dm_queue_mode type = dm_get_md_type(md);
2048 case DM_TYPE_REQUEST_BASED:
2049 r = dm_old_init_request_queue(md, t);
2051 DMERR("Cannot initialize queue for request-based mapped device");
2055 case DM_TYPE_MQ_REQUEST_BASED:
2056 r = dm_mq_init_request_queue(md, t);
2058 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2062 case DM_TYPE_BIO_BASED:
2063 case DM_TYPE_DAX_BIO_BASED:
2064 dm_init_normal_md_queue(md);
2066 * DM handles splitting bios as needed. Free the bio_split bioset
2067 * since it won't be used (saves 1 process per bio-based DM device).
2069 bioset_free(md->queue->bio_split);
2070 md->queue->bio_split = NULL;
2080 struct mapped_device *dm_get_md(dev_t dev)
2082 struct mapped_device *md;
2083 unsigned minor = MINOR(dev);
2085 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2088 spin_lock(&_minor_lock);
2090 md = idr_find(&_minor_idr, minor);
2092 if ((md == MINOR_ALLOCED ||
2093 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2094 dm_deleting_md(md) ||
2095 test_bit(DMF_FREEING, &md->flags))) {
2103 spin_unlock(&_minor_lock);
2107 EXPORT_SYMBOL_GPL(dm_get_md);
2109 void *dm_get_mdptr(struct mapped_device *md)
2111 return md->interface_ptr;
2114 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2116 md->interface_ptr = ptr;
2119 void dm_get(struct mapped_device *md)
2121 atomic_inc(&md->holders);
2122 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2125 int dm_hold(struct mapped_device *md)
2127 spin_lock(&_minor_lock);
2128 if (test_bit(DMF_FREEING, &md->flags)) {
2129 spin_unlock(&_minor_lock);
2133 spin_unlock(&_minor_lock);
2136 EXPORT_SYMBOL_GPL(dm_hold);
2138 const char *dm_device_name(struct mapped_device *md)
2142 EXPORT_SYMBOL_GPL(dm_device_name);
2144 static void __dm_destroy(struct mapped_device *md, bool wait)
2146 struct request_queue *q = dm_get_md_queue(md);
2147 struct dm_table *map;
2152 spin_lock(&_minor_lock);
2153 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2154 set_bit(DMF_FREEING, &md->flags);
2155 spin_unlock(&_minor_lock);
2157 blk_set_queue_dying(q);
2159 if (dm_request_based(md) && md->kworker_task)
2160 kthread_flush_worker(&md->kworker);
2163 * Take suspend_lock so that presuspend and postsuspend methods
2164 * do not race with internal suspend.
2166 mutex_lock(&md->suspend_lock);
2167 map = dm_get_live_table(md, &srcu_idx);
2168 if (!dm_suspended_md(md)) {
2169 dm_table_presuspend_targets(map);
2170 dm_table_postsuspend_targets(map);
2172 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2173 dm_put_live_table(md, srcu_idx);
2174 mutex_unlock(&md->suspend_lock);
2177 * Rare, but there may be I/O requests still going to complete,
2178 * for example. Wait for all references to disappear.
2179 * No one should increment the reference count of the mapped_device,
2180 * after the mapped_device state becomes DMF_FREEING.
2183 while (atomic_read(&md->holders))
2185 else if (atomic_read(&md->holders))
2186 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2187 dm_device_name(md), atomic_read(&md->holders));
2190 dm_table_destroy(__unbind(md));
2194 void dm_destroy(struct mapped_device *md)
2196 __dm_destroy(md, true);
2199 void dm_destroy_immediate(struct mapped_device *md)
2201 __dm_destroy(md, false);
2204 void dm_put(struct mapped_device *md)
2206 atomic_dec(&md->holders);
2208 EXPORT_SYMBOL_GPL(dm_put);
2210 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2216 prepare_to_wait(&md->wait, &wait, task_state);
2218 if (!md_in_flight(md))
2221 if (signal_pending_state(task_state, current)) {
2228 finish_wait(&md->wait, &wait);
2234 * Process the deferred bios
2236 static void dm_wq_work(struct work_struct *work)
2238 struct mapped_device *md = container_of(work, struct mapped_device,
2242 struct dm_table *map;
2244 map = dm_get_live_table(md, &srcu_idx);
2246 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2247 spin_lock_irq(&md->deferred_lock);
2248 c = bio_list_pop(&md->deferred);
2249 spin_unlock_irq(&md->deferred_lock);
2254 if (dm_request_based(md))
2255 generic_make_request(c);
2257 __split_and_process_bio(md, map, c);
2260 dm_put_live_table(md, srcu_idx);
2263 static void dm_queue_flush(struct mapped_device *md)
2265 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2266 smp_mb__after_atomic();
2267 queue_work(md->wq, &md->work);
2271 * Swap in a new table, returning the old one for the caller to destroy.
2273 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2275 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2276 struct queue_limits limits;
2279 mutex_lock(&md->suspend_lock);
2281 /* device must be suspended */
2282 if (!dm_suspended_md(md))
2286 * If the new table has no data devices, retain the existing limits.
2287 * This helps multipath with queue_if_no_path if all paths disappear,
2288 * then new I/O is queued based on these limits, and then some paths
2291 if (dm_table_has_no_data_devices(table)) {
2292 live_map = dm_get_live_table_fast(md);
2294 limits = md->queue->limits;
2295 dm_put_live_table_fast(md);
2299 r = dm_calculate_queue_limits(table, &limits);
2306 map = __bind(md, table, &limits);
2307 dm_issue_global_event();
2310 mutex_unlock(&md->suspend_lock);
2315 * Functions to lock and unlock any filesystem running on the
2318 static int lock_fs(struct mapped_device *md)
2322 WARN_ON(md->frozen_sb);
2324 md->frozen_sb = freeze_bdev(md->bdev);
2325 if (IS_ERR(md->frozen_sb)) {
2326 r = PTR_ERR(md->frozen_sb);
2327 md->frozen_sb = NULL;
2331 set_bit(DMF_FROZEN, &md->flags);
2336 static void unlock_fs(struct mapped_device *md)
2338 if (!test_bit(DMF_FROZEN, &md->flags))
2341 thaw_bdev(md->bdev, md->frozen_sb);
2342 md->frozen_sb = NULL;
2343 clear_bit(DMF_FROZEN, &md->flags);
2347 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2348 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2349 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2351 * If __dm_suspend returns 0, the device is completely quiescent
2352 * now. There is no request-processing activity. All new requests
2353 * are being added to md->deferred list.
2355 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2356 unsigned suspend_flags, long task_state,
2357 int dmf_suspended_flag)
2359 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2360 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2363 lockdep_assert_held(&md->suspend_lock);
2366 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2367 * This flag is cleared before dm_suspend returns.
2370 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2372 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2375 * This gets reverted if there's an error later and the targets
2376 * provide the .presuspend_undo hook.
2378 dm_table_presuspend_targets(map);
2381 * Flush I/O to the device.
2382 * Any I/O submitted after lock_fs() may not be flushed.
2383 * noflush takes precedence over do_lockfs.
2384 * (lock_fs() flushes I/Os and waits for them to complete.)
2386 if (!noflush && do_lockfs) {
2389 dm_table_presuspend_undo_targets(map);
2395 * Here we must make sure that no processes are submitting requests
2396 * to target drivers i.e. no one may be executing
2397 * __split_and_process_bio. This is called from dm_request and
2400 * To get all processes out of __split_and_process_bio in dm_request,
2401 * we take the write lock. To prevent any process from reentering
2402 * __split_and_process_bio from dm_request and quiesce the thread
2403 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2404 * flush_workqueue(md->wq).
2406 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2408 synchronize_srcu(&md->io_barrier);
2411 * Stop md->queue before flushing md->wq in case request-based
2412 * dm defers requests to md->wq from md->queue.
2414 if (dm_request_based(md)) {
2415 dm_stop_queue(md->queue);
2416 if (md->kworker_task)
2417 kthread_flush_worker(&md->kworker);
2420 flush_workqueue(md->wq);
2423 * At this point no more requests are entering target request routines.
2424 * We call dm_wait_for_completion to wait for all existing requests
2427 r = dm_wait_for_completion(md, task_state);
2429 set_bit(dmf_suspended_flag, &md->flags);
2432 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2434 synchronize_srcu(&md->io_barrier);
2436 /* were we interrupted ? */
2440 if (dm_request_based(md))
2441 dm_start_queue(md->queue);
2444 dm_table_presuspend_undo_targets(map);
2445 /* pushback list is already flushed, so skip flush */
2452 * We need to be able to change a mapping table under a mounted
2453 * filesystem. For example we might want to move some data in
2454 * the background. Before the table can be swapped with
2455 * dm_bind_table, dm_suspend must be called to flush any in
2456 * flight bios and ensure that any further io gets deferred.
2459 * Suspend mechanism in request-based dm.
2461 * 1. Flush all I/Os by lock_fs() if needed.
2462 * 2. Stop dispatching any I/O by stopping the request_queue.
2463 * 3. Wait for all in-flight I/Os to be completed or requeued.
2465 * To abort suspend, start the request_queue.
2467 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2469 struct dm_table *map = NULL;
2473 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2475 if (dm_suspended_md(md)) {
2480 if (dm_suspended_internally_md(md)) {
2481 /* already internally suspended, wait for internal resume */
2482 mutex_unlock(&md->suspend_lock);
2483 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2489 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2491 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2495 dm_table_postsuspend_targets(map);
2498 mutex_unlock(&md->suspend_lock);
2502 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2505 int r = dm_table_resume_targets(map);
2513 * Flushing deferred I/Os must be done after targets are resumed
2514 * so that mapping of targets can work correctly.
2515 * Request-based dm is queueing the deferred I/Os in its request_queue.
2517 if (dm_request_based(md))
2518 dm_start_queue(md->queue);
2525 int dm_resume(struct mapped_device *md)
2528 struct dm_table *map = NULL;
2532 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2534 if (!dm_suspended_md(md))
2537 if (dm_suspended_internally_md(md)) {
2538 /* already internally suspended, wait for internal resume */
2539 mutex_unlock(&md->suspend_lock);
2540 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2546 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2547 if (!map || !dm_table_get_size(map))
2550 r = __dm_resume(md, map);
2554 clear_bit(DMF_SUSPENDED, &md->flags);
2556 mutex_unlock(&md->suspend_lock);
2562 * Internal suspend/resume works like userspace-driven suspend. It waits
2563 * until all bios finish and prevents issuing new bios to the target drivers.
2564 * It may be used only from the kernel.
2567 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2569 struct dm_table *map = NULL;
2571 lockdep_assert_held(&md->suspend_lock);
2573 if (md->internal_suspend_count++)
2574 return; /* nested internal suspend */
2576 if (dm_suspended_md(md)) {
2577 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2578 return; /* nest suspend */
2581 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2584 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2585 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2586 * would require changing .presuspend to return an error -- avoid this
2587 * until there is a need for more elaborate variants of internal suspend.
2589 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2590 DMF_SUSPENDED_INTERNALLY);
2592 dm_table_postsuspend_targets(map);
2595 static void __dm_internal_resume(struct mapped_device *md)
2597 BUG_ON(!md->internal_suspend_count);
2599 if (--md->internal_suspend_count)
2600 return; /* resume from nested internal suspend */
2602 if (dm_suspended_md(md))
2603 goto done; /* resume from nested suspend */
2606 * NOTE: existing callers don't need to call dm_table_resume_targets
2607 * (which may fail -- so best to avoid it for now by passing NULL map)
2609 (void) __dm_resume(md, NULL);
2612 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2613 smp_mb__after_atomic();
2614 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2617 void dm_internal_suspend_noflush(struct mapped_device *md)
2619 mutex_lock(&md->suspend_lock);
2620 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2621 mutex_unlock(&md->suspend_lock);
2623 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2625 void dm_internal_resume(struct mapped_device *md)
2627 mutex_lock(&md->suspend_lock);
2628 __dm_internal_resume(md);
2629 mutex_unlock(&md->suspend_lock);
2631 EXPORT_SYMBOL_GPL(dm_internal_resume);
2634 * Fast variants of internal suspend/resume hold md->suspend_lock,
2635 * which prevents interaction with userspace-driven suspend.
2638 void dm_internal_suspend_fast(struct mapped_device *md)
2640 mutex_lock(&md->suspend_lock);
2641 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2644 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2645 synchronize_srcu(&md->io_barrier);
2646 flush_workqueue(md->wq);
2647 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2649 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2651 void dm_internal_resume_fast(struct mapped_device *md)
2653 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2659 mutex_unlock(&md->suspend_lock);
2661 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2663 /*-----------------------------------------------------------------
2664 * Event notification.
2665 *---------------------------------------------------------------*/
2666 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2671 char udev_cookie[DM_COOKIE_LENGTH];
2672 char *envp[] = { udev_cookie, NULL };
2674 noio_flag = memalloc_noio_save();
2677 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2679 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2680 DM_COOKIE_ENV_VAR_NAME, cookie);
2681 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2685 memalloc_noio_restore(noio_flag);
2690 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2692 return atomic_add_return(1, &md->uevent_seq);
2695 uint32_t dm_get_event_nr(struct mapped_device *md)
2697 return atomic_read(&md->event_nr);
2700 int dm_wait_event(struct mapped_device *md, int event_nr)
2702 return wait_event_interruptible(md->eventq,
2703 (event_nr != atomic_read(&md->event_nr)));
2706 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2708 unsigned long flags;
2710 spin_lock_irqsave(&md->uevent_lock, flags);
2711 list_add(elist, &md->uevent_list);
2712 spin_unlock_irqrestore(&md->uevent_lock, flags);
2716 * The gendisk is only valid as long as you have a reference
2719 struct gendisk *dm_disk(struct mapped_device *md)
2723 EXPORT_SYMBOL_GPL(dm_disk);
2725 struct kobject *dm_kobject(struct mapped_device *md)
2727 return &md->kobj_holder.kobj;
2730 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2732 struct mapped_device *md;
2734 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2736 spin_lock(&_minor_lock);
2737 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2743 spin_unlock(&_minor_lock);
2748 int dm_suspended_md(struct mapped_device *md)
2750 return test_bit(DMF_SUSPENDED, &md->flags);
2753 int dm_suspended_internally_md(struct mapped_device *md)
2755 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2758 int dm_test_deferred_remove_flag(struct mapped_device *md)
2760 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2763 int dm_suspended(struct dm_target *ti)
2765 return dm_suspended_md(dm_table_get_md(ti->table));
2767 EXPORT_SYMBOL_GPL(dm_suspended);
2769 int dm_noflush_suspending(struct dm_target *ti)
2771 return __noflush_suspending(dm_table_get_md(ti->table));
2773 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2775 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2776 unsigned integrity, unsigned per_io_data_size)
2778 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2779 unsigned int pool_size = 0;
2780 unsigned int front_pad;
2786 case DM_TYPE_BIO_BASED:
2787 case DM_TYPE_DAX_BIO_BASED:
2788 pool_size = dm_get_reserved_bio_based_ios();
2789 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2791 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2792 if (!pools->io_pool)
2795 case DM_TYPE_REQUEST_BASED:
2796 case DM_TYPE_MQ_REQUEST_BASED:
2797 pool_size = dm_get_reserved_rq_based_ios();
2798 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2799 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2805 pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER);
2809 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2815 dm_free_md_mempools(pools);
2820 void dm_free_md_mempools(struct dm_md_mempools *pools)
2825 mempool_destroy(pools->io_pool);
2828 bioset_free(pools->bs);
2840 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2843 struct mapped_device *md = bdev->bd_disk->private_data;
2844 struct dm_table *table;
2845 struct dm_target *ti;
2846 int ret = -ENOTTY, srcu_idx;
2848 table = dm_get_live_table(md, &srcu_idx);
2849 if (!table || !dm_table_get_size(table))
2852 /* We only support devices that have a single target */
2853 if (dm_table_get_num_targets(table) != 1)
2855 ti = dm_table_get_target(table, 0);
2858 if (!ti->type->iterate_devices)
2861 ret = ti->type->iterate_devices(ti, fn, data);
2863 dm_put_live_table(md, srcu_idx);
2868 * For register / unregister we need to manually call out to every path.
2870 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2871 sector_t start, sector_t len, void *data)
2873 struct dm_pr *pr = data;
2874 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2876 if (!ops || !ops->pr_register)
2878 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2881 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2892 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2893 if (ret && new_key) {
2894 /* unregister all paths if we failed to register any path */
2895 pr.old_key = new_key;
2898 pr.fail_early = false;
2899 dm_call_pr(bdev, __dm_pr_register, &pr);
2905 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2908 struct mapped_device *md = bdev->bd_disk->private_data;
2909 const struct pr_ops *ops;
2913 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2917 ops = bdev->bd_disk->fops->pr_ops;
2918 if (ops && ops->pr_reserve)
2919 r = ops->pr_reserve(bdev, key, type, flags);
2927 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2929 struct mapped_device *md = bdev->bd_disk->private_data;
2930 const struct pr_ops *ops;
2934 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2938 ops = bdev->bd_disk->fops->pr_ops;
2939 if (ops && ops->pr_release)
2940 r = ops->pr_release(bdev, key, type);
2948 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2949 enum pr_type type, bool abort)
2951 struct mapped_device *md = bdev->bd_disk->private_data;
2952 const struct pr_ops *ops;
2956 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2960 ops = bdev->bd_disk->fops->pr_ops;
2961 if (ops && ops->pr_preempt)
2962 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2970 static int dm_pr_clear(struct block_device *bdev, u64 key)
2972 struct mapped_device *md = bdev->bd_disk->private_data;
2973 const struct pr_ops *ops;
2977 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2981 ops = bdev->bd_disk->fops->pr_ops;
2982 if (ops && ops->pr_clear)
2983 r = ops->pr_clear(bdev, key);
2991 static const struct pr_ops dm_pr_ops = {
2992 .pr_register = dm_pr_register,
2993 .pr_reserve = dm_pr_reserve,
2994 .pr_release = dm_pr_release,
2995 .pr_preempt = dm_pr_preempt,
2996 .pr_clear = dm_pr_clear,
2999 static const struct block_device_operations dm_blk_dops = {
3000 .open = dm_blk_open,
3001 .release = dm_blk_close,
3002 .ioctl = dm_blk_ioctl,
3003 .getgeo = dm_blk_getgeo,
3004 .pr_ops = &dm_pr_ops,
3005 .owner = THIS_MODULE
3008 static const struct dax_operations dm_dax_ops = {
3009 .direct_access = dm_dax_direct_access,
3010 .copy_from_iter = dm_dax_copy_from_iter,
3016 module_init(dm_init);
3017 module_exit(dm_exit);
3019 module_param(major, uint, 0);
3020 MODULE_PARM_DESC(major, "The major number of the device mapper");
3022 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3023 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3025 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3026 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3028 MODULE_DESCRIPTION(DM_NAME " driver");
3029 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3030 MODULE_LICENSE("GPL");