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 mapped_device *md, struct bio *bio,
532 unsigned long start_time, struct dm_stats_aux *stats_aux)
534 unsigned long duration = jiffies - start_time;
536 int rw = bio_data_dir(bio);
538 generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, start_time);
540 if (unlikely(dm_stats_used(&md->stats)))
541 dm_stats_account_io(&md->stats, bio_data_dir(bio),
542 bio->bi_iter.bi_sector, bio_sectors(bio),
543 true, duration, stats_aux);
546 * After this is decremented the bio must not be touched if it is
549 pending = atomic_dec_return(&md->pending[rw]);
550 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
551 pending += atomic_read(&md->pending[rw^0x1]);
553 /* nudge anyone waiting on suspend queue */
559 * Add the bio to the list of deferred io.
561 static void queue_io(struct mapped_device *md, struct bio *bio)
565 spin_lock_irqsave(&md->deferred_lock, flags);
566 bio_list_add(&md->deferred, bio);
567 spin_unlock_irqrestore(&md->deferred_lock, flags);
568 queue_work(md->wq, &md->work);
572 * Everyone (including functions in this file), should use this
573 * function to access the md->map field, and make sure they call
574 * dm_put_live_table() when finished.
576 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
578 *srcu_idx = srcu_read_lock(&md->io_barrier);
580 return srcu_dereference(md->map, &md->io_barrier);
583 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
585 srcu_read_unlock(&md->io_barrier, srcu_idx);
588 void dm_sync_table(struct mapped_device *md)
590 synchronize_srcu(&md->io_barrier);
591 synchronize_rcu_expedited();
595 * A fast alternative to dm_get_live_table/dm_put_live_table.
596 * The caller must not block between these two functions.
598 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
601 return rcu_dereference(md->map);
604 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
610 * Open a table device so we can use it as a map destination.
612 static int open_table_device(struct table_device *td, dev_t dev,
613 struct mapped_device *md)
615 static char *_claim_ptr = "I belong to device-mapper";
616 struct block_device *bdev;
620 BUG_ON(td->dm_dev.bdev);
622 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr);
624 return PTR_ERR(bdev);
626 r = bd_link_disk_holder(bdev, dm_disk(md));
628 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
632 td->dm_dev.bdev = bdev;
633 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
638 * Close a table device that we've been using.
640 static void close_table_device(struct table_device *td, struct mapped_device *md)
642 if (!td->dm_dev.bdev)
645 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
646 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
647 put_dax(td->dm_dev.dax_dev);
648 td->dm_dev.bdev = NULL;
649 td->dm_dev.dax_dev = NULL;
652 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
654 struct table_device *td;
656 list_for_each_entry(td, l, list)
657 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
663 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
664 struct dm_dev **result) {
666 struct table_device *td;
668 mutex_lock(&md->table_devices_lock);
669 td = find_table_device(&md->table_devices, dev, mode);
671 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
673 mutex_unlock(&md->table_devices_lock);
677 td->dm_dev.mode = mode;
678 td->dm_dev.bdev = NULL;
680 if ((r = open_table_device(td, dev, md))) {
681 mutex_unlock(&md->table_devices_lock);
686 format_dev_t(td->dm_dev.name, dev);
688 atomic_set(&td->count, 0);
689 list_add(&td->list, &md->table_devices);
691 atomic_inc(&td->count);
692 mutex_unlock(&md->table_devices_lock);
694 *result = &td->dm_dev;
697 EXPORT_SYMBOL_GPL(dm_get_table_device);
699 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
701 struct table_device *td = container_of(d, struct table_device, dm_dev);
703 mutex_lock(&md->table_devices_lock);
704 if (atomic_dec_and_test(&td->count)) {
705 close_table_device(td, md);
709 mutex_unlock(&md->table_devices_lock);
711 EXPORT_SYMBOL(dm_put_table_device);
713 static void free_table_devices(struct list_head *devices)
715 struct list_head *tmp, *next;
717 list_for_each_safe(tmp, next, devices) {
718 struct table_device *td = list_entry(tmp, struct table_device, list);
720 DMWARN("dm_destroy: %s still exists with %d references",
721 td->dm_dev.name, atomic_read(&td->count));
727 * Get the geometry associated with a dm device
729 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
737 * Set the geometry of a device.
739 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
741 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
743 if (geo->start > sz) {
744 DMWARN("Start sector is beyond the geometry limits.");
753 /*-----------------------------------------------------------------
755 * A more elegant soln is in the works that uses the queue
756 * merge fn, unfortunately there are a couple of changes to
757 * the block layer that I want to make for this. So in the
758 * interests of getting something for people to use I give
759 * you this clearly demarcated crap.
760 *---------------------------------------------------------------*/
762 static int __noflush_suspending(struct mapped_device *md)
764 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
768 * Decrements the number of outstanding ios that a bio has been
769 * cloned into, completing the original io if necc.
771 static void dec_pending(struct dm_io *io, blk_status_t error)
774 blk_status_t io_error;
776 struct mapped_device *md = io->md;
777 unsigned long start_time = 0;
778 struct dm_stats_aux stats_aux;
780 /* Push-back supersedes any I/O errors */
781 if (unlikely(error)) {
782 spin_lock_irqsave(&io->endio_lock, flags);
783 if (!(io->status == BLK_STS_DM_REQUEUE &&
784 __noflush_suspending(md)))
786 spin_unlock_irqrestore(&io->endio_lock, flags);
789 if (atomic_dec_and_test(&io->io_count)) {
790 if (io->status == BLK_STS_DM_REQUEUE) {
792 * Target requested pushing back the I/O.
794 spin_lock_irqsave(&md->deferred_lock, flags);
795 if (__noflush_suspending(md))
796 bio_list_add_head(&md->deferred, io->bio);
798 /* noflush suspend was interrupted. */
799 io->status = BLK_STS_IOERR;
800 spin_unlock_irqrestore(&md->deferred_lock, flags);
803 io_error = io->status;
805 start_time = io->start_time;
806 stats_aux = io->stats_aux;
808 end_io_acct(md, bio, start_time, &stats_aux);
810 if (io_error == BLK_STS_DM_REQUEUE)
813 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
815 * Preflush done for flush with data, reissue
816 * without REQ_PREFLUSH.
818 bio->bi_opf &= ~REQ_PREFLUSH;
821 /* done with normal IO or empty flush */
823 bio->bi_status = io_error;
829 void disable_write_same(struct mapped_device *md)
831 struct queue_limits *limits = dm_get_queue_limits(md);
833 /* device doesn't really support WRITE SAME, disable it */
834 limits->max_write_same_sectors = 0;
837 void disable_write_zeroes(struct mapped_device *md)
839 struct queue_limits *limits = dm_get_queue_limits(md);
841 /* device doesn't really support WRITE ZEROES, disable it */
842 limits->max_write_zeroes_sectors = 0;
845 static void clone_endio(struct bio *bio)
847 blk_status_t error = bio->bi_status;
848 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
849 struct dm_io *io = tio->io;
850 struct mapped_device *md = tio->io->md;
851 dm_endio_fn endio = tio->ti->type->end_io;
853 if (unlikely(error == BLK_STS_TARGET)) {
854 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
855 !bio->bi_disk->queue->limits.max_write_same_sectors)
856 disable_write_same(md);
857 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
858 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
859 disable_write_zeroes(md);
863 int r = endio(tio->ti, bio, &error);
865 case DM_ENDIO_REQUEUE:
866 error = BLK_STS_DM_REQUEUE;
870 case DM_ENDIO_INCOMPLETE:
871 /* The target will handle the io */
874 DMWARN("unimplemented target endio return value: %d", r);
880 dec_pending(io, error);
884 * Return maximum size of I/O possible at the supplied sector up to the current
887 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
889 sector_t target_offset = dm_target_offset(ti, sector);
891 return ti->len - target_offset;
894 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
896 sector_t len = max_io_len_target_boundary(sector, ti);
897 sector_t offset, max_len;
900 * Does the target need to split even further?
902 if (ti->max_io_len) {
903 offset = dm_target_offset(ti, sector);
904 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
905 max_len = sector_div(offset, ti->max_io_len);
907 max_len = offset & (ti->max_io_len - 1);
908 max_len = ti->max_io_len - max_len;
917 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
919 if (len > UINT_MAX) {
920 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
921 (unsigned long long)len, UINT_MAX);
922 ti->error = "Maximum size of target IO is too large";
926 ti->max_io_len = (uint32_t) len;
930 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
932 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
933 sector_t sector, int *srcu_idx)
935 struct dm_table *map;
936 struct dm_target *ti;
938 map = dm_get_live_table(md, srcu_idx);
942 ti = dm_table_find_target(map, sector);
943 if (!dm_target_is_valid(ti))
949 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
950 long nr_pages, void **kaddr, pfn_t *pfn)
952 struct mapped_device *md = dax_get_private(dax_dev);
953 sector_t sector = pgoff * PAGE_SECTORS;
954 struct dm_target *ti;
955 long len, ret = -EIO;
958 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
962 if (!ti->type->direct_access)
964 len = max_io_len(sector, ti) / PAGE_SECTORS;
967 nr_pages = min(len, nr_pages);
968 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
971 dm_put_live_table(md, srcu_idx);
976 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
977 void *addr, size_t bytes, struct iov_iter *i)
979 struct mapped_device *md = dax_get_private(dax_dev);
980 sector_t sector = pgoff * PAGE_SECTORS;
981 struct dm_target *ti;
985 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
989 if (!ti->type->dax_copy_from_iter) {
990 ret = copy_from_iter(addr, bytes, i);
993 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
995 dm_put_live_table(md, srcu_idx);
1001 * A target may call dm_accept_partial_bio only from the map routine. It is
1002 * allowed for all bio types except REQ_PREFLUSH.
1004 * dm_accept_partial_bio informs the dm that the target only wants to process
1005 * additional n_sectors sectors of the bio and the rest of the data should be
1006 * sent in a next bio.
1008 * A diagram that explains the arithmetics:
1009 * +--------------------+---------------+-------+
1011 * +--------------------+---------------+-------+
1013 * <-------------- *tio->len_ptr --------------->
1014 * <------- bi_size ------->
1017 * Region 1 was already iterated over with bio_advance or similar function.
1018 * (it may be empty if the target doesn't use bio_advance)
1019 * Region 2 is the remaining bio size that the target wants to process.
1020 * (it may be empty if region 1 is non-empty, although there is no reason
1022 * The target requires that region 3 is to be sent in the next bio.
1024 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1025 * the partially processed part (the sum of regions 1+2) must be the same for all
1026 * copies of the bio.
1028 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1030 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1031 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1032 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1033 BUG_ON(bi_size > *tio->len_ptr);
1034 BUG_ON(n_sectors > bi_size);
1035 *tio->len_ptr -= bi_size - n_sectors;
1036 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1038 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1041 * The zone descriptors obtained with a zone report indicate zone positions
1042 * within the target backing device, regardless of that device is a partition
1043 * and regardless of the target mapping start sector on the device or partition.
1044 * The zone descriptors start sector and write pointer position must be adjusted
1045 * to match their relative position within the dm device.
1046 * A target may call dm_remap_zone_report() after completion of a
1047 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the
1050 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start)
1052 #ifdef CONFIG_BLK_DEV_ZONED
1053 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1054 struct bio *report_bio = tio->io->bio;
1055 struct blk_zone_report_hdr *hdr = NULL;
1056 struct blk_zone *zone;
1057 unsigned int nr_rep = 0;
1059 sector_t part_offset;
1060 struct bio_vec bvec;
1061 struct bvec_iter iter;
1068 * bio sector was incremented by the request size on completion. Taking
1069 * into account the original request sector, the target start offset on
1070 * the backing device and the target mapping offset (ti->begin), the
1071 * start sector of the backing device. The partition offset is always 0
1072 * if the target uses a whole device.
1074 part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio));
1077 * Remap the start sector of the reported zones. For sequential zones,
1078 * also remap the write pointer position.
1080 bio_for_each_segment(bvec, report_bio, iter) {
1081 addr = kmap_atomic(bvec.bv_page);
1083 /* Remember the report header in the first page */
1086 ofst = sizeof(struct blk_zone_report_hdr);
1090 /* Set zones start sector */
1091 while (hdr->nr_zones && ofst < bvec.bv_len) {
1093 zone->start -= part_offset;
1094 if (zone->start >= start + ti->len) {
1098 zone->start = zone->start + ti->begin - start;
1099 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
1100 if (zone->cond == BLK_ZONE_COND_FULL)
1101 zone->wp = zone->start + zone->len;
1102 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1103 zone->wp = zone->start;
1105 zone->wp = zone->wp + ti->begin - start - part_offset;
1107 ofst += sizeof(struct blk_zone);
1113 kunmap_atomic(addr);
1120 hdr->nr_zones = nr_rep;
1124 bio_advance(report_bio, report_bio->bi_iter.bi_size);
1126 #else /* !CONFIG_BLK_DEV_ZONED */
1127 bio->bi_status = BLK_STS_NOTSUPP;
1130 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1133 * Flush current->bio_list when the target map method blocks.
1134 * This fixes deadlocks in snapshot and possibly in other targets.
1137 struct blk_plug plug;
1138 struct blk_plug_cb cb;
1141 static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule)
1143 struct dm_offload *o = container_of(cb, struct dm_offload, cb);
1144 struct bio_list list;
1148 INIT_LIST_HEAD(&o->cb.list);
1150 if (unlikely(!current->bio_list))
1153 for (i = 0; i < 2; i++) {
1154 list = current->bio_list[i];
1155 bio_list_init(¤t->bio_list[i]);
1157 while ((bio = bio_list_pop(&list))) {
1158 struct bio_set *bs = bio->bi_pool;
1159 if (unlikely(!bs) || bs == fs_bio_set ||
1160 !bs->rescue_workqueue) {
1161 bio_list_add(¤t->bio_list[i], bio);
1165 spin_lock(&bs->rescue_lock);
1166 bio_list_add(&bs->rescue_list, bio);
1167 queue_work(bs->rescue_workqueue, &bs->rescue_work);
1168 spin_unlock(&bs->rescue_lock);
1173 static void dm_offload_start(struct dm_offload *o)
1175 blk_start_plug(&o->plug);
1176 o->cb.callback = flush_current_bio_list;
1177 list_add(&o->cb.list, ¤t->plug->cb_list);
1180 static void dm_offload_end(struct dm_offload *o)
1182 list_del(&o->cb.list);
1183 blk_finish_plug(&o->plug);
1186 static void __map_bio(struct dm_target_io *tio)
1190 struct dm_offload o;
1191 struct bio *clone = &tio->clone;
1192 struct dm_target *ti = tio->ti;
1194 clone->bi_end_io = clone_endio;
1197 * Map the clone. If r == 0 we don't need to do
1198 * anything, the target has assumed ownership of
1201 atomic_inc(&tio->io->io_count);
1202 sector = clone->bi_iter.bi_sector;
1204 dm_offload_start(&o);
1205 r = ti->type->map(ti, clone);
1209 case DM_MAPIO_SUBMITTED:
1211 case DM_MAPIO_REMAPPED:
1212 /* the bio has been remapped so dispatch it */
1213 trace_block_bio_remap(clone->bi_disk->queue, clone,
1214 bio_dev(tio->io->bio), sector);
1215 generic_make_request(clone);
1218 dec_pending(tio->io, BLK_STS_IOERR);
1221 case DM_MAPIO_REQUEUE:
1222 dec_pending(tio->io, BLK_STS_DM_REQUEUE);
1226 DMWARN("unimplemented target map return value: %d", r);
1232 struct mapped_device *md;
1233 struct dm_table *map;
1237 unsigned sector_count;
1240 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1242 bio->bi_iter.bi_sector = sector;
1243 bio->bi_iter.bi_size = to_bytes(len);
1247 * Creates a bio that consists of range of complete bvecs.
1249 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1250 sector_t sector, unsigned len)
1252 struct bio *clone = &tio->clone;
1254 __bio_clone_fast(clone, bio);
1256 if (unlikely(bio_integrity(bio) != NULL)) {
1259 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1260 !dm_target_passes_integrity(tio->ti->type))) {
1261 DMWARN("%s: the target %s doesn't support integrity data.",
1262 dm_device_name(tio->io->md),
1263 tio->ti->type->name);
1267 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1272 if (bio_op(bio) != REQ_OP_ZONE_REPORT)
1273 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1274 clone->bi_iter.bi_size = to_bytes(len);
1276 if (unlikely(bio_integrity(bio) != NULL))
1277 bio_integrity_trim(clone);
1282 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1283 struct dm_target *ti,
1284 unsigned target_bio_nr)
1286 struct dm_target_io *tio;
1289 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1290 tio = container_of(clone, struct dm_target_io, clone);
1294 tio->target_bio_nr = target_bio_nr;
1299 static void __clone_and_map_simple_bio(struct clone_info *ci,
1300 struct dm_target *ti,
1301 unsigned target_bio_nr, unsigned *len)
1303 struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr);
1304 struct bio *clone = &tio->clone;
1308 __bio_clone_fast(clone, ci->bio);
1310 bio_setup_sector(clone, ci->sector, *len);
1315 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1316 unsigned num_bios, unsigned *len)
1318 unsigned target_bio_nr;
1320 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1321 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1324 static int __send_empty_flush(struct clone_info *ci)
1326 unsigned target_nr = 0;
1327 struct dm_target *ti;
1329 BUG_ON(bio_has_data(ci->bio));
1330 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1331 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1336 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1337 sector_t sector, unsigned *len)
1339 struct bio *bio = ci->bio;
1340 struct dm_target_io *tio;
1341 unsigned target_bio_nr;
1342 unsigned num_target_bios = 1;
1346 * Does the target want to receive duplicate copies of the bio?
1348 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1349 num_target_bios = ti->num_write_bios(ti, bio);
1351 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1352 tio = alloc_tio(ci, ti, target_bio_nr);
1354 r = clone_bio(tio, bio, sector, *len);
1365 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1367 static unsigned get_num_discard_bios(struct dm_target *ti)
1369 return ti->num_discard_bios;
1372 static unsigned get_num_write_same_bios(struct dm_target *ti)
1374 return ti->num_write_same_bios;
1377 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1379 return ti->num_write_zeroes_bios;
1382 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1384 static bool is_split_required_for_discard(struct dm_target *ti)
1386 return ti->split_discard_bios;
1389 static int __send_changing_extent_only(struct clone_info *ci,
1390 get_num_bios_fn get_num_bios,
1391 is_split_required_fn is_split_required)
1393 struct dm_target *ti;
1398 ti = dm_table_find_target(ci->map, ci->sector);
1399 if (!dm_target_is_valid(ti))
1403 * Even though the device advertised support for this type of
1404 * request, that does not mean every target supports it, and
1405 * reconfiguration might also have changed that since the
1406 * check was performed.
1408 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1412 if (is_split_required && !is_split_required(ti))
1413 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1415 len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti));
1417 __send_duplicate_bios(ci, ti, num_bios, &len);
1420 } while (ci->sector_count -= len);
1425 static int __send_discard(struct clone_info *ci)
1427 return __send_changing_extent_only(ci, get_num_discard_bios,
1428 is_split_required_for_discard);
1431 static int __send_write_same(struct clone_info *ci)
1433 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1436 static int __send_write_zeroes(struct clone_info *ci)
1438 return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL);
1442 * Select the correct strategy for processing a non-flush bio.
1444 static int __split_and_process_non_flush(struct clone_info *ci)
1446 struct bio *bio = ci->bio;
1447 struct dm_target *ti;
1451 if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
1452 return __send_discard(ci);
1453 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1454 return __send_write_same(ci);
1455 else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES))
1456 return __send_write_zeroes(ci);
1458 ti = dm_table_find_target(ci->map, ci->sector);
1459 if (!dm_target_is_valid(ti))
1462 if (bio_op(bio) == REQ_OP_ZONE_REPORT)
1463 len = ci->sector_count;
1465 len = min_t(sector_t, max_io_len(ci->sector, ti),
1468 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1473 ci->sector_count -= len;
1479 * Entry point to split a bio into clones and submit them to the targets.
1481 static void __split_and_process_bio(struct mapped_device *md,
1482 struct dm_table *map, struct bio *bio)
1484 struct clone_info ci;
1487 if (unlikely(!map)) {
1494 ci.io = alloc_io(md);
1496 atomic_set(&ci.io->io_count, 1);
1499 spin_lock_init(&ci.io->endio_lock);
1500 ci.sector = bio->bi_iter.bi_sector;
1502 start_io_acct(ci.io);
1504 if (bio->bi_opf & REQ_PREFLUSH) {
1505 ci.bio = &ci.md->flush_bio;
1506 ci.sector_count = 0;
1507 error = __send_empty_flush(&ci);
1508 /* dec_pending submits any data associated with flush */
1509 } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1511 ci.sector_count = 0;
1512 error = __split_and_process_non_flush(&ci);
1515 ci.sector_count = bio_sectors(bio);
1516 while (ci.sector_count && !error)
1517 error = __split_and_process_non_flush(&ci);
1520 /* drop the extra reference count */
1521 dec_pending(ci.io, errno_to_blk_status(error));
1523 /*-----------------------------------------------------------------
1525 *---------------------------------------------------------------*/
1528 * The request function that just remaps the bio built up by
1531 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1533 int rw = bio_data_dir(bio);
1534 struct mapped_device *md = q->queuedata;
1536 struct dm_table *map;
1538 map = dm_get_live_table(md, &srcu_idx);
1540 generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0);
1542 /* if we're suspended, we have to queue this io for later */
1543 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1544 dm_put_live_table(md, srcu_idx);
1546 if (!(bio->bi_opf & REQ_RAHEAD))
1550 return BLK_QC_T_NONE;
1553 __split_and_process_bio(md, map, bio);
1554 dm_put_live_table(md, srcu_idx);
1555 return BLK_QC_T_NONE;
1558 static int dm_any_congested(void *congested_data, int bdi_bits)
1561 struct mapped_device *md = congested_data;
1562 struct dm_table *map;
1564 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1565 if (dm_request_based(md)) {
1567 * With request-based DM we only need to check the
1568 * top-level queue for congestion.
1570 r = md->queue->backing_dev_info->wb.state & bdi_bits;
1572 map = dm_get_live_table_fast(md);
1574 r = dm_table_any_congested(map, bdi_bits);
1575 dm_put_live_table_fast(md);
1582 /*-----------------------------------------------------------------
1583 * An IDR is used to keep track of allocated minor numbers.
1584 *---------------------------------------------------------------*/
1585 static void free_minor(int minor)
1587 spin_lock(&_minor_lock);
1588 idr_remove(&_minor_idr, minor);
1589 spin_unlock(&_minor_lock);
1593 * See if the device with a specific minor # is free.
1595 static int specific_minor(int minor)
1599 if (minor >= (1 << MINORBITS))
1602 idr_preload(GFP_KERNEL);
1603 spin_lock(&_minor_lock);
1605 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1607 spin_unlock(&_minor_lock);
1610 return r == -ENOSPC ? -EBUSY : r;
1614 static int next_free_minor(int *minor)
1618 idr_preload(GFP_KERNEL);
1619 spin_lock(&_minor_lock);
1621 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1623 spin_unlock(&_minor_lock);
1631 static const struct block_device_operations dm_blk_dops;
1632 static const struct dax_operations dm_dax_ops;
1634 static void dm_wq_work(struct work_struct *work);
1636 void dm_init_md_queue(struct mapped_device *md)
1639 * Request-based dm devices cannot be stacked on top of bio-based dm
1640 * devices. The type of this dm device may not have been decided yet.
1641 * The type is decided at the first table loading time.
1642 * To prevent problematic device stacking, clear the queue flag
1643 * for request stacking support until then.
1645 * This queue is new, so no concurrency on the queue_flags.
1647 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1650 * Initialize data that will only be used by a non-blk-mq DM queue
1651 * - must do so here (in alloc_dev callchain) before queue is used
1653 md->queue->queuedata = md;
1656 void dm_init_normal_md_queue(struct mapped_device *md)
1658 md->use_blk_mq = false;
1659 dm_init_md_queue(md);
1662 * Initialize aspects of queue that aren't relevant for blk-mq
1664 md->queue->backing_dev_info->congested_data = md;
1665 md->queue->backing_dev_info->congested_fn = dm_any_congested;
1668 static void cleanup_mapped_device(struct mapped_device *md)
1671 destroy_workqueue(md->wq);
1672 if (md->kworker_task)
1673 kthread_stop(md->kworker_task);
1674 mempool_destroy(md->io_pool);
1676 bioset_free(md->bs);
1679 kill_dax(md->dax_dev);
1680 put_dax(md->dax_dev);
1685 spin_lock(&_minor_lock);
1686 md->disk->private_data = NULL;
1687 spin_unlock(&_minor_lock);
1688 del_gendisk(md->disk);
1693 blk_cleanup_queue(md->queue);
1695 cleanup_srcu_struct(&md->io_barrier);
1702 dm_mq_cleanup_mapped_device(md);
1706 * Allocate and initialise a blank device with a given minor.
1708 static struct mapped_device *alloc_dev(int minor)
1710 int r, numa_node_id = dm_get_numa_node();
1711 struct dax_device *dax_dev;
1712 struct mapped_device *md;
1715 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1717 DMWARN("unable to allocate device, out of memory.");
1721 if (!try_module_get(THIS_MODULE))
1722 goto bad_module_get;
1724 /* get a minor number for the dev */
1725 if (minor == DM_ANY_MINOR)
1726 r = next_free_minor(&minor);
1728 r = specific_minor(minor);
1732 r = init_srcu_struct(&md->io_barrier);
1734 goto bad_io_barrier;
1736 md->numa_node_id = numa_node_id;
1737 md->use_blk_mq = dm_use_blk_mq_default();
1738 md->init_tio_pdu = false;
1739 md->type = DM_TYPE_NONE;
1740 mutex_init(&md->suspend_lock);
1741 mutex_init(&md->type_lock);
1742 mutex_init(&md->table_devices_lock);
1743 spin_lock_init(&md->deferred_lock);
1744 atomic_set(&md->holders, 1);
1745 atomic_set(&md->open_count, 0);
1746 atomic_set(&md->event_nr, 0);
1747 atomic_set(&md->uevent_seq, 0);
1748 INIT_LIST_HEAD(&md->uevent_list);
1749 INIT_LIST_HEAD(&md->table_devices);
1750 spin_lock_init(&md->uevent_lock);
1752 md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1756 dm_init_md_queue(md);
1758 * default to bio-based required ->make_request_fn until DM
1759 * table is loaded and md->type established. If request-based
1760 * table is loaded: blk-mq will override accordingly.
1762 blk_queue_make_request(md->queue, dm_make_request);
1764 md->disk = alloc_disk_node(1, numa_node_id);
1768 atomic_set(&md->pending[0], 0);
1769 atomic_set(&md->pending[1], 0);
1770 init_waitqueue_head(&md->wait);
1771 INIT_WORK(&md->work, dm_wq_work);
1772 init_waitqueue_head(&md->eventq);
1773 init_completion(&md->kobj_holder.completion);
1774 md->kworker_task = NULL;
1776 md->disk->major = _major;
1777 md->disk->first_minor = minor;
1778 md->disk->fops = &dm_blk_dops;
1779 md->disk->queue = md->queue;
1780 md->disk->private_data = md;
1781 sprintf(md->disk->disk_name, "dm-%d", minor);
1783 dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops);
1786 md->dax_dev = dax_dev;
1789 format_dev_t(md->name, MKDEV(_major, minor));
1791 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1795 md->bdev = bdget_disk(md->disk, 0);
1799 bio_init(&md->flush_bio, NULL, 0);
1800 bio_set_dev(&md->flush_bio, md->bdev);
1801 md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1803 dm_stats_init(&md->stats);
1805 /* Populate the mapping, nobody knows we exist yet */
1806 spin_lock(&_minor_lock);
1807 old_md = idr_replace(&_minor_idr, md, minor);
1808 spin_unlock(&_minor_lock);
1810 BUG_ON(old_md != MINOR_ALLOCED);
1815 cleanup_mapped_device(md);
1819 module_put(THIS_MODULE);
1825 static void unlock_fs(struct mapped_device *md);
1827 static void free_dev(struct mapped_device *md)
1829 int minor = MINOR(disk_devt(md->disk));
1833 cleanup_mapped_device(md);
1835 free_table_devices(&md->table_devices);
1836 dm_stats_cleanup(&md->stats);
1839 module_put(THIS_MODULE);
1843 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1845 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1848 /* The md already has necessary mempools. */
1849 if (dm_table_bio_based(t)) {
1851 * Reload bioset because front_pad may have changed
1852 * because a different table was loaded.
1854 bioset_free(md->bs);
1859 * There's no need to reload with request-based dm
1860 * because the size of front_pad doesn't change.
1861 * Note for future: If you are to reload bioset,
1862 * prep-ed requests in the queue may refer
1863 * to bio from the old bioset, so you must walk
1864 * through the queue to unprep.
1869 BUG_ON(!p || md->io_pool || md->bs);
1871 md->io_pool = p->io_pool;
1877 /* mempool bind completed, no longer need any mempools in the table */
1878 dm_table_free_md_mempools(t);
1882 * Bind a table to the device.
1884 static void event_callback(void *context)
1886 unsigned long flags;
1888 struct mapped_device *md = (struct mapped_device *) context;
1890 spin_lock_irqsave(&md->uevent_lock, flags);
1891 list_splice_init(&md->uevent_list, &uevents);
1892 spin_unlock_irqrestore(&md->uevent_lock, flags);
1894 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1896 atomic_inc(&md->event_nr);
1897 wake_up(&md->eventq);
1898 dm_issue_global_event();
1902 * Protected by md->suspend_lock obtained by dm_swap_table().
1904 static void __set_size(struct mapped_device *md, sector_t size)
1906 lockdep_assert_held(&md->suspend_lock);
1908 set_capacity(md->disk, size);
1910 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1914 * Returns old map, which caller must destroy.
1916 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1917 struct queue_limits *limits)
1919 struct dm_table *old_map;
1920 struct request_queue *q = md->queue;
1923 lockdep_assert_held(&md->suspend_lock);
1925 size = dm_table_get_size(t);
1928 * Wipe any geometry if the size of the table changed.
1930 if (size != dm_get_size(md))
1931 memset(&md->geometry, 0, sizeof(md->geometry));
1933 __set_size(md, size);
1935 dm_table_event_callback(t, event_callback, md);
1938 * The queue hasn't been stopped yet, if the old table type wasn't
1939 * for request-based during suspension. So stop it to prevent
1940 * I/O mapping before resume.
1941 * This must be done before setting the queue restrictions,
1942 * because request-based dm may be run just after the setting.
1944 if (dm_table_request_based(t)) {
1947 * Leverage the fact that request-based DM targets are
1948 * immutable singletons and establish md->immutable_target
1949 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1951 md->immutable_target = dm_table_get_immutable_target(t);
1954 __bind_mempools(md, t);
1956 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1957 rcu_assign_pointer(md->map, (void *)t);
1958 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1960 dm_table_set_restrictions(t, q, limits);
1968 * Returns unbound table for the caller to free.
1970 static struct dm_table *__unbind(struct mapped_device *md)
1972 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1977 dm_table_event_callback(map, NULL, NULL);
1978 RCU_INIT_POINTER(md->map, NULL);
1985 * Constructor for a new device.
1987 int dm_create(int minor, struct mapped_device **result)
1989 struct mapped_device *md;
1991 md = alloc_dev(minor);
2002 * Functions to manage md->type.
2003 * All are required to hold md->type_lock.
2005 void dm_lock_md_type(struct mapped_device *md)
2007 mutex_lock(&md->type_lock);
2010 void dm_unlock_md_type(struct mapped_device *md)
2012 mutex_unlock(&md->type_lock);
2015 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2017 BUG_ON(!mutex_is_locked(&md->type_lock));
2021 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2026 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2028 return md->immutable_target_type;
2032 * The queue_limits are only valid as long as you have a reference
2035 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2037 BUG_ON(!atomic_read(&md->holders));
2038 return &md->queue->limits;
2040 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2043 * Setup the DM device's queue based on md's type
2045 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2048 enum dm_queue_mode type = dm_get_md_type(md);
2051 case DM_TYPE_REQUEST_BASED:
2052 r = dm_old_init_request_queue(md, t);
2054 DMERR("Cannot initialize queue for request-based mapped device");
2058 case DM_TYPE_MQ_REQUEST_BASED:
2059 r = dm_mq_init_request_queue(md, t);
2061 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2065 case DM_TYPE_BIO_BASED:
2066 case DM_TYPE_DAX_BIO_BASED:
2067 dm_init_normal_md_queue(md);
2069 * DM handles splitting bios as needed. Free the bio_split bioset
2070 * since it won't be used (saves 1 process per bio-based DM device).
2072 bioset_free(md->queue->bio_split);
2073 md->queue->bio_split = NULL;
2083 struct mapped_device *dm_get_md(dev_t dev)
2085 struct mapped_device *md;
2086 unsigned minor = MINOR(dev);
2088 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2091 spin_lock(&_minor_lock);
2093 md = idr_find(&_minor_idr, minor);
2095 if ((md == MINOR_ALLOCED ||
2096 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2097 dm_deleting_md(md) ||
2098 test_bit(DMF_FREEING, &md->flags))) {
2106 spin_unlock(&_minor_lock);
2110 EXPORT_SYMBOL_GPL(dm_get_md);
2112 void *dm_get_mdptr(struct mapped_device *md)
2114 return md->interface_ptr;
2117 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2119 md->interface_ptr = ptr;
2122 void dm_get(struct mapped_device *md)
2124 atomic_inc(&md->holders);
2125 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2128 int dm_hold(struct mapped_device *md)
2130 spin_lock(&_minor_lock);
2131 if (test_bit(DMF_FREEING, &md->flags)) {
2132 spin_unlock(&_minor_lock);
2136 spin_unlock(&_minor_lock);
2139 EXPORT_SYMBOL_GPL(dm_hold);
2141 const char *dm_device_name(struct mapped_device *md)
2145 EXPORT_SYMBOL_GPL(dm_device_name);
2147 static void __dm_destroy(struct mapped_device *md, bool wait)
2149 struct request_queue *q = dm_get_md_queue(md);
2150 struct dm_table *map;
2155 spin_lock(&_minor_lock);
2156 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2157 set_bit(DMF_FREEING, &md->flags);
2158 spin_unlock(&_minor_lock);
2160 blk_set_queue_dying(q);
2162 if (dm_request_based(md) && md->kworker_task)
2163 kthread_flush_worker(&md->kworker);
2166 * Take suspend_lock so that presuspend and postsuspend methods
2167 * do not race with internal suspend.
2169 mutex_lock(&md->suspend_lock);
2170 map = dm_get_live_table(md, &srcu_idx);
2171 if (!dm_suspended_md(md)) {
2172 dm_table_presuspend_targets(map);
2173 dm_table_postsuspend_targets(map);
2175 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2176 dm_put_live_table(md, srcu_idx);
2177 mutex_unlock(&md->suspend_lock);
2180 * Rare, but there may be I/O requests still going to complete,
2181 * for example. Wait for all references to disappear.
2182 * No one should increment the reference count of the mapped_device,
2183 * after the mapped_device state becomes DMF_FREEING.
2186 while (atomic_read(&md->holders))
2188 else if (atomic_read(&md->holders))
2189 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2190 dm_device_name(md), atomic_read(&md->holders));
2193 dm_table_destroy(__unbind(md));
2197 void dm_destroy(struct mapped_device *md)
2199 __dm_destroy(md, true);
2202 void dm_destroy_immediate(struct mapped_device *md)
2204 __dm_destroy(md, false);
2207 void dm_put(struct mapped_device *md)
2209 atomic_dec(&md->holders);
2211 EXPORT_SYMBOL_GPL(dm_put);
2213 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2219 prepare_to_wait(&md->wait, &wait, task_state);
2221 if (!md_in_flight(md))
2224 if (signal_pending_state(task_state, current)) {
2231 finish_wait(&md->wait, &wait);
2233 smp_rmb(); /* paired with atomic_dec_return in end_io_acct */
2239 * Process the deferred bios
2241 static void dm_wq_work(struct work_struct *work)
2243 struct mapped_device *md = container_of(work, struct mapped_device,
2247 struct dm_table *map;
2249 map = dm_get_live_table(md, &srcu_idx);
2251 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2252 spin_lock_irq(&md->deferred_lock);
2253 c = bio_list_pop(&md->deferred);
2254 spin_unlock_irq(&md->deferred_lock);
2259 if (dm_request_based(md))
2260 generic_make_request(c);
2262 __split_and_process_bio(md, map, c);
2265 dm_put_live_table(md, srcu_idx);
2268 static void dm_queue_flush(struct mapped_device *md)
2270 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2271 smp_mb__after_atomic();
2272 queue_work(md->wq, &md->work);
2276 * Swap in a new table, returning the old one for the caller to destroy.
2278 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2280 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2281 struct queue_limits limits;
2284 mutex_lock(&md->suspend_lock);
2286 /* device must be suspended */
2287 if (!dm_suspended_md(md))
2291 * If the new table has no data devices, retain the existing limits.
2292 * This helps multipath with queue_if_no_path if all paths disappear,
2293 * then new I/O is queued based on these limits, and then some paths
2296 if (dm_table_has_no_data_devices(table)) {
2297 live_map = dm_get_live_table_fast(md);
2299 limits = md->queue->limits;
2300 dm_put_live_table_fast(md);
2304 r = dm_calculate_queue_limits(table, &limits);
2311 map = __bind(md, table, &limits);
2312 dm_issue_global_event();
2315 mutex_unlock(&md->suspend_lock);
2320 * Functions to lock and unlock any filesystem running on the
2323 static int lock_fs(struct mapped_device *md)
2327 WARN_ON(md->frozen_sb);
2329 md->frozen_sb = freeze_bdev(md->bdev);
2330 if (IS_ERR(md->frozen_sb)) {
2331 r = PTR_ERR(md->frozen_sb);
2332 md->frozen_sb = NULL;
2336 set_bit(DMF_FROZEN, &md->flags);
2341 static void unlock_fs(struct mapped_device *md)
2343 if (!test_bit(DMF_FROZEN, &md->flags))
2346 thaw_bdev(md->bdev, md->frozen_sb);
2347 md->frozen_sb = NULL;
2348 clear_bit(DMF_FROZEN, &md->flags);
2352 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2353 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2354 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2356 * If __dm_suspend returns 0, the device is completely quiescent
2357 * now. There is no request-processing activity. All new requests
2358 * are being added to md->deferred list.
2360 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2361 unsigned suspend_flags, long task_state,
2362 int dmf_suspended_flag)
2364 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2365 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2368 lockdep_assert_held(&md->suspend_lock);
2371 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2372 * This flag is cleared before dm_suspend returns.
2375 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2377 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2380 * This gets reverted if there's an error later and the targets
2381 * provide the .presuspend_undo hook.
2383 dm_table_presuspend_targets(map);
2386 * Flush I/O to the device.
2387 * Any I/O submitted after lock_fs() may not be flushed.
2388 * noflush takes precedence over do_lockfs.
2389 * (lock_fs() flushes I/Os and waits for them to complete.)
2391 if (!noflush && do_lockfs) {
2394 dm_table_presuspend_undo_targets(map);
2400 * Here we must make sure that no processes are submitting requests
2401 * to target drivers i.e. no one may be executing
2402 * __split_and_process_bio. This is called from dm_request and
2405 * To get all processes out of __split_and_process_bio in dm_request,
2406 * we take the write lock. To prevent any process from reentering
2407 * __split_and_process_bio from dm_request and quiesce the thread
2408 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2409 * flush_workqueue(md->wq).
2411 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2413 synchronize_srcu(&md->io_barrier);
2416 * Stop md->queue before flushing md->wq in case request-based
2417 * dm defers requests to md->wq from md->queue.
2419 if (dm_request_based(md)) {
2420 dm_stop_queue(md->queue);
2421 if (md->kworker_task)
2422 kthread_flush_worker(&md->kworker);
2425 flush_workqueue(md->wq);
2428 * At this point no more requests are entering target request routines.
2429 * We call dm_wait_for_completion to wait for all existing requests
2432 r = dm_wait_for_completion(md, task_state);
2434 set_bit(dmf_suspended_flag, &md->flags);
2437 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2439 synchronize_srcu(&md->io_barrier);
2441 /* were we interrupted ? */
2445 if (dm_request_based(md))
2446 dm_start_queue(md->queue);
2449 dm_table_presuspend_undo_targets(map);
2450 /* pushback list is already flushed, so skip flush */
2457 * We need to be able to change a mapping table under a mounted
2458 * filesystem. For example we might want to move some data in
2459 * the background. Before the table can be swapped with
2460 * dm_bind_table, dm_suspend must be called to flush any in
2461 * flight bios and ensure that any further io gets deferred.
2464 * Suspend mechanism in request-based dm.
2466 * 1. Flush all I/Os by lock_fs() if needed.
2467 * 2. Stop dispatching any I/O by stopping the request_queue.
2468 * 3. Wait for all in-flight I/Os to be completed or requeued.
2470 * To abort suspend, start the request_queue.
2472 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2474 struct dm_table *map = NULL;
2478 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2480 if (dm_suspended_md(md)) {
2485 if (dm_suspended_internally_md(md)) {
2486 /* already internally suspended, wait for internal resume */
2487 mutex_unlock(&md->suspend_lock);
2488 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2494 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2496 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2500 dm_table_postsuspend_targets(map);
2503 mutex_unlock(&md->suspend_lock);
2507 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2510 int r = dm_table_resume_targets(map);
2518 * Flushing deferred I/Os must be done after targets are resumed
2519 * so that mapping of targets can work correctly.
2520 * Request-based dm is queueing the deferred I/Os in its request_queue.
2522 if (dm_request_based(md))
2523 dm_start_queue(md->queue);
2530 int dm_resume(struct mapped_device *md)
2533 struct dm_table *map = NULL;
2537 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2539 if (!dm_suspended_md(md))
2542 if (dm_suspended_internally_md(md)) {
2543 /* already internally suspended, wait for internal resume */
2544 mutex_unlock(&md->suspend_lock);
2545 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2551 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2552 if (!map || !dm_table_get_size(map))
2555 r = __dm_resume(md, map);
2559 clear_bit(DMF_SUSPENDED, &md->flags);
2561 mutex_unlock(&md->suspend_lock);
2567 * Internal suspend/resume works like userspace-driven suspend. It waits
2568 * until all bios finish and prevents issuing new bios to the target drivers.
2569 * It may be used only from the kernel.
2572 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2574 struct dm_table *map = NULL;
2576 lockdep_assert_held(&md->suspend_lock);
2578 if (md->internal_suspend_count++)
2579 return; /* nested internal suspend */
2581 if (dm_suspended_md(md)) {
2582 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2583 return; /* nest suspend */
2586 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2589 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2590 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2591 * would require changing .presuspend to return an error -- avoid this
2592 * until there is a need for more elaborate variants of internal suspend.
2594 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2595 DMF_SUSPENDED_INTERNALLY);
2597 dm_table_postsuspend_targets(map);
2600 static void __dm_internal_resume(struct mapped_device *md)
2602 BUG_ON(!md->internal_suspend_count);
2604 if (--md->internal_suspend_count)
2605 return; /* resume from nested internal suspend */
2607 if (dm_suspended_md(md))
2608 goto done; /* resume from nested suspend */
2611 * NOTE: existing callers don't need to call dm_table_resume_targets
2612 * (which may fail -- so best to avoid it for now by passing NULL map)
2614 (void) __dm_resume(md, NULL);
2617 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2618 smp_mb__after_atomic();
2619 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2622 void dm_internal_suspend_noflush(struct mapped_device *md)
2624 mutex_lock(&md->suspend_lock);
2625 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2626 mutex_unlock(&md->suspend_lock);
2628 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2630 void dm_internal_resume(struct mapped_device *md)
2632 mutex_lock(&md->suspend_lock);
2633 __dm_internal_resume(md);
2634 mutex_unlock(&md->suspend_lock);
2636 EXPORT_SYMBOL_GPL(dm_internal_resume);
2639 * Fast variants of internal suspend/resume hold md->suspend_lock,
2640 * which prevents interaction with userspace-driven suspend.
2643 void dm_internal_suspend_fast(struct mapped_device *md)
2645 mutex_lock(&md->suspend_lock);
2646 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2649 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2650 synchronize_srcu(&md->io_barrier);
2651 flush_workqueue(md->wq);
2652 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2654 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2656 void dm_internal_resume_fast(struct mapped_device *md)
2658 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2664 mutex_unlock(&md->suspend_lock);
2666 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2668 /*-----------------------------------------------------------------
2669 * Event notification.
2670 *---------------------------------------------------------------*/
2671 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2676 char udev_cookie[DM_COOKIE_LENGTH];
2677 char *envp[] = { udev_cookie, NULL };
2679 noio_flag = memalloc_noio_save();
2682 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2684 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2685 DM_COOKIE_ENV_VAR_NAME, cookie);
2686 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2690 memalloc_noio_restore(noio_flag);
2695 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2697 return atomic_add_return(1, &md->uevent_seq);
2700 uint32_t dm_get_event_nr(struct mapped_device *md)
2702 return atomic_read(&md->event_nr);
2705 int dm_wait_event(struct mapped_device *md, int event_nr)
2707 return wait_event_interruptible(md->eventq,
2708 (event_nr != atomic_read(&md->event_nr)));
2711 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2713 unsigned long flags;
2715 spin_lock_irqsave(&md->uevent_lock, flags);
2716 list_add(elist, &md->uevent_list);
2717 spin_unlock_irqrestore(&md->uevent_lock, flags);
2721 * The gendisk is only valid as long as you have a reference
2724 struct gendisk *dm_disk(struct mapped_device *md)
2728 EXPORT_SYMBOL_GPL(dm_disk);
2730 struct kobject *dm_kobject(struct mapped_device *md)
2732 return &md->kobj_holder.kobj;
2735 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2737 struct mapped_device *md;
2739 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2741 spin_lock(&_minor_lock);
2742 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2748 spin_unlock(&_minor_lock);
2753 int dm_suspended_md(struct mapped_device *md)
2755 return test_bit(DMF_SUSPENDED, &md->flags);
2758 int dm_suspended_internally_md(struct mapped_device *md)
2760 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2763 int dm_test_deferred_remove_flag(struct mapped_device *md)
2765 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2768 int dm_suspended(struct dm_target *ti)
2770 return dm_suspended_md(dm_table_get_md(ti->table));
2772 EXPORT_SYMBOL_GPL(dm_suspended);
2774 int dm_noflush_suspending(struct dm_target *ti)
2776 return __noflush_suspending(dm_table_get_md(ti->table));
2778 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2780 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2781 unsigned integrity, unsigned per_io_data_size)
2783 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2784 unsigned int pool_size = 0;
2785 unsigned int front_pad;
2791 case DM_TYPE_BIO_BASED:
2792 case DM_TYPE_DAX_BIO_BASED:
2793 pool_size = dm_get_reserved_bio_based_ios();
2794 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2796 pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache);
2797 if (!pools->io_pool)
2800 case DM_TYPE_REQUEST_BASED:
2801 case DM_TYPE_MQ_REQUEST_BASED:
2802 pool_size = dm_get_reserved_rq_based_ios();
2803 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2804 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2810 pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER);
2814 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2820 dm_free_md_mempools(pools);
2825 void dm_free_md_mempools(struct dm_md_mempools *pools)
2830 mempool_destroy(pools->io_pool);
2833 bioset_free(pools->bs);
2845 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2848 struct mapped_device *md = bdev->bd_disk->private_data;
2849 struct dm_table *table;
2850 struct dm_target *ti;
2851 int ret = -ENOTTY, srcu_idx;
2853 table = dm_get_live_table(md, &srcu_idx);
2854 if (!table || !dm_table_get_size(table))
2857 /* We only support devices that have a single target */
2858 if (dm_table_get_num_targets(table) != 1)
2860 ti = dm_table_get_target(table, 0);
2862 if (dm_suspended_md(md)) {
2868 if (!ti->type->iterate_devices)
2871 ret = ti->type->iterate_devices(ti, fn, data);
2873 dm_put_live_table(md, srcu_idx);
2878 * For register / unregister we need to manually call out to every path.
2880 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2881 sector_t start, sector_t len, void *data)
2883 struct dm_pr *pr = data;
2884 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2886 if (!ops || !ops->pr_register)
2888 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2891 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2902 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2903 if (ret && new_key) {
2904 /* unregister all paths if we failed to register any path */
2905 pr.old_key = new_key;
2908 pr.fail_early = false;
2909 dm_call_pr(bdev, __dm_pr_register, &pr);
2915 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2918 struct mapped_device *md = bdev->bd_disk->private_data;
2919 const struct pr_ops *ops;
2923 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2927 ops = bdev->bd_disk->fops->pr_ops;
2928 if (ops && ops->pr_reserve)
2929 r = ops->pr_reserve(bdev, key, type, flags);
2937 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2939 struct mapped_device *md = bdev->bd_disk->private_data;
2940 const struct pr_ops *ops;
2944 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2948 ops = bdev->bd_disk->fops->pr_ops;
2949 if (ops && ops->pr_release)
2950 r = ops->pr_release(bdev, key, type);
2958 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2959 enum pr_type type, bool abort)
2961 struct mapped_device *md = bdev->bd_disk->private_data;
2962 const struct pr_ops *ops;
2966 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2970 ops = bdev->bd_disk->fops->pr_ops;
2971 if (ops && ops->pr_preempt)
2972 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
2980 static int dm_pr_clear(struct block_device *bdev, u64 key)
2982 struct mapped_device *md = bdev->bd_disk->private_data;
2983 const struct pr_ops *ops;
2987 r = dm_grab_bdev_for_ioctl(md, &bdev, &mode);
2991 ops = bdev->bd_disk->fops->pr_ops;
2992 if (ops && ops->pr_clear)
2993 r = ops->pr_clear(bdev, key);
3001 static const struct pr_ops dm_pr_ops = {
3002 .pr_register = dm_pr_register,
3003 .pr_reserve = dm_pr_reserve,
3004 .pr_release = dm_pr_release,
3005 .pr_preempt = dm_pr_preempt,
3006 .pr_clear = dm_pr_clear,
3009 static const struct block_device_operations dm_blk_dops = {
3010 .open = dm_blk_open,
3011 .release = dm_blk_close,
3012 .ioctl = dm_blk_ioctl,
3013 .getgeo = dm_blk_getgeo,
3014 .pr_ops = &dm_pr_ops,
3015 .owner = THIS_MODULE
3018 static const struct dax_operations dm_dax_ops = {
3019 .direct_access = dm_dax_direct_access,
3020 .copy_from_iter = dm_dax_copy_from_iter,
3026 module_init(dm_init);
3027 module_exit(dm_exit);
3029 module_param(major, uint, 0);
3030 MODULE_PARM_DESC(major, "The major number of the device mapper");
3032 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3033 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3035 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3036 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3038 MODULE_DESCRIPTION(DM_NAME " driver");
3039 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3040 MODULE_LICENSE("GPL");