2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
23 #include <linux/wait.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
33 DEFAULT_RATELIMIT_INTERVAL
,
34 DEFAULT_RATELIMIT_BURST
);
35 EXPORT_SYMBOL(dm_ratelimit_state
);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name
= DM_NAME
;
47 static unsigned int major
= 0;
48 static unsigned int _major
= 0;
50 static DEFINE_IDR(_minor_idr
);
52 static DEFINE_SPINLOCK(_minor_lock
);
54 static void do_deferred_remove(struct work_struct
*w
);
56 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
58 static struct workqueue_struct
*deferred_remove_workqueue
;
61 * One of these is allocated per bio.
64 struct mapped_device
*md
;
68 unsigned long start_time
;
69 spinlock_t endio_lock
;
70 struct dm_stats_aux stats_aux
;
73 #define MINOR_ALLOCED ((void *)-1)
76 * Bits for the md->flags field.
78 #define DMF_BLOCK_IO_FOR_SUSPEND 0
79 #define DMF_SUSPENDED 1
82 #define DMF_DELETING 4
83 #define DMF_NOFLUSH_SUSPENDING 5
84 #define DMF_DEFERRED_REMOVE 6
85 #define DMF_SUSPENDED_INTERNALLY 7
87 #define DM_NUMA_NODE NUMA_NO_NODE
88 static int dm_numa_node
= DM_NUMA_NODE
;
91 * For mempools pre-allocation at the table loading time.
93 struct dm_md_mempools
{
99 struct list_head list
;
101 struct dm_dev dm_dev
;
104 static struct kmem_cache
*_io_cache
;
105 static struct kmem_cache
*_rq_tio_cache
;
106 static struct kmem_cache
*_rq_cache
;
109 * Bio-based DM's mempools' reserved IOs set by the user.
111 #define RESERVED_BIO_BASED_IOS 16
112 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
114 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
116 int param
= ACCESS_ONCE(*module_param
);
117 int modified_param
= 0;
118 bool modified
= true;
121 modified_param
= min
;
122 else if (param
> max
)
123 modified_param
= max
;
128 (void)cmpxchg(module_param
, param
, modified_param
);
129 param
= modified_param
;
135 unsigned __dm_get_module_param(unsigned *module_param
,
136 unsigned def
, unsigned max
)
138 unsigned param
= ACCESS_ONCE(*module_param
);
139 unsigned modified_param
= 0;
142 modified_param
= def
;
143 else if (param
> max
)
144 modified_param
= max
;
146 if (modified_param
) {
147 (void)cmpxchg(module_param
, param
, modified_param
);
148 param
= modified_param
;
154 unsigned dm_get_reserved_bio_based_ios(void)
156 return __dm_get_module_param(&reserved_bio_based_ios
,
157 RESERVED_BIO_BASED_IOS
, DM_RESERVED_MAX_IOS
);
159 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
161 static unsigned dm_get_numa_node(void)
163 return __dm_get_module_param_int(&dm_numa_node
,
164 DM_NUMA_NODE
, num_online_nodes() - 1);
167 static int __init
local_init(void)
171 /* allocate a slab for the dm_ios */
172 _io_cache
= KMEM_CACHE(dm_io
, 0);
176 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
178 goto out_free_io_cache
;
180 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
181 __alignof__(struct request
), 0, NULL
);
183 goto out_free_rq_tio_cache
;
185 r
= dm_uevent_init();
187 goto out_free_rq_cache
;
189 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
190 if (!deferred_remove_workqueue
) {
192 goto out_uevent_exit
;
196 r
= register_blkdev(_major
, _name
);
198 goto out_free_workqueue
;
206 destroy_workqueue(deferred_remove_workqueue
);
210 kmem_cache_destroy(_rq_cache
);
211 out_free_rq_tio_cache
:
212 kmem_cache_destroy(_rq_tio_cache
);
214 kmem_cache_destroy(_io_cache
);
219 static void local_exit(void)
221 flush_scheduled_work();
222 destroy_workqueue(deferred_remove_workqueue
);
224 kmem_cache_destroy(_rq_cache
);
225 kmem_cache_destroy(_rq_tio_cache
);
226 kmem_cache_destroy(_io_cache
);
227 unregister_blkdev(_major
, _name
);
232 DMINFO("cleaned up");
235 static int (*_inits
[])(void) __initdata
= {
246 static void (*_exits
[])(void) = {
257 static int __init
dm_init(void)
259 const int count
= ARRAY_SIZE(_inits
);
263 for (i
= 0; i
< count
; i
++) {
278 static void __exit
dm_exit(void)
280 int i
= ARRAY_SIZE(_exits
);
286 * Should be empty by this point.
288 idr_destroy(&_minor_idr
);
292 * Block device functions
294 int dm_deleting_md(struct mapped_device
*md
)
296 return test_bit(DMF_DELETING
, &md
->flags
);
299 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
301 struct mapped_device
*md
;
303 spin_lock(&_minor_lock
);
305 md
= bdev
->bd_disk
->private_data
;
309 if (test_bit(DMF_FREEING
, &md
->flags
) ||
310 dm_deleting_md(md
)) {
316 atomic_inc(&md
->open_count
);
318 spin_unlock(&_minor_lock
);
320 return md
? 0 : -ENXIO
;
323 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
325 struct mapped_device
*md
;
327 spin_lock(&_minor_lock
);
329 md
= disk
->private_data
;
333 if (atomic_dec_and_test(&md
->open_count
) &&
334 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
335 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
339 spin_unlock(&_minor_lock
);
342 int dm_open_count(struct mapped_device
*md
)
344 return atomic_read(&md
->open_count
);
348 * Guarantees nothing is using the device before it's deleted.
350 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
354 spin_lock(&_minor_lock
);
356 if (dm_open_count(md
)) {
359 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
360 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
363 set_bit(DMF_DELETING
, &md
->flags
);
365 spin_unlock(&_minor_lock
);
370 int dm_cancel_deferred_remove(struct mapped_device
*md
)
374 spin_lock(&_minor_lock
);
376 if (test_bit(DMF_DELETING
, &md
->flags
))
379 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
381 spin_unlock(&_minor_lock
);
386 static void do_deferred_remove(struct work_struct
*w
)
388 dm_deferred_remove();
391 sector_t
dm_get_size(struct mapped_device
*md
)
393 return get_capacity(md
->disk
);
396 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
401 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
406 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
408 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
410 return dm_get_geometry(md
, geo
);
413 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
414 struct block_device
**bdev
,
417 struct dm_target
*tgt
;
418 struct dm_table
*map
;
423 map
= dm_get_live_table(md
, &srcu_idx
);
424 if (!map
|| !dm_table_get_size(map
))
427 /* We only support devices that have a single target */
428 if (dm_table_get_num_targets(map
) != 1)
431 tgt
= dm_table_get_target(map
, 0);
432 if (!tgt
->type
->prepare_ioctl
)
435 if (dm_suspended_md(md
)) {
440 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
445 dm_put_live_table(md
, srcu_idx
);
449 dm_put_live_table(md
, srcu_idx
);
450 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
457 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
458 unsigned int cmd
, unsigned long arg
)
460 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
463 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
469 * Target determined this ioctl is being issued against a
470 * subset of the parent bdev; require extra privileges.
472 if (!capable(CAP_SYS_RAWIO
)) {
474 "%s: sending ioctl %x to DM device without required privilege.",
481 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
487 static struct dm_io
*alloc_io(struct mapped_device
*md
)
489 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
492 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
494 mempool_free(io
, md
->io_pool
);
497 static void free_tio(struct dm_target_io
*tio
)
499 bio_put(&tio
->clone
);
502 int md_in_flight(struct mapped_device
*md
)
504 return atomic_read(&md
->pending
[READ
]) +
505 atomic_read(&md
->pending
[WRITE
]);
508 static void start_io_acct(struct dm_io
*io
)
510 struct mapped_device
*md
= io
->md
;
511 struct bio
*bio
= io
->bio
;
513 int rw
= bio_data_dir(bio
);
515 io
->start_time
= jiffies
;
517 cpu
= part_stat_lock();
518 part_round_stats(cpu
, &dm_disk(md
)->part0
);
520 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
521 atomic_inc_return(&md
->pending
[rw
]));
523 if (unlikely(dm_stats_used(&md
->stats
)))
524 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
525 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
526 false, 0, &io
->stats_aux
);
529 static void end_io_acct(struct dm_io
*io
)
531 struct mapped_device
*md
= io
->md
;
532 struct bio
*bio
= io
->bio
;
533 unsigned long duration
= jiffies
- io
->start_time
;
535 int rw
= bio_data_dir(bio
);
537 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
539 if (unlikely(dm_stats_used(&md
->stats
)))
540 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
541 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
542 true, duration
, &io
->stats_aux
);
545 * After this is decremented the bio must not be touched if it is
548 pending
= atomic_dec_return(&md
->pending
[rw
]);
549 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
550 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
552 /* nudge anyone waiting on suspend queue */
558 * Add the bio to the list of deferred io.
560 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
564 spin_lock_irqsave(&md
->deferred_lock
, flags
);
565 bio_list_add(&md
->deferred
, bio
);
566 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
567 queue_work(md
->wq
, &md
->work
);
571 * Everyone (including functions in this file), should use this
572 * function to access the md->map field, and make sure they call
573 * dm_put_live_table() when finished.
575 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
577 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
579 return srcu_dereference(md
->map
, &md
->io_barrier
);
582 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
584 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
587 void dm_sync_table(struct mapped_device
*md
)
589 synchronize_srcu(&md
->io_barrier
);
590 synchronize_rcu_expedited();
594 * A fast alternative to dm_get_live_table/dm_put_live_table.
595 * The caller must not block between these two functions.
597 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
600 return rcu_dereference(md
->map
);
603 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
609 * Open a table device so we can use it as a map destination.
611 static int open_table_device(struct table_device
*td
, dev_t dev
,
612 struct mapped_device
*md
)
614 static char *_claim_ptr
= "I belong to device-mapper";
615 struct block_device
*bdev
;
619 BUG_ON(td
->dm_dev
.bdev
);
621 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
623 return PTR_ERR(bdev
);
625 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
627 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
631 td
->dm_dev
.bdev
= bdev
;
636 * Close a table device that we've been using.
638 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
640 if (!td
->dm_dev
.bdev
)
643 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
644 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
645 td
->dm_dev
.bdev
= NULL
;
648 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
650 struct table_device
*td
;
652 list_for_each_entry(td
, l
, list
)
653 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
659 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
660 struct dm_dev
**result
) {
662 struct table_device
*td
;
664 mutex_lock(&md
->table_devices_lock
);
665 td
= find_table_device(&md
->table_devices
, dev
, mode
);
667 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
669 mutex_unlock(&md
->table_devices_lock
);
673 td
->dm_dev
.mode
= mode
;
674 td
->dm_dev
.bdev
= NULL
;
676 if ((r
= open_table_device(td
, dev
, md
))) {
677 mutex_unlock(&md
->table_devices_lock
);
682 format_dev_t(td
->dm_dev
.name
, dev
);
684 atomic_set(&td
->count
, 0);
685 list_add(&td
->list
, &md
->table_devices
);
687 atomic_inc(&td
->count
);
688 mutex_unlock(&md
->table_devices_lock
);
690 *result
= &td
->dm_dev
;
693 EXPORT_SYMBOL_GPL(dm_get_table_device
);
695 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
697 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
699 mutex_lock(&md
->table_devices_lock
);
700 if (atomic_dec_and_test(&td
->count
)) {
701 close_table_device(td
, md
);
705 mutex_unlock(&md
->table_devices_lock
);
707 EXPORT_SYMBOL(dm_put_table_device
);
709 static void free_table_devices(struct list_head
*devices
)
711 struct list_head
*tmp
, *next
;
713 list_for_each_safe(tmp
, next
, devices
) {
714 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
716 DMWARN("dm_destroy: %s still exists with %d references",
717 td
->dm_dev
.name
, atomic_read(&td
->count
));
723 * Get the geometry associated with a dm device
725 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
733 * Set the geometry of a device.
735 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
737 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
739 if (geo
->start
> sz
) {
740 DMWARN("Start sector is beyond the geometry limits.");
749 /*-----------------------------------------------------------------
751 * A more elegant soln is in the works that uses the queue
752 * merge fn, unfortunately there are a couple of changes to
753 * the block layer that I want to make for this. So in the
754 * interests of getting something for people to use I give
755 * you this clearly demarcated crap.
756 *---------------------------------------------------------------*/
758 static int __noflush_suspending(struct mapped_device
*md
)
760 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
764 * Decrements the number of outstanding ios that a bio has been
765 * cloned into, completing the original io if necc.
767 static void dec_pending(struct dm_io
*io
, int error
)
772 struct mapped_device
*md
= io
->md
;
774 /* Push-back supersedes any I/O errors */
775 if (unlikely(error
)) {
776 spin_lock_irqsave(&io
->endio_lock
, flags
);
777 if (!(io
->error
> 0 && __noflush_suspending(md
)))
779 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
782 if (atomic_dec_and_test(&io
->io_count
)) {
783 if (io
->error
== DM_ENDIO_REQUEUE
) {
785 * Target requested pushing back the I/O.
787 spin_lock_irqsave(&md
->deferred_lock
, flags
);
788 if (__noflush_suspending(md
))
789 bio_list_add_head(&md
->deferred
, io
->bio
);
791 /* noflush suspend was interrupted. */
793 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
796 io_error
= io
->error
;
801 if (io_error
== DM_ENDIO_REQUEUE
)
804 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
806 * Preflush done for flush with data, reissue
807 * without REQ_PREFLUSH.
809 bio
->bi_opf
&= ~REQ_PREFLUSH
;
812 /* done with normal IO or empty flush */
813 bio
->bi_error
= io_error
;
819 void disable_write_same(struct mapped_device
*md
)
821 struct queue_limits
*limits
= dm_get_queue_limits(md
);
823 /* device doesn't really support WRITE SAME, disable it */
824 limits
->max_write_same_sectors
= 0;
827 void disable_write_zeroes(struct mapped_device
*md
)
829 struct queue_limits
*limits
= dm_get_queue_limits(md
);
831 /* device doesn't really support WRITE ZEROES, disable it */
832 limits
->max_write_zeroes_sectors
= 0;
835 static void clone_endio(struct bio
*bio
)
837 int error
= bio
->bi_error
;
839 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
840 struct dm_io
*io
= tio
->io
;
841 struct mapped_device
*md
= tio
->io
->md
;
842 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
845 r
= endio(tio
->ti
, bio
, error
);
846 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
848 * error and requeue request are handled
852 else if (r
== DM_ENDIO_INCOMPLETE
)
853 /* The target will handle the io */
856 DMWARN("unimplemented target endio return value: %d", r
);
861 if (unlikely(r
== -EREMOTEIO
)) {
862 if (bio_op(bio
) == REQ_OP_WRITE_SAME
&&
863 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
)
864 disable_write_same(md
);
865 if (bio_op(bio
) == REQ_OP_WRITE_ZEROES
&&
866 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_zeroes_sectors
)
867 disable_write_zeroes(md
);
871 dec_pending(io
, error
);
875 * Return maximum size of I/O possible at the supplied sector up to the current
878 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
880 sector_t target_offset
= dm_target_offset(ti
, sector
);
882 return ti
->len
- target_offset
;
885 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
887 sector_t len
= max_io_len_target_boundary(sector
, ti
);
888 sector_t offset
, max_len
;
891 * Does the target need to split even further?
893 if (ti
->max_io_len
) {
894 offset
= dm_target_offset(ti
, sector
);
895 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
896 max_len
= sector_div(offset
, ti
->max_io_len
);
898 max_len
= offset
& (ti
->max_io_len
- 1);
899 max_len
= ti
->max_io_len
- max_len
;
908 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
910 if (len
> UINT_MAX
) {
911 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
912 (unsigned long long)len
, UINT_MAX
);
913 ti
->error
= "Maximum size of target IO is too large";
917 ti
->max_io_len
= (uint32_t) len
;
921 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
923 static long dm_blk_direct_access(struct block_device
*bdev
, sector_t sector
,
924 void **kaddr
, pfn_t
*pfn
, long size
)
926 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
927 struct dm_table
*map
;
928 struct dm_target
*ti
;
930 long len
, ret
= -EIO
;
932 map
= dm_get_live_table(md
, &srcu_idx
);
936 ti
= dm_table_find_target(map
, sector
);
937 if (!dm_target_is_valid(ti
))
940 len
= max_io_len(sector
, ti
) << SECTOR_SHIFT
;
941 size
= min(len
, size
);
943 if (ti
->type
->direct_access
)
944 ret
= ti
->type
->direct_access(ti
, sector
, kaddr
, pfn
, size
);
946 dm_put_live_table(md
, srcu_idx
);
947 return min(ret
, size
);
951 * A target may call dm_accept_partial_bio only from the map routine. It is
952 * allowed for all bio types except REQ_PREFLUSH.
954 * dm_accept_partial_bio informs the dm that the target only wants to process
955 * additional n_sectors sectors of the bio and the rest of the data should be
956 * sent in a next bio.
958 * A diagram that explains the arithmetics:
959 * +--------------------+---------------+-------+
961 * +--------------------+---------------+-------+
963 * <-------------- *tio->len_ptr --------------->
964 * <------- bi_size ------->
967 * Region 1 was already iterated over with bio_advance or similar function.
968 * (it may be empty if the target doesn't use bio_advance)
969 * Region 2 is the remaining bio size that the target wants to process.
970 * (it may be empty if region 1 is non-empty, although there is no reason
972 * The target requires that region 3 is to be sent in the next bio.
974 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
975 * the partially processed part (the sum of regions 1+2) must be the same for all
978 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
980 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
981 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
982 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
983 BUG_ON(bi_size
> *tio
->len_ptr
);
984 BUG_ON(n_sectors
> bi_size
);
985 *tio
->len_ptr
-= bi_size
- n_sectors
;
986 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
988 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
991 * Flush current->bio_list when the target map method blocks.
992 * This fixes deadlocks in snapshot and possibly in other targets.
995 struct blk_plug plug
;
996 struct blk_plug_cb cb
;
999 static void flush_current_bio_list(struct blk_plug_cb
*cb
, bool from_schedule
)
1001 struct dm_offload
*o
= container_of(cb
, struct dm_offload
, cb
);
1002 struct bio_list list
;
1006 INIT_LIST_HEAD(&o
->cb
.list
);
1008 if (unlikely(!current
->bio_list
))
1011 for (i
= 0; i
< 2; i
++) {
1012 list
= current
->bio_list
[i
];
1013 bio_list_init(¤t
->bio_list
[i
]);
1015 while ((bio
= bio_list_pop(&list
))) {
1016 struct bio_set
*bs
= bio
->bi_pool
;
1017 if (unlikely(!bs
) || bs
== fs_bio_set
) {
1018 bio_list_add(¤t
->bio_list
[i
], bio
);
1022 spin_lock(&bs
->rescue_lock
);
1023 bio_list_add(&bs
->rescue_list
, bio
);
1024 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
1025 spin_unlock(&bs
->rescue_lock
);
1030 static void dm_offload_start(struct dm_offload
*o
)
1032 blk_start_plug(&o
->plug
);
1033 o
->cb
.callback
= flush_current_bio_list
;
1034 list_add(&o
->cb
.list
, ¤t
->plug
->cb_list
);
1037 static void dm_offload_end(struct dm_offload
*o
)
1039 list_del(&o
->cb
.list
);
1040 blk_finish_plug(&o
->plug
);
1043 static void __map_bio(struct dm_target_io
*tio
)
1047 struct dm_offload o
;
1048 struct bio
*clone
= &tio
->clone
;
1049 struct dm_target
*ti
= tio
->ti
;
1051 clone
->bi_end_io
= clone_endio
;
1054 * Map the clone. If r == 0 we don't need to do
1055 * anything, the target has assumed ownership of
1058 atomic_inc(&tio
->io
->io_count
);
1059 sector
= clone
->bi_iter
.bi_sector
;
1061 dm_offload_start(&o
);
1062 r
= ti
->type
->map(ti
, clone
);
1065 if (r
== DM_MAPIO_REMAPPED
) {
1066 /* the bio has been remapped so dispatch it */
1068 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1069 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1071 generic_make_request(clone
);
1072 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1073 /* error the io and bail out, or requeue it if needed */
1074 dec_pending(tio
->io
, r
);
1076 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1077 DMWARN("unimplemented target map return value: %d", r
);
1083 struct mapped_device
*md
;
1084 struct dm_table
*map
;
1088 unsigned sector_count
;
1091 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1093 bio
->bi_iter
.bi_sector
= sector
;
1094 bio
->bi_iter
.bi_size
= to_bytes(len
);
1098 * Creates a bio that consists of range of complete bvecs.
1100 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1101 sector_t sector
, unsigned len
)
1103 struct bio
*clone
= &tio
->clone
;
1105 __bio_clone_fast(clone
, bio
);
1107 if (bio_integrity(bio
)) {
1108 int r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1113 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1114 clone
->bi_iter
.bi_size
= to_bytes(len
);
1116 if (bio_integrity(bio
))
1117 bio_integrity_trim(clone
, 0, len
);
1122 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1123 struct dm_target
*ti
,
1124 unsigned target_bio_nr
)
1126 struct dm_target_io
*tio
;
1129 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1130 tio
= container_of(clone
, struct dm_target_io
, clone
);
1134 tio
->target_bio_nr
= target_bio_nr
;
1139 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1140 struct dm_target
*ti
,
1141 unsigned target_bio_nr
, unsigned *len
)
1143 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1144 struct bio
*clone
= &tio
->clone
;
1148 __bio_clone_fast(clone
, ci
->bio
);
1150 bio_setup_sector(clone
, ci
->sector
, *len
);
1155 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1156 unsigned num_bios
, unsigned *len
)
1158 unsigned target_bio_nr
;
1160 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1161 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1164 static int __send_empty_flush(struct clone_info
*ci
)
1166 unsigned target_nr
= 0;
1167 struct dm_target
*ti
;
1169 BUG_ON(bio_has_data(ci
->bio
));
1170 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1171 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1176 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1177 sector_t sector
, unsigned *len
)
1179 struct bio
*bio
= ci
->bio
;
1180 struct dm_target_io
*tio
;
1181 unsigned target_bio_nr
;
1182 unsigned num_target_bios
= 1;
1186 * Does the target want to receive duplicate copies of the bio?
1188 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1189 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1191 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1192 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1194 r
= clone_bio(tio
, bio
, sector
, *len
);
1205 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1207 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1209 return ti
->num_discard_bios
;
1212 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1214 return ti
->num_write_same_bios
;
1217 static unsigned get_num_write_zeroes_bios(struct dm_target
*ti
)
1219 return ti
->num_write_zeroes_bios
;
1222 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1224 static bool is_split_required_for_discard(struct dm_target
*ti
)
1226 return ti
->split_discard_bios
;
1229 static int __send_changing_extent_only(struct clone_info
*ci
,
1230 get_num_bios_fn get_num_bios
,
1231 is_split_required_fn is_split_required
)
1233 struct dm_target
*ti
;
1238 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1239 if (!dm_target_is_valid(ti
))
1243 * Even though the device advertised support for this type of
1244 * request, that does not mean every target supports it, and
1245 * reconfiguration might also have changed that since the
1246 * check was performed.
1248 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1252 if (is_split_required
&& !is_split_required(ti
))
1253 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1255 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1257 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1260 } while (ci
->sector_count
-= len
);
1265 static int __send_discard(struct clone_info
*ci
)
1267 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1268 is_split_required_for_discard
);
1271 static int __send_write_same(struct clone_info
*ci
)
1273 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1276 static int __send_write_zeroes(struct clone_info
*ci
)
1278 return __send_changing_extent_only(ci
, get_num_write_zeroes_bios
, NULL
);
1282 * Select the correct strategy for processing a non-flush bio.
1284 static int __split_and_process_non_flush(struct clone_info
*ci
)
1286 struct bio
*bio
= ci
->bio
;
1287 struct dm_target
*ti
;
1291 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1292 return __send_discard(ci
);
1293 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1294 return __send_write_same(ci
);
1295 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_ZEROES
))
1296 return __send_write_zeroes(ci
);
1298 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1299 if (!dm_target_is_valid(ti
))
1302 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1304 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1309 ci
->sector_count
-= len
;
1315 * Entry point to split a bio into clones and submit them to the targets.
1317 static void __split_and_process_bio(struct mapped_device
*md
,
1318 struct dm_table
*map
, struct bio
*bio
)
1320 struct clone_info ci
;
1323 if (unlikely(!map
)) {
1330 ci
.io
= alloc_io(md
);
1332 atomic_set(&ci
.io
->io_count
, 1);
1335 spin_lock_init(&ci
.io
->endio_lock
);
1336 ci
.sector
= bio
->bi_iter
.bi_sector
;
1338 start_io_acct(ci
.io
);
1340 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1341 ci
.bio
= &ci
.md
->flush_bio
;
1342 ci
.sector_count
= 0;
1343 error
= __send_empty_flush(&ci
);
1344 /* dec_pending submits any data associated with flush */
1347 ci
.sector_count
= bio_sectors(bio
);
1348 while (ci
.sector_count
&& !error
)
1349 error
= __split_and_process_non_flush(&ci
);
1352 /* drop the extra reference count */
1353 dec_pending(ci
.io
, error
);
1355 /*-----------------------------------------------------------------
1357 *---------------------------------------------------------------*/
1360 * The request function that just remaps the bio built up by
1363 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1365 int rw
= bio_data_dir(bio
);
1366 struct mapped_device
*md
= q
->queuedata
;
1368 struct dm_table
*map
;
1370 map
= dm_get_live_table(md
, &srcu_idx
);
1372 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1374 /* if we're suspended, we have to queue this io for later */
1375 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1376 dm_put_live_table(md
, srcu_idx
);
1378 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1382 return BLK_QC_T_NONE
;
1385 __split_and_process_bio(md
, map
, bio
);
1386 dm_put_live_table(md
, srcu_idx
);
1387 return BLK_QC_T_NONE
;
1390 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1393 struct mapped_device
*md
= congested_data
;
1394 struct dm_table
*map
;
1396 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1397 if (dm_request_based(md
)) {
1399 * With request-based DM we only need to check the
1400 * top-level queue for congestion.
1402 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1404 map
= dm_get_live_table_fast(md
);
1406 r
= dm_table_any_congested(map
, bdi_bits
);
1407 dm_put_live_table_fast(md
);
1414 /*-----------------------------------------------------------------
1415 * An IDR is used to keep track of allocated minor numbers.
1416 *---------------------------------------------------------------*/
1417 static void free_minor(int minor
)
1419 spin_lock(&_minor_lock
);
1420 idr_remove(&_minor_idr
, minor
);
1421 spin_unlock(&_minor_lock
);
1425 * See if the device with a specific minor # is free.
1427 static int specific_minor(int minor
)
1431 if (minor
>= (1 << MINORBITS
))
1434 idr_preload(GFP_KERNEL
);
1435 spin_lock(&_minor_lock
);
1437 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1439 spin_unlock(&_minor_lock
);
1442 return r
== -ENOSPC
? -EBUSY
: r
;
1446 static int next_free_minor(int *minor
)
1450 idr_preload(GFP_KERNEL
);
1451 spin_lock(&_minor_lock
);
1453 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1455 spin_unlock(&_minor_lock
);
1463 static const struct block_device_operations dm_blk_dops
;
1465 static void dm_wq_work(struct work_struct
*work
);
1467 void dm_init_md_queue(struct mapped_device
*md
)
1470 * Request-based dm devices cannot be stacked on top of bio-based dm
1471 * devices. The type of this dm device may not have been decided yet.
1472 * The type is decided at the first table loading time.
1473 * To prevent problematic device stacking, clear the queue flag
1474 * for request stacking support until then.
1476 * This queue is new, so no concurrency on the queue_flags.
1478 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1481 * Initialize data that will only be used by a non-blk-mq DM queue
1482 * - must do so here (in alloc_dev callchain) before queue is used
1484 md
->queue
->queuedata
= md
;
1485 md
->queue
->backing_dev_info
->congested_data
= md
;
1488 void dm_init_normal_md_queue(struct mapped_device
*md
)
1490 md
->use_blk_mq
= false;
1491 dm_init_md_queue(md
);
1494 * Initialize aspects of queue that aren't relevant for blk-mq
1496 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1497 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1500 static void cleanup_mapped_device(struct mapped_device
*md
)
1503 destroy_workqueue(md
->wq
);
1504 if (md
->kworker_task
)
1505 kthread_stop(md
->kworker_task
);
1506 mempool_destroy(md
->io_pool
);
1508 bioset_free(md
->bs
);
1511 spin_lock(&_minor_lock
);
1512 md
->disk
->private_data
= NULL
;
1513 spin_unlock(&_minor_lock
);
1514 del_gendisk(md
->disk
);
1519 blk_cleanup_queue(md
->queue
);
1521 cleanup_srcu_struct(&md
->io_barrier
);
1528 dm_mq_cleanup_mapped_device(md
);
1532 * Allocate and initialise a blank device with a given minor.
1534 static struct mapped_device
*alloc_dev(int minor
)
1536 int r
, numa_node_id
= dm_get_numa_node();
1537 struct mapped_device
*md
;
1540 md
= kzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1542 DMWARN("unable to allocate device, out of memory.");
1546 if (!try_module_get(THIS_MODULE
))
1547 goto bad_module_get
;
1549 /* get a minor number for the dev */
1550 if (minor
== DM_ANY_MINOR
)
1551 r
= next_free_minor(&minor
);
1553 r
= specific_minor(minor
);
1557 r
= init_srcu_struct(&md
->io_barrier
);
1559 goto bad_io_barrier
;
1561 md
->numa_node_id
= numa_node_id
;
1562 md
->use_blk_mq
= dm_use_blk_mq_default();
1563 md
->init_tio_pdu
= false;
1564 md
->type
= DM_TYPE_NONE
;
1565 mutex_init(&md
->suspend_lock
);
1566 mutex_init(&md
->type_lock
);
1567 mutex_init(&md
->table_devices_lock
);
1568 spin_lock_init(&md
->deferred_lock
);
1569 atomic_set(&md
->holders
, 1);
1570 atomic_set(&md
->open_count
, 0);
1571 atomic_set(&md
->event_nr
, 0);
1572 atomic_set(&md
->uevent_seq
, 0);
1573 INIT_LIST_HEAD(&md
->uevent_list
);
1574 INIT_LIST_HEAD(&md
->table_devices
);
1575 spin_lock_init(&md
->uevent_lock
);
1577 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1581 dm_init_md_queue(md
);
1583 md
->disk
= alloc_disk_node(1, numa_node_id
);
1587 atomic_set(&md
->pending
[0], 0);
1588 atomic_set(&md
->pending
[1], 0);
1589 init_waitqueue_head(&md
->wait
);
1590 INIT_WORK(&md
->work
, dm_wq_work
);
1591 init_waitqueue_head(&md
->eventq
);
1592 init_completion(&md
->kobj_holder
.completion
);
1593 md
->kworker_task
= NULL
;
1595 md
->disk
->major
= _major
;
1596 md
->disk
->first_minor
= minor
;
1597 md
->disk
->fops
= &dm_blk_dops
;
1598 md
->disk
->queue
= md
->queue
;
1599 md
->disk
->private_data
= md
;
1600 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1602 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1604 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1608 md
->bdev
= bdget_disk(md
->disk
, 0);
1612 bio_init(&md
->flush_bio
, NULL
, 0);
1613 md
->flush_bio
.bi_bdev
= md
->bdev
;
1614 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
1616 dm_stats_init(&md
->stats
);
1618 /* Populate the mapping, nobody knows we exist yet */
1619 spin_lock(&_minor_lock
);
1620 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1621 spin_unlock(&_minor_lock
);
1623 BUG_ON(old_md
!= MINOR_ALLOCED
);
1628 cleanup_mapped_device(md
);
1632 module_put(THIS_MODULE
);
1638 static void unlock_fs(struct mapped_device
*md
);
1640 static void free_dev(struct mapped_device
*md
)
1642 int minor
= MINOR(disk_devt(md
->disk
));
1646 cleanup_mapped_device(md
);
1648 free_table_devices(&md
->table_devices
);
1649 dm_stats_cleanup(&md
->stats
);
1652 module_put(THIS_MODULE
);
1656 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1658 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1661 /* The md already has necessary mempools. */
1662 if (dm_table_bio_based(t
)) {
1664 * Reload bioset because front_pad may have changed
1665 * because a different table was loaded.
1667 bioset_free(md
->bs
);
1672 * There's no need to reload with request-based dm
1673 * because the size of front_pad doesn't change.
1674 * Note for future: If you are to reload bioset,
1675 * prep-ed requests in the queue may refer
1676 * to bio from the old bioset, so you must walk
1677 * through the queue to unprep.
1682 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1684 md
->io_pool
= p
->io_pool
;
1690 /* mempool bind completed, no longer need any mempools in the table */
1691 dm_table_free_md_mempools(t
);
1695 * Bind a table to the device.
1697 static void event_callback(void *context
)
1699 unsigned long flags
;
1701 struct mapped_device
*md
= (struct mapped_device
*) context
;
1703 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1704 list_splice_init(&md
->uevent_list
, &uevents
);
1705 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1707 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1709 atomic_inc(&md
->event_nr
);
1710 wake_up(&md
->eventq
);
1714 * Protected by md->suspend_lock obtained by dm_swap_table().
1716 static void __set_size(struct mapped_device
*md
, sector_t size
)
1718 set_capacity(md
->disk
, size
);
1720 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1724 * Returns old map, which caller must destroy.
1726 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1727 struct queue_limits
*limits
)
1729 struct dm_table
*old_map
;
1730 struct request_queue
*q
= md
->queue
;
1733 lockdep_assert_held(&md
->suspend_lock
);
1735 size
= dm_table_get_size(t
);
1738 * Wipe any geometry if the size of the table changed.
1740 if (size
!= dm_get_size(md
))
1741 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1743 __set_size(md
, size
);
1745 dm_table_event_callback(t
, event_callback
, md
);
1748 * The queue hasn't been stopped yet, if the old table type wasn't
1749 * for request-based during suspension. So stop it to prevent
1750 * I/O mapping before resume.
1751 * This must be done before setting the queue restrictions,
1752 * because request-based dm may be run just after the setting.
1754 if (dm_table_request_based(t
)) {
1757 * Leverage the fact that request-based DM targets are
1758 * immutable singletons and establish md->immutable_target
1759 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1761 md
->immutable_target
= dm_table_get_immutable_target(t
);
1764 __bind_mempools(md
, t
);
1766 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1767 rcu_assign_pointer(md
->map
, (void *)t
);
1768 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1770 dm_table_set_restrictions(t
, q
, limits
);
1778 * Returns unbound table for the caller to free.
1780 static struct dm_table
*__unbind(struct mapped_device
*md
)
1782 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1787 dm_table_event_callback(map
, NULL
, NULL
);
1788 RCU_INIT_POINTER(md
->map
, NULL
);
1795 * Constructor for a new device.
1797 int dm_create(int minor
, struct mapped_device
**result
)
1799 struct mapped_device
*md
;
1801 md
= alloc_dev(minor
);
1812 * Functions to manage md->type.
1813 * All are required to hold md->type_lock.
1815 void dm_lock_md_type(struct mapped_device
*md
)
1817 mutex_lock(&md
->type_lock
);
1820 void dm_unlock_md_type(struct mapped_device
*md
)
1822 mutex_unlock(&md
->type_lock
);
1825 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
1827 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1831 unsigned dm_get_md_type(struct mapped_device
*md
)
1836 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
1838 return md
->immutable_target_type
;
1842 * The queue_limits are only valid as long as you have a reference
1845 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
1847 BUG_ON(!atomic_read(&md
->holders
));
1848 return &md
->queue
->limits
;
1850 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
1853 * Setup the DM device's queue based on md's type
1855 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
1858 unsigned type
= dm_get_md_type(md
);
1861 case DM_TYPE_REQUEST_BASED
:
1862 r
= dm_old_init_request_queue(md
, t
);
1864 DMERR("Cannot initialize queue for request-based mapped device");
1868 case DM_TYPE_MQ_REQUEST_BASED
:
1869 r
= dm_mq_init_request_queue(md
, t
);
1871 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1875 case DM_TYPE_BIO_BASED
:
1876 case DM_TYPE_DAX_BIO_BASED
:
1877 dm_init_normal_md_queue(md
);
1878 blk_queue_make_request(md
->queue
, dm_make_request
);
1880 * DM handles splitting bios as needed. Free the bio_split bioset
1881 * since it won't be used (saves 1 process per bio-based DM device).
1883 bioset_free(md
->queue
->bio_split
);
1884 md
->queue
->bio_split
= NULL
;
1886 if (type
== DM_TYPE_DAX_BIO_BASED
)
1887 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
1894 struct mapped_device
*dm_get_md(dev_t dev
)
1896 struct mapped_device
*md
;
1897 unsigned minor
= MINOR(dev
);
1899 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1902 spin_lock(&_minor_lock
);
1904 md
= idr_find(&_minor_idr
, minor
);
1906 if ((md
== MINOR_ALLOCED
||
1907 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1908 dm_deleting_md(md
) ||
1909 test_bit(DMF_FREEING
, &md
->flags
))) {
1917 spin_unlock(&_minor_lock
);
1921 EXPORT_SYMBOL_GPL(dm_get_md
);
1923 void *dm_get_mdptr(struct mapped_device
*md
)
1925 return md
->interface_ptr
;
1928 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1930 md
->interface_ptr
= ptr
;
1933 void dm_get(struct mapped_device
*md
)
1935 atomic_inc(&md
->holders
);
1936 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1939 int dm_hold(struct mapped_device
*md
)
1941 spin_lock(&_minor_lock
);
1942 if (test_bit(DMF_FREEING
, &md
->flags
)) {
1943 spin_unlock(&_minor_lock
);
1947 spin_unlock(&_minor_lock
);
1950 EXPORT_SYMBOL_GPL(dm_hold
);
1952 const char *dm_device_name(struct mapped_device
*md
)
1956 EXPORT_SYMBOL_GPL(dm_device_name
);
1958 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
1960 struct request_queue
*q
= dm_get_md_queue(md
);
1961 struct dm_table
*map
;
1966 spin_lock(&_minor_lock
);
1967 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
1968 set_bit(DMF_FREEING
, &md
->flags
);
1969 spin_unlock(&_minor_lock
);
1971 blk_set_queue_dying(q
);
1973 if (dm_request_based(md
) && md
->kworker_task
)
1974 kthread_flush_worker(&md
->kworker
);
1977 * Take suspend_lock so that presuspend and postsuspend methods
1978 * do not race with internal suspend.
1980 mutex_lock(&md
->suspend_lock
);
1981 map
= dm_get_live_table(md
, &srcu_idx
);
1982 if (!dm_suspended_md(md
)) {
1983 dm_table_presuspend_targets(map
);
1984 dm_table_postsuspend_targets(map
);
1986 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1987 dm_put_live_table(md
, srcu_idx
);
1988 mutex_unlock(&md
->suspend_lock
);
1991 * Rare, but there may be I/O requests still going to complete,
1992 * for example. Wait for all references to disappear.
1993 * No one should increment the reference count of the mapped_device,
1994 * after the mapped_device state becomes DMF_FREEING.
1997 while (atomic_read(&md
->holders
))
1999 else if (atomic_read(&md
->holders
))
2000 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2001 dm_device_name(md
), atomic_read(&md
->holders
));
2004 dm_table_destroy(__unbind(md
));
2008 void dm_destroy(struct mapped_device
*md
)
2010 __dm_destroy(md
, true);
2013 void dm_destroy_immediate(struct mapped_device
*md
)
2015 __dm_destroy(md
, false);
2018 void dm_put(struct mapped_device
*md
)
2020 atomic_dec(&md
->holders
);
2022 EXPORT_SYMBOL_GPL(dm_put
);
2024 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
2030 prepare_to_wait(&md
->wait
, &wait
, task_state
);
2032 if (!md_in_flight(md
))
2035 if (signal_pending_state(task_state
, current
)) {
2042 finish_wait(&md
->wait
, &wait
);
2048 * Process the deferred bios
2050 static void dm_wq_work(struct work_struct
*work
)
2052 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2056 struct dm_table
*map
;
2058 map
= dm_get_live_table(md
, &srcu_idx
);
2060 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2061 spin_lock_irq(&md
->deferred_lock
);
2062 c
= bio_list_pop(&md
->deferred
);
2063 spin_unlock_irq(&md
->deferred_lock
);
2068 if (dm_request_based(md
))
2069 generic_make_request(c
);
2071 __split_and_process_bio(md
, map
, c
);
2074 dm_put_live_table(md
, srcu_idx
);
2077 static void dm_queue_flush(struct mapped_device
*md
)
2079 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2080 smp_mb__after_atomic();
2081 queue_work(md
->wq
, &md
->work
);
2085 * Swap in a new table, returning the old one for the caller to destroy.
2087 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2089 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2090 struct queue_limits limits
;
2093 mutex_lock(&md
->suspend_lock
);
2095 /* device must be suspended */
2096 if (!dm_suspended_md(md
))
2100 * If the new table has no data devices, retain the existing limits.
2101 * This helps multipath with queue_if_no_path if all paths disappear,
2102 * then new I/O is queued based on these limits, and then some paths
2105 if (dm_table_has_no_data_devices(table
)) {
2106 live_map
= dm_get_live_table_fast(md
);
2108 limits
= md
->queue
->limits
;
2109 dm_put_live_table_fast(md
);
2113 r
= dm_calculate_queue_limits(table
, &limits
);
2120 map
= __bind(md
, table
, &limits
);
2123 mutex_unlock(&md
->suspend_lock
);
2128 * Functions to lock and unlock any filesystem running on the
2131 static int lock_fs(struct mapped_device
*md
)
2135 WARN_ON(md
->frozen_sb
);
2137 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2138 if (IS_ERR(md
->frozen_sb
)) {
2139 r
= PTR_ERR(md
->frozen_sb
);
2140 md
->frozen_sb
= NULL
;
2144 set_bit(DMF_FROZEN
, &md
->flags
);
2149 static void unlock_fs(struct mapped_device
*md
)
2151 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2154 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2155 md
->frozen_sb
= NULL
;
2156 clear_bit(DMF_FROZEN
, &md
->flags
);
2160 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2161 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2162 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2164 * If __dm_suspend returns 0, the device is completely quiescent
2165 * now. There is no request-processing activity. All new requests
2166 * are being added to md->deferred list.
2168 * Caller must hold md->suspend_lock
2170 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2171 unsigned suspend_flags
, long task_state
,
2172 int dmf_suspended_flag
)
2174 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2175 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2178 lockdep_assert_held(&md
->suspend_lock
);
2181 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2182 * This flag is cleared before dm_suspend returns.
2185 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2188 * This gets reverted if there's an error later and the targets
2189 * provide the .presuspend_undo hook.
2191 dm_table_presuspend_targets(map
);
2194 * Flush I/O to the device.
2195 * Any I/O submitted after lock_fs() may not be flushed.
2196 * noflush takes precedence over do_lockfs.
2197 * (lock_fs() flushes I/Os and waits for them to complete.)
2199 if (!noflush
&& do_lockfs
) {
2202 dm_table_presuspend_undo_targets(map
);
2208 * Here we must make sure that no processes are submitting requests
2209 * to target drivers i.e. no one may be executing
2210 * __split_and_process_bio. This is called from dm_request and
2213 * To get all processes out of __split_and_process_bio in dm_request,
2214 * we take the write lock. To prevent any process from reentering
2215 * __split_and_process_bio from dm_request and quiesce the thread
2216 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2217 * flush_workqueue(md->wq).
2219 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2221 synchronize_srcu(&md
->io_barrier
);
2224 * Stop md->queue before flushing md->wq in case request-based
2225 * dm defers requests to md->wq from md->queue.
2227 if (dm_request_based(md
)) {
2228 dm_stop_queue(md
->queue
);
2229 if (md
->kworker_task
)
2230 kthread_flush_worker(&md
->kworker
);
2233 flush_workqueue(md
->wq
);
2236 * At this point no more requests are entering target request routines.
2237 * We call dm_wait_for_completion to wait for all existing requests
2240 r
= dm_wait_for_completion(md
, task_state
);
2242 set_bit(dmf_suspended_flag
, &md
->flags
);
2245 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2247 synchronize_srcu(&md
->io_barrier
);
2249 /* were we interrupted ? */
2253 if (dm_request_based(md
))
2254 dm_start_queue(md
->queue
);
2257 dm_table_presuspend_undo_targets(map
);
2258 /* pushback list is already flushed, so skip flush */
2265 * We need to be able to change a mapping table under a mounted
2266 * filesystem. For example we might want to move some data in
2267 * the background. Before the table can be swapped with
2268 * dm_bind_table, dm_suspend must be called to flush any in
2269 * flight bios and ensure that any further io gets deferred.
2272 * Suspend mechanism in request-based dm.
2274 * 1. Flush all I/Os by lock_fs() if needed.
2275 * 2. Stop dispatching any I/O by stopping the request_queue.
2276 * 3. Wait for all in-flight I/Os to be completed or requeued.
2278 * To abort suspend, start the request_queue.
2280 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2282 struct dm_table
*map
= NULL
;
2286 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2288 if (dm_suspended_md(md
)) {
2293 if (dm_suspended_internally_md(md
)) {
2294 /* already internally suspended, wait for internal resume */
2295 mutex_unlock(&md
->suspend_lock
);
2296 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2302 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2304 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2308 dm_table_postsuspend_targets(map
);
2311 mutex_unlock(&md
->suspend_lock
);
2315 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2318 int r
= dm_table_resume_targets(map
);
2326 * Flushing deferred I/Os must be done after targets are resumed
2327 * so that mapping of targets can work correctly.
2328 * Request-based dm is queueing the deferred I/Os in its request_queue.
2330 if (dm_request_based(md
))
2331 dm_start_queue(md
->queue
);
2338 int dm_resume(struct mapped_device
*md
)
2341 struct dm_table
*map
= NULL
;
2345 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2347 if (!dm_suspended_md(md
))
2350 if (dm_suspended_internally_md(md
)) {
2351 /* already internally suspended, wait for internal resume */
2352 mutex_unlock(&md
->suspend_lock
);
2353 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2359 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2360 if (!map
|| !dm_table_get_size(map
))
2363 r
= __dm_resume(md
, map
);
2367 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2369 mutex_unlock(&md
->suspend_lock
);
2375 * Internal suspend/resume works like userspace-driven suspend. It waits
2376 * until all bios finish and prevents issuing new bios to the target drivers.
2377 * It may be used only from the kernel.
2380 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2382 struct dm_table
*map
= NULL
;
2384 if (md
->internal_suspend_count
++)
2385 return; /* nested internal suspend */
2387 if (dm_suspended_md(md
)) {
2388 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2389 return; /* nest suspend */
2392 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2395 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2396 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2397 * would require changing .presuspend to return an error -- avoid this
2398 * until there is a need for more elaborate variants of internal suspend.
2400 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2401 DMF_SUSPENDED_INTERNALLY
);
2403 dm_table_postsuspend_targets(map
);
2406 static void __dm_internal_resume(struct mapped_device
*md
)
2408 BUG_ON(!md
->internal_suspend_count
);
2410 if (--md
->internal_suspend_count
)
2411 return; /* resume from nested internal suspend */
2413 if (dm_suspended_md(md
))
2414 goto done
; /* resume from nested suspend */
2417 * NOTE: existing callers don't need to call dm_table_resume_targets
2418 * (which may fail -- so best to avoid it for now by passing NULL map)
2420 (void) __dm_resume(md
, NULL
);
2423 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2424 smp_mb__after_atomic();
2425 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2428 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2430 mutex_lock(&md
->suspend_lock
);
2431 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2432 mutex_unlock(&md
->suspend_lock
);
2434 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2436 void dm_internal_resume(struct mapped_device
*md
)
2438 mutex_lock(&md
->suspend_lock
);
2439 __dm_internal_resume(md
);
2440 mutex_unlock(&md
->suspend_lock
);
2442 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2445 * Fast variants of internal suspend/resume hold md->suspend_lock,
2446 * which prevents interaction with userspace-driven suspend.
2449 void dm_internal_suspend_fast(struct mapped_device
*md
)
2451 mutex_lock(&md
->suspend_lock
);
2452 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2455 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2456 synchronize_srcu(&md
->io_barrier
);
2457 flush_workqueue(md
->wq
);
2458 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2460 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2462 void dm_internal_resume_fast(struct mapped_device
*md
)
2464 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2470 mutex_unlock(&md
->suspend_lock
);
2472 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2474 /*-----------------------------------------------------------------
2475 * Event notification.
2476 *---------------------------------------------------------------*/
2477 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2480 char udev_cookie
[DM_COOKIE_LENGTH
];
2481 char *envp
[] = { udev_cookie
, NULL
};
2484 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2486 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2487 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2488 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2493 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2495 return atomic_add_return(1, &md
->uevent_seq
);
2498 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2500 return atomic_read(&md
->event_nr
);
2503 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2505 return wait_event_interruptible(md
->eventq
,
2506 (event_nr
!= atomic_read(&md
->event_nr
)));
2509 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2511 unsigned long flags
;
2513 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2514 list_add(elist
, &md
->uevent_list
);
2515 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2519 * The gendisk is only valid as long as you have a reference
2522 struct gendisk
*dm_disk(struct mapped_device
*md
)
2526 EXPORT_SYMBOL_GPL(dm_disk
);
2528 struct kobject
*dm_kobject(struct mapped_device
*md
)
2530 return &md
->kobj_holder
.kobj
;
2533 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2535 struct mapped_device
*md
;
2537 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2539 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2547 int dm_suspended_md(struct mapped_device
*md
)
2549 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2552 int dm_suspended_internally_md(struct mapped_device
*md
)
2554 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2557 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2559 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2562 int dm_suspended(struct dm_target
*ti
)
2564 return dm_suspended_md(dm_table_get_md(ti
->table
));
2566 EXPORT_SYMBOL_GPL(dm_suspended
);
2568 int dm_noflush_suspending(struct dm_target
*ti
)
2570 return __noflush_suspending(dm_table_get_md(ti
->table
));
2572 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2574 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
2575 unsigned integrity
, unsigned per_io_data_size
)
2577 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2578 unsigned int pool_size
= 0;
2579 unsigned int front_pad
;
2585 case DM_TYPE_BIO_BASED
:
2586 case DM_TYPE_DAX_BIO_BASED
:
2587 pool_size
= dm_get_reserved_bio_based_ios();
2588 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2590 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2591 if (!pools
->io_pool
)
2594 case DM_TYPE_REQUEST_BASED
:
2595 case DM_TYPE_MQ_REQUEST_BASED
:
2596 pool_size
= dm_get_reserved_rq_based_ios();
2597 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2598 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2604 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
2608 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2614 dm_free_md_mempools(pools
);
2619 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2624 mempool_destroy(pools
->io_pool
);
2627 bioset_free(pools
->bs
);
2639 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2642 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2643 struct dm_table
*table
;
2644 struct dm_target
*ti
;
2645 int ret
= -ENOTTY
, srcu_idx
;
2647 table
= dm_get_live_table(md
, &srcu_idx
);
2648 if (!table
|| !dm_table_get_size(table
))
2651 /* We only support devices that have a single target */
2652 if (dm_table_get_num_targets(table
) != 1)
2654 ti
= dm_table_get_target(table
, 0);
2657 if (!ti
->type
->iterate_devices
)
2660 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2662 dm_put_live_table(md
, srcu_idx
);
2667 * For register / unregister we need to manually call out to every path.
2669 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2670 sector_t start
, sector_t len
, void *data
)
2672 struct dm_pr
*pr
= data
;
2673 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2675 if (!ops
|| !ops
->pr_register
)
2677 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2680 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2691 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2692 if (ret
&& new_key
) {
2693 /* unregister all paths if we failed to register any path */
2694 pr
.old_key
= new_key
;
2697 pr
.fail_early
= false;
2698 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2704 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2707 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2708 const struct pr_ops
*ops
;
2712 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2716 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2717 if (ops
&& ops
->pr_reserve
)
2718 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2726 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2728 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2729 const struct pr_ops
*ops
;
2733 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2737 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2738 if (ops
&& ops
->pr_release
)
2739 r
= ops
->pr_release(bdev
, key
, type
);
2747 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2748 enum pr_type type
, bool abort
)
2750 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2751 const struct pr_ops
*ops
;
2755 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2759 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2760 if (ops
&& ops
->pr_preempt
)
2761 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2769 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2771 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2772 const struct pr_ops
*ops
;
2776 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2780 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2781 if (ops
&& ops
->pr_clear
)
2782 r
= ops
->pr_clear(bdev
, key
);
2790 static const struct pr_ops dm_pr_ops
= {
2791 .pr_register
= dm_pr_register
,
2792 .pr_reserve
= dm_pr_reserve
,
2793 .pr_release
= dm_pr_release
,
2794 .pr_preempt
= dm_pr_preempt
,
2795 .pr_clear
= dm_pr_clear
,
2798 static const struct block_device_operations dm_blk_dops
= {
2799 .open
= dm_blk_open
,
2800 .release
= dm_blk_close
,
2801 .ioctl
= dm_blk_ioctl
,
2802 .direct_access
= dm_blk_direct_access
,
2803 .getgeo
= dm_blk_getgeo
,
2804 .pr_ops
= &dm_pr_ops
,
2805 .owner
= THIS_MODULE
2811 module_init(dm_init
);
2812 module_exit(dm_exit
);
2814 module_param(major
, uint
, 0);
2815 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2817 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
2818 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
2820 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
2821 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
2823 MODULE_DESCRIPTION(DM_NAME
" driver");
2824 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2825 MODULE_LICENSE("GPL");