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/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
32 DEFAULT_RATELIMIT_INTERVAL
,
33 DEFAULT_RATELIMIT_BURST
);
34 EXPORT_SYMBOL(dm_ratelimit_state
);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name
= DM_NAME
;
46 static unsigned int major
= 0;
47 static unsigned int _major
= 0;
49 static DEFINE_IDR(_minor_idr
);
51 static DEFINE_SPINLOCK(_minor_lock
);
53 static void do_deferred_remove(struct work_struct
*w
);
55 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
57 static struct workqueue_struct
*deferred_remove_workqueue
;
60 * One of these is allocated per bio.
63 struct mapped_device
*md
;
67 unsigned long start_time
;
68 spinlock_t endio_lock
;
69 struct dm_stats_aux stats_aux
;
72 #define MINOR_ALLOCED ((void *)-1)
75 * Bits for the md->flags field.
77 #define DMF_BLOCK_IO_FOR_SUSPEND 0
78 #define DMF_SUSPENDED 1
81 #define DMF_DELETING 4
82 #define DMF_NOFLUSH_SUSPENDING 5
83 #define DMF_DEFERRED_REMOVE 6
84 #define DMF_SUSPENDED_INTERNALLY 7
86 #define DM_NUMA_NODE NUMA_NO_NODE
87 static int dm_numa_node
= DM_NUMA_NODE
;
90 * For mempools pre-allocation at the table loading time.
92 struct dm_md_mempools
{
98 struct list_head list
;
100 struct dm_dev dm_dev
;
103 static struct kmem_cache
*_io_cache
;
104 static struct kmem_cache
*_rq_tio_cache
;
105 static struct kmem_cache
*_rq_cache
;
108 * Bio-based DM's mempools' reserved IOs set by the user.
110 #define RESERVED_BIO_BASED_IOS 16
111 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
113 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
115 int param
= ACCESS_ONCE(*module_param
);
116 int modified_param
= 0;
117 bool modified
= true;
120 modified_param
= min
;
121 else if (param
> max
)
122 modified_param
= max
;
127 (void)cmpxchg(module_param
, param
, modified_param
);
128 param
= modified_param
;
134 unsigned __dm_get_module_param(unsigned *module_param
,
135 unsigned def
, unsigned max
)
137 unsigned param
= ACCESS_ONCE(*module_param
);
138 unsigned modified_param
= 0;
141 modified_param
= def
;
142 else if (param
> max
)
143 modified_param
= max
;
145 if (modified_param
) {
146 (void)cmpxchg(module_param
, param
, modified_param
);
147 param
= modified_param
;
153 unsigned dm_get_reserved_bio_based_ios(void)
155 return __dm_get_module_param(&reserved_bio_based_ios
,
156 RESERVED_BIO_BASED_IOS
, DM_RESERVED_MAX_IOS
);
158 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
160 static unsigned dm_get_numa_node(void)
162 return __dm_get_module_param_int(&dm_numa_node
,
163 DM_NUMA_NODE
, num_online_nodes() - 1);
166 static int __init
local_init(void)
170 /* allocate a slab for the dm_ios */
171 _io_cache
= KMEM_CACHE(dm_io
, 0);
175 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
177 goto out_free_io_cache
;
179 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
180 __alignof__(struct request
), 0, NULL
);
182 goto out_free_rq_tio_cache
;
184 r
= dm_uevent_init();
186 goto out_free_rq_cache
;
188 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
189 if (!deferred_remove_workqueue
) {
191 goto out_uevent_exit
;
195 r
= register_blkdev(_major
, _name
);
197 goto out_free_workqueue
;
205 destroy_workqueue(deferred_remove_workqueue
);
209 kmem_cache_destroy(_rq_cache
);
210 out_free_rq_tio_cache
:
211 kmem_cache_destroy(_rq_tio_cache
);
213 kmem_cache_destroy(_io_cache
);
218 static void local_exit(void)
220 flush_scheduled_work();
221 destroy_workqueue(deferred_remove_workqueue
);
223 kmem_cache_destroy(_rq_cache
);
224 kmem_cache_destroy(_rq_tio_cache
);
225 kmem_cache_destroy(_io_cache
);
226 unregister_blkdev(_major
, _name
);
231 DMINFO("cleaned up");
234 static int (*_inits
[])(void) __initdata
= {
245 static void (*_exits
[])(void) = {
256 static int __init
dm_init(void)
258 const int count
= ARRAY_SIZE(_inits
);
262 for (i
= 0; i
< count
; i
++) {
277 static void __exit
dm_exit(void)
279 int i
= ARRAY_SIZE(_exits
);
285 * Should be empty by this point.
287 idr_destroy(&_minor_idr
);
291 * Block device functions
293 int dm_deleting_md(struct mapped_device
*md
)
295 return test_bit(DMF_DELETING
, &md
->flags
);
298 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
300 struct mapped_device
*md
;
302 spin_lock(&_minor_lock
);
304 md
= bdev
->bd_disk
->private_data
;
308 if (test_bit(DMF_FREEING
, &md
->flags
) ||
309 dm_deleting_md(md
)) {
315 atomic_inc(&md
->open_count
);
317 spin_unlock(&_minor_lock
);
319 return md
? 0 : -ENXIO
;
322 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
324 struct mapped_device
*md
;
326 spin_lock(&_minor_lock
);
328 md
= disk
->private_data
;
332 if (atomic_dec_and_test(&md
->open_count
) &&
333 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
334 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
338 spin_unlock(&_minor_lock
);
341 int dm_open_count(struct mapped_device
*md
)
343 return atomic_read(&md
->open_count
);
347 * Guarantees nothing is using the device before it's deleted.
349 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
353 spin_lock(&_minor_lock
);
355 if (dm_open_count(md
)) {
358 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
359 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
362 set_bit(DMF_DELETING
, &md
->flags
);
364 spin_unlock(&_minor_lock
);
369 int dm_cancel_deferred_remove(struct mapped_device
*md
)
373 spin_lock(&_minor_lock
);
375 if (test_bit(DMF_DELETING
, &md
->flags
))
378 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
380 spin_unlock(&_minor_lock
);
385 static void do_deferred_remove(struct work_struct
*w
)
387 dm_deferred_remove();
390 sector_t
dm_get_size(struct mapped_device
*md
)
392 return get_capacity(md
->disk
);
395 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
400 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
405 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
407 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
409 return dm_get_geometry(md
, geo
);
412 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
413 struct block_device
**bdev
,
416 struct dm_target
*tgt
;
417 struct dm_table
*map
;
422 map
= dm_get_live_table(md
, &srcu_idx
);
423 if (!map
|| !dm_table_get_size(map
))
426 /* We only support devices that have a single target */
427 if (dm_table_get_num_targets(map
) != 1)
430 tgt
= dm_table_get_target(map
, 0);
431 if (!tgt
->type
->prepare_ioctl
)
434 if (dm_suspended_md(md
)) {
439 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
444 dm_put_live_table(md
, srcu_idx
);
448 dm_put_live_table(md
, srcu_idx
);
449 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
456 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
457 unsigned int cmd
, unsigned long arg
)
459 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
462 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
468 * Target determined this ioctl is being issued against a
469 * subset of the parent bdev; require extra privileges.
471 if (!capable(CAP_SYS_RAWIO
)) {
473 "%s: sending ioctl %x to DM device without required privilege.",
480 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
486 static struct dm_io
*alloc_io(struct mapped_device
*md
)
488 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
491 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
493 mempool_free(io
, md
->io_pool
);
496 static void free_tio(struct dm_target_io
*tio
)
498 bio_put(&tio
->clone
);
501 int md_in_flight(struct mapped_device
*md
)
503 return atomic_read(&md
->pending
[READ
]) +
504 atomic_read(&md
->pending
[WRITE
]);
507 static void start_io_acct(struct dm_io
*io
)
509 struct mapped_device
*md
= io
->md
;
510 struct bio
*bio
= io
->bio
;
512 int rw
= bio_data_dir(bio
);
514 io
->start_time
= jiffies
;
516 cpu
= part_stat_lock();
517 part_round_stats(cpu
, &dm_disk(md
)->part0
);
519 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
520 atomic_inc_return(&md
->pending
[rw
]));
522 if (unlikely(dm_stats_used(&md
->stats
)))
523 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
524 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
525 false, 0, &io
->stats_aux
);
528 static void end_io_acct(struct dm_io
*io
)
530 struct mapped_device
*md
= io
->md
;
531 struct bio
*bio
= io
->bio
;
532 unsigned long duration
= jiffies
- io
->start_time
;
534 int rw
= bio_data_dir(bio
);
536 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
538 if (unlikely(dm_stats_used(&md
->stats
)))
539 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
540 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
541 true, duration
, &io
->stats_aux
);
544 * After this is decremented the bio must not be touched if it is
547 pending
= atomic_dec_return(&md
->pending
[rw
]);
548 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
549 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
551 /* nudge anyone waiting on suspend queue */
557 * Add the bio to the list of deferred io.
559 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
563 spin_lock_irqsave(&md
->deferred_lock
, flags
);
564 bio_list_add(&md
->deferred
, bio
);
565 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
566 queue_work(md
->wq
, &md
->work
);
570 * Everyone (including functions in this file), should use this
571 * function to access the md->map field, and make sure they call
572 * dm_put_live_table() when finished.
574 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
576 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
578 return srcu_dereference(md
->map
, &md
->io_barrier
);
581 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
583 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
586 void dm_sync_table(struct mapped_device
*md
)
588 synchronize_srcu(&md
->io_barrier
);
589 synchronize_rcu_expedited();
593 * A fast alternative to dm_get_live_table/dm_put_live_table.
594 * The caller must not block between these two functions.
596 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
599 return rcu_dereference(md
->map
);
602 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
608 * Open a table device so we can use it as a map destination.
610 static int open_table_device(struct table_device
*td
, dev_t dev
,
611 struct mapped_device
*md
)
613 static char *_claim_ptr
= "I belong to device-mapper";
614 struct block_device
*bdev
;
618 BUG_ON(td
->dm_dev
.bdev
);
620 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
622 return PTR_ERR(bdev
);
624 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
626 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
630 td
->dm_dev
.bdev
= bdev
;
635 * Close a table device that we've been using.
637 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
639 if (!td
->dm_dev
.bdev
)
642 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
643 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
644 td
->dm_dev
.bdev
= NULL
;
647 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
649 struct table_device
*td
;
651 list_for_each_entry(td
, l
, list
)
652 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
658 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
659 struct dm_dev
**result
) {
661 struct table_device
*td
;
663 mutex_lock(&md
->table_devices_lock
);
664 td
= find_table_device(&md
->table_devices
, dev
, mode
);
666 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
668 mutex_unlock(&md
->table_devices_lock
);
672 td
->dm_dev
.mode
= mode
;
673 td
->dm_dev
.bdev
= NULL
;
675 if ((r
= open_table_device(td
, dev
, md
))) {
676 mutex_unlock(&md
->table_devices_lock
);
681 format_dev_t(td
->dm_dev
.name
, dev
);
683 atomic_set(&td
->count
, 0);
684 list_add(&td
->list
, &md
->table_devices
);
686 atomic_inc(&td
->count
);
687 mutex_unlock(&md
->table_devices_lock
);
689 *result
= &td
->dm_dev
;
692 EXPORT_SYMBOL_GPL(dm_get_table_device
);
694 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
696 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
698 mutex_lock(&md
->table_devices_lock
);
699 if (atomic_dec_and_test(&td
->count
)) {
700 close_table_device(td
, md
);
704 mutex_unlock(&md
->table_devices_lock
);
706 EXPORT_SYMBOL(dm_put_table_device
);
708 static void free_table_devices(struct list_head
*devices
)
710 struct list_head
*tmp
, *next
;
712 list_for_each_safe(tmp
, next
, devices
) {
713 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
715 DMWARN("dm_destroy: %s still exists with %d references",
716 td
->dm_dev
.name
, atomic_read(&td
->count
));
722 * Get the geometry associated with a dm device
724 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
732 * Set the geometry of a device.
734 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
736 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
738 if (geo
->start
> sz
) {
739 DMWARN("Start sector is beyond the geometry limits.");
748 /*-----------------------------------------------------------------
750 * A more elegant soln is in the works that uses the queue
751 * merge fn, unfortunately there are a couple of changes to
752 * the block layer that I want to make for this. So in the
753 * interests of getting something for people to use I give
754 * you this clearly demarcated crap.
755 *---------------------------------------------------------------*/
757 static int __noflush_suspending(struct mapped_device
*md
)
759 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
763 * Decrements the number of outstanding ios that a bio has been
764 * cloned into, completing the original io if necc.
766 static void dec_pending(struct dm_io
*io
, int error
)
771 struct mapped_device
*md
= io
->md
;
773 /* Push-back supersedes any I/O errors */
774 if (unlikely(error
)) {
775 spin_lock_irqsave(&io
->endio_lock
, flags
);
776 if (!(io
->error
> 0 && __noflush_suspending(md
)))
778 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
781 if (atomic_dec_and_test(&io
->io_count
)) {
782 if (io
->error
== DM_ENDIO_REQUEUE
) {
784 * Target requested pushing back the I/O.
786 spin_lock_irqsave(&md
->deferred_lock
, flags
);
787 if (__noflush_suspending(md
))
788 bio_list_add_head(&md
->deferred
, io
->bio
);
790 /* noflush suspend was interrupted. */
792 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
795 io_error
= io
->error
;
800 if (io_error
== DM_ENDIO_REQUEUE
)
803 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
805 * Preflush done for flush with data, reissue
806 * without REQ_PREFLUSH.
808 bio
->bi_opf
&= ~REQ_PREFLUSH
;
811 /* done with normal IO or empty flush */
812 trace_block_bio_complete(md
->queue
, bio
, io_error
);
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 static void clone_endio(struct bio
*bio
)
829 int error
= bio
->bi_error
;
831 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
832 struct dm_io
*io
= tio
->io
;
833 struct mapped_device
*md
= tio
->io
->md
;
834 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
837 r
= endio(tio
->ti
, bio
, error
);
838 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
840 * error and requeue request are handled
844 else if (r
== DM_ENDIO_INCOMPLETE
)
845 /* The target will handle the io */
848 DMWARN("unimplemented target endio return value: %d", r
);
853 if (unlikely(r
== -EREMOTEIO
&& (bio_op(bio
) == REQ_OP_WRITE_SAME
) &&
854 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
855 disable_write_same(md
);
858 dec_pending(io
, error
);
862 * Return maximum size of I/O possible at the supplied sector up to the current
865 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
867 sector_t target_offset
= dm_target_offset(ti
, sector
);
869 return ti
->len
- target_offset
;
872 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
874 sector_t len
= max_io_len_target_boundary(sector
, ti
);
875 sector_t offset
, max_len
;
878 * Does the target need to split even further?
880 if (ti
->max_io_len
) {
881 offset
= dm_target_offset(ti
, sector
);
882 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
883 max_len
= sector_div(offset
, ti
->max_io_len
);
885 max_len
= offset
& (ti
->max_io_len
- 1);
886 max_len
= ti
->max_io_len
- max_len
;
895 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
897 if (len
> UINT_MAX
) {
898 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
899 (unsigned long long)len
, UINT_MAX
);
900 ti
->error
= "Maximum size of target IO is too large";
904 ti
->max_io_len
= (uint32_t) len
;
908 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
910 static long dm_blk_direct_access(struct block_device
*bdev
, sector_t sector
,
911 void **kaddr
, pfn_t
*pfn
, long size
)
913 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
914 struct dm_table
*map
;
915 struct dm_target
*ti
;
917 long len
, ret
= -EIO
;
919 map
= dm_get_live_table(md
, &srcu_idx
);
923 ti
= dm_table_find_target(map
, sector
);
924 if (!dm_target_is_valid(ti
))
927 len
= max_io_len(sector
, ti
) << SECTOR_SHIFT
;
928 size
= min(len
, size
);
930 if (ti
->type
->direct_access
)
931 ret
= ti
->type
->direct_access(ti
, sector
, kaddr
, pfn
, size
);
933 dm_put_live_table(md
, srcu_idx
);
934 return min(ret
, size
);
938 * A target may call dm_accept_partial_bio only from the map routine. It is
939 * allowed for all bio types except REQ_PREFLUSH.
941 * dm_accept_partial_bio informs the dm that the target only wants to process
942 * additional n_sectors sectors of the bio and the rest of the data should be
943 * sent in a next bio.
945 * A diagram that explains the arithmetics:
946 * +--------------------+---------------+-------+
948 * +--------------------+---------------+-------+
950 * <-------------- *tio->len_ptr --------------->
951 * <------- bi_size ------->
954 * Region 1 was already iterated over with bio_advance or similar function.
955 * (it may be empty if the target doesn't use bio_advance)
956 * Region 2 is the remaining bio size that the target wants to process.
957 * (it may be empty if region 1 is non-empty, although there is no reason
959 * The target requires that region 3 is to be sent in the next bio.
961 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
962 * the partially processed part (the sum of regions 1+2) must be the same for all
965 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
967 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
968 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
969 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
970 BUG_ON(bi_size
> *tio
->len_ptr
);
971 BUG_ON(n_sectors
> bi_size
);
972 *tio
->len_ptr
-= bi_size
- n_sectors
;
973 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
975 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
978 * Flush current->bio_list when the target map method blocks.
979 * This fixes deadlocks in snapshot and possibly in other targets.
982 struct blk_plug plug
;
983 struct blk_plug_cb cb
;
986 static void flush_current_bio_list(struct blk_plug_cb
*cb
, bool from_schedule
)
988 struct dm_offload
*o
= container_of(cb
, struct dm_offload
, cb
);
989 struct bio_list list
;
992 INIT_LIST_HEAD(&o
->cb
.list
);
994 if (unlikely(!current
->bio_list
))
997 list
= *current
->bio_list
;
998 bio_list_init(current
->bio_list
);
1000 while ((bio
= bio_list_pop(&list
))) {
1001 struct bio_set
*bs
= bio
->bi_pool
;
1002 if (unlikely(!bs
) || bs
== fs_bio_set
) {
1003 bio_list_add(current
->bio_list
, bio
);
1007 spin_lock(&bs
->rescue_lock
);
1008 bio_list_add(&bs
->rescue_list
, bio
);
1009 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
1010 spin_unlock(&bs
->rescue_lock
);
1014 static void dm_offload_start(struct dm_offload
*o
)
1016 blk_start_plug(&o
->plug
);
1017 o
->cb
.callback
= flush_current_bio_list
;
1018 list_add(&o
->cb
.list
, ¤t
->plug
->cb_list
);
1021 static void dm_offload_end(struct dm_offload
*o
)
1023 list_del(&o
->cb
.list
);
1024 blk_finish_plug(&o
->plug
);
1027 static void __map_bio(struct dm_target_io
*tio
)
1031 struct dm_offload o
;
1032 struct bio
*clone
= &tio
->clone
;
1033 struct dm_target
*ti
= tio
->ti
;
1035 clone
->bi_end_io
= clone_endio
;
1038 * Map the clone. If r == 0 we don't need to do
1039 * anything, the target has assumed ownership of
1042 atomic_inc(&tio
->io
->io_count
);
1043 sector
= clone
->bi_iter
.bi_sector
;
1045 dm_offload_start(&o
);
1046 r
= ti
->type
->map(ti
, clone
);
1049 if (r
== DM_MAPIO_REMAPPED
) {
1050 /* the bio has been remapped so dispatch it */
1052 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1053 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1055 generic_make_request(clone
);
1056 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1057 /* error the io and bail out, or requeue it if needed */
1058 dec_pending(tio
->io
, r
);
1060 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1061 DMWARN("unimplemented target map return value: %d", r
);
1067 struct mapped_device
*md
;
1068 struct dm_table
*map
;
1072 unsigned sector_count
;
1075 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1077 bio
->bi_iter
.bi_sector
= sector
;
1078 bio
->bi_iter
.bi_size
= to_bytes(len
);
1082 * Creates a bio that consists of range of complete bvecs.
1084 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1085 sector_t sector
, unsigned len
)
1087 struct bio
*clone
= &tio
->clone
;
1089 __bio_clone_fast(clone
, bio
);
1091 if (bio_integrity(bio
)) {
1092 int r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1097 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1098 clone
->bi_iter
.bi_size
= to_bytes(len
);
1100 if (bio_integrity(bio
))
1101 bio_integrity_trim(clone
, 0, len
);
1106 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1107 struct dm_target
*ti
,
1108 unsigned target_bio_nr
)
1110 struct dm_target_io
*tio
;
1113 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1114 tio
= container_of(clone
, struct dm_target_io
, clone
);
1118 tio
->target_bio_nr
= target_bio_nr
;
1123 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1124 struct dm_target
*ti
,
1125 unsigned target_bio_nr
, unsigned *len
)
1127 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1128 struct bio
*clone
= &tio
->clone
;
1132 __bio_clone_fast(clone
, ci
->bio
);
1134 bio_setup_sector(clone
, ci
->sector
, *len
);
1139 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1140 unsigned num_bios
, unsigned *len
)
1142 unsigned target_bio_nr
;
1144 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1145 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1148 static int __send_empty_flush(struct clone_info
*ci
)
1150 unsigned target_nr
= 0;
1151 struct dm_target
*ti
;
1153 BUG_ON(bio_has_data(ci
->bio
));
1154 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1155 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1160 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1161 sector_t sector
, unsigned *len
)
1163 struct bio
*bio
= ci
->bio
;
1164 struct dm_target_io
*tio
;
1165 unsigned target_bio_nr
;
1166 unsigned num_target_bios
= 1;
1170 * Does the target want to receive duplicate copies of the bio?
1172 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1173 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1175 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1176 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1178 r
= clone_bio(tio
, bio
, sector
, *len
);
1189 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1191 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1193 return ti
->num_discard_bios
;
1196 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1198 return ti
->num_write_same_bios
;
1201 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1203 static bool is_split_required_for_discard(struct dm_target
*ti
)
1205 return ti
->split_discard_bios
;
1208 static int __send_changing_extent_only(struct clone_info
*ci
,
1209 get_num_bios_fn get_num_bios
,
1210 is_split_required_fn is_split_required
)
1212 struct dm_target
*ti
;
1217 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1218 if (!dm_target_is_valid(ti
))
1222 * Even though the device advertised support for this type of
1223 * request, that does not mean every target supports it, and
1224 * reconfiguration might also have changed that since the
1225 * check was performed.
1227 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1231 if (is_split_required
&& !is_split_required(ti
))
1232 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1234 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1236 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1239 } while (ci
->sector_count
-= len
);
1244 static int __send_discard(struct clone_info
*ci
)
1246 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1247 is_split_required_for_discard
);
1250 static int __send_write_same(struct clone_info
*ci
)
1252 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1256 * Select the correct strategy for processing a non-flush bio.
1258 static int __split_and_process_non_flush(struct clone_info
*ci
)
1260 struct bio
*bio
= ci
->bio
;
1261 struct dm_target
*ti
;
1265 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1266 return __send_discard(ci
);
1267 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1268 return __send_write_same(ci
);
1270 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1271 if (!dm_target_is_valid(ti
))
1274 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1276 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1281 ci
->sector_count
-= len
;
1287 * Entry point to split a bio into clones and submit them to the targets.
1289 static void __split_and_process_bio(struct mapped_device
*md
,
1290 struct dm_table
*map
, struct bio
*bio
)
1292 struct clone_info ci
;
1295 if (unlikely(!map
)) {
1302 ci
.io
= alloc_io(md
);
1304 atomic_set(&ci
.io
->io_count
, 1);
1307 spin_lock_init(&ci
.io
->endio_lock
);
1308 ci
.sector
= bio
->bi_iter
.bi_sector
;
1310 start_io_acct(ci
.io
);
1312 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1313 ci
.bio
= &ci
.md
->flush_bio
;
1314 ci
.sector_count
= 0;
1315 error
= __send_empty_flush(&ci
);
1316 /* dec_pending submits any data associated with flush */
1319 ci
.sector_count
= bio_sectors(bio
);
1320 while (ci
.sector_count
&& !error
)
1321 error
= __split_and_process_non_flush(&ci
);
1324 /* drop the extra reference count */
1325 dec_pending(ci
.io
, error
);
1327 /*-----------------------------------------------------------------
1329 *---------------------------------------------------------------*/
1332 * The request function that just remaps the bio built up by
1335 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1337 int rw
= bio_data_dir(bio
);
1338 struct mapped_device
*md
= q
->queuedata
;
1340 struct dm_table
*map
;
1342 map
= dm_get_live_table(md
, &srcu_idx
);
1344 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1346 /* if we're suspended, we have to queue this io for later */
1347 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1348 dm_put_live_table(md
, srcu_idx
);
1350 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1354 return BLK_QC_T_NONE
;
1357 __split_and_process_bio(md
, map
, bio
);
1358 dm_put_live_table(md
, srcu_idx
);
1359 return BLK_QC_T_NONE
;
1362 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1365 struct mapped_device
*md
= congested_data
;
1366 struct dm_table
*map
;
1368 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1369 if (dm_request_based(md
)) {
1371 * With request-based DM we only need to check the
1372 * top-level queue for congestion.
1374 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1376 map
= dm_get_live_table_fast(md
);
1378 r
= dm_table_any_congested(map
, bdi_bits
);
1379 dm_put_live_table_fast(md
);
1386 /*-----------------------------------------------------------------
1387 * An IDR is used to keep track of allocated minor numbers.
1388 *---------------------------------------------------------------*/
1389 static void free_minor(int minor
)
1391 spin_lock(&_minor_lock
);
1392 idr_remove(&_minor_idr
, minor
);
1393 spin_unlock(&_minor_lock
);
1397 * See if the device with a specific minor # is free.
1399 static int specific_minor(int minor
)
1403 if (minor
>= (1 << MINORBITS
))
1406 idr_preload(GFP_KERNEL
);
1407 spin_lock(&_minor_lock
);
1409 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1411 spin_unlock(&_minor_lock
);
1414 return r
== -ENOSPC
? -EBUSY
: r
;
1418 static int next_free_minor(int *minor
)
1422 idr_preload(GFP_KERNEL
);
1423 spin_lock(&_minor_lock
);
1425 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1427 spin_unlock(&_minor_lock
);
1435 static const struct block_device_operations dm_blk_dops
;
1437 static void dm_wq_work(struct work_struct
*work
);
1439 void dm_init_md_queue(struct mapped_device
*md
)
1442 * Request-based dm devices cannot be stacked on top of bio-based dm
1443 * devices. The type of this dm device may not have been decided yet.
1444 * The type is decided at the first table loading time.
1445 * To prevent problematic device stacking, clear the queue flag
1446 * for request stacking support until then.
1448 * This queue is new, so no concurrency on the queue_flags.
1450 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1453 * Initialize data that will only be used by a non-blk-mq DM queue
1454 * - must do so here (in alloc_dev callchain) before queue is used
1456 md
->queue
->queuedata
= md
;
1457 md
->queue
->backing_dev_info
->congested_data
= md
;
1460 void dm_init_normal_md_queue(struct mapped_device
*md
)
1462 md
->use_blk_mq
= false;
1463 dm_init_md_queue(md
);
1466 * Initialize aspects of queue that aren't relevant for blk-mq
1468 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1469 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1472 static void cleanup_mapped_device(struct mapped_device
*md
)
1475 destroy_workqueue(md
->wq
);
1476 if (md
->kworker_task
)
1477 kthread_stop(md
->kworker_task
);
1478 mempool_destroy(md
->io_pool
);
1480 bioset_free(md
->bs
);
1483 spin_lock(&_minor_lock
);
1484 md
->disk
->private_data
= NULL
;
1485 spin_unlock(&_minor_lock
);
1486 del_gendisk(md
->disk
);
1491 blk_cleanup_queue(md
->queue
);
1493 cleanup_srcu_struct(&md
->io_barrier
);
1500 dm_mq_cleanup_mapped_device(md
);
1504 * Allocate and initialise a blank device with a given minor.
1506 static struct mapped_device
*alloc_dev(int minor
)
1508 int r
, numa_node_id
= dm_get_numa_node();
1509 struct mapped_device
*md
;
1512 md
= kzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1514 DMWARN("unable to allocate device, out of memory.");
1518 if (!try_module_get(THIS_MODULE
))
1519 goto bad_module_get
;
1521 /* get a minor number for the dev */
1522 if (minor
== DM_ANY_MINOR
)
1523 r
= next_free_minor(&minor
);
1525 r
= specific_minor(minor
);
1529 r
= init_srcu_struct(&md
->io_barrier
);
1531 goto bad_io_barrier
;
1533 md
->numa_node_id
= numa_node_id
;
1534 md
->use_blk_mq
= dm_use_blk_mq_default();
1535 md
->init_tio_pdu
= false;
1536 md
->type
= DM_TYPE_NONE
;
1537 mutex_init(&md
->suspend_lock
);
1538 mutex_init(&md
->type_lock
);
1539 mutex_init(&md
->table_devices_lock
);
1540 spin_lock_init(&md
->deferred_lock
);
1541 atomic_set(&md
->holders
, 1);
1542 atomic_set(&md
->open_count
, 0);
1543 atomic_set(&md
->event_nr
, 0);
1544 atomic_set(&md
->uevent_seq
, 0);
1545 INIT_LIST_HEAD(&md
->uevent_list
);
1546 INIT_LIST_HEAD(&md
->table_devices
);
1547 spin_lock_init(&md
->uevent_lock
);
1549 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1553 dm_init_md_queue(md
);
1555 md
->disk
= alloc_disk_node(1, numa_node_id
);
1559 atomic_set(&md
->pending
[0], 0);
1560 atomic_set(&md
->pending
[1], 0);
1561 init_waitqueue_head(&md
->wait
);
1562 INIT_WORK(&md
->work
, dm_wq_work
);
1563 init_waitqueue_head(&md
->eventq
);
1564 init_completion(&md
->kobj_holder
.completion
);
1565 md
->kworker_task
= NULL
;
1567 md
->disk
->major
= _major
;
1568 md
->disk
->first_minor
= minor
;
1569 md
->disk
->fops
= &dm_blk_dops
;
1570 md
->disk
->queue
= md
->queue
;
1571 md
->disk
->private_data
= md
;
1572 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1574 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1576 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1580 md
->bdev
= bdget_disk(md
->disk
, 0);
1584 bio_init(&md
->flush_bio
, NULL
, 0);
1585 md
->flush_bio
.bi_bdev
= md
->bdev
;
1586 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
1588 dm_stats_init(&md
->stats
);
1590 /* Populate the mapping, nobody knows we exist yet */
1591 spin_lock(&_minor_lock
);
1592 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1593 spin_unlock(&_minor_lock
);
1595 BUG_ON(old_md
!= MINOR_ALLOCED
);
1600 cleanup_mapped_device(md
);
1604 module_put(THIS_MODULE
);
1610 static void unlock_fs(struct mapped_device
*md
);
1612 static void free_dev(struct mapped_device
*md
)
1614 int minor
= MINOR(disk_devt(md
->disk
));
1618 cleanup_mapped_device(md
);
1620 free_table_devices(&md
->table_devices
);
1621 dm_stats_cleanup(&md
->stats
);
1624 module_put(THIS_MODULE
);
1628 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1630 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1633 /* The md already has necessary mempools. */
1634 if (dm_table_bio_based(t
)) {
1636 * Reload bioset because front_pad may have changed
1637 * because a different table was loaded.
1639 bioset_free(md
->bs
);
1644 * There's no need to reload with request-based dm
1645 * because the size of front_pad doesn't change.
1646 * Note for future: If you are to reload bioset,
1647 * prep-ed requests in the queue may refer
1648 * to bio from the old bioset, so you must walk
1649 * through the queue to unprep.
1654 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1656 md
->io_pool
= p
->io_pool
;
1662 /* mempool bind completed, no longer need any mempools in the table */
1663 dm_table_free_md_mempools(t
);
1667 * Bind a table to the device.
1669 static void event_callback(void *context
)
1671 unsigned long flags
;
1673 struct mapped_device
*md
= (struct mapped_device
*) context
;
1675 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1676 list_splice_init(&md
->uevent_list
, &uevents
);
1677 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1679 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1681 atomic_inc(&md
->event_nr
);
1682 wake_up(&md
->eventq
);
1686 * Protected by md->suspend_lock obtained by dm_swap_table().
1688 static void __set_size(struct mapped_device
*md
, sector_t size
)
1690 set_capacity(md
->disk
, size
);
1692 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1696 * Returns old map, which caller must destroy.
1698 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1699 struct queue_limits
*limits
)
1701 struct dm_table
*old_map
;
1702 struct request_queue
*q
= md
->queue
;
1705 lockdep_assert_held(&md
->suspend_lock
);
1707 size
= dm_table_get_size(t
);
1710 * Wipe any geometry if the size of the table changed.
1712 if (size
!= dm_get_size(md
))
1713 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1715 __set_size(md
, size
);
1717 dm_table_event_callback(t
, event_callback
, md
);
1720 * The queue hasn't been stopped yet, if the old table type wasn't
1721 * for request-based during suspension. So stop it to prevent
1722 * I/O mapping before resume.
1723 * This must be done before setting the queue restrictions,
1724 * because request-based dm may be run just after the setting.
1726 if (dm_table_request_based(t
)) {
1729 * Leverage the fact that request-based DM targets are
1730 * immutable singletons and establish md->immutable_target
1731 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1733 md
->immutable_target
= dm_table_get_immutable_target(t
);
1736 __bind_mempools(md
, t
);
1738 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1739 rcu_assign_pointer(md
->map
, (void *)t
);
1740 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1742 dm_table_set_restrictions(t
, q
, limits
);
1750 * Returns unbound table for the caller to free.
1752 static struct dm_table
*__unbind(struct mapped_device
*md
)
1754 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1759 dm_table_event_callback(map
, NULL
, NULL
);
1760 RCU_INIT_POINTER(md
->map
, NULL
);
1767 * Constructor for a new device.
1769 int dm_create(int minor
, struct mapped_device
**result
)
1771 struct mapped_device
*md
;
1773 md
= alloc_dev(minor
);
1784 * Functions to manage md->type.
1785 * All are required to hold md->type_lock.
1787 void dm_lock_md_type(struct mapped_device
*md
)
1789 mutex_lock(&md
->type_lock
);
1792 void dm_unlock_md_type(struct mapped_device
*md
)
1794 mutex_unlock(&md
->type_lock
);
1797 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
1799 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1803 unsigned dm_get_md_type(struct mapped_device
*md
)
1808 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
1810 return md
->immutable_target_type
;
1814 * The queue_limits are only valid as long as you have a reference
1817 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
1819 BUG_ON(!atomic_read(&md
->holders
));
1820 return &md
->queue
->limits
;
1822 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
1825 * Setup the DM device's queue based on md's type
1827 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
1830 unsigned type
= dm_get_md_type(md
);
1833 case DM_TYPE_REQUEST_BASED
:
1834 r
= dm_old_init_request_queue(md
, t
);
1836 DMERR("Cannot initialize queue for request-based mapped device");
1840 case DM_TYPE_MQ_REQUEST_BASED
:
1841 r
= dm_mq_init_request_queue(md
, t
);
1843 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1847 case DM_TYPE_BIO_BASED
:
1848 case DM_TYPE_DAX_BIO_BASED
:
1849 dm_init_normal_md_queue(md
);
1850 blk_queue_make_request(md
->queue
, dm_make_request
);
1852 * DM handles splitting bios as needed. Free the bio_split bioset
1853 * since it won't be used (saves 1 process per bio-based DM device).
1855 bioset_free(md
->queue
->bio_split
);
1856 md
->queue
->bio_split
= NULL
;
1858 if (type
== DM_TYPE_DAX_BIO_BASED
)
1859 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
1866 struct mapped_device
*dm_get_md(dev_t dev
)
1868 struct mapped_device
*md
;
1869 unsigned minor
= MINOR(dev
);
1871 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1874 spin_lock(&_minor_lock
);
1876 md
= idr_find(&_minor_idr
, minor
);
1878 if ((md
== MINOR_ALLOCED
||
1879 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1880 dm_deleting_md(md
) ||
1881 test_bit(DMF_FREEING
, &md
->flags
))) {
1889 spin_unlock(&_minor_lock
);
1893 EXPORT_SYMBOL_GPL(dm_get_md
);
1895 void *dm_get_mdptr(struct mapped_device
*md
)
1897 return md
->interface_ptr
;
1900 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1902 md
->interface_ptr
= ptr
;
1905 void dm_get(struct mapped_device
*md
)
1907 atomic_inc(&md
->holders
);
1908 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1911 int dm_hold(struct mapped_device
*md
)
1913 spin_lock(&_minor_lock
);
1914 if (test_bit(DMF_FREEING
, &md
->flags
)) {
1915 spin_unlock(&_minor_lock
);
1919 spin_unlock(&_minor_lock
);
1922 EXPORT_SYMBOL_GPL(dm_hold
);
1924 const char *dm_device_name(struct mapped_device
*md
)
1928 EXPORT_SYMBOL_GPL(dm_device_name
);
1930 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
1932 struct request_queue
*q
= dm_get_md_queue(md
);
1933 struct dm_table
*map
;
1938 spin_lock(&_minor_lock
);
1939 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
1940 set_bit(DMF_FREEING
, &md
->flags
);
1941 spin_unlock(&_minor_lock
);
1943 blk_set_queue_dying(q
);
1945 if (dm_request_based(md
) && md
->kworker_task
)
1946 kthread_flush_worker(&md
->kworker
);
1949 * Take suspend_lock so that presuspend and postsuspend methods
1950 * do not race with internal suspend.
1952 mutex_lock(&md
->suspend_lock
);
1953 map
= dm_get_live_table(md
, &srcu_idx
);
1954 if (!dm_suspended_md(md
)) {
1955 dm_table_presuspend_targets(map
);
1956 dm_table_postsuspend_targets(map
);
1958 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1959 dm_put_live_table(md
, srcu_idx
);
1960 mutex_unlock(&md
->suspend_lock
);
1963 * Rare, but there may be I/O requests still going to complete,
1964 * for example. Wait for all references to disappear.
1965 * No one should increment the reference count of the mapped_device,
1966 * after the mapped_device state becomes DMF_FREEING.
1969 while (atomic_read(&md
->holders
))
1971 else if (atomic_read(&md
->holders
))
1972 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1973 dm_device_name(md
), atomic_read(&md
->holders
));
1976 dm_table_destroy(__unbind(md
));
1980 void dm_destroy(struct mapped_device
*md
)
1982 __dm_destroy(md
, true);
1985 void dm_destroy_immediate(struct mapped_device
*md
)
1987 __dm_destroy(md
, false);
1990 void dm_put(struct mapped_device
*md
)
1992 atomic_dec(&md
->holders
);
1994 EXPORT_SYMBOL_GPL(dm_put
);
1996 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
2002 prepare_to_wait(&md
->wait
, &wait
, task_state
);
2004 if (!md_in_flight(md
))
2007 if (signal_pending_state(task_state
, current
)) {
2014 finish_wait(&md
->wait
, &wait
);
2020 * Process the deferred bios
2022 static void dm_wq_work(struct work_struct
*work
)
2024 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2028 struct dm_table
*map
;
2030 map
= dm_get_live_table(md
, &srcu_idx
);
2032 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2033 spin_lock_irq(&md
->deferred_lock
);
2034 c
= bio_list_pop(&md
->deferred
);
2035 spin_unlock_irq(&md
->deferred_lock
);
2040 if (dm_request_based(md
))
2041 generic_make_request(c
);
2043 __split_and_process_bio(md
, map
, c
);
2046 dm_put_live_table(md
, srcu_idx
);
2049 static void dm_queue_flush(struct mapped_device
*md
)
2051 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2052 smp_mb__after_atomic();
2053 queue_work(md
->wq
, &md
->work
);
2057 * Swap in a new table, returning the old one for the caller to destroy.
2059 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2061 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2062 struct queue_limits limits
;
2065 mutex_lock(&md
->suspend_lock
);
2067 /* device must be suspended */
2068 if (!dm_suspended_md(md
))
2072 * If the new table has no data devices, retain the existing limits.
2073 * This helps multipath with queue_if_no_path if all paths disappear,
2074 * then new I/O is queued based on these limits, and then some paths
2077 if (dm_table_has_no_data_devices(table
)) {
2078 live_map
= dm_get_live_table_fast(md
);
2080 limits
= md
->queue
->limits
;
2081 dm_put_live_table_fast(md
);
2085 r
= dm_calculate_queue_limits(table
, &limits
);
2092 map
= __bind(md
, table
, &limits
);
2095 mutex_unlock(&md
->suspend_lock
);
2100 * Functions to lock and unlock any filesystem running on the
2103 static int lock_fs(struct mapped_device
*md
)
2107 WARN_ON(md
->frozen_sb
);
2109 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2110 if (IS_ERR(md
->frozen_sb
)) {
2111 r
= PTR_ERR(md
->frozen_sb
);
2112 md
->frozen_sb
= NULL
;
2116 set_bit(DMF_FROZEN
, &md
->flags
);
2121 static void unlock_fs(struct mapped_device
*md
)
2123 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2126 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2127 md
->frozen_sb
= NULL
;
2128 clear_bit(DMF_FROZEN
, &md
->flags
);
2132 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2133 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2134 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2136 * If __dm_suspend returns 0, the device is completely quiescent
2137 * now. There is no request-processing activity. All new requests
2138 * are being added to md->deferred list.
2140 * Caller must hold md->suspend_lock
2142 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2143 unsigned suspend_flags
, long task_state
,
2144 int dmf_suspended_flag
)
2146 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2147 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2150 lockdep_assert_held(&md
->suspend_lock
);
2153 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2154 * This flag is cleared before dm_suspend returns.
2157 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2160 * This gets reverted if there's an error later and the targets
2161 * provide the .presuspend_undo hook.
2163 dm_table_presuspend_targets(map
);
2166 * Flush I/O to the device.
2167 * Any I/O submitted after lock_fs() may not be flushed.
2168 * noflush takes precedence over do_lockfs.
2169 * (lock_fs() flushes I/Os and waits for them to complete.)
2171 if (!noflush
&& do_lockfs
) {
2174 dm_table_presuspend_undo_targets(map
);
2180 * Here we must make sure that no processes are submitting requests
2181 * to target drivers i.e. no one may be executing
2182 * __split_and_process_bio. This is called from dm_request and
2185 * To get all processes out of __split_and_process_bio in dm_request,
2186 * we take the write lock. To prevent any process from reentering
2187 * __split_and_process_bio from dm_request and quiesce the thread
2188 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2189 * flush_workqueue(md->wq).
2191 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2193 synchronize_srcu(&md
->io_barrier
);
2196 * Stop md->queue before flushing md->wq in case request-based
2197 * dm defers requests to md->wq from md->queue.
2199 if (dm_request_based(md
)) {
2200 dm_stop_queue(md
->queue
);
2201 if (md
->kworker_task
)
2202 kthread_flush_worker(&md
->kworker
);
2205 flush_workqueue(md
->wq
);
2208 * At this point no more requests are entering target request routines.
2209 * We call dm_wait_for_completion to wait for all existing requests
2212 r
= dm_wait_for_completion(md
, task_state
);
2214 set_bit(dmf_suspended_flag
, &md
->flags
);
2217 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2219 synchronize_srcu(&md
->io_barrier
);
2221 /* were we interrupted ? */
2225 if (dm_request_based(md
))
2226 dm_start_queue(md
->queue
);
2229 dm_table_presuspend_undo_targets(map
);
2230 /* pushback list is already flushed, so skip flush */
2237 * We need to be able to change a mapping table under a mounted
2238 * filesystem. For example we might want to move some data in
2239 * the background. Before the table can be swapped with
2240 * dm_bind_table, dm_suspend must be called to flush any in
2241 * flight bios and ensure that any further io gets deferred.
2244 * Suspend mechanism in request-based dm.
2246 * 1. Flush all I/Os by lock_fs() if needed.
2247 * 2. Stop dispatching any I/O by stopping the request_queue.
2248 * 3. Wait for all in-flight I/Os to be completed or requeued.
2250 * To abort suspend, start the request_queue.
2252 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2254 struct dm_table
*map
= NULL
;
2258 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2260 if (dm_suspended_md(md
)) {
2265 if (dm_suspended_internally_md(md
)) {
2266 /* already internally suspended, wait for internal resume */
2267 mutex_unlock(&md
->suspend_lock
);
2268 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2274 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2276 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2280 dm_table_postsuspend_targets(map
);
2283 mutex_unlock(&md
->suspend_lock
);
2287 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2290 int r
= dm_table_resume_targets(map
);
2298 * Flushing deferred I/Os must be done after targets are resumed
2299 * so that mapping of targets can work correctly.
2300 * Request-based dm is queueing the deferred I/Os in its request_queue.
2302 if (dm_request_based(md
))
2303 dm_start_queue(md
->queue
);
2310 int dm_resume(struct mapped_device
*md
)
2313 struct dm_table
*map
= NULL
;
2317 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2319 if (!dm_suspended_md(md
))
2322 if (dm_suspended_internally_md(md
)) {
2323 /* already internally suspended, wait for internal resume */
2324 mutex_unlock(&md
->suspend_lock
);
2325 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2331 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2332 if (!map
|| !dm_table_get_size(map
))
2335 r
= __dm_resume(md
, map
);
2339 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2341 mutex_unlock(&md
->suspend_lock
);
2347 * Internal suspend/resume works like userspace-driven suspend. It waits
2348 * until all bios finish and prevents issuing new bios to the target drivers.
2349 * It may be used only from the kernel.
2352 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2354 struct dm_table
*map
= NULL
;
2356 if (md
->internal_suspend_count
++)
2357 return; /* nested internal suspend */
2359 if (dm_suspended_md(md
)) {
2360 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2361 return; /* nest suspend */
2364 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2367 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2368 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2369 * would require changing .presuspend to return an error -- avoid this
2370 * until there is a need for more elaborate variants of internal suspend.
2372 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2373 DMF_SUSPENDED_INTERNALLY
);
2375 dm_table_postsuspend_targets(map
);
2378 static void __dm_internal_resume(struct mapped_device
*md
)
2380 BUG_ON(!md
->internal_suspend_count
);
2382 if (--md
->internal_suspend_count
)
2383 return; /* resume from nested internal suspend */
2385 if (dm_suspended_md(md
))
2386 goto done
; /* resume from nested suspend */
2389 * NOTE: existing callers don't need to call dm_table_resume_targets
2390 * (which may fail -- so best to avoid it for now by passing NULL map)
2392 (void) __dm_resume(md
, NULL
);
2395 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2396 smp_mb__after_atomic();
2397 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2400 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2402 mutex_lock(&md
->suspend_lock
);
2403 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2404 mutex_unlock(&md
->suspend_lock
);
2406 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2408 void dm_internal_resume(struct mapped_device
*md
)
2410 mutex_lock(&md
->suspend_lock
);
2411 __dm_internal_resume(md
);
2412 mutex_unlock(&md
->suspend_lock
);
2414 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2417 * Fast variants of internal suspend/resume hold md->suspend_lock,
2418 * which prevents interaction with userspace-driven suspend.
2421 void dm_internal_suspend_fast(struct mapped_device
*md
)
2423 mutex_lock(&md
->suspend_lock
);
2424 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2427 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2428 synchronize_srcu(&md
->io_barrier
);
2429 flush_workqueue(md
->wq
);
2430 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2432 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2434 void dm_internal_resume_fast(struct mapped_device
*md
)
2436 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2442 mutex_unlock(&md
->suspend_lock
);
2444 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2446 /*-----------------------------------------------------------------
2447 * Event notification.
2448 *---------------------------------------------------------------*/
2449 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2452 char udev_cookie
[DM_COOKIE_LENGTH
];
2453 char *envp
[] = { udev_cookie
, NULL
};
2456 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2458 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2459 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2460 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2465 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2467 return atomic_add_return(1, &md
->uevent_seq
);
2470 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2472 return atomic_read(&md
->event_nr
);
2475 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2477 return wait_event_interruptible(md
->eventq
,
2478 (event_nr
!= atomic_read(&md
->event_nr
)));
2481 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2483 unsigned long flags
;
2485 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2486 list_add(elist
, &md
->uevent_list
);
2487 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2491 * The gendisk is only valid as long as you have a reference
2494 struct gendisk
*dm_disk(struct mapped_device
*md
)
2498 EXPORT_SYMBOL_GPL(dm_disk
);
2500 struct kobject
*dm_kobject(struct mapped_device
*md
)
2502 return &md
->kobj_holder
.kobj
;
2505 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2507 struct mapped_device
*md
;
2509 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2511 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2519 int dm_suspended_md(struct mapped_device
*md
)
2521 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2524 int dm_suspended_internally_md(struct mapped_device
*md
)
2526 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2529 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2531 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2534 int dm_suspended(struct dm_target
*ti
)
2536 return dm_suspended_md(dm_table_get_md(ti
->table
));
2538 EXPORT_SYMBOL_GPL(dm_suspended
);
2540 int dm_noflush_suspending(struct dm_target
*ti
)
2542 return __noflush_suspending(dm_table_get_md(ti
->table
));
2544 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2546 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
2547 unsigned integrity
, unsigned per_io_data_size
)
2549 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2550 unsigned int pool_size
= 0;
2551 unsigned int front_pad
;
2557 case DM_TYPE_BIO_BASED
:
2558 case DM_TYPE_DAX_BIO_BASED
:
2559 pool_size
= dm_get_reserved_bio_based_ios();
2560 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2562 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2563 if (!pools
->io_pool
)
2566 case DM_TYPE_REQUEST_BASED
:
2567 case DM_TYPE_MQ_REQUEST_BASED
:
2568 pool_size
= dm_get_reserved_rq_based_ios();
2569 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2570 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2576 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
2580 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2586 dm_free_md_mempools(pools
);
2591 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2596 mempool_destroy(pools
->io_pool
);
2599 bioset_free(pools
->bs
);
2611 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2614 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2615 struct dm_table
*table
;
2616 struct dm_target
*ti
;
2617 int ret
= -ENOTTY
, srcu_idx
;
2619 table
= dm_get_live_table(md
, &srcu_idx
);
2620 if (!table
|| !dm_table_get_size(table
))
2623 /* We only support devices that have a single target */
2624 if (dm_table_get_num_targets(table
) != 1)
2626 ti
= dm_table_get_target(table
, 0);
2629 if (!ti
->type
->iterate_devices
)
2632 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2634 dm_put_live_table(md
, srcu_idx
);
2639 * For register / unregister we need to manually call out to every path.
2641 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2642 sector_t start
, sector_t len
, void *data
)
2644 struct dm_pr
*pr
= data
;
2645 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2647 if (!ops
|| !ops
->pr_register
)
2649 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2652 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2663 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2664 if (ret
&& new_key
) {
2665 /* unregister all paths if we failed to register any path */
2666 pr
.old_key
= new_key
;
2669 pr
.fail_early
= false;
2670 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2676 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2679 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2680 const struct pr_ops
*ops
;
2684 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2688 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2689 if (ops
&& ops
->pr_reserve
)
2690 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2698 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2700 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2701 const struct pr_ops
*ops
;
2705 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2709 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2710 if (ops
&& ops
->pr_release
)
2711 r
= ops
->pr_release(bdev
, key
, type
);
2719 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2720 enum pr_type type
, bool abort
)
2722 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2723 const struct pr_ops
*ops
;
2727 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2731 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2732 if (ops
&& ops
->pr_preempt
)
2733 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2741 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2743 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2744 const struct pr_ops
*ops
;
2748 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2752 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2753 if (ops
&& ops
->pr_clear
)
2754 r
= ops
->pr_clear(bdev
, key
);
2762 static const struct pr_ops dm_pr_ops
= {
2763 .pr_register
= dm_pr_register
,
2764 .pr_reserve
= dm_pr_reserve
,
2765 .pr_release
= dm_pr_release
,
2766 .pr_preempt
= dm_pr_preempt
,
2767 .pr_clear
= dm_pr_clear
,
2770 static const struct block_device_operations dm_blk_dops
= {
2771 .open
= dm_blk_open
,
2772 .release
= dm_blk_close
,
2773 .ioctl
= dm_blk_ioctl
,
2774 .direct_access
= dm_blk_direct_access
,
2775 .getgeo
= dm_blk_getgeo
,
2776 .pr_ops
= &dm_pr_ops
,
2777 .owner
= THIS_MODULE
2783 module_init(dm_init
);
2784 module_exit(dm_exit
);
2786 module_param(major
, uint
, 0);
2787 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2789 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
2790 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
2792 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
2793 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
2795 MODULE_DESCRIPTION(DM_NAME
" driver");
2796 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2797 MODULE_LICENSE("GPL");