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
469 * a logical partition of the parent bdev; so extra
470 * validation is needed.
472 r
= scsi_verify_blk_ioctl(NULL
, cmd
);
477 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
483 static struct dm_io
*alloc_io(struct mapped_device
*md
)
485 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
488 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
490 mempool_free(io
, md
->io_pool
);
493 static void free_tio(struct dm_target_io
*tio
)
495 bio_put(&tio
->clone
);
498 int md_in_flight(struct mapped_device
*md
)
500 return atomic_read(&md
->pending
[READ
]) +
501 atomic_read(&md
->pending
[WRITE
]);
504 static void start_io_acct(struct dm_io
*io
)
506 struct mapped_device
*md
= io
->md
;
507 struct bio
*bio
= io
->bio
;
509 int rw
= bio_data_dir(bio
);
511 io
->start_time
= jiffies
;
513 cpu
= part_stat_lock();
514 part_round_stats(cpu
, &dm_disk(md
)->part0
);
516 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
517 atomic_inc_return(&md
->pending
[rw
]));
519 if (unlikely(dm_stats_used(&md
->stats
)))
520 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
521 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
522 false, 0, &io
->stats_aux
);
525 static void end_io_acct(struct dm_io
*io
)
527 struct mapped_device
*md
= io
->md
;
528 struct bio
*bio
= io
->bio
;
529 unsigned long duration
= jiffies
- io
->start_time
;
531 int rw
= bio_data_dir(bio
);
533 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
535 if (unlikely(dm_stats_used(&md
->stats
)))
536 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
537 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
538 true, duration
, &io
->stats_aux
);
541 * After this is decremented the bio must not be touched if it is
544 pending
= atomic_dec_return(&md
->pending
[rw
]);
545 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
546 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
548 /* nudge anyone waiting on suspend queue */
554 * Add the bio to the list of deferred io.
556 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
560 spin_lock_irqsave(&md
->deferred_lock
, flags
);
561 bio_list_add(&md
->deferred
, bio
);
562 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
563 queue_work(md
->wq
, &md
->work
);
567 * Everyone (including functions in this file), should use this
568 * function to access the md->map field, and make sure they call
569 * dm_put_live_table() when finished.
571 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
573 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
575 return srcu_dereference(md
->map
, &md
->io_barrier
);
578 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
580 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
583 void dm_sync_table(struct mapped_device
*md
)
585 synchronize_srcu(&md
->io_barrier
);
586 synchronize_rcu_expedited();
590 * A fast alternative to dm_get_live_table/dm_put_live_table.
591 * The caller must not block between these two functions.
593 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
596 return rcu_dereference(md
->map
);
599 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
605 * Open a table device so we can use it as a map destination.
607 static int open_table_device(struct table_device
*td
, dev_t dev
,
608 struct mapped_device
*md
)
610 static char *_claim_ptr
= "I belong to device-mapper";
611 struct block_device
*bdev
;
615 BUG_ON(td
->dm_dev
.bdev
);
617 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
619 return PTR_ERR(bdev
);
621 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
623 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
627 td
->dm_dev
.bdev
= bdev
;
632 * Close a table device that we've been using.
634 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
636 if (!td
->dm_dev
.bdev
)
639 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
640 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
641 td
->dm_dev
.bdev
= NULL
;
644 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
646 struct table_device
*td
;
648 list_for_each_entry(td
, l
, list
)
649 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
655 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
656 struct dm_dev
**result
) {
658 struct table_device
*td
;
660 mutex_lock(&md
->table_devices_lock
);
661 td
= find_table_device(&md
->table_devices
, dev
, mode
);
663 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
665 mutex_unlock(&md
->table_devices_lock
);
669 td
->dm_dev
.mode
= mode
;
670 td
->dm_dev
.bdev
= NULL
;
672 if ((r
= open_table_device(td
, dev
, md
))) {
673 mutex_unlock(&md
->table_devices_lock
);
678 format_dev_t(td
->dm_dev
.name
, dev
);
680 atomic_set(&td
->count
, 0);
681 list_add(&td
->list
, &md
->table_devices
);
683 atomic_inc(&td
->count
);
684 mutex_unlock(&md
->table_devices_lock
);
686 *result
= &td
->dm_dev
;
689 EXPORT_SYMBOL_GPL(dm_get_table_device
);
691 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
693 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
695 mutex_lock(&md
->table_devices_lock
);
696 if (atomic_dec_and_test(&td
->count
)) {
697 close_table_device(td
, md
);
701 mutex_unlock(&md
->table_devices_lock
);
703 EXPORT_SYMBOL(dm_put_table_device
);
705 static void free_table_devices(struct list_head
*devices
)
707 struct list_head
*tmp
, *next
;
709 list_for_each_safe(tmp
, next
, devices
) {
710 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
712 DMWARN("dm_destroy: %s still exists with %d references",
713 td
->dm_dev
.name
, atomic_read(&td
->count
));
719 * Get the geometry associated with a dm device
721 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
729 * Set the geometry of a device.
731 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
733 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
735 if (geo
->start
> sz
) {
736 DMWARN("Start sector is beyond the geometry limits.");
745 /*-----------------------------------------------------------------
747 * A more elegant soln is in the works that uses the queue
748 * merge fn, unfortunately there are a couple of changes to
749 * the block layer that I want to make for this. So in the
750 * interests of getting something for people to use I give
751 * you this clearly demarcated crap.
752 *---------------------------------------------------------------*/
754 static int __noflush_suspending(struct mapped_device
*md
)
756 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
760 * Decrements the number of outstanding ios that a bio has been
761 * cloned into, completing the original io if necc.
763 static void dec_pending(struct dm_io
*io
, int error
)
768 struct mapped_device
*md
= io
->md
;
770 /* Push-back supersedes any I/O errors */
771 if (unlikely(error
)) {
772 spin_lock_irqsave(&io
->endio_lock
, flags
);
773 if (!(io
->error
> 0 && __noflush_suspending(md
)))
775 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
778 if (atomic_dec_and_test(&io
->io_count
)) {
779 if (io
->error
== DM_ENDIO_REQUEUE
) {
781 * Target requested pushing back the I/O.
783 spin_lock_irqsave(&md
->deferred_lock
, flags
);
784 if (__noflush_suspending(md
))
785 bio_list_add_head(&md
->deferred
, io
->bio
);
787 /* noflush suspend was interrupted. */
789 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
792 io_error
= io
->error
;
797 if (io_error
== DM_ENDIO_REQUEUE
)
800 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
802 * Preflush done for flush with data, reissue
803 * without REQ_PREFLUSH.
805 bio
->bi_opf
&= ~REQ_PREFLUSH
;
808 /* done with normal IO or empty flush */
809 trace_block_bio_complete(md
->queue
, bio
, io_error
);
810 bio
->bi_error
= io_error
;
816 void disable_write_same(struct mapped_device
*md
)
818 struct queue_limits
*limits
= dm_get_queue_limits(md
);
820 /* device doesn't really support WRITE SAME, disable it */
821 limits
->max_write_same_sectors
= 0;
824 static void clone_endio(struct bio
*bio
)
826 int error
= bio
->bi_error
;
828 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
829 struct dm_io
*io
= tio
->io
;
830 struct mapped_device
*md
= tio
->io
->md
;
831 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
834 r
= endio(tio
->ti
, bio
, error
);
835 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
837 * error and requeue request are handled
841 else if (r
== DM_ENDIO_INCOMPLETE
)
842 /* The target will handle the io */
845 DMWARN("unimplemented target endio return value: %d", r
);
850 if (unlikely(r
== -EREMOTEIO
&& (bio_op(bio
) == REQ_OP_WRITE_SAME
) &&
851 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
852 disable_write_same(md
);
855 dec_pending(io
, error
);
859 * Return maximum size of I/O possible at the supplied sector up to the current
862 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
864 sector_t target_offset
= dm_target_offset(ti
, sector
);
866 return ti
->len
- target_offset
;
869 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
871 sector_t len
= max_io_len_target_boundary(sector
, ti
);
872 sector_t offset
, max_len
;
875 * Does the target need to split even further?
877 if (ti
->max_io_len
) {
878 offset
= dm_target_offset(ti
, sector
);
879 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
880 max_len
= sector_div(offset
, ti
->max_io_len
);
882 max_len
= offset
& (ti
->max_io_len
- 1);
883 max_len
= ti
->max_io_len
- max_len
;
892 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
894 if (len
> UINT_MAX
) {
895 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
896 (unsigned long long)len
, UINT_MAX
);
897 ti
->error
= "Maximum size of target IO is too large";
901 ti
->max_io_len
= (uint32_t) len
;
905 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
907 static long dm_blk_direct_access(struct block_device
*bdev
, sector_t sector
,
908 void **kaddr
, pfn_t
*pfn
, long size
)
910 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
911 struct dm_table
*map
;
912 struct dm_target
*ti
;
914 long len
, ret
= -EIO
;
916 map
= dm_get_live_table(md
, &srcu_idx
);
920 ti
= dm_table_find_target(map
, sector
);
921 if (!dm_target_is_valid(ti
))
924 len
= max_io_len(sector
, ti
) << SECTOR_SHIFT
;
925 size
= min(len
, size
);
927 if (ti
->type
->direct_access
)
928 ret
= ti
->type
->direct_access(ti
, sector
, kaddr
, pfn
, size
);
930 dm_put_live_table(md
, srcu_idx
);
931 return min(ret
, size
);
935 * A target may call dm_accept_partial_bio only from the map routine. It is
936 * allowed for all bio types except REQ_PREFLUSH.
938 * dm_accept_partial_bio informs the dm that the target only wants to process
939 * additional n_sectors sectors of the bio and the rest of the data should be
940 * sent in a next bio.
942 * A diagram that explains the arithmetics:
943 * +--------------------+---------------+-------+
945 * +--------------------+---------------+-------+
947 * <-------------- *tio->len_ptr --------------->
948 * <------- bi_size ------->
951 * Region 1 was already iterated over with bio_advance or similar function.
952 * (it may be empty if the target doesn't use bio_advance)
953 * Region 2 is the remaining bio size that the target wants to process.
954 * (it may be empty if region 1 is non-empty, although there is no reason
956 * The target requires that region 3 is to be sent in the next bio.
958 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
959 * the partially processed part (the sum of regions 1+2) must be the same for all
962 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
964 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
965 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
966 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
967 BUG_ON(bi_size
> *tio
->len_ptr
);
968 BUG_ON(n_sectors
> bi_size
);
969 *tio
->len_ptr
-= bi_size
- n_sectors
;
970 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
972 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
974 static void __map_bio(struct dm_target_io
*tio
)
978 struct bio
*clone
= &tio
->clone
;
979 struct dm_target
*ti
= tio
->ti
;
981 clone
->bi_end_io
= clone_endio
;
984 * Map the clone. If r == 0 we don't need to do
985 * anything, the target has assumed ownership of
988 atomic_inc(&tio
->io
->io_count
);
989 sector
= clone
->bi_iter
.bi_sector
;
990 r
= ti
->type
->map(ti
, clone
);
991 if (r
== DM_MAPIO_REMAPPED
) {
992 /* the bio has been remapped so dispatch it */
994 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
995 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
997 generic_make_request(clone
);
998 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
999 /* error the io and bail out, or requeue it if needed */
1000 dec_pending(tio
->io
, r
);
1002 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1003 DMWARN("unimplemented target map return value: %d", r
);
1009 struct mapped_device
*md
;
1010 struct dm_table
*map
;
1014 unsigned sector_count
;
1017 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1019 bio
->bi_iter
.bi_sector
= sector
;
1020 bio
->bi_iter
.bi_size
= to_bytes(len
);
1024 * Creates a bio that consists of range of complete bvecs.
1026 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1027 sector_t sector
, unsigned len
)
1029 struct bio
*clone
= &tio
->clone
;
1031 __bio_clone_fast(clone
, bio
);
1033 if (bio_integrity(bio
)) {
1034 int r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1039 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1040 clone
->bi_iter
.bi_size
= to_bytes(len
);
1042 if (bio_integrity(bio
))
1043 bio_integrity_trim(clone
, 0, len
);
1048 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1049 struct dm_target
*ti
,
1050 unsigned target_bio_nr
)
1052 struct dm_target_io
*tio
;
1055 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1056 tio
= container_of(clone
, struct dm_target_io
, clone
);
1060 tio
->target_bio_nr
= target_bio_nr
;
1065 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1066 struct dm_target
*ti
,
1067 unsigned target_bio_nr
, unsigned *len
)
1069 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1070 struct bio
*clone
= &tio
->clone
;
1074 __bio_clone_fast(clone
, ci
->bio
);
1076 bio_setup_sector(clone
, ci
->sector
, *len
);
1081 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1082 unsigned num_bios
, unsigned *len
)
1084 unsigned target_bio_nr
;
1086 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1087 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1090 static int __send_empty_flush(struct clone_info
*ci
)
1092 unsigned target_nr
= 0;
1093 struct dm_target
*ti
;
1095 BUG_ON(bio_has_data(ci
->bio
));
1096 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1097 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1102 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1103 sector_t sector
, unsigned *len
)
1105 struct bio
*bio
= ci
->bio
;
1106 struct dm_target_io
*tio
;
1107 unsigned target_bio_nr
;
1108 unsigned num_target_bios
= 1;
1112 * Does the target want to receive duplicate copies of the bio?
1114 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1115 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1117 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1118 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1120 r
= clone_bio(tio
, bio
, sector
, *len
);
1131 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1133 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1135 return ti
->num_discard_bios
;
1138 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1140 return ti
->num_write_same_bios
;
1143 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1145 static bool is_split_required_for_discard(struct dm_target
*ti
)
1147 return ti
->split_discard_bios
;
1150 static int __send_changing_extent_only(struct clone_info
*ci
,
1151 get_num_bios_fn get_num_bios
,
1152 is_split_required_fn is_split_required
)
1154 struct dm_target
*ti
;
1159 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1160 if (!dm_target_is_valid(ti
))
1164 * Even though the device advertised support for this type of
1165 * request, that does not mean every target supports it, and
1166 * reconfiguration might also have changed that since the
1167 * check was performed.
1169 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1173 if (is_split_required
&& !is_split_required(ti
))
1174 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1176 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1178 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1181 } while (ci
->sector_count
-= len
);
1186 static int __send_discard(struct clone_info
*ci
)
1188 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1189 is_split_required_for_discard
);
1192 static int __send_write_same(struct clone_info
*ci
)
1194 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1198 * Select the correct strategy for processing a non-flush bio.
1200 static int __split_and_process_non_flush(struct clone_info
*ci
)
1202 struct bio
*bio
= ci
->bio
;
1203 struct dm_target
*ti
;
1207 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1208 return __send_discard(ci
);
1209 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1210 return __send_write_same(ci
);
1212 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1213 if (!dm_target_is_valid(ti
))
1216 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1218 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1223 ci
->sector_count
-= len
;
1229 * Entry point to split a bio into clones and submit them to the targets.
1231 static void __split_and_process_bio(struct mapped_device
*md
,
1232 struct dm_table
*map
, struct bio
*bio
)
1234 struct clone_info ci
;
1237 if (unlikely(!map
)) {
1244 ci
.io
= alloc_io(md
);
1246 atomic_set(&ci
.io
->io_count
, 1);
1249 spin_lock_init(&ci
.io
->endio_lock
);
1250 ci
.sector
= bio
->bi_iter
.bi_sector
;
1252 start_io_acct(ci
.io
);
1254 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1255 ci
.bio
= &ci
.md
->flush_bio
;
1256 ci
.sector_count
= 0;
1257 error
= __send_empty_flush(&ci
);
1258 /* dec_pending submits any data associated with flush */
1261 ci
.sector_count
= bio_sectors(bio
);
1262 while (ci
.sector_count
&& !error
)
1263 error
= __split_and_process_non_flush(&ci
);
1266 /* drop the extra reference count */
1267 dec_pending(ci
.io
, error
);
1269 /*-----------------------------------------------------------------
1271 *---------------------------------------------------------------*/
1274 * The request function that just remaps the bio built up by
1277 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1279 int rw
= bio_data_dir(bio
);
1280 struct mapped_device
*md
= q
->queuedata
;
1282 struct dm_table
*map
;
1284 map
= dm_get_live_table(md
, &srcu_idx
);
1286 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1288 /* if we're suspended, we have to queue this io for later */
1289 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1290 dm_put_live_table(md
, srcu_idx
);
1292 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1296 return BLK_QC_T_NONE
;
1299 __split_and_process_bio(md
, map
, bio
);
1300 dm_put_live_table(md
, srcu_idx
);
1301 return BLK_QC_T_NONE
;
1304 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1307 struct mapped_device
*md
= congested_data
;
1308 struct dm_table
*map
;
1310 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1311 if (dm_request_based(md
)) {
1313 * With request-based DM we only need to check the
1314 * top-level queue for congestion.
1316 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1318 map
= dm_get_live_table_fast(md
);
1320 r
= dm_table_any_congested(map
, bdi_bits
);
1321 dm_put_live_table_fast(md
);
1328 /*-----------------------------------------------------------------
1329 * An IDR is used to keep track of allocated minor numbers.
1330 *---------------------------------------------------------------*/
1331 static void free_minor(int minor
)
1333 spin_lock(&_minor_lock
);
1334 idr_remove(&_minor_idr
, minor
);
1335 spin_unlock(&_minor_lock
);
1339 * See if the device with a specific minor # is free.
1341 static int specific_minor(int minor
)
1345 if (minor
>= (1 << MINORBITS
))
1348 idr_preload(GFP_KERNEL
);
1349 spin_lock(&_minor_lock
);
1351 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1353 spin_unlock(&_minor_lock
);
1356 return r
== -ENOSPC
? -EBUSY
: r
;
1360 static int next_free_minor(int *minor
)
1364 idr_preload(GFP_KERNEL
);
1365 spin_lock(&_minor_lock
);
1367 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1369 spin_unlock(&_minor_lock
);
1377 static const struct block_device_operations dm_blk_dops
;
1379 static void dm_wq_work(struct work_struct
*work
);
1381 void dm_init_md_queue(struct mapped_device
*md
)
1384 * Request-based dm devices cannot be stacked on top of bio-based dm
1385 * devices. The type of this dm device may not have been decided yet.
1386 * The type is decided at the first table loading time.
1387 * To prevent problematic device stacking, clear the queue flag
1388 * for request stacking support until then.
1390 * This queue is new, so no concurrency on the queue_flags.
1392 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1395 * Initialize data that will only be used by a non-blk-mq DM queue
1396 * - must do so here (in alloc_dev callchain) before queue is used
1398 md
->queue
->queuedata
= md
;
1399 md
->queue
->backing_dev_info
->congested_data
= md
;
1402 void dm_init_normal_md_queue(struct mapped_device
*md
)
1404 md
->use_blk_mq
= false;
1405 dm_init_md_queue(md
);
1408 * Initialize aspects of queue that aren't relevant for blk-mq
1410 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1411 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1414 static void cleanup_mapped_device(struct mapped_device
*md
)
1417 destroy_workqueue(md
->wq
);
1418 if (md
->kworker_task
)
1419 kthread_stop(md
->kworker_task
);
1420 mempool_destroy(md
->io_pool
);
1422 bioset_free(md
->bs
);
1425 spin_lock(&_minor_lock
);
1426 md
->disk
->private_data
= NULL
;
1427 spin_unlock(&_minor_lock
);
1428 del_gendisk(md
->disk
);
1433 blk_cleanup_queue(md
->queue
);
1435 cleanup_srcu_struct(&md
->io_barrier
);
1442 dm_mq_cleanup_mapped_device(md
);
1446 * Allocate and initialise a blank device with a given minor.
1448 static struct mapped_device
*alloc_dev(int minor
)
1450 int r
, numa_node_id
= dm_get_numa_node();
1451 struct mapped_device
*md
;
1454 md
= kzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1456 DMWARN("unable to allocate device, out of memory.");
1460 if (!try_module_get(THIS_MODULE
))
1461 goto bad_module_get
;
1463 /* get a minor number for the dev */
1464 if (minor
== DM_ANY_MINOR
)
1465 r
= next_free_minor(&minor
);
1467 r
= specific_minor(minor
);
1471 r
= init_srcu_struct(&md
->io_barrier
);
1473 goto bad_io_barrier
;
1475 md
->numa_node_id
= numa_node_id
;
1476 md
->use_blk_mq
= dm_use_blk_mq_default();
1477 md
->init_tio_pdu
= false;
1478 md
->type
= DM_TYPE_NONE
;
1479 mutex_init(&md
->suspend_lock
);
1480 mutex_init(&md
->type_lock
);
1481 mutex_init(&md
->table_devices_lock
);
1482 spin_lock_init(&md
->deferred_lock
);
1483 atomic_set(&md
->holders
, 1);
1484 atomic_set(&md
->open_count
, 0);
1485 atomic_set(&md
->event_nr
, 0);
1486 atomic_set(&md
->uevent_seq
, 0);
1487 INIT_LIST_HEAD(&md
->uevent_list
);
1488 INIT_LIST_HEAD(&md
->table_devices
);
1489 spin_lock_init(&md
->uevent_lock
);
1491 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1495 dm_init_md_queue(md
);
1497 md
->disk
= alloc_disk_node(1, numa_node_id
);
1501 atomic_set(&md
->pending
[0], 0);
1502 atomic_set(&md
->pending
[1], 0);
1503 init_waitqueue_head(&md
->wait
);
1504 INIT_WORK(&md
->work
, dm_wq_work
);
1505 init_waitqueue_head(&md
->eventq
);
1506 init_completion(&md
->kobj_holder
.completion
);
1507 md
->kworker_task
= NULL
;
1509 md
->disk
->major
= _major
;
1510 md
->disk
->first_minor
= minor
;
1511 md
->disk
->fops
= &dm_blk_dops
;
1512 md
->disk
->queue
= md
->queue
;
1513 md
->disk
->private_data
= md
;
1514 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1516 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1518 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1522 md
->bdev
= bdget_disk(md
->disk
, 0);
1526 bio_init(&md
->flush_bio
, NULL
, 0);
1527 md
->flush_bio
.bi_bdev
= md
->bdev
;
1528 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
1530 dm_stats_init(&md
->stats
);
1532 /* Populate the mapping, nobody knows we exist yet */
1533 spin_lock(&_minor_lock
);
1534 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1535 spin_unlock(&_minor_lock
);
1537 BUG_ON(old_md
!= MINOR_ALLOCED
);
1542 cleanup_mapped_device(md
);
1546 module_put(THIS_MODULE
);
1552 static void unlock_fs(struct mapped_device
*md
);
1554 static void free_dev(struct mapped_device
*md
)
1556 int minor
= MINOR(disk_devt(md
->disk
));
1560 cleanup_mapped_device(md
);
1562 free_table_devices(&md
->table_devices
);
1563 dm_stats_cleanup(&md
->stats
);
1566 module_put(THIS_MODULE
);
1570 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1572 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1575 /* The md already has necessary mempools. */
1576 if (dm_table_bio_based(t
)) {
1578 * Reload bioset because front_pad may have changed
1579 * because a different table was loaded.
1581 bioset_free(md
->bs
);
1586 * There's no need to reload with request-based dm
1587 * because the size of front_pad doesn't change.
1588 * Note for future: If you are to reload bioset,
1589 * prep-ed requests in the queue may refer
1590 * to bio from the old bioset, so you must walk
1591 * through the queue to unprep.
1596 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1598 md
->io_pool
= p
->io_pool
;
1604 /* mempool bind completed, no longer need any mempools in the table */
1605 dm_table_free_md_mempools(t
);
1609 * Bind a table to the device.
1611 static void event_callback(void *context
)
1613 unsigned long flags
;
1615 struct mapped_device
*md
= (struct mapped_device
*) context
;
1617 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1618 list_splice_init(&md
->uevent_list
, &uevents
);
1619 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1621 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1623 atomic_inc(&md
->event_nr
);
1624 wake_up(&md
->eventq
);
1628 * Protected by md->suspend_lock obtained by dm_swap_table().
1630 static void __set_size(struct mapped_device
*md
, sector_t size
)
1632 set_capacity(md
->disk
, size
);
1634 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1638 * Returns old map, which caller must destroy.
1640 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1641 struct queue_limits
*limits
)
1643 struct dm_table
*old_map
;
1644 struct request_queue
*q
= md
->queue
;
1647 lockdep_assert_held(&md
->suspend_lock
);
1649 size
= dm_table_get_size(t
);
1652 * Wipe any geometry if the size of the table changed.
1654 if (size
!= dm_get_size(md
))
1655 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1657 __set_size(md
, size
);
1659 dm_table_event_callback(t
, event_callback
, md
);
1662 * The queue hasn't been stopped yet, if the old table type wasn't
1663 * for request-based during suspension. So stop it to prevent
1664 * I/O mapping before resume.
1665 * This must be done before setting the queue restrictions,
1666 * because request-based dm may be run just after the setting.
1668 if (dm_table_request_based(t
)) {
1671 * Leverage the fact that request-based DM targets are
1672 * immutable singletons and establish md->immutable_target
1673 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1675 md
->immutable_target
= dm_table_get_immutable_target(t
);
1678 __bind_mempools(md
, t
);
1680 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1681 rcu_assign_pointer(md
->map
, (void *)t
);
1682 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1684 dm_table_set_restrictions(t
, q
, limits
);
1692 * Returns unbound table for the caller to free.
1694 static struct dm_table
*__unbind(struct mapped_device
*md
)
1696 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1701 dm_table_event_callback(map
, NULL
, NULL
);
1702 RCU_INIT_POINTER(md
->map
, NULL
);
1709 * Constructor for a new device.
1711 int dm_create(int minor
, struct mapped_device
**result
)
1713 struct mapped_device
*md
;
1715 md
= alloc_dev(minor
);
1726 * Functions to manage md->type.
1727 * All are required to hold md->type_lock.
1729 void dm_lock_md_type(struct mapped_device
*md
)
1731 mutex_lock(&md
->type_lock
);
1734 void dm_unlock_md_type(struct mapped_device
*md
)
1736 mutex_unlock(&md
->type_lock
);
1739 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
1741 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1745 unsigned dm_get_md_type(struct mapped_device
*md
)
1750 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
1752 return md
->immutable_target_type
;
1756 * The queue_limits are only valid as long as you have a reference
1759 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
1761 BUG_ON(!atomic_read(&md
->holders
));
1762 return &md
->queue
->limits
;
1764 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
1767 * Setup the DM device's queue based on md's type
1769 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
1772 unsigned type
= dm_get_md_type(md
);
1775 case DM_TYPE_REQUEST_BASED
:
1776 r
= dm_old_init_request_queue(md
, t
);
1778 DMERR("Cannot initialize queue for request-based mapped device");
1782 case DM_TYPE_MQ_REQUEST_BASED
:
1783 r
= dm_mq_init_request_queue(md
, t
);
1785 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1789 case DM_TYPE_BIO_BASED
:
1790 case DM_TYPE_DAX_BIO_BASED
:
1791 dm_init_normal_md_queue(md
);
1792 blk_queue_make_request(md
->queue
, dm_make_request
);
1794 * DM handles splitting bios as needed. Free the bio_split bioset
1795 * since it won't be used (saves 1 process per bio-based DM device).
1797 bioset_free(md
->queue
->bio_split
);
1798 md
->queue
->bio_split
= NULL
;
1800 if (type
== DM_TYPE_DAX_BIO_BASED
)
1801 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
1808 struct mapped_device
*dm_get_md(dev_t dev
)
1810 struct mapped_device
*md
;
1811 unsigned minor
= MINOR(dev
);
1813 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1816 spin_lock(&_minor_lock
);
1818 md
= idr_find(&_minor_idr
, minor
);
1820 if ((md
== MINOR_ALLOCED
||
1821 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1822 dm_deleting_md(md
) ||
1823 test_bit(DMF_FREEING
, &md
->flags
))) {
1831 spin_unlock(&_minor_lock
);
1835 EXPORT_SYMBOL_GPL(dm_get_md
);
1837 void *dm_get_mdptr(struct mapped_device
*md
)
1839 return md
->interface_ptr
;
1842 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1844 md
->interface_ptr
= ptr
;
1847 void dm_get(struct mapped_device
*md
)
1849 atomic_inc(&md
->holders
);
1850 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1853 int dm_hold(struct mapped_device
*md
)
1855 spin_lock(&_minor_lock
);
1856 if (test_bit(DMF_FREEING
, &md
->flags
)) {
1857 spin_unlock(&_minor_lock
);
1861 spin_unlock(&_minor_lock
);
1864 EXPORT_SYMBOL_GPL(dm_hold
);
1866 const char *dm_device_name(struct mapped_device
*md
)
1870 EXPORT_SYMBOL_GPL(dm_device_name
);
1872 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
1874 struct request_queue
*q
= dm_get_md_queue(md
);
1875 struct dm_table
*map
;
1880 spin_lock(&_minor_lock
);
1881 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
1882 set_bit(DMF_FREEING
, &md
->flags
);
1883 spin_unlock(&_minor_lock
);
1885 blk_set_queue_dying(q
);
1887 if (dm_request_based(md
) && md
->kworker_task
)
1888 kthread_flush_worker(&md
->kworker
);
1891 * Take suspend_lock so that presuspend and postsuspend methods
1892 * do not race with internal suspend.
1894 mutex_lock(&md
->suspend_lock
);
1895 map
= dm_get_live_table(md
, &srcu_idx
);
1896 if (!dm_suspended_md(md
)) {
1897 dm_table_presuspend_targets(map
);
1898 dm_table_postsuspend_targets(map
);
1900 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1901 dm_put_live_table(md
, srcu_idx
);
1902 mutex_unlock(&md
->suspend_lock
);
1905 * Rare, but there may be I/O requests still going to complete,
1906 * for example. Wait for all references to disappear.
1907 * No one should increment the reference count of the mapped_device,
1908 * after the mapped_device state becomes DMF_FREEING.
1911 while (atomic_read(&md
->holders
))
1913 else if (atomic_read(&md
->holders
))
1914 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1915 dm_device_name(md
), atomic_read(&md
->holders
));
1918 dm_table_destroy(__unbind(md
));
1922 void dm_destroy(struct mapped_device
*md
)
1924 __dm_destroy(md
, true);
1927 void dm_destroy_immediate(struct mapped_device
*md
)
1929 __dm_destroy(md
, false);
1932 void dm_put(struct mapped_device
*md
)
1934 atomic_dec(&md
->holders
);
1936 EXPORT_SYMBOL_GPL(dm_put
);
1938 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
1944 prepare_to_wait(&md
->wait
, &wait
, task_state
);
1946 if (!md_in_flight(md
))
1949 if (signal_pending_state(task_state
, current
)) {
1956 finish_wait(&md
->wait
, &wait
);
1962 * Process the deferred bios
1964 static void dm_wq_work(struct work_struct
*work
)
1966 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
1970 struct dm_table
*map
;
1972 map
= dm_get_live_table(md
, &srcu_idx
);
1974 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1975 spin_lock_irq(&md
->deferred_lock
);
1976 c
= bio_list_pop(&md
->deferred
);
1977 spin_unlock_irq(&md
->deferred_lock
);
1982 if (dm_request_based(md
))
1983 generic_make_request(c
);
1985 __split_and_process_bio(md
, map
, c
);
1988 dm_put_live_table(md
, srcu_idx
);
1991 static void dm_queue_flush(struct mapped_device
*md
)
1993 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
1994 smp_mb__after_atomic();
1995 queue_work(md
->wq
, &md
->work
);
1999 * Swap in a new table, returning the old one for the caller to destroy.
2001 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2003 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2004 struct queue_limits limits
;
2007 mutex_lock(&md
->suspend_lock
);
2009 /* device must be suspended */
2010 if (!dm_suspended_md(md
))
2014 * If the new table has no data devices, retain the existing limits.
2015 * This helps multipath with queue_if_no_path if all paths disappear,
2016 * then new I/O is queued based on these limits, and then some paths
2019 if (dm_table_has_no_data_devices(table
)) {
2020 live_map
= dm_get_live_table_fast(md
);
2022 limits
= md
->queue
->limits
;
2023 dm_put_live_table_fast(md
);
2027 r
= dm_calculate_queue_limits(table
, &limits
);
2034 map
= __bind(md
, table
, &limits
);
2037 mutex_unlock(&md
->suspend_lock
);
2042 * Functions to lock and unlock any filesystem running on the
2045 static int lock_fs(struct mapped_device
*md
)
2049 WARN_ON(md
->frozen_sb
);
2051 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2052 if (IS_ERR(md
->frozen_sb
)) {
2053 r
= PTR_ERR(md
->frozen_sb
);
2054 md
->frozen_sb
= NULL
;
2058 set_bit(DMF_FROZEN
, &md
->flags
);
2063 static void unlock_fs(struct mapped_device
*md
)
2065 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2068 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2069 md
->frozen_sb
= NULL
;
2070 clear_bit(DMF_FROZEN
, &md
->flags
);
2074 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2075 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2076 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2078 * If __dm_suspend returns 0, the device is completely quiescent
2079 * now. There is no request-processing activity. All new requests
2080 * are being added to md->deferred list.
2082 * Caller must hold md->suspend_lock
2084 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2085 unsigned suspend_flags
, long task_state
,
2086 int dmf_suspended_flag
)
2088 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2089 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2092 lockdep_assert_held(&md
->suspend_lock
);
2095 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2096 * This flag is cleared before dm_suspend returns.
2099 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2102 * This gets reverted if there's an error later and the targets
2103 * provide the .presuspend_undo hook.
2105 dm_table_presuspend_targets(map
);
2108 * Flush I/O to the device.
2109 * Any I/O submitted after lock_fs() may not be flushed.
2110 * noflush takes precedence over do_lockfs.
2111 * (lock_fs() flushes I/Os and waits for them to complete.)
2113 if (!noflush
&& do_lockfs
) {
2116 dm_table_presuspend_undo_targets(map
);
2122 * Here we must make sure that no processes are submitting requests
2123 * to target drivers i.e. no one may be executing
2124 * __split_and_process_bio. This is called from dm_request and
2127 * To get all processes out of __split_and_process_bio in dm_request,
2128 * we take the write lock. To prevent any process from reentering
2129 * __split_and_process_bio from dm_request and quiesce the thread
2130 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2131 * flush_workqueue(md->wq).
2133 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2135 synchronize_srcu(&md
->io_barrier
);
2138 * Stop md->queue before flushing md->wq in case request-based
2139 * dm defers requests to md->wq from md->queue.
2141 if (dm_request_based(md
)) {
2142 dm_stop_queue(md
->queue
);
2143 if (md
->kworker_task
)
2144 kthread_flush_worker(&md
->kworker
);
2147 flush_workqueue(md
->wq
);
2150 * At this point no more requests are entering target request routines.
2151 * We call dm_wait_for_completion to wait for all existing requests
2154 r
= dm_wait_for_completion(md
, task_state
);
2156 set_bit(dmf_suspended_flag
, &md
->flags
);
2159 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2161 synchronize_srcu(&md
->io_barrier
);
2163 /* were we interrupted ? */
2167 if (dm_request_based(md
))
2168 dm_start_queue(md
->queue
);
2171 dm_table_presuspend_undo_targets(map
);
2172 /* pushback list is already flushed, so skip flush */
2179 * We need to be able to change a mapping table under a mounted
2180 * filesystem. For example we might want to move some data in
2181 * the background. Before the table can be swapped with
2182 * dm_bind_table, dm_suspend must be called to flush any in
2183 * flight bios and ensure that any further io gets deferred.
2186 * Suspend mechanism in request-based dm.
2188 * 1. Flush all I/Os by lock_fs() if needed.
2189 * 2. Stop dispatching any I/O by stopping the request_queue.
2190 * 3. Wait for all in-flight I/Os to be completed or requeued.
2192 * To abort suspend, start the request_queue.
2194 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2196 struct dm_table
*map
= NULL
;
2200 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2202 if (dm_suspended_md(md
)) {
2207 if (dm_suspended_internally_md(md
)) {
2208 /* already internally suspended, wait for internal resume */
2209 mutex_unlock(&md
->suspend_lock
);
2210 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2216 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2218 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2222 dm_table_postsuspend_targets(map
);
2225 mutex_unlock(&md
->suspend_lock
);
2229 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2232 int r
= dm_table_resume_targets(map
);
2240 * Flushing deferred I/Os must be done after targets are resumed
2241 * so that mapping of targets can work correctly.
2242 * Request-based dm is queueing the deferred I/Os in its request_queue.
2244 if (dm_request_based(md
))
2245 dm_start_queue(md
->queue
);
2252 int dm_resume(struct mapped_device
*md
)
2255 struct dm_table
*map
= NULL
;
2259 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2261 if (!dm_suspended_md(md
))
2264 if (dm_suspended_internally_md(md
)) {
2265 /* already internally suspended, wait for internal resume */
2266 mutex_unlock(&md
->suspend_lock
);
2267 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2273 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2274 if (!map
|| !dm_table_get_size(map
))
2277 r
= __dm_resume(md
, map
);
2281 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2283 mutex_unlock(&md
->suspend_lock
);
2289 * Internal suspend/resume works like userspace-driven suspend. It waits
2290 * until all bios finish and prevents issuing new bios to the target drivers.
2291 * It may be used only from the kernel.
2294 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2296 struct dm_table
*map
= NULL
;
2298 if (md
->internal_suspend_count
++)
2299 return; /* nested internal suspend */
2301 if (dm_suspended_md(md
)) {
2302 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2303 return; /* nest suspend */
2306 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2309 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2310 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2311 * would require changing .presuspend to return an error -- avoid this
2312 * until there is a need for more elaborate variants of internal suspend.
2314 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2315 DMF_SUSPENDED_INTERNALLY
);
2317 dm_table_postsuspend_targets(map
);
2320 static void __dm_internal_resume(struct mapped_device
*md
)
2322 BUG_ON(!md
->internal_suspend_count
);
2324 if (--md
->internal_suspend_count
)
2325 return; /* resume from nested internal suspend */
2327 if (dm_suspended_md(md
))
2328 goto done
; /* resume from nested suspend */
2331 * NOTE: existing callers don't need to call dm_table_resume_targets
2332 * (which may fail -- so best to avoid it for now by passing NULL map)
2334 (void) __dm_resume(md
, NULL
);
2337 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2338 smp_mb__after_atomic();
2339 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2342 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2344 mutex_lock(&md
->suspend_lock
);
2345 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2346 mutex_unlock(&md
->suspend_lock
);
2348 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2350 void dm_internal_resume(struct mapped_device
*md
)
2352 mutex_lock(&md
->suspend_lock
);
2353 __dm_internal_resume(md
);
2354 mutex_unlock(&md
->suspend_lock
);
2356 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2359 * Fast variants of internal suspend/resume hold md->suspend_lock,
2360 * which prevents interaction with userspace-driven suspend.
2363 void dm_internal_suspend_fast(struct mapped_device
*md
)
2365 mutex_lock(&md
->suspend_lock
);
2366 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2369 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2370 synchronize_srcu(&md
->io_barrier
);
2371 flush_workqueue(md
->wq
);
2372 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2374 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2376 void dm_internal_resume_fast(struct mapped_device
*md
)
2378 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2384 mutex_unlock(&md
->suspend_lock
);
2386 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2388 /*-----------------------------------------------------------------
2389 * Event notification.
2390 *---------------------------------------------------------------*/
2391 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2394 char udev_cookie
[DM_COOKIE_LENGTH
];
2395 char *envp
[] = { udev_cookie
, NULL
};
2398 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2400 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2401 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2402 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2407 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2409 return atomic_add_return(1, &md
->uevent_seq
);
2412 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2414 return atomic_read(&md
->event_nr
);
2417 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2419 return wait_event_interruptible(md
->eventq
,
2420 (event_nr
!= atomic_read(&md
->event_nr
)));
2423 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2425 unsigned long flags
;
2427 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2428 list_add(elist
, &md
->uevent_list
);
2429 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2433 * The gendisk is only valid as long as you have a reference
2436 struct gendisk
*dm_disk(struct mapped_device
*md
)
2440 EXPORT_SYMBOL_GPL(dm_disk
);
2442 struct kobject
*dm_kobject(struct mapped_device
*md
)
2444 return &md
->kobj_holder
.kobj
;
2447 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2449 struct mapped_device
*md
;
2451 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2453 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2461 int dm_suspended_md(struct mapped_device
*md
)
2463 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2466 int dm_suspended_internally_md(struct mapped_device
*md
)
2468 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2471 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2473 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2476 int dm_suspended(struct dm_target
*ti
)
2478 return dm_suspended_md(dm_table_get_md(ti
->table
));
2480 EXPORT_SYMBOL_GPL(dm_suspended
);
2482 int dm_noflush_suspending(struct dm_target
*ti
)
2484 return __noflush_suspending(dm_table_get_md(ti
->table
));
2486 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2488 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
2489 unsigned integrity
, unsigned per_io_data_size
)
2491 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2492 unsigned int pool_size
= 0;
2493 unsigned int front_pad
;
2499 case DM_TYPE_BIO_BASED
:
2500 case DM_TYPE_DAX_BIO_BASED
:
2501 pool_size
= dm_get_reserved_bio_based_ios();
2502 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2504 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2505 if (!pools
->io_pool
)
2508 case DM_TYPE_REQUEST_BASED
:
2509 case DM_TYPE_MQ_REQUEST_BASED
:
2510 pool_size
= dm_get_reserved_rq_based_ios();
2511 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2512 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2518 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
2522 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2528 dm_free_md_mempools(pools
);
2533 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2538 mempool_destroy(pools
->io_pool
);
2541 bioset_free(pools
->bs
);
2553 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2556 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2557 struct dm_table
*table
;
2558 struct dm_target
*ti
;
2559 int ret
= -ENOTTY
, srcu_idx
;
2561 table
= dm_get_live_table(md
, &srcu_idx
);
2562 if (!table
|| !dm_table_get_size(table
))
2565 /* We only support devices that have a single target */
2566 if (dm_table_get_num_targets(table
) != 1)
2568 ti
= dm_table_get_target(table
, 0);
2571 if (!ti
->type
->iterate_devices
)
2574 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2576 dm_put_live_table(md
, srcu_idx
);
2581 * For register / unregister we need to manually call out to every path.
2583 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2584 sector_t start
, sector_t len
, void *data
)
2586 struct dm_pr
*pr
= data
;
2587 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2589 if (!ops
|| !ops
->pr_register
)
2591 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2594 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2605 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2606 if (ret
&& new_key
) {
2607 /* unregister all paths if we failed to register any path */
2608 pr
.old_key
= new_key
;
2611 pr
.fail_early
= false;
2612 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2618 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2621 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2622 const struct pr_ops
*ops
;
2626 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2630 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2631 if (ops
&& ops
->pr_reserve
)
2632 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2640 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2642 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2643 const struct pr_ops
*ops
;
2647 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2651 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2652 if (ops
&& ops
->pr_release
)
2653 r
= ops
->pr_release(bdev
, key
, type
);
2661 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2662 enum pr_type type
, bool abort
)
2664 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2665 const struct pr_ops
*ops
;
2669 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2673 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2674 if (ops
&& ops
->pr_preempt
)
2675 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2683 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2685 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2686 const struct pr_ops
*ops
;
2690 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2694 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2695 if (ops
&& ops
->pr_clear
)
2696 r
= ops
->pr_clear(bdev
, key
);
2704 static const struct pr_ops dm_pr_ops
= {
2705 .pr_register
= dm_pr_register
,
2706 .pr_reserve
= dm_pr_reserve
,
2707 .pr_release
= dm_pr_release
,
2708 .pr_preempt
= dm_pr_preempt
,
2709 .pr_clear
= dm_pr_clear
,
2712 static const struct block_device_operations dm_blk_dops
= {
2713 .open
= dm_blk_open
,
2714 .release
= dm_blk_close
,
2715 .ioctl
= dm_blk_ioctl
,
2716 .direct_access
= dm_blk_direct_access
,
2717 .getgeo
= dm_blk_getgeo
,
2718 .pr_ops
= &dm_pr_ops
,
2719 .owner
= THIS_MODULE
2725 module_init(dm_init
);
2726 module_exit(dm_exit
);
2728 module_param(major
, uint
, 0);
2729 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2731 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
2732 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
2734 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
2735 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
2737 MODULE_DESCRIPTION(DM_NAME
" driver");
2738 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2739 MODULE_LICENSE("GPL");