2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
23 #include <linux/wait.h>
26 #define DM_MSG_PREFIX "core"
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
33 DEFAULT_RATELIMIT_INTERVAL
,
34 DEFAULT_RATELIMIT_BURST
);
35 EXPORT_SYMBOL(dm_ratelimit_state
);
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name
= DM_NAME
;
47 static unsigned int major
= 0;
48 static unsigned int _major
= 0;
50 static DEFINE_IDR(_minor_idr
);
52 static DEFINE_SPINLOCK(_minor_lock
);
54 static void do_deferred_remove(struct work_struct
*w
);
56 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
58 static struct workqueue_struct
*deferred_remove_workqueue
;
61 * One of these is allocated per bio.
64 struct mapped_device
*md
;
68 unsigned long start_time
;
69 spinlock_t endio_lock
;
70 struct dm_stats_aux stats_aux
;
73 #define MINOR_ALLOCED ((void *)-1)
76 * Bits for the md->flags field.
78 #define DMF_BLOCK_IO_FOR_SUSPEND 0
79 #define DMF_SUSPENDED 1
82 #define DMF_DELETING 4
83 #define DMF_NOFLUSH_SUSPENDING 5
84 #define DMF_DEFERRED_REMOVE 6
85 #define DMF_SUSPENDED_INTERNALLY 7
87 #define DM_NUMA_NODE NUMA_NO_NODE
88 static int dm_numa_node
= DM_NUMA_NODE
;
91 * For mempools pre-allocation at the table loading time.
93 struct dm_md_mempools
{
99 struct list_head list
;
101 struct dm_dev dm_dev
;
104 static struct kmem_cache
*_io_cache
;
105 static struct kmem_cache
*_rq_tio_cache
;
106 static struct kmem_cache
*_rq_cache
;
109 * Bio-based DM's mempools' reserved IOs set by the user.
111 #define RESERVED_BIO_BASED_IOS 16
112 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
114 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
116 int param
= ACCESS_ONCE(*module_param
);
117 int modified_param
= 0;
118 bool modified
= true;
121 modified_param
= min
;
122 else if (param
> max
)
123 modified_param
= max
;
128 (void)cmpxchg(module_param
, param
, modified_param
);
129 param
= modified_param
;
135 unsigned __dm_get_module_param(unsigned *module_param
,
136 unsigned def
, unsigned max
)
138 unsigned param
= ACCESS_ONCE(*module_param
);
139 unsigned modified_param
= 0;
142 modified_param
= def
;
143 else if (param
> max
)
144 modified_param
= max
;
146 if (modified_param
) {
147 (void)cmpxchg(module_param
, param
, modified_param
);
148 param
= modified_param
;
154 unsigned dm_get_reserved_bio_based_ios(void)
156 return __dm_get_module_param(&reserved_bio_based_ios
,
157 RESERVED_BIO_BASED_IOS
, DM_RESERVED_MAX_IOS
);
159 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
161 static unsigned dm_get_numa_node(void)
163 return __dm_get_module_param_int(&dm_numa_node
,
164 DM_NUMA_NODE
, num_online_nodes() - 1);
167 static int __init
local_init(void)
171 /* allocate a slab for the dm_ios */
172 _io_cache
= KMEM_CACHE(dm_io
, 0);
176 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
178 goto out_free_io_cache
;
180 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
181 __alignof__(struct request
), 0, NULL
);
183 goto out_free_rq_tio_cache
;
185 r
= dm_uevent_init();
187 goto out_free_rq_cache
;
189 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
190 if (!deferred_remove_workqueue
) {
192 goto out_uevent_exit
;
196 r
= register_blkdev(_major
, _name
);
198 goto out_free_workqueue
;
206 destroy_workqueue(deferred_remove_workqueue
);
210 kmem_cache_destroy(_rq_cache
);
211 out_free_rq_tio_cache
:
212 kmem_cache_destroy(_rq_tio_cache
);
214 kmem_cache_destroy(_io_cache
);
219 static void local_exit(void)
221 flush_scheduled_work();
222 destroy_workqueue(deferred_remove_workqueue
);
224 kmem_cache_destroy(_rq_cache
);
225 kmem_cache_destroy(_rq_tio_cache
);
226 kmem_cache_destroy(_io_cache
);
227 unregister_blkdev(_major
, _name
);
232 DMINFO("cleaned up");
235 static int (*_inits
[])(void) __initdata
= {
246 static void (*_exits
[])(void) = {
257 static int __init
dm_init(void)
259 const int count
= ARRAY_SIZE(_inits
);
263 for (i
= 0; i
< count
; i
++) {
278 static void __exit
dm_exit(void)
280 int i
= ARRAY_SIZE(_exits
);
286 * Should be empty by this point.
288 idr_destroy(&_minor_idr
);
292 * Block device functions
294 int dm_deleting_md(struct mapped_device
*md
)
296 return test_bit(DMF_DELETING
, &md
->flags
);
299 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
301 struct mapped_device
*md
;
303 spin_lock(&_minor_lock
);
305 md
= bdev
->bd_disk
->private_data
;
309 if (test_bit(DMF_FREEING
, &md
->flags
) ||
310 dm_deleting_md(md
)) {
316 atomic_inc(&md
->open_count
);
318 spin_unlock(&_minor_lock
);
320 return md
? 0 : -ENXIO
;
323 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
325 struct mapped_device
*md
;
327 spin_lock(&_minor_lock
);
329 md
= disk
->private_data
;
333 if (atomic_dec_and_test(&md
->open_count
) &&
334 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
335 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
339 spin_unlock(&_minor_lock
);
342 int dm_open_count(struct mapped_device
*md
)
344 return atomic_read(&md
->open_count
);
348 * Guarantees nothing is using the device before it's deleted.
350 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
354 spin_lock(&_minor_lock
);
356 if (dm_open_count(md
)) {
359 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
360 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
363 set_bit(DMF_DELETING
, &md
->flags
);
365 spin_unlock(&_minor_lock
);
370 int dm_cancel_deferred_remove(struct mapped_device
*md
)
374 spin_lock(&_minor_lock
);
376 if (test_bit(DMF_DELETING
, &md
->flags
))
379 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
381 spin_unlock(&_minor_lock
);
386 static void do_deferred_remove(struct work_struct
*w
)
388 dm_deferred_remove();
391 sector_t
dm_get_size(struct mapped_device
*md
)
393 return get_capacity(md
->disk
);
396 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
401 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
406 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
408 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
410 return dm_get_geometry(md
, geo
);
413 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
414 struct block_device
**bdev
,
417 struct dm_target
*tgt
;
418 struct dm_table
*map
;
423 map
= dm_get_live_table(md
, &srcu_idx
);
424 if (!map
|| !dm_table_get_size(map
))
427 /* We only support devices that have a single target */
428 if (dm_table_get_num_targets(map
) != 1)
431 tgt
= dm_table_get_target(map
, 0);
432 if (!tgt
->type
->prepare_ioctl
)
435 if (dm_suspended_md(md
)) {
440 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
445 dm_put_live_table(md
, srcu_idx
);
449 dm_put_live_table(md
, srcu_idx
);
450 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
457 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
458 unsigned int cmd
, unsigned long arg
)
460 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
463 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
469 * Target determined this ioctl is being issued against a
470 * subset of the parent bdev; require extra privileges.
472 if (!capable(CAP_SYS_RAWIO
)) {
474 "%s: sending ioctl %x to DM device without required privilege.",
481 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
487 static struct dm_io
*alloc_io(struct mapped_device
*md
)
489 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
492 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
494 mempool_free(io
, md
->io_pool
);
497 static void free_tio(struct dm_target_io
*tio
)
499 bio_put(&tio
->clone
);
502 int md_in_flight(struct mapped_device
*md
)
504 return atomic_read(&md
->pending
[READ
]) +
505 atomic_read(&md
->pending
[WRITE
]);
508 static void start_io_acct(struct dm_io
*io
)
510 struct mapped_device
*md
= io
->md
;
511 struct bio
*bio
= io
->bio
;
513 int rw
= bio_data_dir(bio
);
515 io
->start_time
= jiffies
;
517 cpu
= part_stat_lock();
518 part_round_stats(cpu
, &dm_disk(md
)->part0
);
520 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
521 atomic_inc_return(&md
->pending
[rw
]));
523 if (unlikely(dm_stats_used(&md
->stats
)))
524 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
525 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
526 false, 0, &io
->stats_aux
);
529 static void end_io_acct(struct dm_io
*io
)
531 struct mapped_device
*md
= io
->md
;
532 struct bio
*bio
= io
->bio
;
533 unsigned long duration
= jiffies
- io
->start_time
;
535 int rw
= bio_data_dir(bio
);
537 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
539 if (unlikely(dm_stats_used(&md
->stats
)))
540 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
541 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
542 true, duration
, &io
->stats_aux
);
545 * After this is decremented the bio must not be touched if it is
548 pending
= atomic_dec_return(&md
->pending
[rw
]);
549 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
550 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
552 /* nudge anyone waiting on suspend queue */
558 * Add the bio to the list of deferred io.
560 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
564 spin_lock_irqsave(&md
->deferred_lock
, flags
);
565 bio_list_add(&md
->deferred
, bio
);
566 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
567 queue_work(md
->wq
, &md
->work
);
571 * Everyone (including functions in this file), should use this
572 * function to access the md->map field, and make sure they call
573 * dm_put_live_table() when finished.
575 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
577 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
579 return srcu_dereference(md
->map
, &md
->io_barrier
);
582 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
584 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
587 void dm_sync_table(struct mapped_device
*md
)
589 synchronize_srcu(&md
->io_barrier
);
590 synchronize_rcu_expedited();
594 * A fast alternative to dm_get_live_table/dm_put_live_table.
595 * The caller must not block between these two functions.
597 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
600 return rcu_dereference(md
->map
);
603 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
609 * Open a table device so we can use it as a map destination.
611 static int open_table_device(struct table_device
*td
, dev_t dev
,
612 struct mapped_device
*md
)
614 static char *_claim_ptr
= "I belong to device-mapper";
615 struct block_device
*bdev
;
619 BUG_ON(td
->dm_dev
.bdev
);
621 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
623 return PTR_ERR(bdev
);
625 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
627 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
631 td
->dm_dev
.bdev
= bdev
;
636 * Close a table device that we've been using.
638 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
640 if (!td
->dm_dev
.bdev
)
643 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
644 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
645 td
->dm_dev
.bdev
= NULL
;
648 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
650 struct table_device
*td
;
652 list_for_each_entry(td
, l
, list
)
653 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
659 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
660 struct dm_dev
**result
) {
662 struct table_device
*td
;
664 mutex_lock(&md
->table_devices_lock
);
665 td
= find_table_device(&md
->table_devices
, dev
, mode
);
667 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
669 mutex_unlock(&md
->table_devices_lock
);
673 td
->dm_dev
.mode
= mode
;
674 td
->dm_dev
.bdev
= NULL
;
676 if ((r
= open_table_device(td
, dev
, md
))) {
677 mutex_unlock(&md
->table_devices_lock
);
682 format_dev_t(td
->dm_dev
.name
, dev
);
684 atomic_set(&td
->count
, 0);
685 list_add(&td
->list
, &md
->table_devices
);
687 atomic_inc(&td
->count
);
688 mutex_unlock(&md
->table_devices_lock
);
690 *result
= &td
->dm_dev
;
693 EXPORT_SYMBOL_GPL(dm_get_table_device
);
695 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
697 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
699 mutex_lock(&md
->table_devices_lock
);
700 if (atomic_dec_and_test(&td
->count
)) {
701 close_table_device(td
, md
);
705 mutex_unlock(&md
->table_devices_lock
);
707 EXPORT_SYMBOL(dm_put_table_device
);
709 static void free_table_devices(struct list_head
*devices
)
711 struct list_head
*tmp
, *next
;
713 list_for_each_safe(tmp
, next
, devices
) {
714 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
716 DMWARN("dm_destroy: %s still exists with %d references",
717 td
->dm_dev
.name
, atomic_read(&td
->count
));
723 * Get the geometry associated with a dm device
725 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
733 * Set the geometry of a device.
735 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
737 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
739 if (geo
->start
> sz
) {
740 DMWARN("Start sector is beyond the geometry limits.");
749 /*-----------------------------------------------------------------
751 * A more elegant soln is in the works that uses the queue
752 * merge fn, unfortunately there are a couple of changes to
753 * the block layer that I want to make for this. So in the
754 * interests of getting something for people to use I give
755 * you this clearly demarcated crap.
756 *---------------------------------------------------------------*/
758 static int __noflush_suspending(struct mapped_device
*md
)
760 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
764 * Decrements the number of outstanding ios that a bio has been
765 * cloned into, completing the original io if necc.
767 static void dec_pending(struct dm_io
*io
, int error
)
772 struct mapped_device
*md
= io
->md
;
774 /* Push-back supersedes any I/O errors */
775 if (unlikely(error
)) {
776 spin_lock_irqsave(&io
->endio_lock
, flags
);
777 if (!(io
->error
> 0 && __noflush_suspending(md
)))
779 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
782 if (atomic_dec_and_test(&io
->io_count
)) {
783 if (io
->error
== DM_ENDIO_REQUEUE
) {
785 * Target requested pushing back the I/O.
787 spin_lock_irqsave(&md
->deferred_lock
, flags
);
788 if (__noflush_suspending(md
))
789 bio_list_add_head(&md
->deferred
, io
->bio
);
791 /* noflush suspend was interrupted. */
793 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
796 io_error
= io
->error
;
801 if (io_error
== DM_ENDIO_REQUEUE
)
804 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
806 * Preflush done for flush with data, reissue
807 * without REQ_PREFLUSH.
809 bio
->bi_opf
&= ~REQ_PREFLUSH
;
812 /* done with normal IO or empty flush */
813 trace_block_bio_complete(md
->queue
, bio
, io_error
);
814 bio
->bi_error
= io_error
;
820 void disable_write_same(struct mapped_device
*md
)
822 struct queue_limits
*limits
= dm_get_queue_limits(md
);
824 /* device doesn't really support WRITE SAME, disable it */
825 limits
->max_write_same_sectors
= 0;
828 static void clone_endio(struct bio
*bio
)
830 int error
= bio
->bi_error
;
832 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
833 struct dm_io
*io
= tio
->io
;
834 struct mapped_device
*md
= tio
->io
->md
;
835 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
838 r
= endio(tio
->ti
, bio
, error
);
839 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
841 * error and requeue request are handled
845 else if (r
== DM_ENDIO_INCOMPLETE
)
846 /* The target will handle the io */
849 DMWARN("unimplemented target endio return value: %d", r
);
854 if (unlikely(r
== -EREMOTEIO
&& (bio_op(bio
) == REQ_OP_WRITE_SAME
) &&
855 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
856 disable_write_same(md
);
859 dec_pending(io
, error
);
863 * Return maximum size of I/O possible at the supplied sector up to the current
866 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
868 sector_t target_offset
= dm_target_offset(ti
, sector
);
870 return ti
->len
- target_offset
;
873 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
875 sector_t len
= max_io_len_target_boundary(sector
, ti
);
876 sector_t offset
, max_len
;
879 * Does the target need to split even further?
881 if (ti
->max_io_len
) {
882 offset
= dm_target_offset(ti
, sector
);
883 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
884 max_len
= sector_div(offset
, ti
->max_io_len
);
886 max_len
= offset
& (ti
->max_io_len
- 1);
887 max_len
= ti
->max_io_len
- max_len
;
896 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
898 if (len
> UINT_MAX
) {
899 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
900 (unsigned long long)len
, UINT_MAX
);
901 ti
->error
= "Maximum size of target IO is too large";
905 ti
->max_io_len
= (uint32_t) len
;
909 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
911 static long dm_blk_direct_access(struct block_device
*bdev
, sector_t sector
,
912 void **kaddr
, pfn_t
*pfn
, long size
)
914 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
915 struct dm_table
*map
;
916 struct dm_target
*ti
;
918 long len
, ret
= -EIO
;
920 map
= dm_get_live_table(md
, &srcu_idx
);
924 ti
= dm_table_find_target(map
, sector
);
925 if (!dm_target_is_valid(ti
))
928 len
= max_io_len(sector
, ti
) << SECTOR_SHIFT
;
929 size
= min(len
, size
);
931 if (ti
->type
->direct_access
)
932 ret
= ti
->type
->direct_access(ti
, sector
, kaddr
, pfn
, size
);
934 dm_put_live_table(md
, srcu_idx
);
935 return min(ret
, size
);
939 * A target may call dm_accept_partial_bio only from the map routine. It is
940 * allowed for all bio types except REQ_PREFLUSH.
942 * dm_accept_partial_bio informs the dm that the target only wants to process
943 * additional n_sectors sectors of the bio and the rest of the data should be
944 * sent in a next bio.
946 * A diagram that explains the arithmetics:
947 * +--------------------+---------------+-------+
949 * +--------------------+---------------+-------+
951 * <-------------- *tio->len_ptr --------------->
952 * <------- bi_size ------->
955 * Region 1 was already iterated over with bio_advance or similar function.
956 * (it may be empty if the target doesn't use bio_advance)
957 * Region 2 is the remaining bio size that the target wants to process.
958 * (it may be empty if region 1 is non-empty, although there is no reason
960 * The target requires that region 3 is to be sent in the next bio.
962 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
963 * the partially processed part (the sum of regions 1+2) must be the same for all
966 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
968 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
969 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
970 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
971 BUG_ON(bi_size
> *tio
->len_ptr
);
972 BUG_ON(n_sectors
> bi_size
);
973 *tio
->len_ptr
-= bi_size
- n_sectors
;
974 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
976 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
979 * Flush current->bio_list when the target map method blocks.
980 * This fixes deadlocks in snapshot and possibly in other targets.
983 struct blk_plug plug
;
984 struct blk_plug_cb cb
;
987 static void flush_current_bio_list(struct blk_plug_cb
*cb
, bool from_schedule
)
989 struct dm_offload
*o
= container_of(cb
, struct dm_offload
, cb
);
990 struct bio_list list
;
994 INIT_LIST_HEAD(&o
->cb
.list
);
996 if (unlikely(!current
->bio_list
))
999 for (i
= 0; i
< 2; i
++) {
1000 list
= current
->bio_list
[i
];
1001 bio_list_init(¤t
->bio_list
[i
]);
1003 while ((bio
= bio_list_pop(&list
))) {
1004 struct bio_set
*bs
= bio
->bi_pool
;
1005 if (unlikely(!bs
) || bs
== fs_bio_set
) {
1006 bio_list_add(¤t
->bio_list
[i
], bio
);
1010 spin_lock(&bs
->rescue_lock
);
1011 bio_list_add(&bs
->rescue_list
, bio
);
1012 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
1013 spin_unlock(&bs
->rescue_lock
);
1018 static void dm_offload_start(struct dm_offload
*o
)
1020 blk_start_plug(&o
->plug
);
1021 o
->cb
.callback
= flush_current_bio_list
;
1022 list_add(&o
->cb
.list
, ¤t
->plug
->cb_list
);
1025 static void dm_offload_end(struct dm_offload
*o
)
1027 list_del(&o
->cb
.list
);
1028 blk_finish_plug(&o
->plug
);
1031 static void __map_bio(struct dm_target_io
*tio
)
1035 struct dm_offload o
;
1036 struct bio
*clone
= &tio
->clone
;
1037 struct dm_target
*ti
= tio
->ti
;
1039 clone
->bi_end_io
= clone_endio
;
1042 * Map the clone. If r == 0 we don't need to do
1043 * anything, the target has assumed ownership of
1046 atomic_inc(&tio
->io
->io_count
);
1047 sector
= clone
->bi_iter
.bi_sector
;
1049 dm_offload_start(&o
);
1050 r
= ti
->type
->map(ti
, clone
);
1053 if (r
== DM_MAPIO_REMAPPED
) {
1054 /* the bio has been remapped so dispatch it */
1056 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1057 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1059 generic_make_request(clone
);
1060 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1061 /* error the io and bail out, or requeue it if needed */
1062 dec_pending(tio
->io
, r
);
1064 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1065 DMWARN("unimplemented target map return value: %d", r
);
1071 struct mapped_device
*md
;
1072 struct dm_table
*map
;
1076 unsigned sector_count
;
1079 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1081 bio
->bi_iter
.bi_sector
= sector
;
1082 bio
->bi_iter
.bi_size
= to_bytes(len
);
1086 * Creates a bio that consists of range of complete bvecs.
1088 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1089 sector_t sector
, unsigned len
)
1091 struct bio
*clone
= &tio
->clone
;
1093 __bio_clone_fast(clone
, bio
);
1095 if (bio_integrity(bio
)) {
1096 int r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1101 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1102 clone
->bi_iter
.bi_size
= to_bytes(len
);
1104 if (bio_integrity(bio
))
1105 bio_integrity_trim(clone
, 0, len
);
1110 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1111 struct dm_target
*ti
,
1112 unsigned target_bio_nr
)
1114 struct dm_target_io
*tio
;
1117 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1118 tio
= container_of(clone
, struct dm_target_io
, clone
);
1122 tio
->target_bio_nr
= target_bio_nr
;
1127 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1128 struct dm_target
*ti
,
1129 unsigned target_bio_nr
, unsigned *len
)
1131 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1132 struct bio
*clone
= &tio
->clone
;
1136 __bio_clone_fast(clone
, ci
->bio
);
1138 bio_setup_sector(clone
, ci
->sector
, *len
);
1143 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1144 unsigned num_bios
, unsigned *len
)
1146 unsigned target_bio_nr
;
1148 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1149 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1152 static int __send_empty_flush(struct clone_info
*ci
)
1154 unsigned target_nr
= 0;
1155 struct dm_target
*ti
;
1157 BUG_ON(bio_has_data(ci
->bio
));
1158 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1159 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1164 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1165 sector_t sector
, unsigned *len
)
1167 struct bio
*bio
= ci
->bio
;
1168 struct dm_target_io
*tio
;
1169 unsigned target_bio_nr
;
1170 unsigned num_target_bios
= 1;
1174 * Does the target want to receive duplicate copies of the bio?
1176 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1177 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1179 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1180 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1182 r
= clone_bio(tio
, bio
, sector
, *len
);
1193 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1195 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1197 return ti
->num_discard_bios
;
1200 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1202 return ti
->num_write_same_bios
;
1205 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1207 static bool is_split_required_for_discard(struct dm_target
*ti
)
1209 return ti
->split_discard_bios
;
1212 static int __send_changing_extent_only(struct clone_info
*ci
,
1213 get_num_bios_fn get_num_bios
,
1214 is_split_required_fn is_split_required
)
1216 struct dm_target
*ti
;
1221 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1222 if (!dm_target_is_valid(ti
))
1226 * Even though the device advertised support for this type of
1227 * request, that does not mean every target supports it, and
1228 * reconfiguration might also have changed that since the
1229 * check was performed.
1231 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1235 if (is_split_required
&& !is_split_required(ti
))
1236 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1238 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1240 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1243 } while (ci
->sector_count
-= len
);
1248 static int __send_discard(struct clone_info
*ci
)
1250 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1251 is_split_required_for_discard
);
1254 static int __send_write_same(struct clone_info
*ci
)
1256 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1260 * Select the correct strategy for processing a non-flush bio.
1262 static int __split_and_process_non_flush(struct clone_info
*ci
)
1264 struct bio
*bio
= ci
->bio
;
1265 struct dm_target
*ti
;
1269 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1270 return __send_discard(ci
);
1271 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1272 return __send_write_same(ci
);
1274 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1275 if (!dm_target_is_valid(ti
))
1278 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1280 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1285 ci
->sector_count
-= len
;
1291 * Entry point to split a bio into clones and submit them to the targets.
1293 static void __split_and_process_bio(struct mapped_device
*md
,
1294 struct dm_table
*map
, struct bio
*bio
)
1296 struct clone_info ci
;
1299 if (unlikely(!map
)) {
1306 ci
.io
= alloc_io(md
);
1308 atomic_set(&ci
.io
->io_count
, 1);
1311 spin_lock_init(&ci
.io
->endio_lock
);
1312 ci
.sector
= bio
->bi_iter
.bi_sector
;
1314 start_io_acct(ci
.io
);
1316 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1317 ci
.bio
= &ci
.md
->flush_bio
;
1318 ci
.sector_count
= 0;
1319 error
= __send_empty_flush(&ci
);
1320 /* dec_pending submits any data associated with flush */
1323 ci
.sector_count
= bio_sectors(bio
);
1324 while (ci
.sector_count
&& !error
)
1325 error
= __split_and_process_non_flush(&ci
);
1328 /* drop the extra reference count */
1329 dec_pending(ci
.io
, error
);
1331 /*-----------------------------------------------------------------
1333 *---------------------------------------------------------------*/
1336 * The request function that just remaps the bio built up by
1339 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1341 int rw
= bio_data_dir(bio
);
1342 struct mapped_device
*md
= q
->queuedata
;
1344 struct dm_table
*map
;
1346 map
= dm_get_live_table(md
, &srcu_idx
);
1348 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1350 /* if we're suspended, we have to queue this io for later */
1351 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1352 dm_put_live_table(md
, srcu_idx
);
1354 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1358 return BLK_QC_T_NONE
;
1361 __split_and_process_bio(md
, map
, bio
);
1362 dm_put_live_table(md
, srcu_idx
);
1363 return BLK_QC_T_NONE
;
1366 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1369 struct mapped_device
*md
= congested_data
;
1370 struct dm_table
*map
;
1372 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1373 if (dm_request_based(md
)) {
1375 * With request-based DM we only need to check the
1376 * top-level queue for congestion.
1378 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1380 map
= dm_get_live_table_fast(md
);
1382 r
= dm_table_any_congested(map
, bdi_bits
);
1383 dm_put_live_table_fast(md
);
1390 /*-----------------------------------------------------------------
1391 * An IDR is used to keep track of allocated minor numbers.
1392 *---------------------------------------------------------------*/
1393 static void free_minor(int minor
)
1395 spin_lock(&_minor_lock
);
1396 idr_remove(&_minor_idr
, minor
);
1397 spin_unlock(&_minor_lock
);
1401 * See if the device with a specific minor # is free.
1403 static int specific_minor(int minor
)
1407 if (minor
>= (1 << MINORBITS
))
1410 idr_preload(GFP_KERNEL
);
1411 spin_lock(&_minor_lock
);
1413 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1415 spin_unlock(&_minor_lock
);
1418 return r
== -ENOSPC
? -EBUSY
: r
;
1422 static int next_free_minor(int *minor
)
1426 idr_preload(GFP_KERNEL
);
1427 spin_lock(&_minor_lock
);
1429 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1431 spin_unlock(&_minor_lock
);
1439 static const struct block_device_operations dm_blk_dops
;
1441 static void dm_wq_work(struct work_struct
*work
);
1443 void dm_init_md_queue(struct mapped_device
*md
)
1446 * Request-based dm devices cannot be stacked on top of bio-based dm
1447 * devices. The type of this dm device may not have been decided yet.
1448 * The type is decided at the first table loading time.
1449 * To prevent problematic device stacking, clear the queue flag
1450 * for request stacking support until then.
1452 * This queue is new, so no concurrency on the queue_flags.
1454 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1457 * Initialize data that will only be used by a non-blk-mq DM queue
1458 * - must do so here (in alloc_dev callchain) before queue is used
1460 md
->queue
->queuedata
= md
;
1461 md
->queue
->backing_dev_info
->congested_data
= md
;
1464 void dm_init_normal_md_queue(struct mapped_device
*md
)
1466 md
->use_blk_mq
= false;
1467 dm_init_md_queue(md
);
1470 * Initialize aspects of queue that aren't relevant for blk-mq
1472 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1473 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1476 static void cleanup_mapped_device(struct mapped_device
*md
)
1479 destroy_workqueue(md
->wq
);
1480 if (md
->kworker_task
)
1481 kthread_stop(md
->kworker_task
);
1482 mempool_destroy(md
->io_pool
);
1484 bioset_free(md
->bs
);
1487 spin_lock(&_minor_lock
);
1488 md
->disk
->private_data
= NULL
;
1489 spin_unlock(&_minor_lock
);
1490 del_gendisk(md
->disk
);
1495 blk_cleanup_queue(md
->queue
);
1497 cleanup_srcu_struct(&md
->io_barrier
);
1504 dm_mq_cleanup_mapped_device(md
);
1508 * Allocate and initialise a blank device with a given minor.
1510 static struct mapped_device
*alloc_dev(int minor
)
1512 int r
, numa_node_id
= dm_get_numa_node();
1513 struct mapped_device
*md
;
1516 md
= kzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1518 DMWARN("unable to allocate device, out of memory.");
1522 if (!try_module_get(THIS_MODULE
))
1523 goto bad_module_get
;
1525 /* get a minor number for the dev */
1526 if (minor
== DM_ANY_MINOR
)
1527 r
= next_free_minor(&minor
);
1529 r
= specific_minor(minor
);
1533 r
= init_srcu_struct(&md
->io_barrier
);
1535 goto bad_io_barrier
;
1537 md
->numa_node_id
= numa_node_id
;
1538 md
->use_blk_mq
= dm_use_blk_mq_default();
1539 md
->init_tio_pdu
= false;
1540 md
->type
= DM_TYPE_NONE
;
1541 mutex_init(&md
->suspend_lock
);
1542 mutex_init(&md
->type_lock
);
1543 mutex_init(&md
->table_devices_lock
);
1544 spin_lock_init(&md
->deferred_lock
);
1545 atomic_set(&md
->holders
, 1);
1546 atomic_set(&md
->open_count
, 0);
1547 atomic_set(&md
->event_nr
, 0);
1548 atomic_set(&md
->uevent_seq
, 0);
1549 INIT_LIST_HEAD(&md
->uevent_list
);
1550 INIT_LIST_HEAD(&md
->table_devices
);
1551 spin_lock_init(&md
->uevent_lock
);
1553 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1557 dm_init_md_queue(md
);
1559 md
->disk
= alloc_disk_node(1, numa_node_id
);
1563 atomic_set(&md
->pending
[0], 0);
1564 atomic_set(&md
->pending
[1], 0);
1565 init_waitqueue_head(&md
->wait
);
1566 INIT_WORK(&md
->work
, dm_wq_work
);
1567 init_waitqueue_head(&md
->eventq
);
1568 init_completion(&md
->kobj_holder
.completion
);
1569 md
->kworker_task
= NULL
;
1571 md
->disk
->major
= _major
;
1572 md
->disk
->first_minor
= minor
;
1573 md
->disk
->fops
= &dm_blk_dops
;
1574 md
->disk
->queue
= md
->queue
;
1575 md
->disk
->private_data
= md
;
1576 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1578 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1580 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1584 md
->bdev
= bdget_disk(md
->disk
, 0);
1588 bio_init(&md
->flush_bio
, NULL
, 0);
1589 md
->flush_bio
.bi_bdev
= md
->bdev
;
1590 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
1592 dm_stats_init(&md
->stats
);
1594 /* Populate the mapping, nobody knows we exist yet */
1595 spin_lock(&_minor_lock
);
1596 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1597 spin_unlock(&_minor_lock
);
1599 BUG_ON(old_md
!= MINOR_ALLOCED
);
1604 cleanup_mapped_device(md
);
1608 module_put(THIS_MODULE
);
1614 static void unlock_fs(struct mapped_device
*md
);
1616 static void free_dev(struct mapped_device
*md
)
1618 int minor
= MINOR(disk_devt(md
->disk
));
1622 cleanup_mapped_device(md
);
1624 free_table_devices(&md
->table_devices
);
1625 dm_stats_cleanup(&md
->stats
);
1628 module_put(THIS_MODULE
);
1632 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1634 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1637 /* The md already has necessary mempools. */
1638 if (dm_table_bio_based(t
)) {
1640 * Reload bioset because front_pad may have changed
1641 * because a different table was loaded.
1643 bioset_free(md
->bs
);
1648 * There's no need to reload with request-based dm
1649 * because the size of front_pad doesn't change.
1650 * Note for future: If you are to reload bioset,
1651 * prep-ed requests in the queue may refer
1652 * to bio from the old bioset, so you must walk
1653 * through the queue to unprep.
1658 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1660 md
->io_pool
= p
->io_pool
;
1666 /* mempool bind completed, no longer need any mempools in the table */
1667 dm_table_free_md_mempools(t
);
1671 * Bind a table to the device.
1673 static void event_callback(void *context
)
1675 unsigned long flags
;
1677 struct mapped_device
*md
= (struct mapped_device
*) context
;
1679 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1680 list_splice_init(&md
->uevent_list
, &uevents
);
1681 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1683 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1685 atomic_inc(&md
->event_nr
);
1686 wake_up(&md
->eventq
);
1690 * Protected by md->suspend_lock obtained by dm_swap_table().
1692 static void __set_size(struct mapped_device
*md
, sector_t size
)
1694 set_capacity(md
->disk
, size
);
1696 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1700 * Returns old map, which caller must destroy.
1702 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1703 struct queue_limits
*limits
)
1705 struct dm_table
*old_map
;
1706 struct request_queue
*q
= md
->queue
;
1709 lockdep_assert_held(&md
->suspend_lock
);
1711 size
= dm_table_get_size(t
);
1714 * Wipe any geometry if the size of the table changed.
1716 if (size
!= dm_get_size(md
))
1717 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1719 __set_size(md
, size
);
1721 dm_table_event_callback(t
, event_callback
, md
);
1724 * The queue hasn't been stopped yet, if the old table type wasn't
1725 * for request-based during suspension. So stop it to prevent
1726 * I/O mapping before resume.
1727 * This must be done before setting the queue restrictions,
1728 * because request-based dm may be run just after the setting.
1730 if (dm_table_request_based(t
)) {
1733 * Leverage the fact that request-based DM targets are
1734 * immutable singletons and establish md->immutable_target
1735 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1737 md
->immutable_target
= dm_table_get_immutable_target(t
);
1740 __bind_mempools(md
, t
);
1742 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1743 rcu_assign_pointer(md
->map
, (void *)t
);
1744 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1746 dm_table_set_restrictions(t
, q
, limits
);
1754 * Returns unbound table for the caller to free.
1756 static struct dm_table
*__unbind(struct mapped_device
*md
)
1758 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1763 dm_table_event_callback(map
, NULL
, NULL
);
1764 RCU_INIT_POINTER(md
->map
, NULL
);
1771 * Constructor for a new device.
1773 int dm_create(int minor
, struct mapped_device
**result
)
1775 struct mapped_device
*md
;
1777 md
= alloc_dev(minor
);
1788 * Functions to manage md->type.
1789 * All are required to hold md->type_lock.
1791 void dm_lock_md_type(struct mapped_device
*md
)
1793 mutex_lock(&md
->type_lock
);
1796 void dm_unlock_md_type(struct mapped_device
*md
)
1798 mutex_unlock(&md
->type_lock
);
1801 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
1803 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1807 unsigned dm_get_md_type(struct mapped_device
*md
)
1812 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
1814 return md
->immutable_target_type
;
1818 * The queue_limits are only valid as long as you have a reference
1821 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
1823 BUG_ON(!atomic_read(&md
->holders
));
1824 return &md
->queue
->limits
;
1826 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
1829 * Setup the DM device's queue based on md's type
1831 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
1834 unsigned type
= dm_get_md_type(md
);
1837 case DM_TYPE_REQUEST_BASED
:
1838 r
= dm_old_init_request_queue(md
, t
);
1840 DMERR("Cannot initialize queue for request-based mapped device");
1844 case DM_TYPE_MQ_REQUEST_BASED
:
1845 r
= dm_mq_init_request_queue(md
, t
);
1847 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
1851 case DM_TYPE_BIO_BASED
:
1852 case DM_TYPE_DAX_BIO_BASED
:
1853 dm_init_normal_md_queue(md
);
1854 blk_queue_make_request(md
->queue
, dm_make_request
);
1856 * DM handles splitting bios as needed. Free the bio_split bioset
1857 * since it won't be used (saves 1 process per bio-based DM device).
1859 bioset_free(md
->queue
->bio_split
);
1860 md
->queue
->bio_split
= NULL
;
1862 if (type
== DM_TYPE_DAX_BIO_BASED
)
1863 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
1870 struct mapped_device
*dm_get_md(dev_t dev
)
1872 struct mapped_device
*md
;
1873 unsigned minor
= MINOR(dev
);
1875 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1878 spin_lock(&_minor_lock
);
1880 md
= idr_find(&_minor_idr
, minor
);
1882 if ((md
== MINOR_ALLOCED
||
1883 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1884 dm_deleting_md(md
) ||
1885 test_bit(DMF_FREEING
, &md
->flags
))) {
1893 spin_unlock(&_minor_lock
);
1897 EXPORT_SYMBOL_GPL(dm_get_md
);
1899 void *dm_get_mdptr(struct mapped_device
*md
)
1901 return md
->interface_ptr
;
1904 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1906 md
->interface_ptr
= ptr
;
1909 void dm_get(struct mapped_device
*md
)
1911 atomic_inc(&md
->holders
);
1912 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1915 int dm_hold(struct mapped_device
*md
)
1917 spin_lock(&_minor_lock
);
1918 if (test_bit(DMF_FREEING
, &md
->flags
)) {
1919 spin_unlock(&_minor_lock
);
1923 spin_unlock(&_minor_lock
);
1926 EXPORT_SYMBOL_GPL(dm_hold
);
1928 const char *dm_device_name(struct mapped_device
*md
)
1932 EXPORT_SYMBOL_GPL(dm_device_name
);
1934 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
1936 struct request_queue
*q
= dm_get_md_queue(md
);
1937 struct dm_table
*map
;
1942 spin_lock(&_minor_lock
);
1943 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
1944 set_bit(DMF_FREEING
, &md
->flags
);
1945 spin_unlock(&_minor_lock
);
1947 blk_set_queue_dying(q
);
1949 if (dm_request_based(md
) && md
->kworker_task
)
1950 kthread_flush_worker(&md
->kworker
);
1953 * Take suspend_lock so that presuspend and postsuspend methods
1954 * do not race with internal suspend.
1956 mutex_lock(&md
->suspend_lock
);
1957 map
= dm_get_live_table(md
, &srcu_idx
);
1958 if (!dm_suspended_md(md
)) {
1959 dm_table_presuspend_targets(map
);
1960 dm_table_postsuspend_targets(map
);
1962 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
1963 dm_put_live_table(md
, srcu_idx
);
1964 mutex_unlock(&md
->suspend_lock
);
1967 * Rare, but there may be I/O requests still going to complete,
1968 * for example. Wait for all references to disappear.
1969 * No one should increment the reference count of the mapped_device,
1970 * after the mapped_device state becomes DMF_FREEING.
1973 while (atomic_read(&md
->holders
))
1975 else if (atomic_read(&md
->holders
))
1976 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
1977 dm_device_name(md
), atomic_read(&md
->holders
));
1980 dm_table_destroy(__unbind(md
));
1984 void dm_destroy(struct mapped_device
*md
)
1986 __dm_destroy(md
, true);
1989 void dm_destroy_immediate(struct mapped_device
*md
)
1991 __dm_destroy(md
, false);
1994 void dm_put(struct mapped_device
*md
)
1996 atomic_dec(&md
->holders
);
1998 EXPORT_SYMBOL_GPL(dm_put
);
2000 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
2006 prepare_to_wait(&md
->wait
, &wait
, task_state
);
2008 if (!md_in_flight(md
))
2011 if (signal_pending_state(task_state
, current
)) {
2018 finish_wait(&md
->wait
, &wait
);
2024 * Process the deferred bios
2026 static void dm_wq_work(struct work_struct
*work
)
2028 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2032 struct dm_table
*map
;
2034 map
= dm_get_live_table(md
, &srcu_idx
);
2036 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2037 spin_lock_irq(&md
->deferred_lock
);
2038 c
= bio_list_pop(&md
->deferred
);
2039 spin_unlock_irq(&md
->deferred_lock
);
2044 if (dm_request_based(md
))
2045 generic_make_request(c
);
2047 __split_and_process_bio(md
, map
, c
);
2050 dm_put_live_table(md
, srcu_idx
);
2053 static void dm_queue_flush(struct mapped_device
*md
)
2055 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2056 smp_mb__after_atomic();
2057 queue_work(md
->wq
, &md
->work
);
2061 * Swap in a new table, returning the old one for the caller to destroy.
2063 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2065 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2066 struct queue_limits limits
;
2069 mutex_lock(&md
->suspend_lock
);
2071 /* device must be suspended */
2072 if (!dm_suspended_md(md
))
2076 * If the new table has no data devices, retain the existing limits.
2077 * This helps multipath with queue_if_no_path if all paths disappear,
2078 * then new I/O is queued based on these limits, and then some paths
2081 if (dm_table_has_no_data_devices(table
)) {
2082 live_map
= dm_get_live_table_fast(md
);
2084 limits
= md
->queue
->limits
;
2085 dm_put_live_table_fast(md
);
2089 r
= dm_calculate_queue_limits(table
, &limits
);
2096 map
= __bind(md
, table
, &limits
);
2099 mutex_unlock(&md
->suspend_lock
);
2104 * Functions to lock and unlock any filesystem running on the
2107 static int lock_fs(struct mapped_device
*md
)
2111 WARN_ON(md
->frozen_sb
);
2113 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2114 if (IS_ERR(md
->frozen_sb
)) {
2115 r
= PTR_ERR(md
->frozen_sb
);
2116 md
->frozen_sb
= NULL
;
2120 set_bit(DMF_FROZEN
, &md
->flags
);
2125 static void unlock_fs(struct mapped_device
*md
)
2127 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2130 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2131 md
->frozen_sb
= NULL
;
2132 clear_bit(DMF_FROZEN
, &md
->flags
);
2136 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2137 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2138 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2140 * If __dm_suspend returns 0, the device is completely quiescent
2141 * now. There is no request-processing activity. All new requests
2142 * are being added to md->deferred list.
2144 * Caller must hold md->suspend_lock
2146 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2147 unsigned suspend_flags
, long task_state
,
2148 int dmf_suspended_flag
)
2150 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2151 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2154 lockdep_assert_held(&md
->suspend_lock
);
2157 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2158 * This flag is cleared before dm_suspend returns.
2161 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2164 * This gets reverted if there's an error later and the targets
2165 * provide the .presuspend_undo hook.
2167 dm_table_presuspend_targets(map
);
2170 * Flush I/O to the device.
2171 * Any I/O submitted after lock_fs() may not be flushed.
2172 * noflush takes precedence over do_lockfs.
2173 * (lock_fs() flushes I/Os and waits for them to complete.)
2175 if (!noflush
&& do_lockfs
) {
2178 dm_table_presuspend_undo_targets(map
);
2184 * Here we must make sure that no processes are submitting requests
2185 * to target drivers i.e. no one may be executing
2186 * __split_and_process_bio. This is called from dm_request and
2189 * To get all processes out of __split_and_process_bio in dm_request,
2190 * we take the write lock. To prevent any process from reentering
2191 * __split_and_process_bio from dm_request and quiesce the thread
2192 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2193 * flush_workqueue(md->wq).
2195 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2197 synchronize_srcu(&md
->io_barrier
);
2200 * Stop md->queue before flushing md->wq in case request-based
2201 * dm defers requests to md->wq from md->queue.
2203 if (dm_request_based(md
)) {
2204 dm_stop_queue(md
->queue
);
2205 if (md
->kworker_task
)
2206 kthread_flush_worker(&md
->kworker
);
2209 flush_workqueue(md
->wq
);
2212 * At this point no more requests are entering target request routines.
2213 * We call dm_wait_for_completion to wait for all existing requests
2216 r
= dm_wait_for_completion(md
, task_state
);
2218 set_bit(dmf_suspended_flag
, &md
->flags
);
2221 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2223 synchronize_srcu(&md
->io_barrier
);
2225 /* were we interrupted ? */
2229 if (dm_request_based(md
))
2230 dm_start_queue(md
->queue
);
2233 dm_table_presuspend_undo_targets(map
);
2234 /* pushback list is already flushed, so skip flush */
2241 * We need to be able to change a mapping table under a mounted
2242 * filesystem. For example we might want to move some data in
2243 * the background. Before the table can be swapped with
2244 * dm_bind_table, dm_suspend must be called to flush any in
2245 * flight bios and ensure that any further io gets deferred.
2248 * Suspend mechanism in request-based dm.
2250 * 1. Flush all I/Os by lock_fs() if needed.
2251 * 2. Stop dispatching any I/O by stopping the request_queue.
2252 * 3. Wait for all in-flight I/Os to be completed or requeued.
2254 * To abort suspend, start the request_queue.
2256 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2258 struct dm_table
*map
= NULL
;
2262 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2264 if (dm_suspended_md(md
)) {
2269 if (dm_suspended_internally_md(md
)) {
2270 /* already internally suspended, wait for internal resume */
2271 mutex_unlock(&md
->suspend_lock
);
2272 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2278 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2280 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2284 dm_table_postsuspend_targets(map
);
2287 mutex_unlock(&md
->suspend_lock
);
2291 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2294 int r
= dm_table_resume_targets(map
);
2302 * Flushing deferred I/Os must be done after targets are resumed
2303 * so that mapping of targets can work correctly.
2304 * Request-based dm is queueing the deferred I/Os in its request_queue.
2306 if (dm_request_based(md
))
2307 dm_start_queue(md
->queue
);
2314 int dm_resume(struct mapped_device
*md
)
2317 struct dm_table
*map
= NULL
;
2321 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2323 if (!dm_suspended_md(md
))
2326 if (dm_suspended_internally_md(md
)) {
2327 /* already internally suspended, wait for internal resume */
2328 mutex_unlock(&md
->suspend_lock
);
2329 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2335 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2336 if (!map
|| !dm_table_get_size(map
))
2339 r
= __dm_resume(md
, map
);
2343 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2345 mutex_unlock(&md
->suspend_lock
);
2351 * Internal suspend/resume works like userspace-driven suspend. It waits
2352 * until all bios finish and prevents issuing new bios to the target drivers.
2353 * It may be used only from the kernel.
2356 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2358 struct dm_table
*map
= NULL
;
2360 if (md
->internal_suspend_count
++)
2361 return; /* nested internal suspend */
2363 if (dm_suspended_md(md
)) {
2364 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2365 return; /* nest suspend */
2368 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2371 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2372 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2373 * would require changing .presuspend to return an error -- avoid this
2374 * until there is a need for more elaborate variants of internal suspend.
2376 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2377 DMF_SUSPENDED_INTERNALLY
);
2379 dm_table_postsuspend_targets(map
);
2382 static void __dm_internal_resume(struct mapped_device
*md
)
2384 BUG_ON(!md
->internal_suspend_count
);
2386 if (--md
->internal_suspend_count
)
2387 return; /* resume from nested internal suspend */
2389 if (dm_suspended_md(md
))
2390 goto done
; /* resume from nested suspend */
2393 * NOTE: existing callers don't need to call dm_table_resume_targets
2394 * (which may fail -- so best to avoid it for now by passing NULL map)
2396 (void) __dm_resume(md
, NULL
);
2399 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2400 smp_mb__after_atomic();
2401 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2404 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2406 mutex_lock(&md
->suspend_lock
);
2407 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2408 mutex_unlock(&md
->suspend_lock
);
2410 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2412 void dm_internal_resume(struct mapped_device
*md
)
2414 mutex_lock(&md
->suspend_lock
);
2415 __dm_internal_resume(md
);
2416 mutex_unlock(&md
->suspend_lock
);
2418 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2421 * Fast variants of internal suspend/resume hold md->suspend_lock,
2422 * which prevents interaction with userspace-driven suspend.
2425 void dm_internal_suspend_fast(struct mapped_device
*md
)
2427 mutex_lock(&md
->suspend_lock
);
2428 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2431 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2432 synchronize_srcu(&md
->io_barrier
);
2433 flush_workqueue(md
->wq
);
2434 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2436 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2438 void dm_internal_resume_fast(struct mapped_device
*md
)
2440 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2446 mutex_unlock(&md
->suspend_lock
);
2448 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2450 /*-----------------------------------------------------------------
2451 * Event notification.
2452 *---------------------------------------------------------------*/
2453 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2456 char udev_cookie
[DM_COOKIE_LENGTH
];
2457 char *envp
[] = { udev_cookie
, NULL
};
2460 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2462 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2463 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2464 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2469 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2471 return atomic_add_return(1, &md
->uevent_seq
);
2474 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2476 return atomic_read(&md
->event_nr
);
2479 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2481 return wait_event_interruptible(md
->eventq
,
2482 (event_nr
!= atomic_read(&md
->event_nr
)));
2485 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2487 unsigned long flags
;
2489 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2490 list_add(elist
, &md
->uevent_list
);
2491 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2495 * The gendisk is only valid as long as you have a reference
2498 struct gendisk
*dm_disk(struct mapped_device
*md
)
2502 EXPORT_SYMBOL_GPL(dm_disk
);
2504 struct kobject
*dm_kobject(struct mapped_device
*md
)
2506 return &md
->kobj_holder
.kobj
;
2509 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2511 struct mapped_device
*md
;
2513 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2515 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2523 int dm_suspended_md(struct mapped_device
*md
)
2525 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2528 int dm_suspended_internally_md(struct mapped_device
*md
)
2530 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2533 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2535 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2538 int dm_suspended(struct dm_target
*ti
)
2540 return dm_suspended_md(dm_table_get_md(ti
->table
));
2542 EXPORT_SYMBOL_GPL(dm_suspended
);
2544 int dm_noflush_suspending(struct dm_target
*ti
)
2546 return __noflush_suspending(dm_table_get_md(ti
->table
));
2548 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2550 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
2551 unsigned integrity
, unsigned per_io_data_size
)
2553 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2554 unsigned int pool_size
= 0;
2555 unsigned int front_pad
;
2561 case DM_TYPE_BIO_BASED
:
2562 case DM_TYPE_DAX_BIO_BASED
:
2563 pool_size
= dm_get_reserved_bio_based_ios();
2564 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2566 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2567 if (!pools
->io_pool
)
2570 case DM_TYPE_REQUEST_BASED
:
2571 case DM_TYPE_MQ_REQUEST_BASED
:
2572 pool_size
= dm_get_reserved_rq_based_ios();
2573 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2574 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2580 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
2584 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2590 dm_free_md_mempools(pools
);
2595 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2600 mempool_destroy(pools
->io_pool
);
2603 bioset_free(pools
->bs
);
2615 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2618 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2619 struct dm_table
*table
;
2620 struct dm_target
*ti
;
2621 int ret
= -ENOTTY
, srcu_idx
;
2623 table
= dm_get_live_table(md
, &srcu_idx
);
2624 if (!table
|| !dm_table_get_size(table
))
2627 /* We only support devices that have a single target */
2628 if (dm_table_get_num_targets(table
) != 1)
2630 ti
= dm_table_get_target(table
, 0);
2633 if (!ti
->type
->iterate_devices
)
2636 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2638 dm_put_live_table(md
, srcu_idx
);
2643 * For register / unregister we need to manually call out to every path.
2645 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2646 sector_t start
, sector_t len
, void *data
)
2648 struct dm_pr
*pr
= data
;
2649 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2651 if (!ops
|| !ops
->pr_register
)
2653 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2656 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2667 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2668 if (ret
&& new_key
) {
2669 /* unregister all paths if we failed to register any path */
2670 pr
.old_key
= new_key
;
2673 pr
.fail_early
= false;
2674 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2680 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2683 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2684 const struct pr_ops
*ops
;
2688 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2692 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2693 if (ops
&& ops
->pr_reserve
)
2694 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2702 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2704 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2705 const struct pr_ops
*ops
;
2709 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2713 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2714 if (ops
&& ops
->pr_release
)
2715 r
= ops
->pr_release(bdev
, key
, type
);
2723 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2724 enum pr_type type
, bool abort
)
2726 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2727 const struct pr_ops
*ops
;
2731 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2735 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2736 if (ops
&& ops
->pr_preempt
)
2737 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2745 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2747 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2748 const struct pr_ops
*ops
;
2752 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2756 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2757 if (ops
&& ops
->pr_clear
)
2758 r
= ops
->pr_clear(bdev
, key
);
2766 static const struct pr_ops dm_pr_ops
= {
2767 .pr_register
= dm_pr_register
,
2768 .pr_reserve
= dm_pr_reserve
,
2769 .pr_release
= dm_pr_release
,
2770 .pr_preempt
= dm_pr_preempt
,
2771 .pr_clear
= dm_pr_clear
,
2774 static const struct block_device_operations dm_blk_dops
= {
2775 .open
= dm_blk_open
,
2776 .release
= dm_blk_close
,
2777 .ioctl
= dm_blk_ioctl
,
2778 .direct_access
= dm_blk_direct_access
,
2779 .getgeo
= dm_blk_getgeo
,
2780 .pr_ops
= &dm_pr_ops
,
2781 .owner
= THIS_MODULE
2787 module_init(dm_init
);
2788 module_exit(dm_exit
);
2790 module_param(major
, uint
, 0);
2791 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2793 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
2794 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
2796 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
2797 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
2799 MODULE_DESCRIPTION(DM_NAME
" driver");
2800 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2801 MODULE_LICENSE("GPL");