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.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.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>
23 #include <linux/kthread.h>
24 #include <linux/ktime.h>
25 #include <linux/elevator.h> /* for rq_end_sector() */
26 #include <linux/blk-mq.h>
28 #include <trace/events/block.h>
30 #define DM_MSG_PREFIX "core"
34 * ratelimit state to be used in DMXXX_LIMIT().
36 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
37 DEFAULT_RATELIMIT_INTERVAL
,
38 DEFAULT_RATELIMIT_BURST
);
39 EXPORT_SYMBOL(dm_ratelimit_state
);
43 * Cookies are numeric values sent with CHANGE and REMOVE
44 * uevents while resuming, removing or renaming the device.
46 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
47 #define DM_COOKIE_LENGTH 24
49 static const char *_name
= DM_NAME
;
51 static unsigned int major
= 0;
52 static unsigned int _major
= 0;
54 static DEFINE_IDR(_minor_idr
);
56 static DEFINE_SPINLOCK(_minor_lock
);
58 static void do_deferred_remove(struct work_struct
*w
);
60 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
62 static struct workqueue_struct
*deferred_remove_workqueue
;
66 * One of these is allocated per bio.
69 struct mapped_device
*md
;
73 unsigned long start_time
;
74 spinlock_t endio_lock
;
75 struct dm_stats_aux stats_aux
;
79 * For request-based dm.
80 * One of these is allocated per request.
82 struct dm_rq_target_io
{
83 struct mapped_device
*md
;
85 struct request
*orig
, *clone
;
86 struct kthread_work work
;
92 * For request-based dm - the bio clones we allocate are embedded in these
95 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
96 * the bioset is created - this means the bio has to come at the end of the
99 struct dm_rq_clone_bio_info
{
101 struct dm_rq_target_io
*tio
;
105 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
107 if (rq
&& rq
->end_io_data
)
108 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
125 #define DMF_DEFERRED_REMOVE 7
126 #define DMF_SUSPENDED_INTERNALLY 8
129 * A dummy definition to make RCU happy.
130 * struct dm_table should never be dereferenced in this file.
137 * Work processed by per-device workqueue.
139 struct mapped_device
{
140 struct srcu_struct io_barrier
;
141 struct mutex suspend_lock
;
146 * The current mapping.
147 * Use dm_get_live_table{_fast} or take suspend_lock for
150 struct dm_table __rcu
*map
;
152 struct list_head table_devices
;
153 struct mutex table_devices_lock
;
157 struct request_queue
*queue
;
159 /* Protect queue and type against concurrent access. */
160 struct mutex type_lock
;
162 struct target_type
*immutable_target_type
;
164 struct gendisk
*disk
;
170 * A list of ios that arrived while we were suspended.
173 wait_queue_head_t wait
;
174 struct work_struct work
;
175 struct bio_list deferred
;
176 spinlock_t deferred_lock
;
179 * Processing queue (flush)
181 struct workqueue_struct
*wq
;
184 * io objects are allocated from here.
195 wait_queue_head_t eventq
;
197 struct list_head uevent_list
;
198 spinlock_t uevent_lock
; /* Protect access to uevent_list */
201 * freeze/thaw support require holding onto a super block
203 struct super_block
*frozen_sb
;
204 struct block_device
*bdev
;
206 /* forced geometry settings */
207 struct hd_geometry geometry
;
209 /* kobject and completion */
210 struct dm_kobject_holder kobj_holder
;
212 /* zero-length flush that will be cloned and submitted to targets */
213 struct bio flush_bio
;
215 /* the number of internal suspends */
216 unsigned internal_suspend_count
;
218 struct dm_stats stats
;
220 struct kthread_worker kworker
;
221 struct task_struct
*kworker_task
;
223 /* for request-based merge heuristic in dm_request_fn() */
224 unsigned seq_rq_merge_deadline_usecs
;
226 sector_t last_rq_pos
;
227 ktime_t last_rq_start_time
;
229 /* for blk-mq request-based DM support */
230 struct blk_mq_tag_set tag_set
;
234 #ifdef CONFIG_DM_MQ_DEFAULT
235 static bool use_blk_mq
= true;
237 static bool use_blk_mq
= false;
240 bool dm_use_blk_mq(struct mapped_device
*md
)
242 return md
->use_blk_mq
;
246 * For mempools pre-allocation at the table loading time.
248 struct dm_md_mempools
{
254 struct table_device
{
255 struct list_head list
;
257 struct dm_dev dm_dev
;
260 #define RESERVED_BIO_BASED_IOS 16
261 #define RESERVED_REQUEST_BASED_IOS 256
262 #define RESERVED_MAX_IOS 1024
263 static struct kmem_cache
*_io_cache
;
264 static struct kmem_cache
*_rq_tio_cache
;
265 static struct kmem_cache
*_rq_cache
;
268 * Bio-based DM's mempools' reserved IOs set by the user.
270 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
273 * Request-based DM's mempools' reserved IOs set by the user.
275 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
277 static unsigned __dm_get_module_param(unsigned *module_param
,
278 unsigned def
, unsigned max
)
280 unsigned param
= ACCESS_ONCE(*module_param
);
281 unsigned modified_param
= 0;
284 modified_param
= def
;
285 else if (param
> max
)
286 modified_param
= max
;
288 if (modified_param
) {
289 (void)cmpxchg(module_param
, param
, modified_param
);
290 param
= modified_param
;
296 unsigned dm_get_reserved_bio_based_ios(void)
298 return __dm_get_module_param(&reserved_bio_based_ios
,
299 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
301 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
303 unsigned dm_get_reserved_rq_based_ios(void)
305 return __dm_get_module_param(&reserved_rq_based_ios
,
306 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
308 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
310 static int __init
local_init(void)
314 /* allocate a slab for the dm_ios */
315 _io_cache
= KMEM_CACHE(dm_io
, 0);
319 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
321 goto out_free_io_cache
;
323 _rq_cache
= kmem_cache_create("dm_clone_request", sizeof(struct request
),
324 __alignof__(struct request
), 0, NULL
);
326 goto out_free_rq_tio_cache
;
328 r
= dm_uevent_init();
330 goto out_free_rq_cache
;
332 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
333 if (!deferred_remove_workqueue
) {
335 goto out_uevent_exit
;
339 r
= register_blkdev(_major
, _name
);
341 goto out_free_workqueue
;
349 destroy_workqueue(deferred_remove_workqueue
);
353 kmem_cache_destroy(_rq_cache
);
354 out_free_rq_tio_cache
:
355 kmem_cache_destroy(_rq_tio_cache
);
357 kmem_cache_destroy(_io_cache
);
362 static void local_exit(void)
364 flush_scheduled_work();
365 destroy_workqueue(deferred_remove_workqueue
);
367 kmem_cache_destroy(_rq_cache
);
368 kmem_cache_destroy(_rq_tio_cache
);
369 kmem_cache_destroy(_io_cache
);
370 unregister_blkdev(_major
, _name
);
375 DMINFO("cleaned up");
378 static int (*_inits
[])(void) __initdata
= {
389 static void (*_exits
[])(void) = {
400 static int __init
dm_init(void)
402 const int count
= ARRAY_SIZE(_inits
);
406 for (i
= 0; i
< count
; i
++) {
421 static void __exit
dm_exit(void)
423 int i
= ARRAY_SIZE(_exits
);
429 * Should be empty by this point.
431 idr_destroy(&_minor_idr
);
435 * Block device functions
437 int dm_deleting_md(struct mapped_device
*md
)
439 return test_bit(DMF_DELETING
, &md
->flags
);
442 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
444 struct mapped_device
*md
;
446 spin_lock(&_minor_lock
);
448 md
= bdev
->bd_disk
->private_data
;
452 if (test_bit(DMF_FREEING
, &md
->flags
) ||
453 dm_deleting_md(md
)) {
459 atomic_inc(&md
->open_count
);
461 spin_unlock(&_minor_lock
);
463 return md
? 0 : -ENXIO
;
466 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
468 struct mapped_device
*md
;
470 spin_lock(&_minor_lock
);
472 md
= disk
->private_data
;
476 if (atomic_dec_and_test(&md
->open_count
) &&
477 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
478 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
482 spin_unlock(&_minor_lock
);
485 int dm_open_count(struct mapped_device
*md
)
487 return atomic_read(&md
->open_count
);
491 * Guarantees nothing is using the device before it's deleted.
493 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
497 spin_lock(&_minor_lock
);
499 if (dm_open_count(md
)) {
502 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
503 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
506 set_bit(DMF_DELETING
, &md
->flags
);
508 spin_unlock(&_minor_lock
);
513 int dm_cancel_deferred_remove(struct mapped_device
*md
)
517 spin_lock(&_minor_lock
);
519 if (test_bit(DMF_DELETING
, &md
->flags
))
522 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
524 spin_unlock(&_minor_lock
);
529 static void do_deferred_remove(struct work_struct
*w
)
531 dm_deferred_remove();
534 sector_t
dm_get_size(struct mapped_device
*md
)
536 return get_capacity(md
->disk
);
539 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
544 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
549 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
551 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
553 return dm_get_geometry(md
, geo
);
556 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
557 unsigned int cmd
, unsigned long arg
)
559 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
561 struct dm_table
*map
;
562 struct dm_target
*tgt
;
566 map
= dm_get_live_table(md
, &srcu_idx
);
568 if (!map
|| !dm_table_get_size(map
))
571 /* We only support devices that have a single target */
572 if (dm_table_get_num_targets(map
) != 1)
575 tgt
= dm_table_get_target(map
, 0);
576 if (!tgt
->type
->ioctl
)
579 if (dm_suspended_md(md
)) {
584 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
587 dm_put_live_table(md
, srcu_idx
);
589 if (r
== -ENOTCONN
) {
597 static struct dm_io
*alloc_io(struct mapped_device
*md
)
599 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
602 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
604 mempool_free(io
, md
->io_pool
);
607 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
609 bio_put(&tio
->clone
);
612 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
615 return mempool_alloc(md
->io_pool
, gfp_mask
);
618 static void free_rq_tio(struct dm_rq_target_io
*tio
)
620 mempool_free(tio
, tio
->md
->io_pool
);
623 static struct request
*alloc_clone_request(struct mapped_device
*md
,
626 return mempool_alloc(md
->rq_pool
, gfp_mask
);
629 static void free_clone_request(struct mapped_device
*md
, struct request
*rq
)
631 mempool_free(rq
, md
->rq_pool
);
634 static int md_in_flight(struct mapped_device
*md
)
636 return atomic_read(&md
->pending
[READ
]) +
637 atomic_read(&md
->pending
[WRITE
]);
640 static void start_io_acct(struct dm_io
*io
)
642 struct mapped_device
*md
= io
->md
;
643 struct bio
*bio
= io
->bio
;
645 int rw
= bio_data_dir(bio
);
647 io
->start_time
= jiffies
;
649 cpu
= part_stat_lock();
650 part_round_stats(cpu
, &dm_disk(md
)->part0
);
652 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
653 atomic_inc_return(&md
->pending
[rw
]));
655 if (unlikely(dm_stats_used(&md
->stats
)))
656 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
657 bio_sectors(bio
), false, 0, &io
->stats_aux
);
660 static void end_io_acct(struct dm_io
*io
)
662 struct mapped_device
*md
= io
->md
;
663 struct bio
*bio
= io
->bio
;
664 unsigned long duration
= jiffies
- io
->start_time
;
666 int rw
= bio_data_dir(bio
);
668 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
670 if (unlikely(dm_stats_used(&md
->stats
)))
671 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
672 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
675 * After this is decremented the bio must not be touched if it is
678 pending
= atomic_dec_return(&md
->pending
[rw
]);
679 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
680 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
682 /* nudge anyone waiting on suspend queue */
688 * Add the bio to the list of deferred io.
690 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
694 spin_lock_irqsave(&md
->deferred_lock
, flags
);
695 bio_list_add(&md
->deferred
, bio
);
696 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
697 queue_work(md
->wq
, &md
->work
);
701 * Everyone (including functions in this file), should use this
702 * function to access the md->map field, and make sure they call
703 * dm_put_live_table() when finished.
705 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
707 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
709 return srcu_dereference(md
->map
, &md
->io_barrier
);
712 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
714 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
717 void dm_sync_table(struct mapped_device
*md
)
719 synchronize_srcu(&md
->io_barrier
);
720 synchronize_rcu_expedited();
724 * A fast alternative to dm_get_live_table/dm_put_live_table.
725 * The caller must not block between these two functions.
727 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
730 return rcu_dereference(md
->map
);
733 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
739 * Open a table device so we can use it as a map destination.
741 static int open_table_device(struct table_device
*td
, dev_t dev
,
742 struct mapped_device
*md
)
744 static char *_claim_ptr
= "I belong to device-mapper";
745 struct block_device
*bdev
;
749 BUG_ON(td
->dm_dev
.bdev
);
751 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
753 return PTR_ERR(bdev
);
755 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
757 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
761 td
->dm_dev
.bdev
= bdev
;
766 * Close a table device that we've been using.
768 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
770 if (!td
->dm_dev
.bdev
)
773 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
774 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
775 td
->dm_dev
.bdev
= NULL
;
778 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
780 struct table_device
*td
;
782 list_for_each_entry(td
, l
, list
)
783 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
789 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
790 struct dm_dev
**result
) {
792 struct table_device
*td
;
794 mutex_lock(&md
->table_devices_lock
);
795 td
= find_table_device(&md
->table_devices
, dev
, mode
);
797 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
799 mutex_unlock(&md
->table_devices_lock
);
803 td
->dm_dev
.mode
= mode
;
804 td
->dm_dev
.bdev
= NULL
;
806 if ((r
= open_table_device(td
, dev
, md
))) {
807 mutex_unlock(&md
->table_devices_lock
);
812 format_dev_t(td
->dm_dev
.name
, dev
);
814 atomic_set(&td
->count
, 0);
815 list_add(&td
->list
, &md
->table_devices
);
817 atomic_inc(&td
->count
);
818 mutex_unlock(&md
->table_devices_lock
);
820 *result
= &td
->dm_dev
;
823 EXPORT_SYMBOL_GPL(dm_get_table_device
);
825 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
827 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
829 mutex_lock(&md
->table_devices_lock
);
830 if (atomic_dec_and_test(&td
->count
)) {
831 close_table_device(td
, md
);
835 mutex_unlock(&md
->table_devices_lock
);
837 EXPORT_SYMBOL(dm_put_table_device
);
839 static void free_table_devices(struct list_head
*devices
)
841 struct list_head
*tmp
, *next
;
843 list_for_each_safe(tmp
, next
, devices
) {
844 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
846 DMWARN("dm_destroy: %s still exists with %d references",
847 td
->dm_dev
.name
, atomic_read(&td
->count
));
853 * Get the geometry associated with a dm device
855 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
863 * Set the geometry of a device.
865 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
867 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
869 if (geo
->start
> sz
) {
870 DMWARN("Start sector is beyond the geometry limits.");
879 /*-----------------------------------------------------------------
881 * A more elegant soln is in the works that uses the queue
882 * merge fn, unfortunately there are a couple of changes to
883 * the block layer that I want to make for this. So in the
884 * interests of getting something for people to use I give
885 * you this clearly demarcated crap.
886 *---------------------------------------------------------------*/
888 static int __noflush_suspending(struct mapped_device
*md
)
890 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
894 * Decrements the number of outstanding ios that a bio has been
895 * cloned into, completing the original io if necc.
897 static void dec_pending(struct dm_io
*io
, int error
)
902 struct mapped_device
*md
= io
->md
;
904 /* Push-back supersedes any I/O errors */
905 if (unlikely(error
)) {
906 spin_lock_irqsave(&io
->endio_lock
, flags
);
907 if (!(io
->error
> 0 && __noflush_suspending(md
)))
909 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
912 if (atomic_dec_and_test(&io
->io_count
)) {
913 if (io
->error
== DM_ENDIO_REQUEUE
) {
915 * Target requested pushing back the I/O.
917 spin_lock_irqsave(&md
->deferred_lock
, flags
);
918 if (__noflush_suspending(md
))
919 bio_list_add_head(&md
->deferred
, io
->bio
);
921 /* noflush suspend was interrupted. */
923 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
926 io_error
= io
->error
;
931 if (io_error
== DM_ENDIO_REQUEUE
)
934 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
936 * Preflush done for flush with data, reissue
939 bio
->bi_rw
&= ~REQ_FLUSH
;
942 /* done with normal IO or empty flush */
943 trace_block_bio_complete(md
->queue
, bio
, io_error
);
944 bio_endio(bio
, io_error
);
949 static void disable_write_same(struct mapped_device
*md
)
951 struct queue_limits
*limits
= dm_get_queue_limits(md
);
953 /* device doesn't really support WRITE SAME, disable it */
954 limits
->max_write_same_sectors
= 0;
957 static void clone_endio(struct bio
*bio
, int error
)
960 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
961 struct dm_io
*io
= tio
->io
;
962 struct mapped_device
*md
= tio
->io
->md
;
963 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
965 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
969 r
= endio(tio
->ti
, bio
, error
);
970 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
972 * error and requeue request are handled
976 else if (r
== DM_ENDIO_INCOMPLETE
)
977 /* The target will handle the io */
980 DMWARN("unimplemented target endio return value: %d", r
);
985 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
986 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
987 disable_write_same(md
);
990 dec_pending(io
, error
);
993 static struct dm_rq_target_io
*tio_from_request(struct request
*rq
)
995 return (rq
->q
->mq_ops
? blk_mq_rq_to_pdu(rq
) : rq
->special
);
999 * Don't touch any member of the md after calling this function because
1000 * the md may be freed in dm_put() at the end of this function.
1001 * Or do dm_get() before calling this function and dm_put() later.
1003 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1005 int nr_requests_pending
;
1007 atomic_dec(&md
->pending
[rw
]);
1009 /* nudge anyone waiting on suspend queue */
1010 nr_requests_pending
= md_in_flight(md
);
1011 if (!nr_requests_pending
)
1015 * Run this off this callpath, as drivers could invoke end_io while
1016 * inside their request_fn (and holding the queue lock). Calling
1017 * back into ->request_fn() could deadlock attempting to grab the
1021 if (md
->queue
->mq_ops
)
1022 blk_mq_run_hw_queues(md
->queue
, true);
1023 else if (!nr_requests_pending
||
1024 (nr_requests_pending
>= md
->queue
->nr_congestion_on
))
1025 blk_run_queue_async(md
->queue
);
1029 * dm_put() must be at the end of this function. See the comment above
1034 static void free_rq_clone(struct request
*clone
, bool must_be_mapped
)
1036 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1037 struct mapped_device
*md
= tio
->md
;
1039 WARN_ON_ONCE(must_be_mapped
&& !clone
->q
);
1041 if (md
->type
== DM_TYPE_MQ_REQUEST_BASED
)
1042 /* stacked on blk-mq queue(s) */
1043 tio
->ti
->type
->release_clone_rq(clone
);
1044 else if (!md
->queue
->mq_ops
)
1045 /* request_fn queue stacked on request_fn queue(s) */
1046 free_clone_request(md
, clone
);
1048 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1049 * no need to call free_clone_request() because we leverage blk-mq by
1050 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1053 if (!md
->queue
->mq_ops
)
1058 * Complete the clone and the original request.
1059 * Must be called without clone's queue lock held,
1060 * see end_clone_request() for more details.
1062 static void dm_end_request(struct request
*clone
, int error
)
1064 int rw
= rq_data_dir(clone
);
1065 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1066 struct mapped_device
*md
= tio
->md
;
1067 struct request
*rq
= tio
->orig
;
1069 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1070 rq
->errors
= clone
->errors
;
1071 rq
->resid_len
= clone
->resid_len
;
1075 * We are using the sense buffer of the original
1077 * So setting the length of the sense data is enough.
1079 rq
->sense_len
= clone
->sense_len
;
1082 free_rq_clone(clone
, true);
1084 blk_end_request_all(rq
, error
);
1086 blk_mq_end_request(rq
, error
);
1087 rq_completed(md
, rw
, true);
1090 static void dm_unprep_request(struct request
*rq
)
1092 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1093 struct request
*clone
= tio
->clone
;
1095 if (!rq
->q
->mq_ops
) {
1097 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1101 free_rq_clone(clone
, false);
1105 * Requeue the original request of a clone.
1107 static void old_requeue_request(struct request
*rq
)
1109 struct request_queue
*q
= rq
->q
;
1110 unsigned long flags
;
1112 spin_lock_irqsave(q
->queue_lock
, flags
);
1113 blk_requeue_request(q
, rq
);
1114 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1117 static void dm_requeue_unmapped_original_request(struct mapped_device
*md
,
1120 int rw
= rq_data_dir(rq
);
1122 dm_unprep_request(rq
);
1125 old_requeue_request(rq
);
1127 blk_mq_requeue_request(rq
);
1128 blk_mq_kick_requeue_list(rq
->q
);
1131 rq_completed(md
, rw
, false);
1134 static void dm_requeue_unmapped_request(struct request
*clone
)
1136 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1138 dm_requeue_unmapped_original_request(tio
->md
, tio
->orig
);
1141 static void old_stop_queue(struct request_queue
*q
)
1143 unsigned long flags
;
1145 if (blk_queue_stopped(q
))
1148 spin_lock_irqsave(q
->queue_lock
, flags
);
1150 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1153 static void stop_queue(struct request_queue
*q
)
1158 blk_mq_stop_hw_queues(q
);
1161 static void old_start_queue(struct request_queue
*q
)
1163 unsigned long flags
;
1165 spin_lock_irqsave(q
->queue_lock
, flags
);
1166 if (blk_queue_stopped(q
))
1168 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1171 static void start_queue(struct request_queue
*q
)
1176 blk_mq_start_stopped_hw_queues(q
, true);
1179 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1182 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1183 dm_request_endio_fn rq_end_io
= NULL
;
1186 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1188 if (mapped
&& rq_end_io
)
1189 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1192 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1193 !clone
->q
->limits
.max_write_same_sectors
))
1194 disable_write_same(tio
->md
);
1197 /* The target wants to complete the I/O */
1198 dm_end_request(clone
, r
);
1199 else if (r
== DM_ENDIO_INCOMPLETE
)
1200 /* The target will handle the I/O */
1202 else if (r
== DM_ENDIO_REQUEUE
)
1203 /* The target wants to requeue the I/O */
1204 dm_requeue_unmapped_request(clone
);
1206 DMWARN("unimplemented target endio return value: %d", r
);
1212 * Request completion handler for request-based dm
1214 static void dm_softirq_done(struct request
*rq
)
1217 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1218 struct request
*clone
= tio
->clone
;
1222 rw
= rq_data_dir(rq
);
1223 if (!rq
->q
->mq_ops
) {
1224 blk_end_request_all(rq
, tio
->error
);
1225 rq_completed(tio
->md
, rw
, false);
1228 blk_mq_end_request(rq
, tio
->error
);
1229 rq_completed(tio
->md
, rw
, false);
1234 if (rq
->cmd_flags
& REQ_FAILED
)
1237 dm_done(clone
, tio
->error
, mapped
);
1241 * Complete the clone and the original request with the error status
1242 * through softirq context.
1244 static void dm_complete_request(struct request
*rq
, int error
)
1246 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1249 blk_complete_request(rq
);
1253 * Complete the not-mapped clone and the original request with the error status
1254 * through softirq context.
1255 * Target's rq_end_io() function isn't called.
1256 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1258 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1260 rq
->cmd_flags
|= REQ_FAILED
;
1261 dm_complete_request(rq
, error
);
1265 * Called with the clone's queue lock held (for non-blk-mq)
1267 static void end_clone_request(struct request
*clone
, int error
)
1269 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1271 if (!clone
->q
->mq_ops
) {
1273 * For just cleaning up the information of the queue in which
1274 * the clone was dispatched.
1275 * The clone is *NOT* freed actually here because it is alloced
1276 * from dm own mempool (REQ_ALLOCED isn't set).
1278 __blk_put_request(clone
->q
, clone
);
1282 * Actual request completion is done in a softirq context which doesn't
1283 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1284 * - another request may be submitted by the upper level driver
1285 * of the stacking during the completion
1286 * - the submission which requires queue lock may be done
1287 * against this clone's queue
1289 dm_complete_request(tio
->orig
, error
);
1293 * Return maximum size of I/O possible at the supplied sector up to the current
1296 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1298 sector_t target_offset
= dm_target_offset(ti
, sector
);
1300 return ti
->len
- target_offset
;
1303 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1305 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1306 sector_t offset
, max_len
;
1309 * Does the target need to split even further?
1311 if (ti
->max_io_len
) {
1312 offset
= dm_target_offset(ti
, sector
);
1313 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1314 max_len
= sector_div(offset
, ti
->max_io_len
);
1316 max_len
= offset
& (ti
->max_io_len
- 1);
1317 max_len
= ti
->max_io_len
- max_len
;
1326 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1328 if (len
> UINT_MAX
) {
1329 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1330 (unsigned long long)len
, UINT_MAX
);
1331 ti
->error
= "Maximum size of target IO is too large";
1335 ti
->max_io_len
= (uint32_t) len
;
1339 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1342 * A target may call dm_accept_partial_bio only from the map routine. It is
1343 * allowed for all bio types except REQ_FLUSH.
1345 * dm_accept_partial_bio informs the dm that the target only wants to process
1346 * additional n_sectors sectors of the bio and the rest of the data should be
1347 * sent in a next bio.
1349 * A diagram that explains the arithmetics:
1350 * +--------------------+---------------+-------+
1352 * +--------------------+---------------+-------+
1354 * <-------------- *tio->len_ptr --------------->
1355 * <------- bi_size ------->
1358 * Region 1 was already iterated over with bio_advance or similar function.
1359 * (it may be empty if the target doesn't use bio_advance)
1360 * Region 2 is the remaining bio size that the target wants to process.
1361 * (it may be empty if region 1 is non-empty, although there is no reason
1363 * The target requires that region 3 is to be sent in the next bio.
1365 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1366 * the partially processed part (the sum of regions 1+2) must be the same for all
1367 * copies of the bio.
1369 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1371 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1372 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1373 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1374 BUG_ON(bi_size
> *tio
->len_ptr
);
1375 BUG_ON(n_sectors
> bi_size
);
1376 *tio
->len_ptr
-= bi_size
- n_sectors
;
1377 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1379 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1381 static void __map_bio(struct dm_target_io
*tio
)
1385 struct mapped_device
*md
;
1386 struct bio
*clone
= &tio
->clone
;
1387 struct dm_target
*ti
= tio
->ti
;
1389 clone
->bi_end_io
= clone_endio
;
1392 * Map the clone. If r == 0 we don't need to do
1393 * anything, the target has assumed ownership of
1396 atomic_inc(&tio
->io
->io_count
);
1397 sector
= clone
->bi_iter
.bi_sector
;
1398 r
= ti
->type
->map(ti
, clone
);
1399 if (r
== DM_MAPIO_REMAPPED
) {
1400 /* the bio has been remapped so dispatch it */
1402 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1403 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1405 generic_make_request(clone
);
1406 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1407 /* error the io and bail out, or requeue it if needed */
1409 dec_pending(tio
->io
, r
);
1412 DMWARN("unimplemented target map return value: %d", r
);
1418 struct mapped_device
*md
;
1419 struct dm_table
*map
;
1423 unsigned sector_count
;
1426 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1428 bio
->bi_iter
.bi_sector
= sector
;
1429 bio
->bi_iter
.bi_size
= to_bytes(len
);
1433 * Creates a bio that consists of range of complete bvecs.
1435 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1436 sector_t sector
, unsigned len
)
1438 struct bio
*clone
= &tio
->clone
;
1440 __bio_clone_fast(clone
, bio
);
1442 if (bio_integrity(bio
))
1443 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1445 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1446 clone
->bi_iter
.bi_size
= to_bytes(len
);
1448 if (bio_integrity(bio
))
1449 bio_integrity_trim(clone
, 0, len
);
1452 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1453 struct dm_target
*ti
,
1454 unsigned target_bio_nr
)
1456 struct dm_target_io
*tio
;
1459 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1460 tio
= container_of(clone
, struct dm_target_io
, clone
);
1464 tio
->target_bio_nr
= target_bio_nr
;
1469 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1470 struct dm_target
*ti
,
1471 unsigned target_bio_nr
, unsigned *len
)
1473 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1474 struct bio
*clone
= &tio
->clone
;
1478 __bio_clone_fast(clone
, ci
->bio
);
1480 bio_setup_sector(clone
, ci
->sector
, *len
);
1485 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1486 unsigned num_bios
, unsigned *len
)
1488 unsigned target_bio_nr
;
1490 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1491 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1494 static int __send_empty_flush(struct clone_info
*ci
)
1496 unsigned target_nr
= 0;
1497 struct dm_target
*ti
;
1499 BUG_ON(bio_has_data(ci
->bio
));
1500 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1501 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1506 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1507 sector_t sector
, unsigned *len
)
1509 struct bio
*bio
= ci
->bio
;
1510 struct dm_target_io
*tio
;
1511 unsigned target_bio_nr
;
1512 unsigned num_target_bios
= 1;
1515 * Does the target want to receive duplicate copies of the bio?
1517 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1518 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1520 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1521 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1523 clone_bio(tio
, bio
, sector
, *len
);
1528 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1530 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1532 return ti
->num_discard_bios
;
1535 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1537 return ti
->num_write_same_bios
;
1540 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1542 static bool is_split_required_for_discard(struct dm_target
*ti
)
1544 return ti
->split_discard_bios
;
1547 static int __send_changing_extent_only(struct clone_info
*ci
,
1548 get_num_bios_fn get_num_bios
,
1549 is_split_required_fn is_split_required
)
1551 struct dm_target
*ti
;
1556 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1557 if (!dm_target_is_valid(ti
))
1561 * Even though the device advertised support for this type of
1562 * request, that does not mean every target supports it, and
1563 * reconfiguration might also have changed that since the
1564 * check was performed.
1566 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1570 if (is_split_required
&& !is_split_required(ti
))
1571 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1573 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1575 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1578 } while (ci
->sector_count
-= len
);
1583 static int __send_discard(struct clone_info
*ci
)
1585 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1586 is_split_required_for_discard
);
1589 static int __send_write_same(struct clone_info
*ci
)
1591 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1595 * Select the correct strategy for processing a non-flush bio.
1597 static int __split_and_process_non_flush(struct clone_info
*ci
)
1599 struct bio
*bio
= ci
->bio
;
1600 struct dm_target
*ti
;
1603 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1604 return __send_discard(ci
);
1605 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1606 return __send_write_same(ci
);
1608 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1609 if (!dm_target_is_valid(ti
))
1612 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1614 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1617 ci
->sector_count
-= len
;
1623 * Entry point to split a bio into clones and submit them to the targets.
1625 static void __split_and_process_bio(struct mapped_device
*md
,
1626 struct dm_table
*map
, struct bio
*bio
)
1628 struct clone_info ci
;
1631 if (unlikely(!map
)) {
1638 ci
.io
= alloc_io(md
);
1640 atomic_set(&ci
.io
->io_count
, 1);
1643 spin_lock_init(&ci
.io
->endio_lock
);
1644 ci
.sector
= bio
->bi_iter
.bi_sector
;
1646 start_io_acct(ci
.io
);
1648 if (bio
->bi_rw
& REQ_FLUSH
) {
1649 ci
.bio
= &ci
.md
->flush_bio
;
1650 ci
.sector_count
= 0;
1651 error
= __send_empty_flush(&ci
);
1652 /* dec_pending submits any data associated with flush */
1655 ci
.sector_count
= bio_sectors(bio
);
1656 while (ci
.sector_count
&& !error
)
1657 error
= __split_and_process_non_flush(&ci
);
1660 /* drop the extra reference count */
1661 dec_pending(ci
.io
, error
);
1663 /*-----------------------------------------------------------------
1665 *---------------------------------------------------------------*/
1667 static int dm_merge_bvec(struct request_queue
*q
,
1668 struct bvec_merge_data
*bvm
,
1669 struct bio_vec
*biovec
)
1671 struct mapped_device
*md
= q
->queuedata
;
1672 struct dm_table
*map
= dm_get_live_table_fast(md
);
1673 struct dm_target
*ti
;
1674 sector_t max_sectors
;
1680 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1681 if (!dm_target_is_valid(ti
))
1685 * Find maximum amount of I/O that won't need splitting
1687 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1688 (sector_t
) queue_max_sectors(q
));
1689 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1690 if (unlikely(max_size
< 0)) /* this shouldn't _ever_ happen */
1694 * merge_bvec_fn() returns number of bytes
1695 * it can accept at this offset
1696 * max is precomputed maximal io size
1698 if (max_size
&& ti
->type
->merge
)
1699 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1701 * If the target doesn't support merge method and some of the devices
1702 * provided their merge_bvec method (we know this by looking for the
1703 * max_hw_sectors that dm_set_device_limits may set), then we can't
1704 * allow bios with multiple vector entries. So always set max_size
1705 * to 0, and the code below allows just one page.
1707 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1711 dm_put_live_table_fast(md
);
1713 * Always allow an entire first page
1715 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1716 max_size
= biovec
->bv_len
;
1722 * The request function that just remaps the bio built up by
1725 static void dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1727 int rw
= bio_data_dir(bio
);
1728 struct mapped_device
*md
= q
->queuedata
;
1730 struct dm_table
*map
;
1732 map
= dm_get_live_table(md
, &srcu_idx
);
1734 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1736 /* if we're suspended, we have to queue this io for later */
1737 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1738 dm_put_live_table(md
, srcu_idx
);
1740 if (bio_rw(bio
) != READA
)
1747 __split_and_process_bio(md
, map
, bio
);
1748 dm_put_live_table(md
, srcu_idx
);
1752 int dm_request_based(struct mapped_device
*md
)
1754 return blk_queue_stackable(md
->queue
);
1757 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1761 if (blk_queue_io_stat(clone
->q
))
1762 clone
->cmd_flags
|= REQ_IO_STAT
;
1764 clone
->start_time
= jiffies
;
1765 r
= blk_insert_cloned_request(clone
->q
, clone
);
1767 /* must complete clone in terms of original request */
1768 dm_complete_request(rq
, r
);
1771 static void setup_clone(struct request
*clone
, struct request
*rq
,
1772 struct dm_rq_target_io
*tio
)
1774 blk_rq_prep_clone(clone
, rq
);
1775 clone
->end_io
= end_clone_request
;
1776 clone
->end_io_data
= tio
;
1780 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1781 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1784 * Do not allocate a clone if tio->clone was already set
1785 * (see: dm_mq_queue_rq).
1787 bool alloc_clone
= !tio
->clone
;
1788 struct request
*clone
;
1791 clone
= alloc_clone_request(md
, gfp_mask
);
1797 blk_rq_init(NULL
, clone
);
1798 setup_clone(clone
, rq
, tio
);
1803 static void map_tio_request(struct kthread_work
*work
);
1805 static void init_tio(struct dm_rq_target_io
*tio
, struct request
*rq
,
1806 struct mapped_device
*md
)
1813 memset(&tio
->info
, 0, sizeof(tio
->info
));
1814 if (md
->kworker_task
)
1815 init_kthread_work(&tio
->work
, map_tio_request
);
1818 static struct dm_rq_target_io
*prep_tio(struct request
*rq
,
1819 struct mapped_device
*md
, gfp_t gfp_mask
)
1821 struct dm_rq_target_io
*tio
;
1823 struct dm_table
*table
;
1825 tio
= alloc_rq_tio(md
, gfp_mask
);
1829 init_tio(tio
, rq
, md
);
1831 table
= dm_get_live_table(md
, &srcu_idx
);
1832 if (!dm_table_mq_request_based(table
)) {
1833 if (!clone_rq(rq
, md
, tio
, gfp_mask
)) {
1834 dm_put_live_table(md
, srcu_idx
);
1839 dm_put_live_table(md
, srcu_idx
);
1845 * Called with the queue lock held.
1847 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1849 struct mapped_device
*md
= q
->queuedata
;
1850 struct dm_rq_target_io
*tio
;
1852 if (unlikely(rq
->special
)) {
1853 DMWARN("Already has something in rq->special.");
1854 return BLKPREP_KILL
;
1857 tio
= prep_tio(rq
, md
, GFP_ATOMIC
);
1859 return BLKPREP_DEFER
;
1862 rq
->cmd_flags
|= REQ_DONTPREP
;
1869 * 0 : the request has been processed
1870 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1871 * < 0 : the request was completed due to failure
1873 static int map_request(struct dm_rq_target_io
*tio
, struct request
*rq
,
1874 struct mapped_device
*md
)
1877 struct dm_target
*ti
= tio
->ti
;
1878 struct request
*clone
= NULL
;
1882 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1884 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
1886 /* The target wants to complete the I/O */
1887 dm_kill_unmapped_request(rq
, r
);
1891 return DM_MAPIO_REQUEUE
;
1892 setup_clone(clone
, rq
, tio
);
1896 case DM_MAPIO_SUBMITTED
:
1897 /* The target has taken the I/O to submit by itself later */
1899 case DM_MAPIO_REMAPPED
:
1900 /* The target has remapped the I/O so dispatch it */
1901 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1903 dm_dispatch_clone_request(clone
, rq
);
1905 case DM_MAPIO_REQUEUE
:
1906 /* The target wants to requeue the I/O */
1907 dm_requeue_unmapped_request(clone
);
1911 DMWARN("unimplemented target map return value: %d", r
);
1915 /* The target wants to complete the I/O */
1916 dm_kill_unmapped_request(rq
, r
);
1923 static void map_tio_request(struct kthread_work
*work
)
1925 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
1926 struct request
*rq
= tio
->orig
;
1927 struct mapped_device
*md
= tio
->md
;
1929 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
)
1930 dm_requeue_unmapped_original_request(md
, rq
);
1933 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1935 if (!orig
->q
->mq_ops
)
1936 blk_start_request(orig
);
1938 blk_mq_start_request(orig
);
1939 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
1941 if (md
->seq_rq_merge_deadline_usecs
) {
1942 md
->last_rq_pos
= rq_end_sector(orig
);
1943 md
->last_rq_rw
= rq_data_dir(orig
);
1944 md
->last_rq_start_time
= ktime_get();
1948 * Hold the md reference here for the in-flight I/O.
1949 * We can't rely on the reference count by device opener,
1950 * because the device may be closed during the request completion
1951 * when all bios are completed.
1952 * See the comment in rq_completed() too.
1957 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
1959 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device
*md
, char *buf
)
1961 return sprintf(buf
, "%u\n", md
->seq_rq_merge_deadline_usecs
);
1964 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device
*md
,
1965 const char *buf
, size_t count
)
1969 if (!dm_request_based(md
) || md
->use_blk_mq
)
1972 if (kstrtouint(buf
, 10, &deadline
))
1975 if (deadline
> MAX_SEQ_RQ_MERGE_DEADLINE_USECS
)
1976 deadline
= MAX_SEQ_RQ_MERGE_DEADLINE_USECS
;
1978 md
->seq_rq_merge_deadline_usecs
= deadline
;
1983 static bool dm_request_peeked_before_merge_deadline(struct mapped_device
*md
)
1985 ktime_t kt_deadline
;
1987 if (!md
->seq_rq_merge_deadline_usecs
)
1990 kt_deadline
= ns_to_ktime((u64
)md
->seq_rq_merge_deadline_usecs
* NSEC_PER_USEC
);
1991 kt_deadline
= ktime_add_safe(md
->last_rq_start_time
, kt_deadline
);
1993 return !ktime_after(ktime_get(), kt_deadline
);
1997 * q->request_fn for request-based dm.
1998 * Called with the queue lock held.
2000 static void dm_request_fn(struct request_queue
*q
)
2002 struct mapped_device
*md
= q
->queuedata
;
2004 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2005 struct dm_target
*ti
;
2007 struct dm_rq_target_io
*tio
;
2011 * For suspend, check blk_queue_stopped() and increment
2012 * ->pending within a single queue_lock not to increment the
2013 * number of in-flight I/Os after the queue is stopped in
2016 while (!blk_queue_stopped(q
)) {
2017 rq
= blk_peek_request(q
);
2021 /* always use block 0 to find the target for flushes for now */
2023 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2024 pos
= blk_rq_pos(rq
);
2026 ti
= dm_table_find_target(map
, pos
);
2027 if (!dm_target_is_valid(ti
)) {
2029 * Must perform setup, that rq_completed() requires,
2030 * before calling dm_kill_unmapped_request
2032 DMERR_LIMIT("request attempted access beyond the end of device");
2033 dm_start_request(md
, rq
);
2034 dm_kill_unmapped_request(rq
, -EIO
);
2038 if (dm_request_peeked_before_merge_deadline(md
) &&
2039 md_in_flight(md
) && rq
->bio
&& rq
->bio
->bi_vcnt
== 1 &&
2040 md
->last_rq_pos
== pos
&& md
->last_rq_rw
== rq_data_dir(rq
))
2043 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2046 dm_start_request(md
, rq
);
2048 tio
= tio_from_request(rq
);
2049 /* Establish tio->ti before queuing work (map_tio_request) */
2051 queue_kthread_work(&md
->kworker
, &tio
->work
);
2052 BUG_ON(!irqs_disabled());
2058 blk_delay_queue(q
, HZ
/ 100);
2060 dm_put_live_table(md
, srcu_idx
);
2063 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2066 struct mapped_device
*md
= congested_data
;
2067 struct dm_table
*map
;
2069 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2070 map
= dm_get_live_table_fast(md
);
2073 * Request-based dm cares about only own queue for
2074 * the query about congestion status of request_queue
2076 if (dm_request_based(md
))
2077 r
= md
->queue
->backing_dev_info
.state
&
2080 r
= dm_table_any_congested(map
, bdi_bits
);
2082 dm_put_live_table_fast(md
);
2088 /*-----------------------------------------------------------------
2089 * An IDR is used to keep track of allocated minor numbers.
2090 *---------------------------------------------------------------*/
2091 static void free_minor(int minor
)
2093 spin_lock(&_minor_lock
);
2094 idr_remove(&_minor_idr
, minor
);
2095 spin_unlock(&_minor_lock
);
2099 * See if the device with a specific minor # is free.
2101 static int specific_minor(int minor
)
2105 if (minor
>= (1 << MINORBITS
))
2108 idr_preload(GFP_KERNEL
);
2109 spin_lock(&_minor_lock
);
2111 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2113 spin_unlock(&_minor_lock
);
2116 return r
== -ENOSPC
? -EBUSY
: r
;
2120 static int next_free_minor(int *minor
)
2124 idr_preload(GFP_KERNEL
);
2125 spin_lock(&_minor_lock
);
2127 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2129 spin_unlock(&_minor_lock
);
2137 static const struct block_device_operations dm_blk_dops
;
2139 static void dm_wq_work(struct work_struct
*work
);
2141 static void dm_init_md_queue(struct mapped_device
*md
)
2144 * Request-based dm devices cannot be stacked on top of bio-based dm
2145 * devices. The type of this dm device may not have been decided yet.
2146 * The type is decided at the first table loading time.
2147 * To prevent problematic device stacking, clear the queue flag
2148 * for request stacking support until then.
2150 * This queue is new, so no concurrency on the queue_flags.
2152 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2155 static void dm_init_old_md_queue(struct mapped_device
*md
)
2157 md
->use_blk_mq
= false;
2158 dm_init_md_queue(md
);
2161 * Initialize aspects of queue that aren't relevant for blk-mq
2163 md
->queue
->queuedata
= md
;
2164 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2165 md
->queue
->backing_dev_info
.congested_data
= md
;
2167 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2171 * Allocate and initialise a blank device with a given minor.
2173 static struct mapped_device
*alloc_dev(int minor
)
2176 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2180 DMWARN("unable to allocate device, out of memory.");
2184 if (!try_module_get(THIS_MODULE
))
2185 goto bad_module_get
;
2187 /* get a minor number for the dev */
2188 if (minor
== DM_ANY_MINOR
)
2189 r
= next_free_minor(&minor
);
2191 r
= specific_minor(minor
);
2195 r
= init_srcu_struct(&md
->io_barrier
);
2197 goto bad_io_barrier
;
2199 md
->use_blk_mq
= use_blk_mq
;
2200 md
->type
= DM_TYPE_NONE
;
2201 mutex_init(&md
->suspend_lock
);
2202 mutex_init(&md
->type_lock
);
2203 mutex_init(&md
->table_devices_lock
);
2204 spin_lock_init(&md
->deferred_lock
);
2205 atomic_set(&md
->holders
, 1);
2206 atomic_set(&md
->open_count
, 0);
2207 atomic_set(&md
->event_nr
, 0);
2208 atomic_set(&md
->uevent_seq
, 0);
2209 INIT_LIST_HEAD(&md
->uevent_list
);
2210 INIT_LIST_HEAD(&md
->table_devices
);
2211 spin_lock_init(&md
->uevent_lock
);
2213 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2217 dm_init_md_queue(md
);
2219 md
->disk
= alloc_disk(1);
2223 atomic_set(&md
->pending
[0], 0);
2224 atomic_set(&md
->pending
[1], 0);
2225 init_waitqueue_head(&md
->wait
);
2226 INIT_WORK(&md
->work
, dm_wq_work
);
2227 init_waitqueue_head(&md
->eventq
);
2228 init_completion(&md
->kobj_holder
.completion
);
2229 md
->kworker_task
= NULL
;
2231 md
->disk
->major
= _major
;
2232 md
->disk
->first_minor
= minor
;
2233 md
->disk
->fops
= &dm_blk_dops
;
2234 md
->disk
->queue
= md
->queue
;
2235 md
->disk
->private_data
= md
;
2236 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2238 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2240 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2244 md
->bdev
= bdget_disk(md
->disk
, 0);
2248 bio_init(&md
->flush_bio
);
2249 md
->flush_bio
.bi_bdev
= md
->bdev
;
2250 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2252 dm_stats_init(&md
->stats
);
2254 /* Populate the mapping, nobody knows we exist yet */
2255 spin_lock(&_minor_lock
);
2256 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2257 spin_unlock(&_minor_lock
);
2259 BUG_ON(old_md
!= MINOR_ALLOCED
);
2264 destroy_workqueue(md
->wq
);
2266 del_gendisk(md
->disk
);
2269 blk_cleanup_queue(md
->queue
);
2271 cleanup_srcu_struct(&md
->io_barrier
);
2275 module_put(THIS_MODULE
);
2281 static void unlock_fs(struct mapped_device
*md
);
2283 static void free_dev(struct mapped_device
*md
)
2285 int minor
= MINOR(disk_devt(md
->disk
));
2288 destroy_workqueue(md
->wq
);
2290 if (md
->kworker_task
)
2291 kthread_stop(md
->kworker_task
);
2293 mempool_destroy(md
->io_pool
);
2295 mempool_destroy(md
->rq_pool
);
2297 bioset_free(md
->bs
);
2299 cleanup_srcu_struct(&md
->io_barrier
);
2300 free_table_devices(&md
->table_devices
);
2301 dm_stats_cleanup(&md
->stats
);
2303 spin_lock(&_minor_lock
);
2304 md
->disk
->private_data
= NULL
;
2305 spin_unlock(&_minor_lock
);
2306 if (blk_get_integrity(md
->disk
))
2307 blk_integrity_unregister(md
->disk
);
2308 del_gendisk(md
->disk
);
2310 blk_cleanup_queue(md
->queue
);
2312 blk_mq_free_tag_set(&md
->tag_set
);
2316 module_put(THIS_MODULE
);
2320 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2322 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2325 /* The md already has necessary mempools. */
2326 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2328 * Reload bioset because front_pad may have changed
2329 * because a different table was loaded.
2331 bioset_free(md
->bs
);
2336 * There's no need to reload with request-based dm
2337 * because the size of front_pad doesn't change.
2338 * Note for future: If you are to reload bioset,
2339 * prep-ed requests in the queue may refer
2340 * to bio from the old bioset, so you must walk
2341 * through the queue to unprep.
2346 md
->io_pool
= p
->io_pool
;
2348 md
->rq_pool
= p
->rq_pool
;
2354 /* mempool bind completed, no longer need any mempools in the table */
2355 dm_table_free_md_mempools(t
);
2359 * Bind a table to the device.
2361 static void event_callback(void *context
)
2363 unsigned long flags
;
2365 struct mapped_device
*md
= (struct mapped_device
*) context
;
2367 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2368 list_splice_init(&md
->uevent_list
, &uevents
);
2369 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2371 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2373 atomic_inc(&md
->event_nr
);
2374 wake_up(&md
->eventq
);
2378 * Protected by md->suspend_lock obtained by dm_swap_table().
2380 static void __set_size(struct mapped_device
*md
, sector_t size
)
2382 set_capacity(md
->disk
, size
);
2384 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2388 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2390 * If this function returns 0, then the device is either a non-dm
2391 * device without a merge_bvec_fn, or it is a dm device that is
2392 * able to split any bios it receives that are too big.
2394 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2396 struct mapped_device
*dev_md
;
2398 if (!q
->merge_bvec_fn
)
2401 if (q
->make_request_fn
== dm_make_request
) {
2402 dev_md
= q
->queuedata
;
2403 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2410 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2411 struct dm_dev
*dev
, sector_t start
,
2412 sector_t len
, void *data
)
2414 struct block_device
*bdev
= dev
->bdev
;
2415 struct request_queue
*q
= bdev_get_queue(bdev
);
2417 return dm_queue_merge_is_compulsory(q
);
2421 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2422 * on the properties of the underlying devices.
2424 static int dm_table_merge_is_optional(struct dm_table
*table
)
2427 struct dm_target
*ti
;
2429 while (i
< dm_table_get_num_targets(table
)) {
2430 ti
= dm_table_get_target(table
, i
++);
2432 if (ti
->type
->iterate_devices
&&
2433 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2441 * Returns old map, which caller must destroy.
2443 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2444 struct queue_limits
*limits
)
2446 struct dm_table
*old_map
;
2447 struct request_queue
*q
= md
->queue
;
2449 int merge_is_optional
;
2451 size
= dm_table_get_size(t
);
2454 * Wipe any geometry if the size of the table changed.
2456 if (size
!= dm_get_size(md
))
2457 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2459 __set_size(md
, size
);
2461 dm_table_event_callback(t
, event_callback
, md
);
2464 * The queue hasn't been stopped yet, if the old table type wasn't
2465 * for request-based during suspension. So stop it to prevent
2466 * I/O mapping before resume.
2467 * This must be done before setting the queue restrictions,
2468 * because request-based dm may be run just after the setting.
2470 if (dm_table_request_based(t
))
2473 __bind_mempools(md
, t
);
2475 merge_is_optional
= dm_table_merge_is_optional(t
);
2477 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2478 rcu_assign_pointer(md
->map
, t
);
2479 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2481 dm_table_set_restrictions(t
, q
, limits
);
2482 if (merge_is_optional
)
2483 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2485 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2493 * Returns unbound table for the caller to free.
2495 static struct dm_table
*__unbind(struct mapped_device
*md
)
2497 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2502 dm_table_event_callback(map
, NULL
, NULL
);
2503 RCU_INIT_POINTER(md
->map
, NULL
);
2510 * Constructor for a new device.
2512 int dm_create(int minor
, struct mapped_device
**result
)
2514 struct mapped_device
*md
;
2516 md
= alloc_dev(minor
);
2527 * Functions to manage md->type.
2528 * All are required to hold md->type_lock.
2530 void dm_lock_md_type(struct mapped_device
*md
)
2532 mutex_lock(&md
->type_lock
);
2535 void dm_unlock_md_type(struct mapped_device
*md
)
2537 mutex_unlock(&md
->type_lock
);
2540 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2542 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2546 unsigned dm_get_md_type(struct mapped_device
*md
)
2548 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2552 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2554 return md
->immutable_target_type
;
2558 * The queue_limits are only valid as long as you have a reference
2561 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2563 BUG_ON(!atomic_read(&md
->holders
));
2564 return &md
->queue
->limits
;
2566 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2568 static void init_rq_based_worker_thread(struct mapped_device
*md
)
2570 /* Initialize the request-based DM worker thread */
2571 init_kthread_worker(&md
->kworker
);
2572 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2573 "kdmwork-%s", dm_device_name(md
));
2577 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2579 static int dm_init_request_based_queue(struct mapped_device
*md
)
2581 struct request_queue
*q
= NULL
;
2583 /* Fully initialize the queue */
2584 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2588 /* disable dm_request_fn's merge heuristic by default */
2589 md
->seq_rq_merge_deadline_usecs
= 0;
2592 dm_init_old_md_queue(md
);
2593 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2594 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2596 init_rq_based_worker_thread(md
);
2598 elv_register_queue(md
->queue
);
2603 static int dm_mq_init_request(void *data
, struct request
*rq
,
2604 unsigned int hctx_idx
, unsigned int request_idx
,
2605 unsigned int numa_node
)
2607 struct mapped_device
*md
= data
;
2608 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2611 * Must initialize md member of tio, otherwise it won't
2612 * be available in dm_mq_queue_rq.
2619 static int dm_mq_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2620 const struct blk_mq_queue_data
*bd
)
2622 struct request
*rq
= bd
->rq
;
2623 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2624 struct mapped_device
*md
= tio
->md
;
2626 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2627 struct dm_target
*ti
;
2630 /* always use block 0 to find the target for flushes for now */
2632 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2633 pos
= blk_rq_pos(rq
);
2635 ti
= dm_table_find_target(map
, pos
);
2636 if (!dm_target_is_valid(ti
)) {
2637 dm_put_live_table(md
, srcu_idx
);
2638 DMERR_LIMIT("request attempted access beyond the end of device");
2640 * Must perform setup, that rq_completed() requires,
2641 * before returning BLK_MQ_RQ_QUEUE_ERROR
2643 dm_start_request(md
, rq
);
2644 return BLK_MQ_RQ_QUEUE_ERROR
;
2646 dm_put_live_table(md
, srcu_idx
);
2648 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2649 return BLK_MQ_RQ_QUEUE_BUSY
;
2651 dm_start_request(md
, rq
);
2653 /* Init tio using md established in .init_request */
2654 init_tio(tio
, rq
, md
);
2657 * Establish tio->ti before queuing work (map_tio_request)
2658 * or making direct call to map_request().
2662 /* Clone the request if underlying devices aren't blk-mq */
2663 if (dm_table_get_type(map
) == DM_TYPE_REQUEST_BASED
) {
2664 /* clone request is allocated at the end of the pdu */
2665 tio
->clone
= (void *)blk_mq_rq_to_pdu(rq
) + sizeof(struct dm_rq_target_io
);
2666 if (!clone_rq(rq
, md
, tio
, GFP_ATOMIC
))
2667 return BLK_MQ_RQ_QUEUE_BUSY
;
2668 queue_kthread_work(&md
->kworker
, &tio
->work
);
2670 /* Direct call is fine since .queue_rq allows allocations */
2671 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
)
2672 dm_requeue_unmapped_original_request(md
, rq
);
2675 return BLK_MQ_RQ_QUEUE_OK
;
2678 static struct blk_mq_ops dm_mq_ops
= {
2679 .queue_rq
= dm_mq_queue_rq
,
2680 .map_queue
= blk_mq_map_queue
,
2681 .complete
= dm_softirq_done
,
2682 .init_request
= dm_mq_init_request
,
2685 static int dm_init_request_based_blk_mq_queue(struct mapped_device
*md
)
2687 unsigned md_type
= dm_get_md_type(md
);
2688 struct request_queue
*q
;
2691 memset(&md
->tag_set
, 0, sizeof(md
->tag_set
));
2692 md
->tag_set
.ops
= &dm_mq_ops
;
2693 md
->tag_set
.queue_depth
= BLKDEV_MAX_RQ
;
2694 md
->tag_set
.numa_node
= NUMA_NO_NODE
;
2695 md
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2696 md
->tag_set
.nr_hw_queues
= 1;
2697 if (md_type
== DM_TYPE_REQUEST_BASED
) {
2698 /* make the memory for non-blk-mq clone part of the pdu */
2699 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
) + sizeof(struct request
);
2701 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
);
2702 md
->tag_set
.driver_data
= md
;
2704 err
= blk_mq_alloc_tag_set(&md
->tag_set
);
2708 q
= blk_mq_init_allocated_queue(&md
->tag_set
, md
->queue
);
2714 dm_init_md_queue(md
);
2716 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2717 blk_mq_register_disk(md
->disk
);
2719 if (md_type
== DM_TYPE_REQUEST_BASED
)
2720 init_rq_based_worker_thread(md
);
2725 blk_mq_free_tag_set(&md
->tag_set
);
2729 static unsigned filter_md_type(unsigned type
, struct mapped_device
*md
)
2731 if (type
== DM_TYPE_BIO_BASED
)
2734 return !md
->use_blk_mq
? DM_TYPE_REQUEST_BASED
: DM_TYPE_MQ_REQUEST_BASED
;
2738 * Setup the DM device's queue based on md's type
2740 int dm_setup_md_queue(struct mapped_device
*md
)
2743 unsigned md_type
= filter_md_type(dm_get_md_type(md
), md
);
2746 case DM_TYPE_REQUEST_BASED
:
2747 r
= dm_init_request_based_queue(md
);
2749 DMWARN("Cannot initialize queue for request-based mapped device");
2753 case DM_TYPE_MQ_REQUEST_BASED
:
2754 r
= dm_init_request_based_blk_mq_queue(md
);
2756 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2760 case DM_TYPE_BIO_BASED
:
2761 dm_init_old_md_queue(md
);
2762 blk_queue_make_request(md
->queue
, dm_make_request
);
2763 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
2770 struct mapped_device
*dm_get_md(dev_t dev
)
2772 struct mapped_device
*md
;
2773 unsigned minor
= MINOR(dev
);
2775 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2778 spin_lock(&_minor_lock
);
2780 md
= idr_find(&_minor_idr
, minor
);
2782 if ((md
== MINOR_ALLOCED
||
2783 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2784 dm_deleting_md(md
) ||
2785 test_bit(DMF_FREEING
, &md
->flags
))) {
2793 spin_unlock(&_minor_lock
);
2797 EXPORT_SYMBOL_GPL(dm_get_md
);
2799 void *dm_get_mdptr(struct mapped_device
*md
)
2801 return md
->interface_ptr
;
2804 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2806 md
->interface_ptr
= ptr
;
2809 void dm_get(struct mapped_device
*md
)
2811 atomic_inc(&md
->holders
);
2812 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2815 int dm_hold(struct mapped_device
*md
)
2817 spin_lock(&_minor_lock
);
2818 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2819 spin_unlock(&_minor_lock
);
2823 spin_unlock(&_minor_lock
);
2826 EXPORT_SYMBOL_GPL(dm_hold
);
2828 const char *dm_device_name(struct mapped_device
*md
)
2832 EXPORT_SYMBOL_GPL(dm_device_name
);
2834 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2836 struct dm_table
*map
;
2841 map
= dm_get_live_table(md
, &srcu_idx
);
2843 spin_lock(&_minor_lock
);
2844 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2845 set_bit(DMF_FREEING
, &md
->flags
);
2846 spin_unlock(&_minor_lock
);
2848 if (dm_request_based(md
) && md
->kworker_task
)
2849 flush_kthread_worker(&md
->kworker
);
2852 * Take suspend_lock so that presuspend and postsuspend methods
2853 * do not race with internal suspend.
2855 mutex_lock(&md
->suspend_lock
);
2856 if (!dm_suspended_md(md
)) {
2857 dm_table_presuspend_targets(map
);
2858 dm_table_postsuspend_targets(map
);
2860 mutex_unlock(&md
->suspend_lock
);
2862 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2863 dm_put_live_table(md
, srcu_idx
);
2866 * Rare, but there may be I/O requests still going to complete,
2867 * for example. Wait for all references to disappear.
2868 * No one should increment the reference count of the mapped_device,
2869 * after the mapped_device state becomes DMF_FREEING.
2872 while (atomic_read(&md
->holders
))
2874 else if (atomic_read(&md
->holders
))
2875 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2876 dm_device_name(md
), atomic_read(&md
->holders
));
2879 dm_table_destroy(__unbind(md
));
2883 void dm_destroy(struct mapped_device
*md
)
2885 __dm_destroy(md
, true);
2888 void dm_destroy_immediate(struct mapped_device
*md
)
2890 __dm_destroy(md
, false);
2893 void dm_put(struct mapped_device
*md
)
2895 atomic_dec(&md
->holders
);
2897 EXPORT_SYMBOL_GPL(dm_put
);
2899 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2902 DECLARE_WAITQUEUE(wait
, current
);
2904 add_wait_queue(&md
->wait
, &wait
);
2907 set_current_state(interruptible
);
2909 if (!md_in_flight(md
))
2912 if (interruptible
== TASK_INTERRUPTIBLE
&&
2913 signal_pending(current
)) {
2920 set_current_state(TASK_RUNNING
);
2922 remove_wait_queue(&md
->wait
, &wait
);
2928 * Process the deferred bios
2930 static void dm_wq_work(struct work_struct
*work
)
2932 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2936 struct dm_table
*map
;
2938 map
= dm_get_live_table(md
, &srcu_idx
);
2940 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2941 spin_lock_irq(&md
->deferred_lock
);
2942 c
= bio_list_pop(&md
->deferred
);
2943 spin_unlock_irq(&md
->deferred_lock
);
2948 if (dm_request_based(md
))
2949 generic_make_request(c
);
2951 __split_and_process_bio(md
, map
, c
);
2954 dm_put_live_table(md
, srcu_idx
);
2957 static void dm_queue_flush(struct mapped_device
*md
)
2959 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2960 smp_mb__after_atomic();
2961 queue_work(md
->wq
, &md
->work
);
2965 * Swap in a new table, returning the old one for the caller to destroy.
2967 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2969 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2970 struct queue_limits limits
;
2973 mutex_lock(&md
->suspend_lock
);
2975 /* device must be suspended */
2976 if (!dm_suspended_md(md
))
2980 * If the new table has no data devices, retain the existing limits.
2981 * This helps multipath with queue_if_no_path if all paths disappear,
2982 * then new I/O is queued based on these limits, and then some paths
2985 if (dm_table_has_no_data_devices(table
)) {
2986 live_map
= dm_get_live_table_fast(md
);
2988 limits
= md
->queue
->limits
;
2989 dm_put_live_table_fast(md
);
2993 r
= dm_calculate_queue_limits(table
, &limits
);
3000 map
= __bind(md
, table
, &limits
);
3003 mutex_unlock(&md
->suspend_lock
);
3008 * Functions to lock and unlock any filesystem running on the
3011 static int lock_fs(struct mapped_device
*md
)
3015 WARN_ON(md
->frozen_sb
);
3017 md
->frozen_sb
= freeze_bdev(md
->bdev
);
3018 if (IS_ERR(md
->frozen_sb
)) {
3019 r
= PTR_ERR(md
->frozen_sb
);
3020 md
->frozen_sb
= NULL
;
3024 set_bit(DMF_FROZEN
, &md
->flags
);
3029 static void unlock_fs(struct mapped_device
*md
)
3031 if (!test_bit(DMF_FROZEN
, &md
->flags
))
3034 thaw_bdev(md
->bdev
, md
->frozen_sb
);
3035 md
->frozen_sb
= NULL
;
3036 clear_bit(DMF_FROZEN
, &md
->flags
);
3040 * If __dm_suspend returns 0, the device is completely quiescent
3041 * now. There is no request-processing activity. All new requests
3042 * are being added to md->deferred list.
3044 * Caller must hold md->suspend_lock
3046 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
3047 unsigned suspend_flags
, int interruptible
)
3049 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
3050 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
3054 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3055 * This flag is cleared before dm_suspend returns.
3058 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3061 * This gets reverted if there's an error later and the targets
3062 * provide the .presuspend_undo hook.
3064 dm_table_presuspend_targets(map
);
3067 * Flush I/O to the device.
3068 * Any I/O submitted after lock_fs() may not be flushed.
3069 * noflush takes precedence over do_lockfs.
3070 * (lock_fs() flushes I/Os and waits for them to complete.)
3072 if (!noflush
&& do_lockfs
) {
3075 dm_table_presuspend_undo_targets(map
);
3081 * Here we must make sure that no processes are submitting requests
3082 * to target drivers i.e. no one may be executing
3083 * __split_and_process_bio. This is called from dm_request and
3086 * To get all processes out of __split_and_process_bio in dm_request,
3087 * we take the write lock. To prevent any process from reentering
3088 * __split_and_process_bio from dm_request and quiesce the thread
3089 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3090 * flush_workqueue(md->wq).
3092 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3094 synchronize_srcu(&md
->io_barrier
);
3097 * Stop md->queue before flushing md->wq in case request-based
3098 * dm defers requests to md->wq from md->queue.
3100 if (dm_request_based(md
)) {
3101 stop_queue(md
->queue
);
3102 if (md
->kworker_task
)
3103 flush_kthread_worker(&md
->kworker
);
3106 flush_workqueue(md
->wq
);
3109 * At this point no more requests are entering target request routines.
3110 * We call dm_wait_for_completion to wait for all existing requests
3113 r
= dm_wait_for_completion(md
, interruptible
);
3116 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3118 synchronize_srcu(&md
->io_barrier
);
3120 /* were we interrupted ? */
3124 if (dm_request_based(md
))
3125 start_queue(md
->queue
);
3128 dm_table_presuspend_undo_targets(map
);
3129 /* pushback list is already flushed, so skip flush */
3136 * We need to be able to change a mapping table under a mounted
3137 * filesystem. For example we might want to move some data in
3138 * the background. Before the table can be swapped with
3139 * dm_bind_table, dm_suspend must be called to flush any in
3140 * flight bios and ensure that any further io gets deferred.
3143 * Suspend mechanism in request-based dm.
3145 * 1. Flush all I/Os by lock_fs() if needed.
3146 * 2. Stop dispatching any I/O by stopping the request_queue.
3147 * 3. Wait for all in-flight I/Os to be completed or requeued.
3149 * To abort suspend, start the request_queue.
3151 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3153 struct dm_table
*map
= NULL
;
3157 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3159 if (dm_suspended_md(md
)) {
3164 if (dm_suspended_internally_md(md
)) {
3165 /* already internally suspended, wait for internal resume */
3166 mutex_unlock(&md
->suspend_lock
);
3167 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3173 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3175 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
3179 set_bit(DMF_SUSPENDED
, &md
->flags
);
3181 dm_table_postsuspend_targets(map
);
3184 mutex_unlock(&md
->suspend_lock
);
3188 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
3191 int r
= dm_table_resume_targets(map
);
3199 * Flushing deferred I/Os must be done after targets are resumed
3200 * so that mapping of targets can work correctly.
3201 * Request-based dm is queueing the deferred I/Os in its request_queue.
3203 if (dm_request_based(md
))
3204 start_queue(md
->queue
);
3211 int dm_resume(struct mapped_device
*md
)
3214 struct dm_table
*map
= NULL
;
3217 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3219 if (!dm_suspended_md(md
))
3222 if (dm_suspended_internally_md(md
)) {
3223 /* already internally suspended, wait for internal resume */
3224 mutex_unlock(&md
->suspend_lock
);
3225 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3231 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3232 if (!map
|| !dm_table_get_size(map
))
3235 r
= __dm_resume(md
, map
);
3239 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3243 mutex_unlock(&md
->suspend_lock
);
3249 * Internal suspend/resume works like userspace-driven suspend. It waits
3250 * until all bios finish and prevents issuing new bios to the target drivers.
3251 * It may be used only from the kernel.
3254 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3256 struct dm_table
*map
= NULL
;
3258 if (md
->internal_suspend_count
++)
3259 return; /* nested internal suspend */
3261 if (dm_suspended_md(md
)) {
3262 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3263 return; /* nest suspend */
3266 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3269 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3270 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3271 * would require changing .presuspend to return an error -- avoid this
3272 * until there is a need for more elaborate variants of internal suspend.
3274 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3276 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3278 dm_table_postsuspend_targets(map
);
3281 static void __dm_internal_resume(struct mapped_device
*md
)
3283 BUG_ON(!md
->internal_suspend_count
);
3285 if (--md
->internal_suspend_count
)
3286 return; /* resume from nested internal suspend */
3288 if (dm_suspended_md(md
))
3289 goto done
; /* resume from nested suspend */
3292 * NOTE: existing callers don't need to call dm_table_resume_targets
3293 * (which may fail -- so best to avoid it for now by passing NULL map)
3295 (void) __dm_resume(md
, NULL
);
3298 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3299 smp_mb__after_atomic();
3300 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3303 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3305 mutex_lock(&md
->suspend_lock
);
3306 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3307 mutex_unlock(&md
->suspend_lock
);
3309 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3311 void dm_internal_resume(struct mapped_device
*md
)
3313 mutex_lock(&md
->suspend_lock
);
3314 __dm_internal_resume(md
);
3315 mutex_unlock(&md
->suspend_lock
);
3317 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3320 * Fast variants of internal suspend/resume hold md->suspend_lock,
3321 * which prevents interaction with userspace-driven suspend.
3324 void dm_internal_suspend_fast(struct mapped_device
*md
)
3326 mutex_lock(&md
->suspend_lock
);
3327 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3330 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3331 synchronize_srcu(&md
->io_barrier
);
3332 flush_workqueue(md
->wq
);
3333 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3335 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
3337 void dm_internal_resume_fast(struct mapped_device
*md
)
3339 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3345 mutex_unlock(&md
->suspend_lock
);
3347 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
3349 /*-----------------------------------------------------------------
3350 * Event notification.
3351 *---------------------------------------------------------------*/
3352 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3355 char udev_cookie
[DM_COOKIE_LENGTH
];
3356 char *envp
[] = { udev_cookie
, NULL
};
3359 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3361 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3362 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3363 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3368 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3370 return atomic_add_return(1, &md
->uevent_seq
);
3373 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3375 return atomic_read(&md
->event_nr
);
3378 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3380 return wait_event_interruptible(md
->eventq
,
3381 (event_nr
!= atomic_read(&md
->event_nr
)));
3384 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3386 unsigned long flags
;
3388 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3389 list_add(elist
, &md
->uevent_list
);
3390 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3394 * The gendisk is only valid as long as you have a reference
3397 struct gendisk
*dm_disk(struct mapped_device
*md
)
3401 EXPORT_SYMBOL_GPL(dm_disk
);
3403 struct kobject
*dm_kobject(struct mapped_device
*md
)
3405 return &md
->kobj_holder
.kobj
;
3408 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3410 struct mapped_device
*md
;
3412 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3414 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3422 int dm_suspended_md(struct mapped_device
*md
)
3424 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3427 int dm_suspended_internally_md(struct mapped_device
*md
)
3429 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3432 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3434 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3437 int dm_suspended(struct dm_target
*ti
)
3439 return dm_suspended_md(dm_table_get_md(ti
->table
));
3441 EXPORT_SYMBOL_GPL(dm_suspended
);
3443 int dm_noflush_suspending(struct dm_target
*ti
)
3445 return __noflush_suspending(dm_table_get_md(ti
->table
));
3447 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3449 struct dm_md_mempools
*dm_alloc_bio_mempools(unsigned integrity
,
3450 unsigned per_bio_data_size
)
3452 struct dm_md_mempools
*pools
;
3453 unsigned int pool_size
= dm_get_reserved_bio_based_ios();
3454 unsigned int front_pad
;
3456 pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3460 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) +
3461 offsetof(struct dm_target_io
, clone
);
3463 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
3464 if (!pools
->io_pool
)
3467 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3471 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3476 dm_free_md_mempools(pools
);
3480 struct dm_md_mempools
*dm_alloc_rq_mempools(struct mapped_device
*md
,
3483 unsigned int pool_size
= dm_get_reserved_rq_based_ios();
3484 struct dm_md_mempools
*pools
;
3486 pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3490 if (filter_md_type(type
, md
) == DM_TYPE_REQUEST_BASED
) {
3491 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3492 if (!pools
->rq_pool
)
3496 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _rq_tio_cache
);
3497 if (!pools
->io_pool
)
3502 dm_free_md_mempools(pools
);
3506 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3512 mempool_destroy(pools
->io_pool
);
3515 mempool_destroy(pools
->rq_pool
);
3518 bioset_free(pools
->bs
);
3523 static const struct block_device_operations dm_blk_dops
= {
3524 .open
= dm_blk_open
,
3525 .release
= dm_blk_close
,
3526 .ioctl
= dm_blk_ioctl
,
3527 .getgeo
= dm_blk_getgeo
,
3528 .owner
= THIS_MODULE
3534 module_init(dm_init
);
3535 module_exit(dm_exit
);
3537 module_param(major
, uint
, 0);
3538 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3540 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3541 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3543 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3544 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3546 module_param(use_blk_mq
, bool, S_IRUGO
| S_IWUSR
);
3547 MODULE_PARM_DESC(use_blk_mq
, "Use block multiqueue for request-based DM devices");
3549 MODULE_DESCRIPTION(DM_NAME
" driver");
3550 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3551 MODULE_LICENSE("GPL");