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>
29 #include <trace/events/block.h>
31 #define DM_MSG_PREFIX "core"
35 * ratelimit state to be used in DMXXX_LIMIT().
37 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
38 DEFAULT_RATELIMIT_INTERVAL
,
39 DEFAULT_RATELIMIT_BURST
);
40 EXPORT_SYMBOL(dm_ratelimit_state
);
44 * Cookies are numeric values sent with CHANGE and REMOVE
45 * uevents while resuming, removing or renaming the device.
47 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
48 #define DM_COOKIE_LENGTH 24
50 static const char *_name
= DM_NAME
;
52 static unsigned int major
= 0;
53 static unsigned int _major
= 0;
55 static DEFINE_IDR(_minor_idr
);
57 static DEFINE_SPINLOCK(_minor_lock
);
59 static void do_deferred_remove(struct work_struct
*w
);
61 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
63 static struct workqueue_struct
*deferred_remove_workqueue
;
67 * One of these is allocated per bio.
70 struct mapped_device
*md
;
74 unsigned long start_time
;
75 spinlock_t endio_lock
;
76 struct dm_stats_aux stats_aux
;
80 * For request-based dm.
81 * One of these is allocated per request.
83 struct dm_rq_target_io
{
84 struct mapped_device
*md
;
86 struct request
*orig
, *clone
;
87 struct kthread_work work
;
90 struct dm_stats_aux stats_aux
;
91 unsigned long duration_jiffies
;
96 * For request-based dm - the bio clones we allocate are embedded in these
99 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
100 * the bioset is created - this means the bio has to come at the end of the
103 struct dm_rq_clone_bio_info
{
105 struct dm_rq_target_io
*tio
;
109 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
111 if (rq
&& rq
->end_io_data
)
112 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
115 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
117 #define MINOR_ALLOCED ((void *)-1)
120 * Bits for the md->flags field.
122 #define DMF_BLOCK_IO_FOR_SUSPEND 0
123 #define DMF_SUSPENDED 1
125 #define DMF_FREEING 3
126 #define DMF_DELETING 4
127 #define DMF_NOFLUSH_SUSPENDING 5
128 #define DMF_DEFERRED_REMOVE 6
129 #define DMF_SUSPENDED_INTERNALLY 7
132 * A dummy definition to make RCU happy.
133 * struct dm_table should never be dereferenced in this file.
140 * Work processed by per-device workqueue.
142 struct mapped_device
{
143 struct srcu_struct io_barrier
;
144 struct mutex suspend_lock
;
149 * The current mapping.
150 * Use dm_get_live_table{_fast} or take suspend_lock for
153 struct dm_table __rcu
*map
;
155 struct list_head table_devices
;
156 struct mutex table_devices_lock
;
160 struct request_queue
*queue
;
162 /* Protect queue and type against concurrent access. */
163 struct mutex type_lock
;
165 struct target_type
*immutable_target_type
;
167 struct gendisk
*disk
;
173 * A list of ios that arrived while we were suspended.
176 wait_queue_head_t wait
;
177 struct work_struct work
;
178 struct bio_list deferred
;
179 spinlock_t deferred_lock
;
182 * Processing queue (flush)
184 struct workqueue_struct
*wq
;
187 * io objects are allocated from here.
198 wait_queue_head_t eventq
;
200 struct list_head uevent_list
;
201 spinlock_t uevent_lock
; /* Protect access to uevent_list */
204 * freeze/thaw support require holding onto a super block
206 struct super_block
*frozen_sb
;
207 struct block_device
*bdev
;
209 /* forced geometry settings */
210 struct hd_geometry geometry
;
212 /* kobject and completion */
213 struct dm_kobject_holder kobj_holder
;
215 /* zero-length flush that will be cloned and submitted to targets */
216 struct bio flush_bio
;
218 /* the number of internal suspends */
219 unsigned internal_suspend_count
;
221 struct dm_stats stats
;
223 struct kthread_worker kworker
;
224 struct task_struct
*kworker_task
;
226 /* for request-based merge heuristic in dm_request_fn() */
227 unsigned seq_rq_merge_deadline_usecs
;
229 sector_t last_rq_pos
;
230 ktime_t last_rq_start_time
;
232 /* for blk-mq request-based DM support */
233 struct blk_mq_tag_set tag_set
;
237 #ifdef CONFIG_DM_MQ_DEFAULT
238 static bool use_blk_mq
= true;
240 static bool use_blk_mq
= false;
243 bool dm_use_blk_mq(struct mapped_device
*md
)
245 return md
->use_blk_mq
;
249 * For mempools pre-allocation at the table loading time.
251 struct dm_md_mempools
{
257 struct table_device
{
258 struct list_head list
;
260 struct dm_dev dm_dev
;
263 #define RESERVED_BIO_BASED_IOS 16
264 #define RESERVED_REQUEST_BASED_IOS 256
265 #define RESERVED_MAX_IOS 1024
266 static struct kmem_cache
*_io_cache
;
267 static struct kmem_cache
*_rq_tio_cache
;
268 static struct kmem_cache
*_rq_cache
;
271 * Bio-based DM's mempools' reserved IOs set by the user.
273 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
276 * Request-based DM's mempools' reserved IOs set by the user.
278 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
280 static unsigned __dm_get_module_param(unsigned *module_param
,
281 unsigned def
, unsigned max
)
283 unsigned param
= ACCESS_ONCE(*module_param
);
284 unsigned modified_param
= 0;
287 modified_param
= def
;
288 else if (param
> max
)
289 modified_param
= max
;
291 if (modified_param
) {
292 (void)cmpxchg(module_param
, param
, modified_param
);
293 param
= modified_param
;
299 unsigned dm_get_reserved_bio_based_ios(void)
301 return __dm_get_module_param(&reserved_bio_based_ios
,
302 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
304 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
306 unsigned dm_get_reserved_rq_based_ios(void)
308 return __dm_get_module_param(&reserved_rq_based_ios
,
309 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
311 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
313 static int __init
local_init(void)
317 /* allocate a slab for the dm_ios */
318 _io_cache
= KMEM_CACHE(dm_io
, 0);
322 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
324 goto out_free_io_cache
;
326 _rq_cache
= kmem_cache_create("dm_clone_request", sizeof(struct request
),
327 __alignof__(struct request
), 0, NULL
);
329 goto out_free_rq_tio_cache
;
331 r
= dm_uevent_init();
333 goto out_free_rq_cache
;
335 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
336 if (!deferred_remove_workqueue
) {
338 goto out_uevent_exit
;
342 r
= register_blkdev(_major
, _name
);
344 goto out_free_workqueue
;
352 destroy_workqueue(deferred_remove_workqueue
);
356 kmem_cache_destroy(_rq_cache
);
357 out_free_rq_tio_cache
:
358 kmem_cache_destroy(_rq_tio_cache
);
360 kmem_cache_destroy(_io_cache
);
365 static void local_exit(void)
367 flush_scheduled_work();
368 destroy_workqueue(deferred_remove_workqueue
);
370 kmem_cache_destroy(_rq_cache
);
371 kmem_cache_destroy(_rq_tio_cache
);
372 kmem_cache_destroy(_io_cache
);
373 unregister_blkdev(_major
, _name
);
378 DMINFO("cleaned up");
381 static int (*_inits
[])(void) __initdata
= {
392 static void (*_exits
[])(void) = {
403 static int __init
dm_init(void)
405 const int count
= ARRAY_SIZE(_inits
);
409 for (i
= 0; i
< count
; i
++) {
424 static void __exit
dm_exit(void)
426 int i
= ARRAY_SIZE(_exits
);
432 * Should be empty by this point.
434 idr_destroy(&_minor_idr
);
438 * Block device functions
440 int dm_deleting_md(struct mapped_device
*md
)
442 return test_bit(DMF_DELETING
, &md
->flags
);
445 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
447 struct mapped_device
*md
;
449 spin_lock(&_minor_lock
);
451 md
= bdev
->bd_disk
->private_data
;
455 if (test_bit(DMF_FREEING
, &md
->flags
) ||
456 dm_deleting_md(md
)) {
462 atomic_inc(&md
->open_count
);
464 spin_unlock(&_minor_lock
);
466 return md
? 0 : -ENXIO
;
469 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
471 struct mapped_device
*md
;
473 spin_lock(&_minor_lock
);
475 md
= disk
->private_data
;
479 if (atomic_dec_and_test(&md
->open_count
) &&
480 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
481 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
485 spin_unlock(&_minor_lock
);
488 int dm_open_count(struct mapped_device
*md
)
490 return atomic_read(&md
->open_count
);
494 * Guarantees nothing is using the device before it's deleted.
496 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
500 spin_lock(&_minor_lock
);
502 if (dm_open_count(md
)) {
505 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
506 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
509 set_bit(DMF_DELETING
, &md
->flags
);
511 spin_unlock(&_minor_lock
);
516 int dm_cancel_deferred_remove(struct mapped_device
*md
)
520 spin_lock(&_minor_lock
);
522 if (test_bit(DMF_DELETING
, &md
->flags
))
525 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
527 spin_unlock(&_minor_lock
);
532 static void do_deferred_remove(struct work_struct
*w
)
534 dm_deferred_remove();
537 sector_t
dm_get_size(struct mapped_device
*md
)
539 return get_capacity(md
->disk
);
542 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
547 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
552 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
554 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
556 return dm_get_geometry(md
, geo
);
559 static int dm_get_live_table_for_ioctl(struct mapped_device
*md
,
560 struct dm_target
**tgt
, struct block_device
**bdev
,
561 fmode_t
*mode
, int *srcu_idx
)
563 struct dm_table
*map
;
568 map
= dm_get_live_table(md
, srcu_idx
);
569 if (!map
|| !dm_table_get_size(map
))
572 /* We only support devices that have a single target */
573 if (dm_table_get_num_targets(map
) != 1)
576 *tgt
= dm_table_get_target(map
, 0);
578 if (!(*tgt
)->type
->prepare_ioctl
)
581 if (dm_suspended_md(md
)) {
586 r
= (*tgt
)->type
->prepare_ioctl(*tgt
, bdev
, mode
);
593 dm_put_live_table(md
, *srcu_idx
);
594 if (r
== -ENOTCONN
) {
601 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
602 unsigned int cmd
, unsigned long arg
)
604 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
605 struct dm_target
*tgt
;
608 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
614 * Target determined this ioctl is being issued against
615 * a logical partition of the parent bdev; so extra
616 * validation is needed.
618 r
= scsi_verify_blk_ioctl(NULL
, cmd
);
623 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
625 dm_put_live_table(md
, srcu_idx
);
629 static struct dm_io
*alloc_io(struct mapped_device
*md
)
631 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
634 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
636 mempool_free(io
, md
->io_pool
);
639 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
641 bio_put(&tio
->clone
);
644 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
647 return mempool_alloc(md
->io_pool
, gfp_mask
);
650 static void free_rq_tio(struct dm_rq_target_io
*tio
)
652 mempool_free(tio
, tio
->md
->io_pool
);
655 static struct request
*alloc_clone_request(struct mapped_device
*md
,
658 return mempool_alloc(md
->rq_pool
, gfp_mask
);
661 static void free_clone_request(struct mapped_device
*md
, struct request
*rq
)
663 mempool_free(rq
, md
->rq_pool
);
666 static int md_in_flight(struct mapped_device
*md
)
668 return atomic_read(&md
->pending
[READ
]) +
669 atomic_read(&md
->pending
[WRITE
]);
672 static void start_io_acct(struct dm_io
*io
)
674 struct mapped_device
*md
= io
->md
;
675 struct bio
*bio
= io
->bio
;
677 int rw
= bio_data_dir(bio
);
679 io
->start_time
= jiffies
;
681 cpu
= part_stat_lock();
682 part_round_stats(cpu
, &dm_disk(md
)->part0
);
684 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
685 atomic_inc_return(&md
->pending
[rw
]));
687 if (unlikely(dm_stats_used(&md
->stats
)))
688 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
689 bio_sectors(bio
), false, 0, &io
->stats_aux
);
692 static void end_io_acct(struct dm_io
*io
)
694 struct mapped_device
*md
= io
->md
;
695 struct bio
*bio
= io
->bio
;
696 unsigned long duration
= jiffies
- io
->start_time
;
698 int rw
= bio_data_dir(bio
);
700 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
702 if (unlikely(dm_stats_used(&md
->stats
)))
703 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
704 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
707 * After this is decremented the bio must not be touched if it is
710 pending
= atomic_dec_return(&md
->pending
[rw
]);
711 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
712 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
714 /* nudge anyone waiting on suspend queue */
720 * Add the bio to the list of deferred io.
722 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
726 spin_lock_irqsave(&md
->deferred_lock
, flags
);
727 bio_list_add(&md
->deferred
, bio
);
728 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
729 queue_work(md
->wq
, &md
->work
);
733 * Everyone (including functions in this file), should use this
734 * function to access the md->map field, and make sure they call
735 * dm_put_live_table() when finished.
737 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
739 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
741 return srcu_dereference(md
->map
, &md
->io_barrier
);
744 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
746 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
749 void dm_sync_table(struct mapped_device
*md
)
751 synchronize_srcu(&md
->io_barrier
);
752 synchronize_rcu_expedited();
756 * A fast alternative to dm_get_live_table/dm_put_live_table.
757 * The caller must not block between these two functions.
759 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
762 return rcu_dereference(md
->map
);
765 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
771 * Open a table device so we can use it as a map destination.
773 static int open_table_device(struct table_device
*td
, dev_t dev
,
774 struct mapped_device
*md
)
776 static char *_claim_ptr
= "I belong to device-mapper";
777 struct block_device
*bdev
;
781 BUG_ON(td
->dm_dev
.bdev
);
783 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
785 return PTR_ERR(bdev
);
787 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
789 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
793 td
->dm_dev
.bdev
= bdev
;
798 * Close a table device that we've been using.
800 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
802 if (!td
->dm_dev
.bdev
)
805 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
806 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
807 td
->dm_dev
.bdev
= NULL
;
810 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
812 struct table_device
*td
;
814 list_for_each_entry(td
, l
, list
)
815 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
821 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
822 struct dm_dev
**result
) {
824 struct table_device
*td
;
826 mutex_lock(&md
->table_devices_lock
);
827 td
= find_table_device(&md
->table_devices
, dev
, mode
);
829 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
831 mutex_unlock(&md
->table_devices_lock
);
835 td
->dm_dev
.mode
= mode
;
836 td
->dm_dev
.bdev
= NULL
;
838 if ((r
= open_table_device(td
, dev
, md
))) {
839 mutex_unlock(&md
->table_devices_lock
);
844 format_dev_t(td
->dm_dev
.name
, dev
);
846 atomic_set(&td
->count
, 0);
847 list_add(&td
->list
, &md
->table_devices
);
849 atomic_inc(&td
->count
);
850 mutex_unlock(&md
->table_devices_lock
);
852 *result
= &td
->dm_dev
;
855 EXPORT_SYMBOL_GPL(dm_get_table_device
);
857 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
859 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
861 mutex_lock(&md
->table_devices_lock
);
862 if (atomic_dec_and_test(&td
->count
)) {
863 close_table_device(td
, md
);
867 mutex_unlock(&md
->table_devices_lock
);
869 EXPORT_SYMBOL(dm_put_table_device
);
871 static void free_table_devices(struct list_head
*devices
)
873 struct list_head
*tmp
, *next
;
875 list_for_each_safe(tmp
, next
, devices
) {
876 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
878 DMWARN("dm_destroy: %s still exists with %d references",
879 td
->dm_dev
.name
, atomic_read(&td
->count
));
885 * Get the geometry associated with a dm device
887 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
895 * Set the geometry of a device.
897 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
899 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
901 if (geo
->start
> sz
) {
902 DMWARN("Start sector is beyond the geometry limits.");
911 /*-----------------------------------------------------------------
913 * A more elegant soln is in the works that uses the queue
914 * merge fn, unfortunately there are a couple of changes to
915 * the block layer that I want to make for this. So in the
916 * interests of getting something for people to use I give
917 * you this clearly demarcated crap.
918 *---------------------------------------------------------------*/
920 static int __noflush_suspending(struct mapped_device
*md
)
922 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
926 * Decrements the number of outstanding ios that a bio has been
927 * cloned into, completing the original io if necc.
929 static void dec_pending(struct dm_io
*io
, int error
)
934 struct mapped_device
*md
= io
->md
;
936 /* Push-back supersedes any I/O errors */
937 if (unlikely(error
)) {
938 spin_lock_irqsave(&io
->endio_lock
, flags
);
939 if (!(io
->error
> 0 && __noflush_suspending(md
)))
941 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
944 if (atomic_dec_and_test(&io
->io_count
)) {
945 if (io
->error
== DM_ENDIO_REQUEUE
) {
947 * Target requested pushing back the I/O.
949 spin_lock_irqsave(&md
->deferred_lock
, flags
);
950 if (__noflush_suspending(md
))
951 bio_list_add_head(&md
->deferred
, io
->bio
);
953 /* noflush suspend was interrupted. */
955 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
958 io_error
= io
->error
;
963 if (io_error
== DM_ENDIO_REQUEUE
)
966 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
968 * Preflush done for flush with data, reissue
971 bio
->bi_rw
&= ~REQ_FLUSH
;
974 /* done with normal IO or empty flush */
975 trace_block_bio_complete(md
->queue
, bio
, io_error
);
976 bio
->bi_error
= io_error
;
982 static void disable_write_same(struct mapped_device
*md
)
984 struct queue_limits
*limits
= dm_get_queue_limits(md
);
986 /* device doesn't really support WRITE SAME, disable it */
987 limits
->max_write_same_sectors
= 0;
990 static void clone_endio(struct bio
*bio
)
992 int error
= bio
->bi_error
;
994 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
995 struct dm_io
*io
= tio
->io
;
996 struct mapped_device
*md
= tio
->io
->md
;
997 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
1000 r
= endio(tio
->ti
, bio
, error
);
1001 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
1003 * error and requeue request are handled
1007 else if (r
== DM_ENDIO_INCOMPLETE
)
1008 /* The target will handle the io */
1011 DMWARN("unimplemented target endio return value: %d", r
);
1016 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
1017 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
1018 disable_write_same(md
);
1021 dec_pending(io
, error
);
1025 * Partial completion handling for request-based dm
1027 static void end_clone_bio(struct bio
*clone
)
1029 struct dm_rq_clone_bio_info
*info
=
1030 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
1031 struct dm_rq_target_io
*tio
= info
->tio
;
1032 struct bio
*bio
= info
->orig
;
1033 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
1034 int error
= clone
->bi_error
;
1040 * An error has already been detected on the request.
1041 * Once error occurred, just let clone->end_io() handle
1047 * Don't notice the error to the upper layer yet.
1048 * The error handling decision is made by the target driver,
1049 * when the request is completed.
1056 * I/O for the bio successfully completed.
1057 * Notice the data completion to the upper layer.
1061 * bios are processed from the head of the list.
1062 * So the completing bio should always be rq->bio.
1063 * If it's not, something wrong is happening.
1065 if (tio
->orig
->bio
!= bio
)
1066 DMERR("bio completion is going in the middle of the request");
1069 * Update the original request.
1070 * Do not use blk_end_request() here, because it may complete
1071 * the original request before the clone, and break the ordering.
1073 blk_update_request(tio
->orig
, 0, nr_bytes
);
1076 static struct dm_rq_target_io
*tio_from_request(struct request
*rq
)
1078 return (rq
->q
->mq_ops
? blk_mq_rq_to_pdu(rq
) : rq
->special
);
1081 static void rq_end_stats(struct mapped_device
*md
, struct request
*orig
)
1083 if (unlikely(dm_stats_used(&md
->stats
))) {
1084 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
1085 tio
->duration_jiffies
= jiffies
- tio
->duration_jiffies
;
1086 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
1087 tio
->n_sectors
, true, tio
->duration_jiffies
,
1093 * Don't touch any member of the md after calling this function because
1094 * the md may be freed in dm_put() at the end of this function.
1095 * Or do dm_get() before calling this function and dm_put() later.
1097 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1099 atomic_dec(&md
->pending
[rw
]);
1101 /* nudge anyone waiting on suspend queue */
1102 if (!md_in_flight(md
))
1106 * Run this off this callpath, as drivers could invoke end_io while
1107 * inside their request_fn (and holding the queue lock). Calling
1108 * back into ->request_fn() could deadlock attempting to grab the
1112 if (md
->queue
->mq_ops
)
1113 blk_mq_run_hw_queues(md
->queue
, true);
1115 blk_run_queue_async(md
->queue
);
1119 * dm_put() must be at the end of this function. See the comment above
1124 static void free_rq_clone(struct request
*clone
)
1126 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1127 struct mapped_device
*md
= tio
->md
;
1129 blk_rq_unprep_clone(clone
);
1131 if (md
->type
== DM_TYPE_MQ_REQUEST_BASED
)
1132 /* stacked on blk-mq queue(s) */
1133 tio
->ti
->type
->release_clone_rq(clone
);
1134 else if (!md
->queue
->mq_ops
)
1135 /* request_fn queue stacked on request_fn queue(s) */
1136 free_clone_request(md
, clone
);
1138 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1139 * no need to call free_clone_request() because we leverage blk-mq by
1140 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1143 if (!md
->queue
->mq_ops
)
1148 * Complete the clone and the original request.
1149 * Must be called without clone's queue lock held,
1150 * see end_clone_request() for more details.
1152 static void dm_end_request(struct request
*clone
, int error
)
1154 int rw
= rq_data_dir(clone
);
1155 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1156 struct mapped_device
*md
= tio
->md
;
1157 struct request
*rq
= tio
->orig
;
1159 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1160 rq
->errors
= clone
->errors
;
1161 rq
->resid_len
= clone
->resid_len
;
1165 * We are using the sense buffer of the original
1167 * So setting the length of the sense data is enough.
1169 rq
->sense_len
= clone
->sense_len
;
1172 free_rq_clone(clone
);
1173 rq_end_stats(md
, rq
);
1175 blk_end_request_all(rq
, error
);
1177 blk_mq_end_request(rq
, error
);
1178 rq_completed(md
, rw
, true);
1181 static void dm_unprep_request(struct request
*rq
)
1183 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1184 struct request
*clone
= tio
->clone
;
1186 if (!rq
->q
->mq_ops
) {
1188 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1192 free_rq_clone(clone
);
1196 * Requeue the original request of a clone.
1198 static void old_requeue_request(struct request
*rq
)
1200 struct request_queue
*q
= rq
->q
;
1201 unsigned long flags
;
1203 spin_lock_irqsave(q
->queue_lock
, flags
);
1204 blk_requeue_request(q
, rq
);
1205 blk_run_queue_async(q
);
1206 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1209 static void dm_requeue_original_request(struct mapped_device
*md
,
1212 int rw
= rq_data_dir(rq
);
1214 dm_unprep_request(rq
);
1216 rq_end_stats(md
, rq
);
1218 old_requeue_request(rq
);
1220 blk_mq_requeue_request(rq
);
1221 blk_mq_kick_requeue_list(rq
->q
);
1224 rq_completed(md
, rw
, false);
1227 static void old_stop_queue(struct request_queue
*q
)
1229 unsigned long flags
;
1231 if (blk_queue_stopped(q
))
1234 spin_lock_irqsave(q
->queue_lock
, flags
);
1236 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1239 static void stop_queue(struct request_queue
*q
)
1244 blk_mq_stop_hw_queues(q
);
1247 static void old_start_queue(struct request_queue
*q
)
1249 unsigned long flags
;
1251 spin_lock_irqsave(q
->queue_lock
, flags
);
1252 if (blk_queue_stopped(q
))
1254 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1257 static void start_queue(struct request_queue
*q
)
1262 blk_mq_start_stopped_hw_queues(q
, true);
1265 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1268 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1269 dm_request_endio_fn rq_end_io
= NULL
;
1272 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1274 if (mapped
&& rq_end_io
)
1275 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1278 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1279 !clone
->q
->limits
.max_write_same_sectors
))
1280 disable_write_same(tio
->md
);
1283 /* The target wants to complete the I/O */
1284 dm_end_request(clone
, r
);
1285 else if (r
== DM_ENDIO_INCOMPLETE
)
1286 /* The target will handle the I/O */
1288 else if (r
== DM_ENDIO_REQUEUE
)
1289 /* The target wants to requeue the I/O */
1290 dm_requeue_original_request(tio
->md
, tio
->orig
);
1292 DMWARN("unimplemented target endio return value: %d", r
);
1298 * Request completion handler for request-based dm
1300 static void dm_softirq_done(struct request
*rq
)
1303 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1304 struct request
*clone
= tio
->clone
;
1308 rq_end_stats(tio
->md
, rq
);
1309 rw
= rq_data_dir(rq
);
1310 if (!rq
->q
->mq_ops
) {
1311 blk_end_request_all(rq
, tio
->error
);
1312 rq_completed(tio
->md
, rw
, false);
1315 blk_mq_end_request(rq
, tio
->error
);
1316 rq_completed(tio
->md
, rw
, false);
1321 if (rq
->cmd_flags
& REQ_FAILED
)
1324 dm_done(clone
, tio
->error
, mapped
);
1328 * Complete the clone and the original request with the error status
1329 * through softirq context.
1331 static void dm_complete_request(struct request
*rq
, int error
)
1333 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1336 blk_complete_request(rq
);
1340 * Complete the not-mapped clone and the original request with the error status
1341 * through softirq context.
1342 * Target's rq_end_io() function isn't called.
1343 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1345 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1347 rq
->cmd_flags
|= REQ_FAILED
;
1348 dm_complete_request(rq
, error
);
1352 * Called with the clone's queue lock held (for non-blk-mq)
1354 static void end_clone_request(struct request
*clone
, int error
)
1356 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1358 if (!clone
->q
->mq_ops
) {
1360 * For just cleaning up the information of the queue in which
1361 * the clone was dispatched.
1362 * The clone is *NOT* freed actually here because it is alloced
1363 * from dm own mempool (REQ_ALLOCED isn't set).
1365 __blk_put_request(clone
->q
, clone
);
1369 * Actual request completion is done in a softirq context which doesn't
1370 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1371 * - another request may be submitted by the upper level driver
1372 * of the stacking during the completion
1373 * - the submission which requires queue lock may be done
1374 * against this clone's queue
1376 dm_complete_request(tio
->orig
, error
);
1380 * Return maximum size of I/O possible at the supplied sector up to the current
1383 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1385 sector_t target_offset
= dm_target_offset(ti
, sector
);
1387 return ti
->len
- target_offset
;
1390 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1392 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1393 sector_t offset
, max_len
;
1396 * Does the target need to split even further?
1398 if (ti
->max_io_len
) {
1399 offset
= dm_target_offset(ti
, sector
);
1400 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1401 max_len
= sector_div(offset
, ti
->max_io_len
);
1403 max_len
= offset
& (ti
->max_io_len
- 1);
1404 max_len
= ti
->max_io_len
- max_len
;
1413 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1415 if (len
> UINT_MAX
) {
1416 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1417 (unsigned long long)len
, UINT_MAX
);
1418 ti
->error
= "Maximum size of target IO is too large";
1422 ti
->max_io_len
= (uint32_t) len
;
1426 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1429 * A target may call dm_accept_partial_bio only from the map routine. It is
1430 * allowed for all bio types except REQ_FLUSH.
1432 * dm_accept_partial_bio informs the dm that the target only wants to process
1433 * additional n_sectors sectors of the bio and the rest of the data should be
1434 * sent in a next bio.
1436 * A diagram that explains the arithmetics:
1437 * +--------------------+---------------+-------+
1439 * +--------------------+---------------+-------+
1441 * <-------------- *tio->len_ptr --------------->
1442 * <------- bi_size ------->
1445 * Region 1 was already iterated over with bio_advance or similar function.
1446 * (it may be empty if the target doesn't use bio_advance)
1447 * Region 2 is the remaining bio size that the target wants to process.
1448 * (it may be empty if region 1 is non-empty, although there is no reason
1450 * The target requires that region 3 is to be sent in the next bio.
1452 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1453 * the partially processed part (the sum of regions 1+2) must be the same for all
1454 * copies of the bio.
1456 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1458 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1459 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1460 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1461 BUG_ON(bi_size
> *tio
->len_ptr
);
1462 BUG_ON(n_sectors
> bi_size
);
1463 *tio
->len_ptr
-= bi_size
- n_sectors
;
1464 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1466 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1468 static void __map_bio(struct dm_target_io
*tio
)
1472 struct mapped_device
*md
;
1473 struct bio
*clone
= &tio
->clone
;
1474 struct dm_target
*ti
= tio
->ti
;
1476 clone
->bi_end_io
= clone_endio
;
1479 * Map the clone. If r == 0 we don't need to do
1480 * anything, the target has assumed ownership of
1483 atomic_inc(&tio
->io
->io_count
);
1484 sector
= clone
->bi_iter
.bi_sector
;
1485 r
= ti
->type
->map(ti
, clone
);
1486 if (r
== DM_MAPIO_REMAPPED
) {
1487 /* the bio has been remapped so dispatch it */
1489 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1490 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1492 generic_make_request(clone
);
1493 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1494 /* error the io and bail out, or requeue it if needed */
1496 dec_pending(tio
->io
, r
);
1498 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1499 DMWARN("unimplemented target map return value: %d", r
);
1505 struct mapped_device
*md
;
1506 struct dm_table
*map
;
1510 unsigned sector_count
;
1513 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1515 bio
->bi_iter
.bi_sector
= sector
;
1516 bio
->bi_iter
.bi_size
= to_bytes(len
);
1520 * Creates a bio that consists of range of complete bvecs.
1522 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1523 sector_t sector
, unsigned len
)
1525 struct bio
*clone
= &tio
->clone
;
1527 __bio_clone_fast(clone
, bio
);
1529 if (bio_integrity(bio
))
1530 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1532 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1533 clone
->bi_iter
.bi_size
= to_bytes(len
);
1535 if (bio_integrity(bio
))
1536 bio_integrity_trim(clone
, 0, len
);
1539 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1540 struct dm_target
*ti
,
1541 unsigned target_bio_nr
)
1543 struct dm_target_io
*tio
;
1546 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1547 tio
= container_of(clone
, struct dm_target_io
, clone
);
1551 tio
->target_bio_nr
= target_bio_nr
;
1556 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1557 struct dm_target
*ti
,
1558 unsigned target_bio_nr
, unsigned *len
)
1560 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1561 struct bio
*clone
= &tio
->clone
;
1565 __bio_clone_fast(clone
, ci
->bio
);
1567 bio_setup_sector(clone
, ci
->sector
, *len
);
1572 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1573 unsigned num_bios
, unsigned *len
)
1575 unsigned target_bio_nr
;
1577 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1578 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1581 static int __send_empty_flush(struct clone_info
*ci
)
1583 unsigned target_nr
= 0;
1584 struct dm_target
*ti
;
1586 BUG_ON(bio_has_data(ci
->bio
));
1587 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1588 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1593 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1594 sector_t sector
, unsigned *len
)
1596 struct bio
*bio
= ci
->bio
;
1597 struct dm_target_io
*tio
;
1598 unsigned target_bio_nr
;
1599 unsigned num_target_bios
= 1;
1602 * Does the target want to receive duplicate copies of the bio?
1604 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1605 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1607 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1608 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1610 clone_bio(tio
, bio
, sector
, *len
);
1615 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1617 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1619 return ti
->num_discard_bios
;
1622 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1624 return ti
->num_write_same_bios
;
1627 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1629 static bool is_split_required_for_discard(struct dm_target
*ti
)
1631 return ti
->split_discard_bios
;
1634 static int __send_changing_extent_only(struct clone_info
*ci
,
1635 get_num_bios_fn get_num_bios
,
1636 is_split_required_fn is_split_required
)
1638 struct dm_target
*ti
;
1643 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1644 if (!dm_target_is_valid(ti
))
1648 * Even though the device advertised support for this type of
1649 * request, that does not mean every target supports it, and
1650 * reconfiguration might also have changed that since the
1651 * check was performed.
1653 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1657 if (is_split_required
&& !is_split_required(ti
))
1658 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1660 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1662 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1665 } while (ci
->sector_count
-= len
);
1670 static int __send_discard(struct clone_info
*ci
)
1672 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1673 is_split_required_for_discard
);
1676 static int __send_write_same(struct clone_info
*ci
)
1678 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1682 * Select the correct strategy for processing a non-flush bio.
1684 static int __split_and_process_non_flush(struct clone_info
*ci
)
1686 struct bio
*bio
= ci
->bio
;
1687 struct dm_target
*ti
;
1690 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1691 return __send_discard(ci
);
1692 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1693 return __send_write_same(ci
);
1695 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1696 if (!dm_target_is_valid(ti
))
1699 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1701 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1704 ci
->sector_count
-= len
;
1710 * Entry point to split a bio into clones and submit them to the targets.
1712 static void __split_and_process_bio(struct mapped_device
*md
,
1713 struct dm_table
*map
, struct bio
*bio
)
1715 struct clone_info ci
;
1718 if (unlikely(!map
)) {
1725 ci
.io
= alloc_io(md
);
1727 atomic_set(&ci
.io
->io_count
, 1);
1730 spin_lock_init(&ci
.io
->endio_lock
);
1731 ci
.sector
= bio
->bi_iter
.bi_sector
;
1733 start_io_acct(ci
.io
);
1735 if (bio
->bi_rw
& REQ_FLUSH
) {
1736 ci
.bio
= &ci
.md
->flush_bio
;
1737 ci
.sector_count
= 0;
1738 error
= __send_empty_flush(&ci
);
1739 /* dec_pending submits any data associated with flush */
1742 ci
.sector_count
= bio_sectors(bio
);
1743 while (ci
.sector_count
&& !error
)
1744 error
= __split_and_process_non_flush(&ci
);
1747 /* drop the extra reference count */
1748 dec_pending(ci
.io
, error
);
1750 /*-----------------------------------------------------------------
1752 *---------------------------------------------------------------*/
1755 * The request function that just remaps the bio built up by
1758 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1760 int rw
= bio_data_dir(bio
);
1761 struct mapped_device
*md
= q
->queuedata
;
1763 struct dm_table
*map
;
1765 map
= dm_get_live_table(md
, &srcu_idx
);
1767 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1769 /* if we're suspended, we have to queue this io for later */
1770 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1771 dm_put_live_table(md
, srcu_idx
);
1773 if (bio_rw(bio
) != READA
)
1777 return BLK_QC_T_NONE
;
1780 __split_and_process_bio(md
, map
, bio
);
1781 dm_put_live_table(md
, srcu_idx
);
1782 return BLK_QC_T_NONE
;
1785 int dm_request_based(struct mapped_device
*md
)
1787 return blk_queue_stackable(md
->queue
);
1790 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1794 if (blk_queue_io_stat(clone
->q
))
1795 clone
->cmd_flags
|= REQ_IO_STAT
;
1797 clone
->start_time
= jiffies
;
1798 r
= blk_insert_cloned_request(clone
->q
, clone
);
1800 /* must complete clone in terms of original request */
1801 dm_complete_request(rq
, r
);
1804 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1807 struct dm_rq_target_io
*tio
= data
;
1808 struct dm_rq_clone_bio_info
*info
=
1809 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1811 info
->orig
= bio_orig
;
1813 bio
->bi_end_io
= end_clone_bio
;
1818 static int setup_clone(struct request
*clone
, struct request
*rq
,
1819 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1823 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, gfp_mask
,
1824 dm_rq_bio_constructor
, tio
);
1828 clone
->cmd
= rq
->cmd
;
1829 clone
->cmd_len
= rq
->cmd_len
;
1830 clone
->sense
= rq
->sense
;
1831 clone
->end_io
= end_clone_request
;
1832 clone
->end_io_data
= tio
;
1839 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1840 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1843 * Do not allocate a clone if tio->clone was already set
1844 * (see: dm_mq_queue_rq).
1846 bool alloc_clone
= !tio
->clone
;
1847 struct request
*clone
;
1850 clone
= alloc_clone_request(md
, gfp_mask
);
1856 blk_rq_init(NULL
, clone
);
1857 if (setup_clone(clone
, rq
, tio
, gfp_mask
)) {
1860 free_clone_request(md
, clone
);
1867 static void map_tio_request(struct kthread_work
*work
);
1869 static void init_tio(struct dm_rq_target_io
*tio
, struct request
*rq
,
1870 struct mapped_device
*md
)
1877 memset(&tio
->info
, 0, sizeof(tio
->info
));
1878 if (md
->kworker_task
)
1879 init_kthread_work(&tio
->work
, map_tio_request
);
1882 static struct dm_rq_target_io
*prep_tio(struct request
*rq
,
1883 struct mapped_device
*md
, gfp_t gfp_mask
)
1885 struct dm_rq_target_io
*tio
;
1887 struct dm_table
*table
;
1889 tio
= alloc_rq_tio(md
, gfp_mask
);
1893 init_tio(tio
, rq
, md
);
1895 table
= dm_get_live_table(md
, &srcu_idx
);
1896 if (!dm_table_mq_request_based(table
)) {
1897 if (!clone_rq(rq
, md
, tio
, gfp_mask
)) {
1898 dm_put_live_table(md
, srcu_idx
);
1903 dm_put_live_table(md
, srcu_idx
);
1909 * Called with the queue lock held.
1911 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1913 struct mapped_device
*md
= q
->queuedata
;
1914 struct dm_rq_target_io
*tio
;
1916 if (unlikely(rq
->special
)) {
1917 DMWARN("Already has something in rq->special.");
1918 return BLKPREP_KILL
;
1921 tio
= prep_tio(rq
, md
, GFP_ATOMIC
);
1923 return BLKPREP_DEFER
;
1926 rq
->cmd_flags
|= REQ_DONTPREP
;
1933 * 0 : the request has been processed
1934 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1935 * < 0 : the request was completed due to failure
1937 static int map_request(struct dm_rq_target_io
*tio
, struct request
*rq
,
1938 struct mapped_device
*md
)
1941 struct dm_target
*ti
= tio
->ti
;
1942 struct request
*clone
= NULL
;
1946 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1948 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
1950 /* The target wants to complete the I/O */
1951 dm_kill_unmapped_request(rq
, r
);
1954 if (r
!= DM_MAPIO_REMAPPED
)
1956 if (setup_clone(clone
, rq
, tio
, GFP_ATOMIC
)) {
1958 ti
->type
->release_clone_rq(clone
);
1959 return DM_MAPIO_REQUEUE
;
1964 case DM_MAPIO_SUBMITTED
:
1965 /* The target has taken the I/O to submit by itself later */
1967 case DM_MAPIO_REMAPPED
:
1968 /* The target has remapped the I/O so dispatch it */
1969 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1971 dm_dispatch_clone_request(clone
, rq
);
1973 case DM_MAPIO_REQUEUE
:
1974 /* The target wants to requeue the I/O */
1975 dm_requeue_original_request(md
, tio
->orig
);
1979 DMWARN("unimplemented target map return value: %d", r
);
1983 /* The target wants to complete the I/O */
1984 dm_kill_unmapped_request(rq
, r
);
1991 static void map_tio_request(struct kthread_work
*work
)
1993 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
1994 struct request
*rq
= tio
->orig
;
1995 struct mapped_device
*md
= tio
->md
;
1997 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
)
1998 dm_requeue_original_request(md
, rq
);
2001 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
2003 if (!orig
->q
->mq_ops
)
2004 blk_start_request(orig
);
2006 blk_mq_start_request(orig
);
2007 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
2009 if (md
->seq_rq_merge_deadline_usecs
) {
2010 md
->last_rq_pos
= rq_end_sector(orig
);
2011 md
->last_rq_rw
= rq_data_dir(orig
);
2012 md
->last_rq_start_time
= ktime_get();
2015 if (unlikely(dm_stats_used(&md
->stats
))) {
2016 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
2017 tio
->duration_jiffies
= jiffies
;
2018 tio
->n_sectors
= blk_rq_sectors(orig
);
2019 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
2020 tio
->n_sectors
, false, 0, &tio
->stats_aux
);
2024 * Hold the md reference here for the in-flight I/O.
2025 * We can't rely on the reference count by device opener,
2026 * because the device may be closed during the request completion
2027 * when all bios are completed.
2028 * See the comment in rq_completed() too.
2033 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2035 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device
*md
, char *buf
)
2037 return sprintf(buf
, "%u\n", md
->seq_rq_merge_deadline_usecs
);
2040 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device
*md
,
2041 const char *buf
, size_t count
)
2045 if (!dm_request_based(md
) || md
->use_blk_mq
)
2048 if (kstrtouint(buf
, 10, &deadline
))
2051 if (deadline
> MAX_SEQ_RQ_MERGE_DEADLINE_USECS
)
2052 deadline
= MAX_SEQ_RQ_MERGE_DEADLINE_USECS
;
2054 md
->seq_rq_merge_deadline_usecs
= deadline
;
2059 static bool dm_request_peeked_before_merge_deadline(struct mapped_device
*md
)
2061 ktime_t kt_deadline
;
2063 if (!md
->seq_rq_merge_deadline_usecs
)
2066 kt_deadline
= ns_to_ktime((u64
)md
->seq_rq_merge_deadline_usecs
* NSEC_PER_USEC
);
2067 kt_deadline
= ktime_add_safe(md
->last_rq_start_time
, kt_deadline
);
2069 return !ktime_after(ktime_get(), kt_deadline
);
2073 * q->request_fn for request-based dm.
2074 * Called with the queue lock held.
2076 static void dm_request_fn(struct request_queue
*q
)
2078 struct mapped_device
*md
= q
->queuedata
;
2080 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2081 struct dm_target
*ti
;
2083 struct dm_rq_target_io
*tio
;
2087 * For suspend, check blk_queue_stopped() and increment
2088 * ->pending within a single queue_lock not to increment the
2089 * number of in-flight I/Os after the queue is stopped in
2092 while (!blk_queue_stopped(q
)) {
2093 rq
= blk_peek_request(q
);
2097 /* always use block 0 to find the target for flushes for now */
2099 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2100 pos
= blk_rq_pos(rq
);
2102 ti
= dm_table_find_target(map
, pos
);
2103 if (!dm_target_is_valid(ti
)) {
2105 * Must perform setup, that rq_completed() requires,
2106 * before calling dm_kill_unmapped_request
2108 DMERR_LIMIT("request attempted access beyond the end of device");
2109 dm_start_request(md
, rq
);
2110 dm_kill_unmapped_request(rq
, -EIO
);
2114 if (dm_request_peeked_before_merge_deadline(md
) &&
2115 md_in_flight(md
) && rq
->bio
&& rq
->bio
->bi_vcnt
== 1 &&
2116 md
->last_rq_pos
== pos
&& md
->last_rq_rw
== rq_data_dir(rq
))
2119 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2122 dm_start_request(md
, rq
);
2124 tio
= tio_from_request(rq
);
2125 /* Establish tio->ti before queuing work (map_tio_request) */
2127 queue_kthread_work(&md
->kworker
, &tio
->work
);
2128 BUG_ON(!irqs_disabled());
2134 blk_delay_queue(q
, HZ
/ 100);
2136 dm_put_live_table(md
, srcu_idx
);
2139 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2142 struct mapped_device
*md
= congested_data
;
2143 struct dm_table
*map
;
2145 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2146 map
= dm_get_live_table_fast(md
);
2149 * Request-based dm cares about only own queue for
2150 * the query about congestion status of request_queue
2152 if (dm_request_based(md
))
2153 r
= md
->queue
->backing_dev_info
.wb
.state
&
2156 r
= dm_table_any_congested(map
, bdi_bits
);
2158 dm_put_live_table_fast(md
);
2164 /*-----------------------------------------------------------------
2165 * An IDR is used to keep track of allocated minor numbers.
2166 *---------------------------------------------------------------*/
2167 static void free_minor(int minor
)
2169 spin_lock(&_minor_lock
);
2170 idr_remove(&_minor_idr
, minor
);
2171 spin_unlock(&_minor_lock
);
2175 * See if the device with a specific minor # is free.
2177 static int specific_minor(int minor
)
2181 if (minor
>= (1 << MINORBITS
))
2184 idr_preload(GFP_KERNEL
);
2185 spin_lock(&_minor_lock
);
2187 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2189 spin_unlock(&_minor_lock
);
2192 return r
== -ENOSPC
? -EBUSY
: r
;
2196 static int next_free_minor(int *minor
)
2200 idr_preload(GFP_KERNEL
);
2201 spin_lock(&_minor_lock
);
2203 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2205 spin_unlock(&_minor_lock
);
2213 static const struct block_device_operations dm_blk_dops
;
2215 static void dm_wq_work(struct work_struct
*work
);
2217 static void dm_init_md_queue(struct mapped_device
*md
)
2220 * Request-based dm devices cannot be stacked on top of bio-based dm
2221 * devices. The type of this dm device may not have been decided yet.
2222 * The type is decided at the first table loading time.
2223 * To prevent problematic device stacking, clear the queue flag
2224 * for request stacking support until then.
2226 * This queue is new, so no concurrency on the queue_flags.
2228 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2231 * Initialize data that will only be used by a non-blk-mq DM queue
2232 * - must do so here (in alloc_dev callchain) before queue is used
2234 md
->queue
->queuedata
= md
;
2235 md
->queue
->backing_dev_info
.congested_data
= md
;
2238 static void dm_init_old_md_queue(struct mapped_device
*md
)
2240 md
->use_blk_mq
= false;
2241 dm_init_md_queue(md
);
2244 * Initialize aspects of queue that aren't relevant for blk-mq
2246 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2247 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2250 static void cleanup_mapped_device(struct mapped_device
*md
)
2253 destroy_workqueue(md
->wq
);
2254 if (md
->kworker_task
)
2255 kthread_stop(md
->kworker_task
);
2256 mempool_destroy(md
->io_pool
);
2257 mempool_destroy(md
->rq_pool
);
2259 bioset_free(md
->bs
);
2261 cleanup_srcu_struct(&md
->io_barrier
);
2264 spin_lock(&_minor_lock
);
2265 md
->disk
->private_data
= NULL
;
2266 spin_unlock(&_minor_lock
);
2267 del_gendisk(md
->disk
);
2272 blk_cleanup_queue(md
->queue
);
2281 * Allocate and initialise a blank device with a given minor.
2283 static struct mapped_device
*alloc_dev(int minor
)
2286 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2290 DMWARN("unable to allocate device, out of memory.");
2294 if (!try_module_get(THIS_MODULE
))
2295 goto bad_module_get
;
2297 /* get a minor number for the dev */
2298 if (minor
== DM_ANY_MINOR
)
2299 r
= next_free_minor(&minor
);
2301 r
= specific_minor(minor
);
2305 r
= init_srcu_struct(&md
->io_barrier
);
2307 goto bad_io_barrier
;
2309 md
->use_blk_mq
= use_blk_mq
;
2310 md
->type
= DM_TYPE_NONE
;
2311 mutex_init(&md
->suspend_lock
);
2312 mutex_init(&md
->type_lock
);
2313 mutex_init(&md
->table_devices_lock
);
2314 spin_lock_init(&md
->deferred_lock
);
2315 atomic_set(&md
->holders
, 1);
2316 atomic_set(&md
->open_count
, 0);
2317 atomic_set(&md
->event_nr
, 0);
2318 atomic_set(&md
->uevent_seq
, 0);
2319 INIT_LIST_HEAD(&md
->uevent_list
);
2320 INIT_LIST_HEAD(&md
->table_devices
);
2321 spin_lock_init(&md
->uevent_lock
);
2323 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2327 dm_init_md_queue(md
);
2329 md
->disk
= alloc_disk(1);
2333 atomic_set(&md
->pending
[0], 0);
2334 atomic_set(&md
->pending
[1], 0);
2335 init_waitqueue_head(&md
->wait
);
2336 INIT_WORK(&md
->work
, dm_wq_work
);
2337 init_waitqueue_head(&md
->eventq
);
2338 init_completion(&md
->kobj_holder
.completion
);
2339 md
->kworker_task
= NULL
;
2341 md
->disk
->major
= _major
;
2342 md
->disk
->first_minor
= minor
;
2343 md
->disk
->fops
= &dm_blk_dops
;
2344 md
->disk
->queue
= md
->queue
;
2345 md
->disk
->private_data
= md
;
2346 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2348 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2350 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2354 md
->bdev
= bdget_disk(md
->disk
, 0);
2358 bio_init(&md
->flush_bio
);
2359 md
->flush_bio
.bi_bdev
= md
->bdev
;
2360 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2362 dm_stats_init(&md
->stats
);
2364 /* Populate the mapping, nobody knows we exist yet */
2365 spin_lock(&_minor_lock
);
2366 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2367 spin_unlock(&_minor_lock
);
2369 BUG_ON(old_md
!= MINOR_ALLOCED
);
2374 cleanup_mapped_device(md
);
2378 module_put(THIS_MODULE
);
2384 static void unlock_fs(struct mapped_device
*md
);
2386 static void free_dev(struct mapped_device
*md
)
2388 int minor
= MINOR(disk_devt(md
->disk
));
2392 cleanup_mapped_device(md
);
2394 blk_mq_free_tag_set(&md
->tag_set
);
2396 free_table_devices(&md
->table_devices
);
2397 dm_stats_cleanup(&md
->stats
);
2400 module_put(THIS_MODULE
);
2404 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2406 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2409 /* The md already has necessary mempools. */
2410 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2412 * Reload bioset because front_pad may have changed
2413 * because a different table was loaded.
2415 bioset_free(md
->bs
);
2420 * There's no need to reload with request-based dm
2421 * because the size of front_pad doesn't change.
2422 * Note for future: If you are to reload bioset,
2423 * prep-ed requests in the queue may refer
2424 * to bio from the old bioset, so you must walk
2425 * through the queue to unprep.
2430 BUG_ON(!p
|| md
->io_pool
|| md
->rq_pool
|| md
->bs
);
2432 md
->io_pool
= p
->io_pool
;
2434 md
->rq_pool
= p
->rq_pool
;
2440 /* mempool bind completed, no longer need any mempools in the table */
2441 dm_table_free_md_mempools(t
);
2445 * Bind a table to the device.
2447 static void event_callback(void *context
)
2449 unsigned long flags
;
2451 struct mapped_device
*md
= (struct mapped_device
*) context
;
2453 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2454 list_splice_init(&md
->uevent_list
, &uevents
);
2455 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2457 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2459 atomic_inc(&md
->event_nr
);
2460 wake_up(&md
->eventq
);
2464 * Protected by md->suspend_lock obtained by dm_swap_table().
2466 static void __set_size(struct mapped_device
*md
, sector_t size
)
2468 set_capacity(md
->disk
, size
);
2470 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2474 * Returns old map, which caller must destroy.
2476 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2477 struct queue_limits
*limits
)
2479 struct dm_table
*old_map
;
2480 struct request_queue
*q
= md
->queue
;
2483 size
= dm_table_get_size(t
);
2486 * Wipe any geometry if the size of the table changed.
2488 if (size
!= dm_get_size(md
))
2489 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2491 __set_size(md
, size
);
2493 dm_table_event_callback(t
, event_callback
, md
);
2496 * The queue hasn't been stopped yet, if the old table type wasn't
2497 * for request-based during suspension. So stop it to prevent
2498 * I/O mapping before resume.
2499 * This must be done before setting the queue restrictions,
2500 * because request-based dm may be run just after the setting.
2502 if (dm_table_request_based(t
))
2505 __bind_mempools(md
, t
);
2507 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2508 rcu_assign_pointer(md
->map
, t
);
2509 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2511 dm_table_set_restrictions(t
, q
, limits
);
2519 * Returns unbound table for the caller to free.
2521 static struct dm_table
*__unbind(struct mapped_device
*md
)
2523 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2528 dm_table_event_callback(map
, NULL
, NULL
);
2529 RCU_INIT_POINTER(md
->map
, NULL
);
2536 * Constructor for a new device.
2538 int dm_create(int minor
, struct mapped_device
**result
)
2540 struct mapped_device
*md
;
2542 md
= alloc_dev(minor
);
2553 * Functions to manage md->type.
2554 * All are required to hold md->type_lock.
2556 void dm_lock_md_type(struct mapped_device
*md
)
2558 mutex_lock(&md
->type_lock
);
2561 void dm_unlock_md_type(struct mapped_device
*md
)
2563 mutex_unlock(&md
->type_lock
);
2566 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2568 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2572 unsigned dm_get_md_type(struct mapped_device
*md
)
2574 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2578 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2580 return md
->immutable_target_type
;
2584 * The queue_limits are only valid as long as you have a reference
2587 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2589 BUG_ON(!atomic_read(&md
->holders
));
2590 return &md
->queue
->limits
;
2592 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2594 static void init_rq_based_worker_thread(struct mapped_device
*md
)
2596 /* Initialize the request-based DM worker thread */
2597 init_kthread_worker(&md
->kworker
);
2598 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2599 "kdmwork-%s", dm_device_name(md
));
2603 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2605 static int dm_init_request_based_queue(struct mapped_device
*md
)
2607 struct request_queue
*q
= NULL
;
2609 /* Fully initialize the queue */
2610 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2614 /* disable dm_request_fn's merge heuristic by default */
2615 md
->seq_rq_merge_deadline_usecs
= 0;
2618 dm_init_old_md_queue(md
);
2619 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2620 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2622 init_rq_based_worker_thread(md
);
2624 elv_register_queue(md
->queue
);
2629 static int dm_mq_init_request(void *data
, struct request
*rq
,
2630 unsigned int hctx_idx
, unsigned int request_idx
,
2631 unsigned int numa_node
)
2633 struct mapped_device
*md
= data
;
2634 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2637 * Must initialize md member of tio, otherwise it won't
2638 * be available in dm_mq_queue_rq.
2645 static int dm_mq_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2646 const struct blk_mq_queue_data
*bd
)
2648 struct request
*rq
= bd
->rq
;
2649 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2650 struct mapped_device
*md
= tio
->md
;
2652 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2653 struct dm_target
*ti
;
2656 /* always use block 0 to find the target for flushes for now */
2658 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2659 pos
= blk_rq_pos(rq
);
2661 ti
= dm_table_find_target(map
, pos
);
2662 if (!dm_target_is_valid(ti
)) {
2663 dm_put_live_table(md
, srcu_idx
);
2664 DMERR_LIMIT("request attempted access beyond the end of device");
2666 * Must perform setup, that rq_completed() requires,
2667 * before returning BLK_MQ_RQ_QUEUE_ERROR
2669 dm_start_request(md
, rq
);
2670 return BLK_MQ_RQ_QUEUE_ERROR
;
2672 dm_put_live_table(md
, srcu_idx
);
2674 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2675 return BLK_MQ_RQ_QUEUE_BUSY
;
2677 dm_start_request(md
, rq
);
2679 /* Init tio using md established in .init_request */
2680 init_tio(tio
, rq
, md
);
2683 * Establish tio->ti before queuing work (map_tio_request)
2684 * or making direct call to map_request().
2688 /* Clone the request if underlying devices aren't blk-mq */
2689 if (dm_table_get_type(map
) == DM_TYPE_REQUEST_BASED
) {
2690 /* clone request is allocated at the end of the pdu */
2691 tio
->clone
= (void *)blk_mq_rq_to_pdu(rq
) + sizeof(struct dm_rq_target_io
);
2692 (void) clone_rq(rq
, md
, tio
, GFP_ATOMIC
);
2693 queue_kthread_work(&md
->kworker
, &tio
->work
);
2695 /* Direct call is fine since .queue_rq allows allocations */
2696 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
) {
2697 /* Undo dm_start_request() before requeuing */
2698 rq_end_stats(md
, rq
);
2699 rq_completed(md
, rq_data_dir(rq
), false);
2700 return BLK_MQ_RQ_QUEUE_BUSY
;
2704 return BLK_MQ_RQ_QUEUE_OK
;
2707 static struct blk_mq_ops dm_mq_ops
= {
2708 .queue_rq
= dm_mq_queue_rq
,
2709 .map_queue
= blk_mq_map_queue
,
2710 .complete
= dm_softirq_done
,
2711 .init_request
= dm_mq_init_request
,
2714 static int dm_init_request_based_blk_mq_queue(struct mapped_device
*md
)
2716 unsigned md_type
= dm_get_md_type(md
);
2717 struct request_queue
*q
;
2720 memset(&md
->tag_set
, 0, sizeof(md
->tag_set
));
2721 md
->tag_set
.ops
= &dm_mq_ops
;
2722 md
->tag_set
.queue_depth
= BLKDEV_MAX_RQ
;
2723 md
->tag_set
.numa_node
= NUMA_NO_NODE
;
2724 md
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2725 md
->tag_set
.nr_hw_queues
= 1;
2726 if (md_type
== DM_TYPE_REQUEST_BASED
) {
2727 /* make the memory for non-blk-mq clone part of the pdu */
2728 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
) + sizeof(struct request
);
2730 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
);
2731 md
->tag_set
.driver_data
= md
;
2733 err
= blk_mq_alloc_tag_set(&md
->tag_set
);
2737 q
= blk_mq_init_allocated_queue(&md
->tag_set
, md
->queue
);
2743 dm_init_md_queue(md
);
2745 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2746 blk_mq_register_disk(md
->disk
);
2748 if (md_type
== DM_TYPE_REQUEST_BASED
)
2749 init_rq_based_worker_thread(md
);
2754 blk_mq_free_tag_set(&md
->tag_set
);
2758 static unsigned filter_md_type(unsigned type
, struct mapped_device
*md
)
2760 if (type
== DM_TYPE_BIO_BASED
)
2763 return !md
->use_blk_mq
? DM_TYPE_REQUEST_BASED
: DM_TYPE_MQ_REQUEST_BASED
;
2767 * Setup the DM device's queue based on md's type
2769 int dm_setup_md_queue(struct mapped_device
*md
)
2772 unsigned md_type
= filter_md_type(dm_get_md_type(md
), md
);
2775 case DM_TYPE_REQUEST_BASED
:
2776 r
= dm_init_request_based_queue(md
);
2778 DMWARN("Cannot initialize queue for request-based mapped device");
2782 case DM_TYPE_MQ_REQUEST_BASED
:
2783 r
= dm_init_request_based_blk_mq_queue(md
);
2785 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2789 case DM_TYPE_BIO_BASED
:
2790 dm_init_old_md_queue(md
);
2791 blk_queue_make_request(md
->queue
, dm_make_request
);
2793 * DM handles splitting bios as needed. Free the bio_split bioset
2794 * since it won't be used (saves 1 process per bio-based DM device).
2796 bioset_free(md
->queue
->bio_split
);
2797 md
->queue
->bio_split
= NULL
;
2804 struct mapped_device
*dm_get_md(dev_t dev
)
2806 struct mapped_device
*md
;
2807 unsigned minor
= MINOR(dev
);
2809 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2812 spin_lock(&_minor_lock
);
2814 md
= idr_find(&_minor_idr
, minor
);
2816 if ((md
== MINOR_ALLOCED
||
2817 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2818 dm_deleting_md(md
) ||
2819 test_bit(DMF_FREEING
, &md
->flags
))) {
2827 spin_unlock(&_minor_lock
);
2831 EXPORT_SYMBOL_GPL(dm_get_md
);
2833 void *dm_get_mdptr(struct mapped_device
*md
)
2835 return md
->interface_ptr
;
2838 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2840 md
->interface_ptr
= ptr
;
2843 void dm_get(struct mapped_device
*md
)
2845 atomic_inc(&md
->holders
);
2846 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2849 int dm_hold(struct mapped_device
*md
)
2851 spin_lock(&_minor_lock
);
2852 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2853 spin_unlock(&_minor_lock
);
2857 spin_unlock(&_minor_lock
);
2860 EXPORT_SYMBOL_GPL(dm_hold
);
2862 const char *dm_device_name(struct mapped_device
*md
)
2866 EXPORT_SYMBOL_GPL(dm_device_name
);
2868 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2870 struct dm_table
*map
;
2875 spin_lock(&_minor_lock
);
2876 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2877 set_bit(DMF_FREEING
, &md
->flags
);
2878 spin_unlock(&_minor_lock
);
2880 if (dm_request_based(md
) && md
->kworker_task
)
2881 flush_kthread_worker(&md
->kworker
);
2884 * Take suspend_lock so that presuspend and postsuspend methods
2885 * do not race with internal suspend.
2887 mutex_lock(&md
->suspend_lock
);
2888 map
= dm_get_live_table(md
, &srcu_idx
);
2889 if (!dm_suspended_md(md
)) {
2890 dm_table_presuspend_targets(map
);
2891 dm_table_postsuspend_targets(map
);
2893 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2894 dm_put_live_table(md
, srcu_idx
);
2895 mutex_unlock(&md
->suspend_lock
);
2898 * Rare, but there may be I/O requests still going to complete,
2899 * for example. Wait for all references to disappear.
2900 * No one should increment the reference count of the mapped_device,
2901 * after the mapped_device state becomes DMF_FREEING.
2904 while (atomic_read(&md
->holders
))
2906 else if (atomic_read(&md
->holders
))
2907 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2908 dm_device_name(md
), atomic_read(&md
->holders
));
2911 dm_table_destroy(__unbind(md
));
2915 void dm_destroy(struct mapped_device
*md
)
2917 __dm_destroy(md
, true);
2920 void dm_destroy_immediate(struct mapped_device
*md
)
2922 __dm_destroy(md
, false);
2925 void dm_put(struct mapped_device
*md
)
2927 atomic_dec(&md
->holders
);
2929 EXPORT_SYMBOL_GPL(dm_put
);
2931 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2934 DECLARE_WAITQUEUE(wait
, current
);
2936 add_wait_queue(&md
->wait
, &wait
);
2939 set_current_state(interruptible
);
2941 if (!md_in_flight(md
))
2944 if (interruptible
== TASK_INTERRUPTIBLE
&&
2945 signal_pending(current
)) {
2952 set_current_state(TASK_RUNNING
);
2954 remove_wait_queue(&md
->wait
, &wait
);
2960 * Process the deferred bios
2962 static void dm_wq_work(struct work_struct
*work
)
2964 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2968 struct dm_table
*map
;
2970 map
= dm_get_live_table(md
, &srcu_idx
);
2972 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2973 spin_lock_irq(&md
->deferred_lock
);
2974 c
= bio_list_pop(&md
->deferred
);
2975 spin_unlock_irq(&md
->deferred_lock
);
2980 if (dm_request_based(md
))
2981 generic_make_request(c
);
2983 __split_and_process_bio(md
, map
, c
);
2986 dm_put_live_table(md
, srcu_idx
);
2989 static void dm_queue_flush(struct mapped_device
*md
)
2991 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2992 smp_mb__after_atomic();
2993 queue_work(md
->wq
, &md
->work
);
2997 * Swap in a new table, returning the old one for the caller to destroy.
2999 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
3001 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
3002 struct queue_limits limits
;
3005 mutex_lock(&md
->suspend_lock
);
3007 /* device must be suspended */
3008 if (!dm_suspended_md(md
))
3012 * If the new table has no data devices, retain the existing limits.
3013 * This helps multipath with queue_if_no_path if all paths disappear,
3014 * then new I/O is queued based on these limits, and then some paths
3017 if (dm_table_has_no_data_devices(table
)) {
3018 live_map
= dm_get_live_table_fast(md
);
3020 limits
= md
->queue
->limits
;
3021 dm_put_live_table_fast(md
);
3025 r
= dm_calculate_queue_limits(table
, &limits
);
3032 map
= __bind(md
, table
, &limits
);
3035 mutex_unlock(&md
->suspend_lock
);
3040 * Functions to lock and unlock any filesystem running on the
3043 static int lock_fs(struct mapped_device
*md
)
3047 WARN_ON(md
->frozen_sb
);
3049 md
->frozen_sb
= freeze_bdev(md
->bdev
);
3050 if (IS_ERR(md
->frozen_sb
)) {
3051 r
= PTR_ERR(md
->frozen_sb
);
3052 md
->frozen_sb
= NULL
;
3056 set_bit(DMF_FROZEN
, &md
->flags
);
3061 static void unlock_fs(struct mapped_device
*md
)
3063 if (!test_bit(DMF_FROZEN
, &md
->flags
))
3066 thaw_bdev(md
->bdev
, md
->frozen_sb
);
3067 md
->frozen_sb
= NULL
;
3068 clear_bit(DMF_FROZEN
, &md
->flags
);
3072 * If __dm_suspend returns 0, the device is completely quiescent
3073 * now. There is no request-processing activity. All new requests
3074 * are being added to md->deferred list.
3076 * Caller must hold md->suspend_lock
3078 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
3079 unsigned suspend_flags
, int interruptible
)
3081 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
3082 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
3086 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3087 * This flag is cleared before dm_suspend returns.
3090 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3093 * This gets reverted if there's an error later and the targets
3094 * provide the .presuspend_undo hook.
3096 dm_table_presuspend_targets(map
);
3099 * Flush I/O to the device.
3100 * Any I/O submitted after lock_fs() may not be flushed.
3101 * noflush takes precedence over do_lockfs.
3102 * (lock_fs() flushes I/Os and waits for them to complete.)
3104 if (!noflush
&& do_lockfs
) {
3107 dm_table_presuspend_undo_targets(map
);
3113 * Here we must make sure that no processes are submitting requests
3114 * to target drivers i.e. no one may be executing
3115 * __split_and_process_bio. This is called from dm_request and
3118 * To get all processes out of __split_and_process_bio in dm_request,
3119 * we take the write lock. To prevent any process from reentering
3120 * __split_and_process_bio from dm_request and quiesce the thread
3121 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3122 * flush_workqueue(md->wq).
3124 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3126 synchronize_srcu(&md
->io_barrier
);
3129 * Stop md->queue before flushing md->wq in case request-based
3130 * dm defers requests to md->wq from md->queue.
3132 if (dm_request_based(md
)) {
3133 stop_queue(md
->queue
);
3134 if (md
->kworker_task
)
3135 flush_kthread_worker(&md
->kworker
);
3138 flush_workqueue(md
->wq
);
3141 * At this point no more requests are entering target request routines.
3142 * We call dm_wait_for_completion to wait for all existing requests
3145 r
= dm_wait_for_completion(md
, interruptible
);
3148 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3150 synchronize_srcu(&md
->io_barrier
);
3152 /* were we interrupted ? */
3156 if (dm_request_based(md
))
3157 start_queue(md
->queue
);
3160 dm_table_presuspend_undo_targets(map
);
3161 /* pushback list is already flushed, so skip flush */
3168 * We need to be able to change a mapping table under a mounted
3169 * filesystem. For example we might want to move some data in
3170 * the background. Before the table can be swapped with
3171 * dm_bind_table, dm_suspend must be called to flush any in
3172 * flight bios and ensure that any further io gets deferred.
3175 * Suspend mechanism in request-based dm.
3177 * 1. Flush all I/Os by lock_fs() if needed.
3178 * 2. Stop dispatching any I/O by stopping the request_queue.
3179 * 3. Wait for all in-flight I/Os to be completed or requeued.
3181 * To abort suspend, start the request_queue.
3183 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3185 struct dm_table
*map
= NULL
;
3189 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3191 if (dm_suspended_md(md
)) {
3196 if (dm_suspended_internally_md(md
)) {
3197 /* already internally suspended, wait for internal resume */
3198 mutex_unlock(&md
->suspend_lock
);
3199 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3205 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3207 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
3211 set_bit(DMF_SUSPENDED
, &md
->flags
);
3213 dm_table_postsuspend_targets(map
);
3216 mutex_unlock(&md
->suspend_lock
);
3220 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
3223 int r
= dm_table_resume_targets(map
);
3231 * Flushing deferred I/Os must be done after targets are resumed
3232 * so that mapping of targets can work correctly.
3233 * Request-based dm is queueing the deferred I/Os in its request_queue.
3235 if (dm_request_based(md
))
3236 start_queue(md
->queue
);
3243 int dm_resume(struct mapped_device
*md
)
3246 struct dm_table
*map
= NULL
;
3249 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3251 if (!dm_suspended_md(md
))
3254 if (dm_suspended_internally_md(md
)) {
3255 /* already internally suspended, wait for internal resume */
3256 mutex_unlock(&md
->suspend_lock
);
3257 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3263 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3264 if (!map
|| !dm_table_get_size(map
))
3267 r
= __dm_resume(md
, map
);
3271 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3275 mutex_unlock(&md
->suspend_lock
);
3281 * Internal suspend/resume works like userspace-driven suspend. It waits
3282 * until all bios finish and prevents issuing new bios to the target drivers.
3283 * It may be used only from the kernel.
3286 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3288 struct dm_table
*map
= NULL
;
3290 if (md
->internal_suspend_count
++)
3291 return; /* nested internal suspend */
3293 if (dm_suspended_md(md
)) {
3294 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3295 return; /* nest suspend */
3298 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3301 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3302 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3303 * would require changing .presuspend to return an error -- avoid this
3304 * until there is a need for more elaborate variants of internal suspend.
3306 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3308 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3310 dm_table_postsuspend_targets(map
);
3313 static void __dm_internal_resume(struct mapped_device
*md
)
3315 BUG_ON(!md
->internal_suspend_count
);
3317 if (--md
->internal_suspend_count
)
3318 return; /* resume from nested internal suspend */
3320 if (dm_suspended_md(md
))
3321 goto done
; /* resume from nested suspend */
3324 * NOTE: existing callers don't need to call dm_table_resume_targets
3325 * (which may fail -- so best to avoid it for now by passing NULL map)
3327 (void) __dm_resume(md
, NULL
);
3330 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3331 smp_mb__after_atomic();
3332 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3335 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3337 mutex_lock(&md
->suspend_lock
);
3338 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3339 mutex_unlock(&md
->suspend_lock
);
3341 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3343 void dm_internal_resume(struct mapped_device
*md
)
3345 mutex_lock(&md
->suspend_lock
);
3346 __dm_internal_resume(md
);
3347 mutex_unlock(&md
->suspend_lock
);
3349 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3352 * Fast variants of internal suspend/resume hold md->suspend_lock,
3353 * which prevents interaction with userspace-driven suspend.
3356 void dm_internal_suspend_fast(struct mapped_device
*md
)
3358 mutex_lock(&md
->suspend_lock
);
3359 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3362 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3363 synchronize_srcu(&md
->io_barrier
);
3364 flush_workqueue(md
->wq
);
3365 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3367 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
3369 void dm_internal_resume_fast(struct mapped_device
*md
)
3371 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3377 mutex_unlock(&md
->suspend_lock
);
3379 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
3381 /*-----------------------------------------------------------------
3382 * Event notification.
3383 *---------------------------------------------------------------*/
3384 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3387 char udev_cookie
[DM_COOKIE_LENGTH
];
3388 char *envp
[] = { udev_cookie
, NULL
};
3391 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3393 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3394 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3395 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3400 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3402 return atomic_add_return(1, &md
->uevent_seq
);
3405 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3407 return atomic_read(&md
->event_nr
);
3410 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3412 return wait_event_interruptible(md
->eventq
,
3413 (event_nr
!= atomic_read(&md
->event_nr
)));
3416 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3418 unsigned long flags
;
3420 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3421 list_add(elist
, &md
->uevent_list
);
3422 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3426 * The gendisk is only valid as long as you have a reference
3429 struct gendisk
*dm_disk(struct mapped_device
*md
)
3433 EXPORT_SYMBOL_GPL(dm_disk
);
3435 struct kobject
*dm_kobject(struct mapped_device
*md
)
3437 return &md
->kobj_holder
.kobj
;
3440 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3442 struct mapped_device
*md
;
3444 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3446 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3454 int dm_suspended_md(struct mapped_device
*md
)
3456 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3459 int dm_suspended_internally_md(struct mapped_device
*md
)
3461 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3464 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3466 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3469 int dm_suspended(struct dm_target
*ti
)
3471 return dm_suspended_md(dm_table_get_md(ti
->table
));
3473 EXPORT_SYMBOL_GPL(dm_suspended
);
3475 int dm_noflush_suspending(struct dm_target
*ti
)
3477 return __noflush_suspending(dm_table_get_md(ti
->table
));
3479 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3481 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
3482 unsigned integrity
, unsigned per_bio_data_size
)
3484 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3485 struct kmem_cache
*cachep
= NULL
;
3486 unsigned int pool_size
= 0;
3487 unsigned int front_pad
;
3492 type
= filter_md_type(type
, md
);
3495 case DM_TYPE_BIO_BASED
:
3497 pool_size
= dm_get_reserved_bio_based_ios();
3498 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3500 case DM_TYPE_REQUEST_BASED
:
3501 cachep
= _rq_tio_cache
;
3502 pool_size
= dm_get_reserved_rq_based_ios();
3503 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3504 if (!pools
->rq_pool
)
3506 /* fall through to setup remaining rq-based pools */
3507 case DM_TYPE_MQ_REQUEST_BASED
:
3509 pool_size
= dm_get_reserved_rq_based_ios();
3510 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3511 /* per_bio_data_size is not used. See __bind_mempools(). */
3512 WARN_ON(per_bio_data_size
!= 0);
3519 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3520 if (!pools
->io_pool
)
3524 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3528 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3534 dm_free_md_mempools(pools
);
3539 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3544 mempool_destroy(pools
->io_pool
);
3545 mempool_destroy(pools
->rq_pool
);
3548 bioset_free(pools
->bs
);
3553 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
3556 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3557 const struct pr_ops
*ops
;
3558 struct dm_target
*tgt
;
3562 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
3566 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3567 if (ops
&& ops
->pr_register
)
3568 r
= ops
->pr_register(bdev
, old_key
, new_key
, flags
);
3572 dm_put_live_table(md
, srcu_idx
);
3576 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
3579 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3580 const struct pr_ops
*ops
;
3581 struct dm_target
*tgt
;
3585 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
3589 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3590 if (ops
&& ops
->pr_reserve
)
3591 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
3595 dm_put_live_table(md
, srcu_idx
);
3599 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
3601 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3602 const struct pr_ops
*ops
;
3603 struct dm_target
*tgt
;
3607 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
3611 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3612 if (ops
&& ops
->pr_release
)
3613 r
= ops
->pr_release(bdev
, key
, type
);
3617 dm_put_live_table(md
, srcu_idx
);
3621 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
3622 enum pr_type type
, bool abort
)
3624 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3625 const struct pr_ops
*ops
;
3626 struct dm_target
*tgt
;
3630 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
3634 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3635 if (ops
&& ops
->pr_preempt
)
3636 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
3640 dm_put_live_table(md
, srcu_idx
);
3644 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
3646 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3647 const struct pr_ops
*ops
;
3648 struct dm_target
*tgt
;
3652 r
= dm_get_live_table_for_ioctl(md
, &tgt
, &bdev
, &mode
, &srcu_idx
);
3656 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3657 if (ops
&& ops
->pr_clear
)
3658 r
= ops
->pr_clear(bdev
, key
);
3662 dm_put_live_table(md
, srcu_idx
);
3666 static const struct pr_ops dm_pr_ops
= {
3667 .pr_register
= dm_pr_register
,
3668 .pr_reserve
= dm_pr_reserve
,
3669 .pr_release
= dm_pr_release
,
3670 .pr_preempt
= dm_pr_preempt
,
3671 .pr_clear
= dm_pr_clear
,
3674 static const struct block_device_operations dm_blk_dops
= {
3675 .open
= dm_blk_open
,
3676 .release
= dm_blk_close
,
3677 .ioctl
= dm_blk_ioctl
,
3678 .getgeo
= dm_blk_getgeo
,
3679 .pr_ops
= &dm_pr_ops
,
3680 .owner
= THIS_MODULE
3686 module_init(dm_init
);
3687 module_exit(dm_exit
);
3689 module_param(major
, uint
, 0);
3690 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3692 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3693 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3695 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3696 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3698 module_param(use_blk_mq
, bool, S_IRUGO
| S_IWUSR
);
3699 MODULE_PARM_DESC(use_blk_mq
, "Use block multiqueue for request-based DM devices");
3701 MODULE_DESCRIPTION(DM_NAME
" driver");
3702 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3703 MODULE_LICENSE("GPL");