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
;
89 struct dm_stats_aux stats_aux
;
90 unsigned long duration_jiffies
;
95 * For request-based dm - the bio clones we allocate are embedded in these
98 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
99 * the bioset is created - this means the bio has to come at the end of the
102 struct dm_rq_clone_bio_info
{
104 struct dm_rq_target_io
*tio
;
108 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
110 if (rq
&& rq
->end_io_data
)
111 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
114 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
116 #define MINOR_ALLOCED ((void *)-1)
119 * Bits for the md->flags field.
121 #define DMF_BLOCK_IO_FOR_SUSPEND 0
122 #define DMF_SUSPENDED 1
124 #define DMF_FREEING 3
125 #define DMF_DELETING 4
126 #define DMF_NOFLUSH_SUSPENDING 5
127 #define DMF_MERGE_IS_OPTIONAL 6
128 #define DMF_DEFERRED_REMOVE 7
129 #define DMF_SUSPENDED_INTERNALLY 8
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_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
560 unsigned int cmd
, unsigned long arg
)
562 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
564 struct dm_table
*map
;
565 struct dm_target
*tgt
;
569 map
= dm_get_live_table(md
, &srcu_idx
);
571 if (!map
|| !dm_table_get_size(map
))
574 /* We only support devices that have a single target */
575 if (dm_table_get_num_targets(map
) != 1)
578 tgt
= dm_table_get_target(map
, 0);
579 if (!tgt
->type
->ioctl
)
582 if (dm_suspended_md(md
)) {
587 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
590 dm_put_live_table(md
, srcu_idx
);
592 if (r
== -ENOTCONN
) {
600 static struct dm_io
*alloc_io(struct mapped_device
*md
)
602 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
605 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
607 mempool_free(io
, md
->io_pool
);
610 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
612 bio_put(&tio
->clone
);
615 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
618 return mempool_alloc(md
->io_pool
, gfp_mask
);
621 static void free_rq_tio(struct dm_rq_target_io
*tio
)
623 mempool_free(tio
, tio
->md
->io_pool
);
626 static struct request
*alloc_clone_request(struct mapped_device
*md
,
629 return mempool_alloc(md
->rq_pool
, gfp_mask
);
632 static void free_clone_request(struct mapped_device
*md
, struct request
*rq
)
634 mempool_free(rq
, md
->rq_pool
);
637 static int md_in_flight(struct mapped_device
*md
)
639 return atomic_read(&md
->pending
[READ
]) +
640 atomic_read(&md
->pending
[WRITE
]);
643 static void start_io_acct(struct dm_io
*io
)
645 struct mapped_device
*md
= io
->md
;
646 struct bio
*bio
= io
->bio
;
648 int rw
= bio_data_dir(bio
);
650 io
->start_time
= jiffies
;
652 cpu
= part_stat_lock();
653 part_round_stats(cpu
, &dm_disk(md
)->part0
);
655 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
656 atomic_inc_return(&md
->pending
[rw
]));
658 if (unlikely(dm_stats_used(&md
->stats
)))
659 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
660 bio_sectors(bio
), false, 0, &io
->stats_aux
);
663 static void end_io_acct(struct dm_io
*io
)
665 struct mapped_device
*md
= io
->md
;
666 struct bio
*bio
= io
->bio
;
667 unsigned long duration
= jiffies
- io
->start_time
;
669 int rw
= bio_data_dir(bio
);
671 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
673 if (unlikely(dm_stats_used(&md
->stats
)))
674 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
675 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
678 * After this is decremented the bio must not be touched if it is
681 pending
= atomic_dec_return(&md
->pending
[rw
]);
682 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
683 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
685 /* nudge anyone waiting on suspend queue */
691 * Add the bio to the list of deferred io.
693 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
697 spin_lock_irqsave(&md
->deferred_lock
, flags
);
698 bio_list_add(&md
->deferred
, bio
);
699 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
700 queue_work(md
->wq
, &md
->work
);
704 * Everyone (including functions in this file), should use this
705 * function to access the md->map field, and make sure they call
706 * dm_put_live_table() when finished.
708 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
710 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
712 return srcu_dereference(md
->map
, &md
->io_barrier
);
715 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
717 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
720 void dm_sync_table(struct mapped_device
*md
)
722 synchronize_srcu(&md
->io_barrier
);
723 synchronize_rcu_expedited();
727 * A fast alternative to dm_get_live_table/dm_put_live_table.
728 * The caller must not block between these two functions.
730 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
733 return rcu_dereference(md
->map
);
736 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
742 * Open a table device so we can use it as a map destination.
744 static int open_table_device(struct table_device
*td
, dev_t dev
,
745 struct mapped_device
*md
)
747 static char *_claim_ptr
= "I belong to device-mapper";
748 struct block_device
*bdev
;
752 BUG_ON(td
->dm_dev
.bdev
);
754 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
756 return PTR_ERR(bdev
);
758 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
760 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
764 td
->dm_dev
.bdev
= bdev
;
769 * Close a table device that we've been using.
771 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
773 if (!td
->dm_dev
.bdev
)
776 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
777 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
778 td
->dm_dev
.bdev
= NULL
;
781 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
783 struct table_device
*td
;
785 list_for_each_entry(td
, l
, list
)
786 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
792 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
793 struct dm_dev
**result
) {
795 struct table_device
*td
;
797 mutex_lock(&md
->table_devices_lock
);
798 td
= find_table_device(&md
->table_devices
, dev
, mode
);
800 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
802 mutex_unlock(&md
->table_devices_lock
);
806 td
->dm_dev
.mode
= mode
;
807 td
->dm_dev
.bdev
= NULL
;
809 if ((r
= open_table_device(td
, dev
, md
))) {
810 mutex_unlock(&md
->table_devices_lock
);
815 format_dev_t(td
->dm_dev
.name
, dev
);
817 atomic_set(&td
->count
, 0);
818 list_add(&td
->list
, &md
->table_devices
);
820 atomic_inc(&td
->count
);
821 mutex_unlock(&md
->table_devices_lock
);
823 *result
= &td
->dm_dev
;
826 EXPORT_SYMBOL_GPL(dm_get_table_device
);
828 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
830 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
832 mutex_lock(&md
->table_devices_lock
);
833 if (atomic_dec_and_test(&td
->count
)) {
834 close_table_device(td
, md
);
838 mutex_unlock(&md
->table_devices_lock
);
840 EXPORT_SYMBOL(dm_put_table_device
);
842 static void free_table_devices(struct list_head
*devices
)
844 struct list_head
*tmp
, *next
;
846 list_for_each_safe(tmp
, next
, devices
) {
847 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
849 DMWARN("dm_destroy: %s still exists with %d references",
850 td
->dm_dev
.name
, atomic_read(&td
->count
));
856 * Get the geometry associated with a dm device
858 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
866 * Set the geometry of a device.
868 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
870 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
872 if (geo
->start
> sz
) {
873 DMWARN("Start sector is beyond the geometry limits.");
882 /*-----------------------------------------------------------------
884 * A more elegant soln is in the works that uses the queue
885 * merge fn, unfortunately there are a couple of changes to
886 * the block layer that I want to make for this. So in the
887 * interests of getting something for people to use I give
888 * you this clearly demarcated crap.
889 *---------------------------------------------------------------*/
891 static int __noflush_suspending(struct mapped_device
*md
)
893 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
897 * Decrements the number of outstanding ios that a bio has been
898 * cloned into, completing the original io if necc.
900 static void dec_pending(struct dm_io
*io
, int error
)
905 struct mapped_device
*md
= io
->md
;
907 /* Push-back supersedes any I/O errors */
908 if (unlikely(error
)) {
909 spin_lock_irqsave(&io
->endio_lock
, flags
);
910 if (!(io
->error
> 0 && __noflush_suspending(md
)))
912 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
915 if (atomic_dec_and_test(&io
->io_count
)) {
916 if (io
->error
== DM_ENDIO_REQUEUE
) {
918 * Target requested pushing back the I/O.
920 spin_lock_irqsave(&md
->deferred_lock
, flags
);
921 if (__noflush_suspending(md
))
922 bio_list_add_head(&md
->deferred
, io
->bio
);
924 /* noflush suspend was interrupted. */
926 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
929 io_error
= io
->error
;
934 if (io_error
== DM_ENDIO_REQUEUE
)
937 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
939 * Preflush done for flush with data, reissue
942 bio
->bi_rw
&= ~REQ_FLUSH
;
945 /* done with normal IO or empty flush */
946 trace_block_bio_complete(md
->queue
, bio
, io_error
);
947 bio_endio(bio
, io_error
);
952 static void disable_write_same(struct mapped_device
*md
)
954 struct queue_limits
*limits
= dm_get_queue_limits(md
);
956 /* device doesn't really support WRITE SAME, disable it */
957 limits
->max_write_same_sectors
= 0;
960 static void clone_endio(struct bio
*bio
, int error
)
963 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
964 struct dm_io
*io
= tio
->io
;
965 struct mapped_device
*md
= tio
->io
->md
;
966 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
968 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
972 r
= endio(tio
->ti
, bio
, error
);
973 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
975 * error and requeue request are handled
979 else if (r
== DM_ENDIO_INCOMPLETE
)
980 /* The target will handle the io */
983 DMWARN("unimplemented target endio return value: %d", r
);
988 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
989 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
990 disable_write_same(md
);
993 dec_pending(io
, error
);
997 * Partial completion handling for request-based dm
999 static void end_clone_bio(struct bio
*clone
, int error
)
1001 struct dm_rq_clone_bio_info
*info
=
1002 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
1003 struct dm_rq_target_io
*tio
= info
->tio
;
1004 struct bio
*bio
= info
->orig
;
1005 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
1011 * An error has already been detected on the request.
1012 * Once error occurred, just let clone->end_io() handle
1018 * Don't notice the error to the upper layer yet.
1019 * The error handling decision is made by the target driver,
1020 * when the request is completed.
1027 * I/O for the bio successfully completed.
1028 * Notice the data completion to the upper layer.
1032 * bios are processed from the head of the list.
1033 * So the completing bio should always be rq->bio.
1034 * If it's not, something wrong is happening.
1036 if (tio
->orig
->bio
!= bio
)
1037 DMERR("bio completion is going in the middle of the request");
1040 * Update the original request.
1041 * Do not use blk_end_request() here, because it may complete
1042 * the original request before the clone, and break the ordering.
1044 blk_update_request(tio
->orig
, 0, nr_bytes
);
1047 static struct dm_rq_target_io
*tio_from_request(struct request
*rq
)
1049 return (rq
->q
->mq_ops
? blk_mq_rq_to_pdu(rq
) : rq
->special
);
1052 static void rq_end_stats(struct mapped_device
*md
, struct request
*orig
)
1054 if (unlikely(dm_stats_used(&md
->stats
))) {
1055 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
1056 tio
->duration_jiffies
= jiffies
- tio
->duration_jiffies
;
1057 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
1058 tio
->n_sectors
, true, tio
->duration_jiffies
,
1064 * Don't touch any member of the md after calling this function because
1065 * the md may be freed in dm_put() at the end of this function.
1066 * Or do dm_get() before calling this function and dm_put() later.
1068 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1070 atomic_dec(&md
->pending
[rw
]);
1072 /* nudge anyone waiting on suspend queue */
1073 if (!md_in_flight(md
))
1077 * Run this off this callpath, as drivers could invoke end_io while
1078 * inside their request_fn (and holding the queue lock). Calling
1079 * back into ->request_fn() could deadlock attempting to grab the
1083 if (md
->queue
->mq_ops
)
1084 blk_mq_run_hw_queues(md
->queue
, true);
1086 blk_run_queue_async(md
->queue
);
1090 * dm_put() must be at the end of this function. See the comment above
1095 static void free_rq_clone(struct request
*clone
)
1097 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1098 struct mapped_device
*md
= tio
->md
;
1100 blk_rq_unprep_clone(clone
);
1102 if (md
->type
== DM_TYPE_MQ_REQUEST_BASED
)
1103 /* stacked on blk-mq queue(s) */
1104 tio
->ti
->type
->release_clone_rq(clone
);
1105 else if (!md
->queue
->mq_ops
)
1106 /* request_fn queue stacked on request_fn queue(s) */
1107 free_clone_request(md
, clone
);
1109 * NOTE: for the blk-mq queue stacked on request_fn queue(s) case:
1110 * no need to call free_clone_request() because we leverage blk-mq by
1111 * allocating the clone at the end of the blk-mq pdu (see: clone_rq)
1114 if (!md
->queue
->mq_ops
)
1119 * Complete the clone and the original request.
1120 * Must be called without clone's queue lock held,
1121 * see end_clone_request() for more details.
1123 static void dm_end_request(struct request
*clone
, int error
)
1125 int rw
= rq_data_dir(clone
);
1126 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1127 struct mapped_device
*md
= tio
->md
;
1128 struct request
*rq
= tio
->orig
;
1130 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1131 rq
->errors
= clone
->errors
;
1132 rq
->resid_len
= clone
->resid_len
;
1136 * We are using the sense buffer of the original
1138 * So setting the length of the sense data is enough.
1140 rq
->sense_len
= clone
->sense_len
;
1143 free_rq_clone(clone
);
1144 rq_end_stats(md
, rq
);
1146 blk_end_request_all(rq
, error
);
1148 blk_mq_end_request(rq
, error
);
1149 rq_completed(md
, rw
, true);
1152 static void dm_unprep_request(struct request
*rq
)
1154 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1155 struct request
*clone
= tio
->clone
;
1157 if (!rq
->q
->mq_ops
) {
1159 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1163 free_rq_clone(clone
);
1167 * Requeue the original request of a clone.
1169 static void old_requeue_request(struct request
*rq
)
1171 struct request_queue
*q
= rq
->q
;
1172 unsigned long flags
;
1174 spin_lock_irqsave(q
->queue_lock
, flags
);
1175 blk_requeue_request(q
, rq
);
1176 blk_run_queue_async(q
);
1177 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1180 static void dm_requeue_original_request(struct mapped_device
*md
,
1183 int rw
= rq_data_dir(rq
);
1185 dm_unprep_request(rq
);
1187 rq_end_stats(md
, rq
);
1189 old_requeue_request(rq
);
1191 blk_mq_requeue_request(rq
);
1192 blk_mq_kick_requeue_list(rq
->q
);
1195 rq_completed(md
, rw
, false);
1198 static void old_stop_queue(struct request_queue
*q
)
1200 unsigned long flags
;
1202 if (blk_queue_stopped(q
))
1205 spin_lock_irqsave(q
->queue_lock
, flags
);
1207 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1210 static void stop_queue(struct request_queue
*q
)
1215 blk_mq_stop_hw_queues(q
);
1218 static void old_start_queue(struct request_queue
*q
)
1220 unsigned long flags
;
1222 spin_lock_irqsave(q
->queue_lock
, flags
);
1223 if (blk_queue_stopped(q
))
1225 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1228 static void start_queue(struct request_queue
*q
)
1233 blk_mq_start_stopped_hw_queues(q
, true);
1236 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1239 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1240 dm_request_endio_fn rq_end_io
= NULL
;
1243 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1245 if (mapped
&& rq_end_io
)
1246 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1249 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1250 !clone
->q
->limits
.max_write_same_sectors
))
1251 disable_write_same(tio
->md
);
1254 /* The target wants to complete the I/O */
1255 dm_end_request(clone
, r
);
1256 else if (r
== DM_ENDIO_INCOMPLETE
)
1257 /* The target will handle the I/O */
1259 else if (r
== DM_ENDIO_REQUEUE
)
1260 /* The target wants to requeue the I/O */
1261 dm_requeue_original_request(tio
->md
, tio
->orig
);
1263 DMWARN("unimplemented target endio return value: %d", r
);
1269 * Request completion handler for request-based dm
1271 static void dm_softirq_done(struct request
*rq
)
1274 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1275 struct request
*clone
= tio
->clone
;
1279 rq_end_stats(tio
->md
, rq
);
1280 rw
= rq_data_dir(rq
);
1281 if (!rq
->q
->mq_ops
) {
1282 blk_end_request_all(rq
, tio
->error
);
1283 rq_completed(tio
->md
, rw
, false);
1286 blk_mq_end_request(rq
, tio
->error
);
1287 rq_completed(tio
->md
, rw
, false);
1292 if (rq
->cmd_flags
& REQ_FAILED
)
1295 dm_done(clone
, tio
->error
, mapped
);
1299 * Complete the clone and the original request with the error status
1300 * through softirq context.
1302 static void dm_complete_request(struct request
*rq
, int error
)
1304 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1307 blk_complete_request(rq
);
1311 * Complete the not-mapped clone and the original request with the error status
1312 * through softirq context.
1313 * Target's rq_end_io() function isn't called.
1314 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1316 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1318 rq
->cmd_flags
|= REQ_FAILED
;
1319 dm_complete_request(rq
, error
);
1323 * Called with the clone's queue lock held (for non-blk-mq)
1325 static void end_clone_request(struct request
*clone
, int error
)
1327 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1329 if (!clone
->q
->mq_ops
) {
1331 * For just cleaning up the information of the queue in which
1332 * the clone was dispatched.
1333 * The clone is *NOT* freed actually here because it is alloced
1334 * from dm own mempool (REQ_ALLOCED isn't set).
1336 __blk_put_request(clone
->q
, clone
);
1340 * Actual request completion is done in a softirq context which doesn't
1341 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1342 * - another request may be submitted by the upper level driver
1343 * of the stacking during the completion
1344 * - the submission which requires queue lock may be done
1345 * against this clone's queue
1347 dm_complete_request(tio
->orig
, error
);
1351 * Return maximum size of I/O possible at the supplied sector up to the current
1354 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1356 sector_t target_offset
= dm_target_offset(ti
, sector
);
1358 return ti
->len
- target_offset
;
1361 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1363 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1364 sector_t offset
, max_len
;
1367 * Does the target need to split even further?
1369 if (ti
->max_io_len
) {
1370 offset
= dm_target_offset(ti
, sector
);
1371 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1372 max_len
= sector_div(offset
, ti
->max_io_len
);
1374 max_len
= offset
& (ti
->max_io_len
- 1);
1375 max_len
= ti
->max_io_len
- max_len
;
1384 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1386 if (len
> UINT_MAX
) {
1387 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1388 (unsigned long long)len
, UINT_MAX
);
1389 ti
->error
= "Maximum size of target IO is too large";
1393 ti
->max_io_len
= (uint32_t) len
;
1397 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1400 * A target may call dm_accept_partial_bio only from the map routine. It is
1401 * allowed for all bio types except REQ_FLUSH.
1403 * dm_accept_partial_bio informs the dm that the target only wants to process
1404 * additional n_sectors sectors of the bio and the rest of the data should be
1405 * sent in a next bio.
1407 * A diagram that explains the arithmetics:
1408 * +--------------------+---------------+-------+
1410 * +--------------------+---------------+-------+
1412 * <-------------- *tio->len_ptr --------------->
1413 * <------- bi_size ------->
1416 * Region 1 was already iterated over with bio_advance or similar function.
1417 * (it may be empty if the target doesn't use bio_advance)
1418 * Region 2 is the remaining bio size that the target wants to process.
1419 * (it may be empty if region 1 is non-empty, although there is no reason
1421 * The target requires that region 3 is to be sent in the next bio.
1423 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1424 * the partially processed part (the sum of regions 1+2) must be the same for all
1425 * copies of the bio.
1427 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1429 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1430 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1431 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1432 BUG_ON(bi_size
> *tio
->len_ptr
);
1433 BUG_ON(n_sectors
> bi_size
);
1434 *tio
->len_ptr
-= bi_size
- n_sectors
;
1435 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1437 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1439 static void __map_bio(struct dm_target_io
*tio
)
1443 struct mapped_device
*md
;
1444 struct bio
*clone
= &tio
->clone
;
1445 struct dm_target
*ti
= tio
->ti
;
1447 clone
->bi_end_io
= clone_endio
;
1450 * Map the clone. If r == 0 we don't need to do
1451 * anything, the target has assumed ownership of
1454 atomic_inc(&tio
->io
->io_count
);
1455 sector
= clone
->bi_iter
.bi_sector
;
1456 r
= ti
->type
->map(ti
, clone
);
1457 if (r
== DM_MAPIO_REMAPPED
) {
1458 /* the bio has been remapped so dispatch it */
1460 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1461 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1463 generic_make_request(clone
);
1464 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1465 /* error the io and bail out, or requeue it if needed */
1467 dec_pending(tio
->io
, r
);
1470 DMWARN("unimplemented target map return value: %d", r
);
1476 struct mapped_device
*md
;
1477 struct dm_table
*map
;
1481 unsigned sector_count
;
1484 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1486 bio
->bi_iter
.bi_sector
= sector
;
1487 bio
->bi_iter
.bi_size
= to_bytes(len
);
1491 * Creates a bio that consists of range of complete bvecs.
1493 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1494 sector_t sector
, unsigned len
)
1496 struct bio
*clone
= &tio
->clone
;
1498 __bio_clone_fast(clone
, bio
);
1500 if (bio_integrity(bio
))
1501 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1503 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1504 clone
->bi_iter
.bi_size
= to_bytes(len
);
1506 if (bio_integrity(bio
))
1507 bio_integrity_trim(clone
, 0, len
);
1510 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1511 struct dm_target
*ti
,
1512 unsigned target_bio_nr
)
1514 struct dm_target_io
*tio
;
1517 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1518 tio
= container_of(clone
, struct dm_target_io
, clone
);
1522 tio
->target_bio_nr
= target_bio_nr
;
1527 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1528 struct dm_target
*ti
,
1529 unsigned target_bio_nr
, unsigned *len
)
1531 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1532 struct bio
*clone
= &tio
->clone
;
1536 __bio_clone_fast(clone
, ci
->bio
);
1538 bio_setup_sector(clone
, ci
->sector
, *len
);
1543 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1544 unsigned num_bios
, unsigned *len
)
1546 unsigned target_bio_nr
;
1548 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1549 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1552 static int __send_empty_flush(struct clone_info
*ci
)
1554 unsigned target_nr
= 0;
1555 struct dm_target
*ti
;
1557 BUG_ON(bio_has_data(ci
->bio
));
1558 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1559 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1564 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1565 sector_t sector
, unsigned *len
)
1567 struct bio
*bio
= ci
->bio
;
1568 struct dm_target_io
*tio
;
1569 unsigned target_bio_nr
;
1570 unsigned num_target_bios
= 1;
1573 * Does the target want to receive duplicate copies of the bio?
1575 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1576 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1578 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1579 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1581 clone_bio(tio
, bio
, sector
, *len
);
1586 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1588 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1590 return ti
->num_discard_bios
;
1593 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1595 return ti
->num_write_same_bios
;
1598 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1600 static bool is_split_required_for_discard(struct dm_target
*ti
)
1602 return ti
->split_discard_bios
;
1605 static int __send_changing_extent_only(struct clone_info
*ci
,
1606 get_num_bios_fn get_num_bios
,
1607 is_split_required_fn is_split_required
)
1609 struct dm_target
*ti
;
1614 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1615 if (!dm_target_is_valid(ti
))
1619 * Even though the device advertised support for this type of
1620 * request, that does not mean every target supports it, and
1621 * reconfiguration might also have changed that since the
1622 * check was performed.
1624 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1628 if (is_split_required
&& !is_split_required(ti
))
1629 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1631 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1633 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1636 } while (ci
->sector_count
-= len
);
1641 static int __send_discard(struct clone_info
*ci
)
1643 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1644 is_split_required_for_discard
);
1647 static int __send_write_same(struct clone_info
*ci
)
1649 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1653 * Select the correct strategy for processing a non-flush bio.
1655 static int __split_and_process_non_flush(struct clone_info
*ci
)
1657 struct bio
*bio
= ci
->bio
;
1658 struct dm_target
*ti
;
1661 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1662 return __send_discard(ci
);
1663 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1664 return __send_write_same(ci
);
1666 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1667 if (!dm_target_is_valid(ti
))
1670 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1672 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1675 ci
->sector_count
-= len
;
1681 * Entry point to split a bio into clones and submit them to the targets.
1683 static void __split_and_process_bio(struct mapped_device
*md
,
1684 struct dm_table
*map
, struct bio
*bio
)
1686 struct clone_info ci
;
1689 if (unlikely(!map
)) {
1696 ci
.io
= alloc_io(md
);
1698 atomic_set(&ci
.io
->io_count
, 1);
1701 spin_lock_init(&ci
.io
->endio_lock
);
1702 ci
.sector
= bio
->bi_iter
.bi_sector
;
1704 start_io_acct(ci
.io
);
1706 if (bio
->bi_rw
& REQ_FLUSH
) {
1707 ci
.bio
= &ci
.md
->flush_bio
;
1708 ci
.sector_count
= 0;
1709 error
= __send_empty_flush(&ci
);
1710 /* dec_pending submits any data associated with flush */
1713 ci
.sector_count
= bio_sectors(bio
);
1714 while (ci
.sector_count
&& !error
)
1715 error
= __split_and_process_non_flush(&ci
);
1718 /* drop the extra reference count */
1719 dec_pending(ci
.io
, error
);
1721 /*-----------------------------------------------------------------
1723 *---------------------------------------------------------------*/
1725 static int dm_merge_bvec(struct request_queue
*q
,
1726 struct bvec_merge_data
*bvm
,
1727 struct bio_vec
*biovec
)
1729 struct mapped_device
*md
= q
->queuedata
;
1730 struct dm_table
*map
= dm_get_live_table_fast(md
);
1731 struct dm_target
*ti
;
1732 sector_t max_sectors
;
1738 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1739 if (!dm_target_is_valid(ti
))
1743 * Find maximum amount of I/O that won't need splitting
1745 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1746 (sector_t
) BIO_MAX_SECTORS
);
1747 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1752 * merge_bvec_fn() returns number of bytes
1753 * it can accept at this offset
1754 * max is precomputed maximal io size
1756 if (max_size
&& ti
->type
->merge
)
1757 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1759 * If the target doesn't support merge method and some of the devices
1760 * provided their merge_bvec method (we know this by looking at
1761 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1762 * entries. So always set max_size to 0, and the code below allows
1765 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1769 dm_put_live_table_fast(md
);
1771 * Always allow an entire first page
1773 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1774 max_size
= biovec
->bv_len
;
1780 * The request function that just remaps the bio built up by
1783 static void dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1785 int rw
= bio_data_dir(bio
);
1786 struct mapped_device
*md
= q
->queuedata
;
1788 struct dm_table
*map
;
1790 map
= dm_get_live_table(md
, &srcu_idx
);
1792 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1794 /* if we're suspended, we have to queue this io for later */
1795 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1796 dm_put_live_table(md
, srcu_idx
);
1798 if (bio_rw(bio
) != READA
)
1805 __split_and_process_bio(md
, map
, bio
);
1806 dm_put_live_table(md
, srcu_idx
);
1810 int dm_request_based(struct mapped_device
*md
)
1812 return blk_queue_stackable(md
->queue
);
1815 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1819 if (blk_queue_io_stat(clone
->q
))
1820 clone
->cmd_flags
|= REQ_IO_STAT
;
1822 clone
->start_time
= jiffies
;
1823 r
= blk_insert_cloned_request(clone
->q
, clone
);
1825 /* must complete clone in terms of original request */
1826 dm_complete_request(rq
, r
);
1829 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1832 struct dm_rq_target_io
*tio
= data
;
1833 struct dm_rq_clone_bio_info
*info
=
1834 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1836 info
->orig
= bio_orig
;
1838 bio
->bi_end_io
= end_clone_bio
;
1843 static int setup_clone(struct request
*clone
, struct request
*rq
,
1844 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1848 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, gfp_mask
,
1849 dm_rq_bio_constructor
, tio
);
1853 clone
->cmd
= rq
->cmd
;
1854 clone
->cmd_len
= rq
->cmd_len
;
1855 clone
->sense
= rq
->sense
;
1856 clone
->end_io
= end_clone_request
;
1857 clone
->end_io_data
= tio
;
1864 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1865 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1868 * Do not allocate a clone if tio->clone was already set
1869 * (see: dm_mq_queue_rq).
1871 bool alloc_clone
= !tio
->clone
;
1872 struct request
*clone
;
1875 clone
= alloc_clone_request(md
, gfp_mask
);
1881 blk_rq_init(NULL
, clone
);
1882 if (setup_clone(clone
, rq
, tio
, gfp_mask
)) {
1885 free_clone_request(md
, clone
);
1892 static void map_tio_request(struct kthread_work
*work
);
1894 static void init_tio(struct dm_rq_target_io
*tio
, struct request
*rq
,
1895 struct mapped_device
*md
)
1902 memset(&tio
->info
, 0, sizeof(tio
->info
));
1903 if (md
->kworker_task
)
1904 init_kthread_work(&tio
->work
, map_tio_request
);
1907 static struct dm_rq_target_io
*prep_tio(struct request
*rq
,
1908 struct mapped_device
*md
, gfp_t gfp_mask
)
1910 struct dm_rq_target_io
*tio
;
1912 struct dm_table
*table
;
1914 tio
= alloc_rq_tio(md
, gfp_mask
);
1918 init_tio(tio
, rq
, md
);
1920 table
= dm_get_live_table(md
, &srcu_idx
);
1921 if (!dm_table_mq_request_based(table
)) {
1922 if (!clone_rq(rq
, md
, tio
, gfp_mask
)) {
1923 dm_put_live_table(md
, srcu_idx
);
1928 dm_put_live_table(md
, srcu_idx
);
1934 * Called with the queue lock held.
1936 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1938 struct mapped_device
*md
= q
->queuedata
;
1939 struct dm_rq_target_io
*tio
;
1941 if (unlikely(rq
->special
)) {
1942 DMWARN("Already has something in rq->special.");
1943 return BLKPREP_KILL
;
1946 tio
= prep_tio(rq
, md
, GFP_ATOMIC
);
1948 return BLKPREP_DEFER
;
1951 rq
->cmd_flags
|= REQ_DONTPREP
;
1958 * 0 : the request has been processed
1959 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1960 * < 0 : the request was completed due to failure
1962 static int map_request(struct dm_rq_target_io
*tio
, struct request
*rq
,
1963 struct mapped_device
*md
)
1966 struct dm_target
*ti
= tio
->ti
;
1967 struct request
*clone
= NULL
;
1971 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1973 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
1975 /* The target wants to complete the I/O */
1976 dm_kill_unmapped_request(rq
, r
);
1979 if (r
!= DM_MAPIO_REMAPPED
)
1981 if (setup_clone(clone
, rq
, tio
, GFP_ATOMIC
)) {
1983 ti
->type
->release_clone_rq(clone
);
1984 return DM_MAPIO_REQUEUE
;
1989 case DM_MAPIO_SUBMITTED
:
1990 /* The target has taken the I/O to submit by itself later */
1992 case DM_MAPIO_REMAPPED
:
1993 /* The target has remapped the I/O so dispatch it */
1994 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1996 dm_dispatch_clone_request(clone
, rq
);
1998 case DM_MAPIO_REQUEUE
:
1999 /* The target wants to requeue the I/O */
2000 dm_requeue_original_request(md
, tio
->orig
);
2004 DMWARN("unimplemented target map return value: %d", r
);
2008 /* The target wants to complete the I/O */
2009 dm_kill_unmapped_request(rq
, r
);
2016 static void map_tio_request(struct kthread_work
*work
)
2018 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
2019 struct request
*rq
= tio
->orig
;
2020 struct mapped_device
*md
= tio
->md
;
2022 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
)
2023 dm_requeue_original_request(md
, rq
);
2026 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
2028 if (!orig
->q
->mq_ops
)
2029 blk_start_request(orig
);
2031 blk_mq_start_request(orig
);
2032 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
2034 if (md
->seq_rq_merge_deadline_usecs
) {
2035 md
->last_rq_pos
= rq_end_sector(orig
);
2036 md
->last_rq_rw
= rq_data_dir(orig
);
2037 md
->last_rq_start_time
= ktime_get();
2040 if (unlikely(dm_stats_used(&md
->stats
))) {
2041 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
2042 tio
->duration_jiffies
= jiffies
;
2043 tio
->n_sectors
= blk_rq_sectors(orig
);
2044 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
2045 tio
->n_sectors
, false, 0, &tio
->stats_aux
);
2049 * Hold the md reference here for the in-flight I/O.
2050 * We can't rely on the reference count by device opener,
2051 * because the device may be closed during the request completion
2052 * when all bios are completed.
2053 * See the comment in rq_completed() too.
2058 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2060 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device
*md
, char *buf
)
2062 return sprintf(buf
, "%u\n", md
->seq_rq_merge_deadline_usecs
);
2065 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device
*md
,
2066 const char *buf
, size_t count
)
2070 if (!dm_request_based(md
) || md
->use_blk_mq
)
2073 if (kstrtouint(buf
, 10, &deadline
))
2076 if (deadline
> MAX_SEQ_RQ_MERGE_DEADLINE_USECS
)
2077 deadline
= MAX_SEQ_RQ_MERGE_DEADLINE_USECS
;
2079 md
->seq_rq_merge_deadline_usecs
= deadline
;
2084 static bool dm_request_peeked_before_merge_deadline(struct mapped_device
*md
)
2086 ktime_t kt_deadline
;
2088 if (!md
->seq_rq_merge_deadline_usecs
)
2091 kt_deadline
= ns_to_ktime((u64
)md
->seq_rq_merge_deadline_usecs
* NSEC_PER_USEC
);
2092 kt_deadline
= ktime_add_safe(md
->last_rq_start_time
, kt_deadline
);
2094 return !ktime_after(ktime_get(), kt_deadline
);
2098 * q->request_fn for request-based dm.
2099 * Called with the queue lock held.
2101 static void dm_request_fn(struct request_queue
*q
)
2103 struct mapped_device
*md
= q
->queuedata
;
2105 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2106 struct dm_target
*ti
;
2108 struct dm_rq_target_io
*tio
;
2112 * For suspend, check blk_queue_stopped() and increment
2113 * ->pending within a single queue_lock not to increment the
2114 * number of in-flight I/Os after the queue is stopped in
2117 while (!blk_queue_stopped(q
)) {
2118 rq
= blk_peek_request(q
);
2122 /* always use block 0 to find the target for flushes for now */
2124 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2125 pos
= blk_rq_pos(rq
);
2127 ti
= dm_table_find_target(map
, pos
);
2128 if (!dm_target_is_valid(ti
)) {
2130 * Must perform setup, that rq_completed() requires,
2131 * before calling dm_kill_unmapped_request
2133 DMERR_LIMIT("request attempted access beyond the end of device");
2134 dm_start_request(md
, rq
);
2135 dm_kill_unmapped_request(rq
, -EIO
);
2139 if (dm_request_peeked_before_merge_deadline(md
) &&
2140 md_in_flight(md
) && rq
->bio
&& rq
->bio
->bi_vcnt
== 1 &&
2141 md
->last_rq_pos
== pos
&& md
->last_rq_rw
== rq_data_dir(rq
))
2144 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2147 dm_start_request(md
, rq
);
2149 tio
= tio_from_request(rq
);
2150 /* Establish tio->ti before queuing work (map_tio_request) */
2152 queue_kthread_work(&md
->kworker
, &tio
->work
);
2153 BUG_ON(!irqs_disabled());
2159 blk_delay_queue(q
, HZ
/ 100);
2161 dm_put_live_table(md
, srcu_idx
);
2164 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2167 struct mapped_device
*md
= congested_data
;
2168 struct dm_table
*map
;
2170 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2171 map
= dm_get_live_table_fast(md
);
2174 * Request-based dm cares about only own queue for
2175 * the query about congestion status of request_queue
2177 if (dm_request_based(md
))
2178 r
= md
->queue
->backing_dev_info
.wb
.state
&
2181 r
= dm_table_any_congested(map
, bdi_bits
);
2183 dm_put_live_table_fast(md
);
2189 /*-----------------------------------------------------------------
2190 * An IDR is used to keep track of allocated minor numbers.
2191 *---------------------------------------------------------------*/
2192 static void free_minor(int minor
)
2194 spin_lock(&_minor_lock
);
2195 idr_remove(&_minor_idr
, minor
);
2196 spin_unlock(&_minor_lock
);
2200 * See if the device with a specific minor # is free.
2202 static int specific_minor(int minor
)
2206 if (minor
>= (1 << MINORBITS
))
2209 idr_preload(GFP_KERNEL
);
2210 spin_lock(&_minor_lock
);
2212 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2214 spin_unlock(&_minor_lock
);
2217 return r
== -ENOSPC
? -EBUSY
: r
;
2221 static int next_free_minor(int *minor
)
2225 idr_preload(GFP_KERNEL
);
2226 spin_lock(&_minor_lock
);
2228 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2230 spin_unlock(&_minor_lock
);
2238 static const struct block_device_operations dm_blk_dops
;
2240 static void dm_wq_work(struct work_struct
*work
);
2242 static void dm_init_md_queue(struct mapped_device
*md
)
2245 * Request-based dm devices cannot be stacked on top of bio-based dm
2246 * devices. The type of this dm device may not have been decided yet.
2247 * The type is decided at the first table loading time.
2248 * To prevent problematic device stacking, clear the queue flag
2249 * for request stacking support until then.
2251 * This queue is new, so no concurrency on the queue_flags.
2253 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2256 static void dm_init_old_md_queue(struct mapped_device
*md
)
2258 md
->use_blk_mq
= false;
2259 dm_init_md_queue(md
);
2262 * Initialize aspects of queue that aren't relevant for blk-mq
2264 md
->queue
->queuedata
= md
;
2265 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2266 md
->queue
->backing_dev_info
.congested_data
= md
;
2268 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2271 static void cleanup_mapped_device(struct mapped_device
*md
)
2274 destroy_workqueue(md
->wq
);
2275 if (md
->kworker_task
)
2276 kthread_stop(md
->kworker_task
);
2278 mempool_destroy(md
->io_pool
);
2280 mempool_destroy(md
->rq_pool
);
2282 bioset_free(md
->bs
);
2284 cleanup_srcu_struct(&md
->io_barrier
);
2287 spin_lock(&_minor_lock
);
2288 md
->disk
->private_data
= NULL
;
2289 spin_unlock(&_minor_lock
);
2290 if (blk_get_integrity(md
->disk
))
2291 blk_integrity_unregister(md
->disk
);
2292 del_gendisk(md
->disk
);
2297 blk_cleanup_queue(md
->queue
);
2306 * Allocate and initialise a blank device with a given minor.
2308 static struct mapped_device
*alloc_dev(int minor
)
2311 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2315 DMWARN("unable to allocate device, out of memory.");
2319 if (!try_module_get(THIS_MODULE
))
2320 goto bad_module_get
;
2322 /* get a minor number for the dev */
2323 if (minor
== DM_ANY_MINOR
)
2324 r
= next_free_minor(&minor
);
2326 r
= specific_minor(minor
);
2330 r
= init_srcu_struct(&md
->io_barrier
);
2332 goto bad_io_barrier
;
2334 md
->use_blk_mq
= use_blk_mq
;
2335 md
->type
= DM_TYPE_NONE
;
2336 mutex_init(&md
->suspend_lock
);
2337 mutex_init(&md
->type_lock
);
2338 mutex_init(&md
->table_devices_lock
);
2339 spin_lock_init(&md
->deferred_lock
);
2340 atomic_set(&md
->holders
, 1);
2341 atomic_set(&md
->open_count
, 0);
2342 atomic_set(&md
->event_nr
, 0);
2343 atomic_set(&md
->uevent_seq
, 0);
2344 INIT_LIST_HEAD(&md
->uevent_list
);
2345 INIT_LIST_HEAD(&md
->table_devices
);
2346 spin_lock_init(&md
->uevent_lock
);
2348 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2352 dm_init_md_queue(md
);
2354 md
->disk
= alloc_disk(1);
2358 atomic_set(&md
->pending
[0], 0);
2359 atomic_set(&md
->pending
[1], 0);
2360 init_waitqueue_head(&md
->wait
);
2361 INIT_WORK(&md
->work
, dm_wq_work
);
2362 init_waitqueue_head(&md
->eventq
);
2363 init_completion(&md
->kobj_holder
.completion
);
2364 md
->kworker_task
= NULL
;
2366 md
->disk
->major
= _major
;
2367 md
->disk
->first_minor
= minor
;
2368 md
->disk
->fops
= &dm_blk_dops
;
2369 md
->disk
->queue
= md
->queue
;
2370 md
->disk
->private_data
= md
;
2371 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2373 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2375 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2379 md
->bdev
= bdget_disk(md
->disk
, 0);
2383 bio_init(&md
->flush_bio
);
2384 md
->flush_bio
.bi_bdev
= md
->bdev
;
2385 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2387 dm_stats_init(&md
->stats
);
2389 /* Populate the mapping, nobody knows we exist yet */
2390 spin_lock(&_minor_lock
);
2391 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2392 spin_unlock(&_minor_lock
);
2394 BUG_ON(old_md
!= MINOR_ALLOCED
);
2399 cleanup_mapped_device(md
);
2403 module_put(THIS_MODULE
);
2409 static void unlock_fs(struct mapped_device
*md
);
2411 static void free_dev(struct mapped_device
*md
)
2413 int minor
= MINOR(disk_devt(md
->disk
));
2417 cleanup_mapped_device(md
);
2419 blk_mq_free_tag_set(&md
->tag_set
);
2421 free_table_devices(&md
->table_devices
);
2422 dm_stats_cleanup(&md
->stats
);
2425 module_put(THIS_MODULE
);
2429 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2431 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2434 /* The md already has necessary mempools. */
2435 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2437 * Reload bioset because front_pad may have changed
2438 * because a different table was loaded.
2440 bioset_free(md
->bs
);
2445 * There's no need to reload with request-based dm
2446 * because the size of front_pad doesn't change.
2447 * Note for future: If you are to reload bioset,
2448 * prep-ed requests in the queue may refer
2449 * to bio from the old bioset, so you must walk
2450 * through the queue to unprep.
2455 BUG_ON(!p
|| md
->io_pool
|| md
->rq_pool
|| md
->bs
);
2457 md
->io_pool
= p
->io_pool
;
2459 md
->rq_pool
= p
->rq_pool
;
2465 /* mempool bind completed, no longer need any mempools in the table */
2466 dm_table_free_md_mempools(t
);
2470 * Bind a table to the device.
2472 static void event_callback(void *context
)
2474 unsigned long flags
;
2476 struct mapped_device
*md
= (struct mapped_device
*) context
;
2478 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2479 list_splice_init(&md
->uevent_list
, &uevents
);
2480 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2482 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2484 atomic_inc(&md
->event_nr
);
2485 wake_up(&md
->eventq
);
2489 * Protected by md->suspend_lock obtained by dm_swap_table().
2491 static void __set_size(struct mapped_device
*md
, sector_t size
)
2493 set_capacity(md
->disk
, size
);
2495 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2499 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2501 * If this function returns 0, then the device is either a non-dm
2502 * device without a merge_bvec_fn, or it is a dm device that is
2503 * able to split any bios it receives that are too big.
2505 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2507 struct mapped_device
*dev_md
;
2509 if (!q
->merge_bvec_fn
)
2512 if (q
->make_request_fn
== dm_make_request
) {
2513 dev_md
= q
->queuedata
;
2514 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2521 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2522 struct dm_dev
*dev
, sector_t start
,
2523 sector_t len
, void *data
)
2525 struct block_device
*bdev
= dev
->bdev
;
2526 struct request_queue
*q
= bdev_get_queue(bdev
);
2528 return dm_queue_merge_is_compulsory(q
);
2532 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2533 * on the properties of the underlying devices.
2535 static int dm_table_merge_is_optional(struct dm_table
*table
)
2538 struct dm_target
*ti
;
2540 while (i
< dm_table_get_num_targets(table
)) {
2541 ti
= dm_table_get_target(table
, i
++);
2543 if (ti
->type
->iterate_devices
&&
2544 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2552 * Returns old map, which caller must destroy.
2554 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2555 struct queue_limits
*limits
)
2557 struct dm_table
*old_map
;
2558 struct request_queue
*q
= md
->queue
;
2560 int merge_is_optional
;
2562 size
= dm_table_get_size(t
);
2565 * Wipe any geometry if the size of the table changed.
2567 if (size
!= dm_get_size(md
))
2568 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2570 __set_size(md
, size
);
2572 dm_table_event_callback(t
, event_callback
, md
);
2575 * The queue hasn't been stopped yet, if the old table type wasn't
2576 * for request-based during suspension. So stop it to prevent
2577 * I/O mapping before resume.
2578 * This must be done before setting the queue restrictions,
2579 * because request-based dm may be run just after the setting.
2581 if (dm_table_request_based(t
))
2584 __bind_mempools(md
, t
);
2586 merge_is_optional
= dm_table_merge_is_optional(t
);
2588 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2589 rcu_assign_pointer(md
->map
, t
);
2590 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2592 dm_table_set_restrictions(t
, q
, limits
);
2593 if (merge_is_optional
)
2594 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2596 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2604 * Returns unbound table for the caller to free.
2606 static struct dm_table
*__unbind(struct mapped_device
*md
)
2608 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2613 dm_table_event_callback(map
, NULL
, NULL
);
2614 RCU_INIT_POINTER(md
->map
, NULL
);
2621 * Constructor for a new device.
2623 int dm_create(int minor
, struct mapped_device
**result
)
2625 struct mapped_device
*md
;
2627 md
= alloc_dev(minor
);
2638 * Functions to manage md->type.
2639 * All are required to hold md->type_lock.
2641 void dm_lock_md_type(struct mapped_device
*md
)
2643 mutex_lock(&md
->type_lock
);
2646 void dm_unlock_md_type(struct mapped_device
*md
)
2648 mutex_unlock(&md
->type_lock
);
2651 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2653 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2657 unsigned dm_get_md_type(struct mapped_device
*md
)
2659 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2663 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2665 return md
->immutable_target_type
;
2669 * The queue_limits are only valid as long as you have a reference
2672 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2674 BUG_ON(!atomic_read(&md
->holders
));
2675 return &md
->queue
->limits
;
2677 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2679 static void init_rq_based_worker_thread(struct mapped_device
*md
)
2681 /* Initialize the request-based DM worker thread */
2682 init_kthread_worker(&md
->kworker
);
2683 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2684 "kdmwork-%s", dm_device_name(md
));
2688 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2690 static int dm_init_request_based_queue(struct mapped_device
*md
)
2692 struct request_queue
*q
= NULL
;
2694 /* Fully initialize the queue */
2695 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2699 /* disable dm_request_fn's merge heuristic by default */
2700 md
->seq_rq_merge_deadline_usecs
= 0;
2703 dm_init_old_md_queue(md
);
2704 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2705 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2707 init_rq_based_worker_thread(md
);
2709 elv_register_queue(md
->queue
);
2714 static int dm_mq_init_request(void *data
, struct request
*rq
,
2715 unsigned int hctx_idx
, unsigned int request_idx
,
2716 unsigned int numa_node
)
2718 struct mapped_device
*md
= data
;
2719 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2722 * Must initialize md member of tio, otherwise it won't
2723 * be available in dm_mq_queue_rq.
2730 static int dm_mq_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2731 const struct blk_mq_queue_data
*bd
)
2733 struct request
*rq
= bd
->rq
;
2734 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2735 struct mapped_device
*md
= tio
->md
;
2737 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2738 struct dm_target
*ti
;
2741 /* always use block 0 to find the target for flushes for now */
2743 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2744 pos
= blk_rq_pos(rq
);
2746 ti
= dm_table_find_target(map
, pos
);
2747 if (!dm_target_is_valid(ti
)) {
2748 dm_put_live_table(md
, srcu_idx
);
2749 DMERR_LIMIT("request attempted access beyond the end of device");
2751 * Must perform setup, that rq_completed() requires,
2752 * before returning BLK_MQ_RQ_QUEUE_ERROR
2754 dm_start_request(md
, rq
);
2755 return BLK_MQ_RQ_QUEUE_ERROR
;
2757 dm_put_live_table(md
, srcu_idx
);
2759 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2760 return BLK_MQ_RQ_QUEUE_BUSY
;
2762 dm_start_request(md
, rq
);
2764 /* Init tio using md established in .init_request */
2765 init_tio(tio
, rq
, md
);
2768 * Establish tio->ti before queuing work (map_tio_request)
2769 * or making direct call to map_request().
2773 /* Clone the request if underlying devices aren't blk-mq */
2774 if (dm_table_get_type(map
) == DM_TYPE_REQUEST_BASED
) {
2775 /* clone request is allocated at the end of the pdu */
2776 tio
->clone
= (void *)blk_mq_rq_to_pdu(rq
) + sizeof(struct dm_rq_target_io
);
2777 (void) clone_rq(rq
, md
, tio
, GFP_ATOMIC
);
2778 queue_kthread_work(&md
->kworker
, &tio
->work
);
2780 /* Direct call is fine since .queue_rq allows allocations */
2781 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
) {
2782 /* Undo dm_start_request() before requeuing */
2783 rq_end_stats(md
, rq
);
2784 rq_completed(md
, rq_data_dir(rq
), false);
2785 return BLK_MQ_RQ_QUEUE_BUSY
;
2789 return BLK_MQ_RQ_QUEUE_OK
;
2792 static struct blk_mq_ops dm_mq_ops
= {
2793 .queue_rq
= dm_mq_queue_rq
,
2794 .map_queue
= blk_mq_map_queue
,
2795 .complete
= dm_softirq_done
,
2796 .init_request
= dm_mq_init_request
,
2799 static int dm_init_request_based_blk_mq_queue(struct mapped_device
*md
)
2801 unsigned md_type
= dm_get_md_type(md
);
2802 struct request_queue
*q
;
2805 memset(&md
->tag_set
, 0, sizeof(md
->tag_set
));
2806 md
->tag_set
.ops
= &dm_mq_ops
;
2807 md
->tag_set
.queue_depth
= BLKDEV_MAX_RQ
;
2808 md
->tag_set
.numa_node
= NUMA_NO_NODE
;
2809 md
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2810 md
->tag_set
.nr_hw_queues
= 1;
2811 if (md_type
== DM_TYPE_REQUEST_BASED
) {
2812 /* make the memory for non-blk-mq clone part of the pdu */
2813 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
) + sizeof(struct request
);
2815 md
->tag_set
.cmd_size
= sizeof(struct dm_rq_target_io
);
2816 md
->tag_set
.driver_data
= md
;
2818 err
= blk_mq_alloc_tag_set(&md
->tag_set
);
2822 q
= blk_mq_init_allocated_queue(&md
->tag_set
, md
->queue
);
2828 dm_init_md_queue(md
);
2830 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2831 blk_mq_register_disk(md
->disk
);
2833 if (md_type
== DM_TYPE_REQUEST_BASED
)
2834 init_rq_based_worker_thread(md
);
2839 blk_mq_free_tag_set(&md
->tag_set
);
2843 static unsigned filter_md_type(unsigned type
, struct mapped_device
*md
)
2845 if (type
== DM_TYPE_BIO_BASED
)
2848 return !md
->use_blk_mq
? DM_TYPE_REQUEST_BASED
: DM_TYPE_MQ_REQUEST_BASED
;
2852 * Setup the DM device's queue based on md's type
2854 int dm_setup_md_queue(struct mapped_device
*md
)
2857 unsigned md_type
= filter_md_type(dm_get_md_type(md
), md
);
2860 case DM_TYPE_REQUEST_BASED
:
2861 r
= dm_init_request_based_queue(md
);
2863 DMWARN("Cannot initialize queue for request-based mapped device");
2867 case DM_TYPE_MQ_REQUEST_BASED
:
2868 r
= dm_init_request_based_blk_mq_queue(md
);
2870 DMWARN("Cannot initialize queue for request-based blk-mq mapped device");
2874 case DM_TYPE_BIO_BASED
:
2875 dm_init_old_md_queue(md
);
2876 blk_queue_make_request(md
->queue
, dm_make_request
);
2877 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
2884 struct mapped_device
*dm_get_md(dev_t dev
)
2886 struct mapped_device
*md
;
2887 unsigned minor
= MINOR(dev
);
2889 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2892 spin_lock(&_minor_lock
);
2894 md
= idr_find(&_minor_idr
, minor
);
2896 if ((md
== MINOR_ALLOCED
||
2897 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2898 dm_deleting_md(md
) ||
2899 test_bit(DMF_FREEING
, &md
->flags
))) {
2907 spin_unlock(&_minor_lock
);
2911 EXPORT_SYMBOL_GPL(dm_get_md
);
2913 void *dm_get_mdptr(struct mapped_device
*md
)
2915 return md
->interface_ptr
;
2918 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2920 md
->interface_ptr
= ptr
;
2923 void dm_get(struct mapped_device
*md
)
2925 atomic_inc(&md
->holders
);
2926 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2929 int dm_hold(struct mapped_device
*md
)
2931 spin_lock(&_minor_lock
);
2932 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2933 spin_unlock(&_minor_lock
);
2937 spin_unlock(&_minor_lock
);
2940 EXPORT_SYMBOL_GPL(dm_hold
);
2942 const char *dm_device_name(struct mapped_device
*md
)
2946 EXPORT_SYMBOL_GPL(dm_device_name
);
2948 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2950 struct dm_table
*map
;
2955 map
= dm_get_live_table(md
, &srcu_idx
);
2957 spin_lock(&_minor_lock
);
2958 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2959 set_bit(DMF_FREEING
, &md
->flags
);
2960 spin_unlock(&_minor_lock
);
2962 if (dm_request_based(md
) && md
->kworker_task
)
2963 flush_kthread_worker(&md
->kworker
);
2966 * Take suspend_lock so that presuspend and postsuspend methods
2967 * do not race with internal suspend.
2969 mutex_lock(&md
->suspend_lock
);
2970 if (!dm_suspended_md(md
)) {
2971 dm_table_presuspend_targets(map
);
2972 dm_table_postsuspend_targets(map
);
2974 mutex_unlock(&md
->suspend_lock
);
2976 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2977 dm_put_live_table(md
, srcu_idx
);
2980 * Rare, but there may be I/O requests still going to complete,
2981 * for example. Wait for all references to disappear.
2982 * No one should increment the reference count of the mapped_device,
2983 * after the mapped_device state becomes DMF_FREEING.
2986 while (atomic_read(&md
->holders
))
2988 else if (atomic_read(&md
->holders
))
2989 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2990 dm_device_name(md
), atomic_read(&md
->holders
));
2993 dm_table_destroy(__unbind(md
));
2997 void dm_destroy(struct mapped_device
*md
)
2999 __dm_destroy(md
, true);
3002 void dm_destroy_immediate(struct mapped_device
*md
)
3004 __dm_destroy(md
, false);
3007 void dm_put(struct mapped_device
*md
)
3009 atomic_dec(&md
->holders
);
3011 EXPORT_SYMBOL_GPL(dm_put
);
3013 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
3016 DECLARE_WAITQUEUE(wait
, current
);
3018 add_wait_queue(&md
->wait
, &wait
);
3021 set_current_state(interruptible
);
3023 if (!md_in_flight(md
))
3026 if (interruptible
== TASK_INTERRUPTIBLE
&&
3027 signal_pending(current
)) {
3034 set_current_state(TASK_RUNNING
);
3036 remove_wait_queue(&md
->wait
, &wait
);
3042 * Process the deferred bios
3044 static void dm_wq_work(struct work_struct
*work
)
3046 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
3050 struct dm_table
*map
;
3052 map
= dm_get_live_table(md
, &srcu_idx
);
3054 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
3055 spin_lock_irq(&md
->deferred_lock
);
3056 c
= bio_list_pop(&md
->deferred
);
3057 spin_unlock_irq(&md
->deferred_lock
);
3062 if (dm_request_based(md
))
3063 generic_make_request(c
);
3065 __split_and_process_bio(md
, map
, c
);
3068 dm_put_live_table(md
, srcu_idx
);
3071 static void dm_queue_flush(struct mapped_device
*md
)
3073 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3074 smp_mb__after_atomic();
3075 queue_work(md
->wq
, &md
->work
);
3079 * Swap in a new table, returning the old one for the caller to destroy.
3081 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
3083 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
3084 struct queue_limits limits
;
3087 mutex_lock(&md
->suspend_lock
);
3089 /* device must be suspended */
3090 if (!dm_suspended_md(md
))
3094 * If the new table has no data devices, retain the existing limits.
3095 * This helps multipath with queue_if_no_path if all paths disappear,
3096 * then new I/O is queued based on these limits, and then some paths
3099 if (dm_table_has_no_data_devices(table
)) {
3100 live_map
= dm_get_live_table_fast(md
);
3102 limits
= md
->queue
->limits
;
3103 dm_put_live_table_fast(md
);
3107 r
= dm_calculate_queue_limits(table
, &limits
);
3114 map
= __bind(md
, table
, &limits
);
3117 mutex_unlock(&md
->suspend_lock
);
3122 * Functions to lock and unlock any filesystem running on the
3125 static int lock_fs(struct mapped_device
*md
)
3129 WARN_ON(md
->frozen_sb
);
3131 md
->frozen_sb
= freeze_bdev(md
->bdev
);
3132 if (IS_ERR(md
->frozen_sb
)) {
3133 r
= PTR_ERR(md
->frozen_sb
);
3134 md
->frozen_sb
= NULL
;
3138 set_bit(DMF_FROZEN
, &md
->flags
);
3143 static void unlock_fs(struct mapped_device
*md
)
3145 if (!test_bit(DMF_FROZEN
, &md
->flags
))
3148 thaw_bdev(md
->bdev
, md
->frozen_sb
);
3149 md
->frozen_sb
= NULL
;
3150 clear_bit(DMF_FROZEN
, &md
->flags
);
3154 * If __dm_suspend returns 0, the device is completely quiescent
3155 * now. There is no request-processing activity. All new requests
3156 * are being added to md->deferred list.
3158 * Caller must hold md->suspend_lock
3160 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
3161 unsigned suspend_flags
, int interruptible
)
3163 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
3164 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
3168 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3169 * This flag is cleared before dm_suspend returns.
3172 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3175 * This gets reverted if there's an error later and the targets
3176 * provide the .presuspend_undo hook.
3178 dm_table_presuspend_targets(map
);
3181 * Flush I/O to the device.
3182 * Any I/O submitted after lock_fs() may not be flushed.
3183 * noflush takes precedence over do_lockfs.
3184 * (lock_fs() flushes I/Os and waits for them to complete.)
3186 if (!noflush
&& do_lockfs
) {
3189 dm_table_presuspend_undo_targets(map
);
3195 * Here we must make sure that no processes are submitting requests
3196 * to target drivers i.e. no one may be executing
3197 * __split_and_process_bio. This is called from dm_request and
3200 * To get all processes out of __split_and_process_bio in dm_request,
3201 * we take the write lock. To prevent any process from reentering
3202 * __split_and_process_bio from dm_request and quiesce the thread
3203 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3204 * flush_workqueue(md->wq).
3206 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3208 synchronize_srcu(&md
->io_barrier
);
3211 * Stop md->queue before flushing md->wq in case request-based
3212 * dm defers requests to md->wq from md->queue.
3214 if (dm_request_based(md
)) {
3215 stop_queue(md
->queue
);
3216 if (md
->kworker_task
)
3217 flush_kthread_worker(&md
->kworker
);
3220 flush_workqueue(md
->wq
);
3223 * At this point no more requests are entering target request routines.
3224 * We call dm_wait_for_completion to wait for all existing requests
3227 r
= dm_wait_for_completion(md
, interruptible
);
3230 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3232 synchronize_srcu(&md
->io_barrier
);
3234 /* were we interrupted ? */
3238 if (dm_request_based(md
))
3239 start_queue(md
->queue
);
3242 dm_table_presuspend_undo_targets(map
);
3243 /* pushback list is already flushed, so skip flush */
3250 * We need to be able to change a mapping table under a mounted
3251 * filesystem. For example we might want to move some data in
3252 * the background. Before the table can be swapped with
3253 * dm_bind_table, dm_suspend must be called to flush any in
3254 * flight bios and ensure that any further io gets deferred.
3257 * Suspend mechanism in request-based dm.
3259 * 1. Flush all I/Os by lock_fs() if needed.
3260 * 2. Stop dispatching any I/O by stopping the request_queue.
3261 * 3. Wait for all in-flight I/Os to be completed or requeued.
3263 * To abort suspend, start the request_queue.
3265 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3267 struct dm_table
*map
= NULL
;
3271 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3273 if (dm_suspended_md(md
)) {
3278 if (dm_suspended_internally_md(md
)) {
3279 /* already internally suspended, wait for internal resume */
3280 mutex_unlock(&md
->suspend_lock
);
3281 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3287 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3289 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
3293 set_bit(DMF_SUSPENDED
, &md
->flags
);
3295 dm_table_postsuspend_targets(map
);
3298 mutex_unlock(&md
->suspend_lock
);
3302 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
3305 int r
= dm_table_resume_targets(map
);
3313 * Flushing deferred I/Os must be done after targets are resumed
3314 * so that mapping of targets can work correctly.
3315 * Request-based dm is queueing the deferred I/Os in its request_queue.
3317 if (dm_request_based(md
))
3318 start_queue(md
->queue
);
3325 int dm_resume(struct mapped_device
*md
)
3328 struct dm_table
*map
= NULL
;
3331 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3333 if (!dm_suspended_md(md
))
3336 if (dm_suspended_internally_md(md
)) {
3337 /* already internally suspended, wait for internal resume */
3338 mutex_unlock(&md
->suspend_lock
);
3339 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3345 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3346 if (!map
|| !dm_table_get_size(map
))
3349 r
= __dm_resume(md
, map
);
3353 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3357 mutex_unlock(&md
->suspend_lock
);
3363 * Internal suspend/resume works like userspace-driven suspend. It waits
3364 * until all bios finish and prevents issuing new bios to the target drivers.
3365 * It may be used only from the kernel.
3368 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3370 struct dm_table
*map
= NULL
;
3372 if (md
->internal_suspend_count
++)
3373 return; /* nested internal suspend */
3375 if (dm_suspended_md(md
)) {
3376 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3377 return; /* nest suspend */
3380 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3383 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3384 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3385 * would require changing .presuspend to return an error -- avoid this
3386 * until there is a need for more elaborate variants of internal suspend.
3388 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3390 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3392 dm_table_postsuspend_targets(map
);
3395 static void __dm_internal_resume(struct mapped_device
*md
)
3397 BUG_ON(!md
->internal_suspend_count
);
3399 if (--md
->internal_suspend_count
)
3400 return; /* resume from nested internal suspend */
3402 if (dm_suspended_md(md
))
3403 goto done
; /* resume from nested suspend */
3406 * NOTE: existing callers don't need to call dm_table_resume_targets
3407 * (which may fail -- so best to avoid it for now by passing NULL map)
3409 (void) __dm_resume(md
, NULL
);
3412 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3413 smp_mb__after_atomic();
3414 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3417 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3419 mutex_lock(&md
->suspend_lock
);
3420 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3421 mutex_unlock(&md
->suspend_lock
);
3423 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3425 void dm_internal_resume(struct mapped_device
*md
)
3427 mutex_lock(&md
->suspend_lock
);
3428 __dm_internal_resume(md
);
3429 mutex_unlock(&md
->suspend_lock
);
3431 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3434 * Fast variants of internal suspend/resume hold md->suspend_lock,
3435 * which prevents interaction with userspace-driven suspend.
3438 void dm_internal_suspend_fast(struct mapped_device
*md
)
3440 mutex_lock(&md
->suspend_lock
);
3441 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3444 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3445 synchronize_srcu(&md
->io_barrier
);
3446 flush_workqueue(md
->wq
);
3447 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3449 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
3451 void dm_internal_resume_fast(struct mapped_device
*md
)
3453 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3459 mutex_unlock(&md
->suspend_lock
);
3461 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
3463 /*-----------------------------------------------------------------
3464 * Event notification.
3465 *---------------------------------------------------------------*/
3466 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3469 char udev_cookie
[DM_COOKIE_LENGTH
];
3470 char *envp
[] = { udev_cookie
, NULL
};
3473 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3475 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3476 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3477 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3482 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3484 return atomic_add_return(1, &md
->uevent_seq
);
3487 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3489 return atomic_read(&md
->event_nr
);
3492 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3494 return wait_event_interruptible(md
->eventq
,
3495 (event_nr
!= atomic_read(&md
->event_nr
)));
3498 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3500 unsigned long flags
;
3502 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3503 list_add(elist
, &md
->uevent_list
);
3504 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3508 * The gendisk is only valid as long as you have a reference
3511 struct gendisk
*dm_disk(struct mapped_device
*md
)
3515 EXPORT_SYMBOL_GPL(dm_disk
);
3517 struct kobject
*dm_kobject(struct mapped_device
*md
)
3519 return &md
->kobj_holder
.kobj
;
3522 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3524 struct mapped_device
*md
;
3526 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3528 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3536 int dm_suspended_md(struct mapped_device
*md
)
3538 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3541 int dm_suspended_internally_md(struct mapped_device
*md
)
3543 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3546 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3548 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3551 int dm_suspended(struct dm_target
*ti
)
3553 return dm_suspended_md(dm_table_get_md(ti
->table
));
3555 EXPORT_SYMBOL_GPL(dm_suspended
);
3557 int dm_noflush_suspending(struct dm_target
*ti
)
3559 return __noflush_suspending(dm_table_get_md(ti
->table
));
3561 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3563 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
3564 unsigned integrity
, unsigned per_bio_data_size
)
3566 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3567 struct kmem_cache
*cachep
= NULL
;
3568 unsigned int pool_size
= 0;
3569 unsigned int front_pad
;
3574 type
= filter_md_type(type
, md
);
3577 case DM_TYPE_BIO_BASED
:
3579 pool_size
= dm_get_reserved_bio_based_ios();
3580 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3582 case DM_TYPE_REQUEST_BASED
:
3583 cachep
= _rq_tio_cache
;
3584 pool_size
= dm_get_reserved_rq_based_ios();
3585 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3586 if (!pools
->rq_pool
)
3588 /* fall through to setup remaining rq-based pools */
3589 case DM_TYPE_MQ_REQUEST_BASED
:
3591 pool_size
= dm_get_reserved_rq_based_ios();
3592 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3593 /* per_bio_data_size is not used. See __bind_mempools(). */
3594 WARN_ON(per_bio_data_size
!= 0);
3601 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3602 if (!pools
->io_pool
)
3606 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3610 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3616 dm_free_md_mempools(pools
);
3621 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3627 mempool_destroy(pools
->io_pool
);
3630 mempool_destroy(pools
->rq_pool
);
3633 bioset_free(pools
->bs
);
3638 static const struct block_device_operations dm_blk_dops
= {
3639 .open
= dm_blk_open
,
3640 .release
= dm_blk_close
,
3641 .ioctl
= dm_blk_ioctl
,
3642 .getgeo
= dm_blk_getgeo
,
3643 .owner
= THIS_MODULE
3649 module_init(dm_init
);
3650 module_exit(dm_exit
);
3652 module_param(major
, uint
, 0);
3653 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3655 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3656 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3658 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3659 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3661 module_param(use_blk_mq
, bool, S_IRUGO
| S_IWUSR
);
3662 MODULE_PARM_DESC(use_blk_mq
, "Use block multiqueue for request-based DM devices");
3664 MODULE_DESCRIPTION(DM_NAME
" driver");
3665 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3666 MODULE_LICENSE("GPL");