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>
25 #include <trace/events/block.h>
27 #define DM_MSG_PREFIX "core"
31 * ratelimit state to be used in DMXXX_LIMIT().
33 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
34 DEFAULT_RATELIMIT_INTERVAL
,
35 DEFAULT_RATELIMIT_BURST
);
36 EXPORT_SYMBOL(dm_ratelimit_state
);
40 * Cookies are numeric values sent with CHANGE and REMOVE
41 * uevents while resuming, removing or renaming the device.
43 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
44 #define DM_COOKIE_LENGTH 24
46 static const char *_name
= DM_NAME
;
48 static unsigned int major
= 0;
49 static unsigned int _major
= 0;
51 static DEFINE_IDR(_minor_idr
);
53 static DEFINE_SPINLOCK(_minor_lock
);
55 static void do_deferred_remove(struct work_struct
*w
);
57 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
59 static struct workqueue_struct
*deferred_remove_workqueue
;
63 * One of these is allocated per bio.
66 struct mapped_device
*md
;
70 unsigned long start_time
;
71 spinlock_t endio_lock
;
72 struct dm_stats_aux stats_aux
;
76 * For request-based dm.
77 * One of these is allocated per request.
79 struct dm_rq_target_io
{
80 struct mapped_device
*md
;
82 struct request
*orig
, *clone
;
83 struct kthread_work work
;
89 * For request-based dm - the bio clones we allocate are embedded in these
92 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93 * the bioset is created - this means the bio has to come at the end of the
96 struct dm_rq_clone_bio_info
{
98 struct dm_rq_target_io
*tio
;
102 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
104 if (rq
&& rq
->end_io_data
)
105 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
108 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
110 #define MINOR_ALLOCED ((void *)-1)
113 * Bits for the md->flags field.
115 #define DMF_BLOCK_IO_FOR_SUSPEND 0
116 #define DMF_SUSPENDED 1
118 #define DMF_FREEING 3
119 #define DMF_DELETING 4
120 #define DMF_NOFLUSH_SUSPENDING 5
121 #define DMF_MERGE_IS_OPTIONAL 6
122 #define DMF_DEFERRED_REMOVE 7
123 #define DMF_SUSPENDED_INTERNALLY 8
126 * A dummy definition to make RCU happy.
127 * struct dm_table should never be dereferenced in this file.
134 * Work processed by per-device workqueue.
136 struct mapped_device
{
137 struct srcu_struct io_barrier
;
138 struct mutex suspend_lock
;
143 * The current mapping.
144 * Use dm_get_live_table{_fast} or take suspend_lock for
147 struct dm_table __rcu
*map
;
149 struct list_head table_devices
;
150 struct mutex table_devices_lock
;
154 struct request_queue
*queue
;
156 /* Protect queue and type against concurrent access. */
157 struct mutex type_lock
;
159 struct target_type
*immutable_target_type
;
161 struct gendisk
*disk
;
167 * A list of ios that arrived while we were suspended.
170 wait_queue_head_t wait
;
171 struct work_struct work
;
172 struct bio_list deferred
;
173 spinlock_t deferred_lock
;
176 * Processing queue (flush)
178 struct workqueue_struct
*wq
;
181 * io objects are allocated from here.
192 wait_queue_head_t eventq
;
194 struct list_head uevent_list
;
195 spinlock_t uevent_lock
; /* Protect access to uevent_list */
198 * freeze/thaw support require holding onto a super block
200 struct super_block
*frozen_sb
;
201 struct block_device
*bdev
;
203 /* forced geometry settings */
204 struct hd_geometry geometry
;
206 /* kobject and completion */
207 struct dm_kobject_holder kobj_holder
;
209 /* zero-length flush that will be cloned and submitted to targets */
210 struct bio flush_bio
;
212 /* the number of internal suspends */
213 unsigned internal_suspend_count
;
215 struct dm_stats stats
;
217 struct kthread_worker kworker
;
218 struct task_struct
*kworker_task
;
222 * For mempools pre-allocation at the table loading time.
224 struct dm_md_mempools
{
230 struct table_device
{
231 struct list_head list
;
233 struct dm_dev dm_dev
;
236 #define RESERVED_BIO_BASED_IOS 16
237 #define RESERVED_REQUEST_BASED_IOS 256
238 #define RESERVED_MAX_IOS 1024
239 static struct kmem_cache
*_io_cache
;
240 static struct kmem_cache
*_rq_tio_cache
;
241 static struct kmem_cache
*_rq_cache
;
244 * Bio-based DM's mempools' reserved IOs set by the user.
246 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
249 * Request-based DM's mempools' reserved IOs set by the user.
251 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
253 static unsigned __dm_get_reserved_ios(unsigned *reserved_ios
,
254 unsigned def
, unsigned max
)
256 unsigned ios
= ACCESS_ONCE(*reserved_ios
);
257 unsigned modified_ios
= 0;
265 (void)cmpxchg(reserved_ios
, ios
, modified_ios
);
272 unsigned dm_get_reserved_bio_based_ios(void)
274 return __dm_get_reserved_ios(&reserved_bio_based_ios
,
275 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
277 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
279 unsigned dm_get_reserved_rq_based_ios(void)
281 return __dm_get_reserved_ios(&reserved_rq_based_ios
,
282 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
284 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
286 static int __init
local_init(void)
290 /* allocate a slab for the dm_ios */
291 _io_cache
= KMEM_CACHE(dm_io
, 0);
295 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
297 goto out_free_io_cache
;
299 _rq_cache
= kmem_cache_create("dm_clone_request", sizeof(struct request
),
300 __alignof__(struct request
), 0, NULL
);
302 goto out_free_rq_tio_cache
;
304 r
= dm_uevent_init();
306 goto out_free_rq_cache
;
308 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
309 if (!deferred_remove_workqueue
) {
311 goto out_uevent_exit
;
315 r
= register_blkdev(_major
, _name
);
317 goto out_free_workqueue
;
325 destroy_workqueue(deferred_remove_workqueue
);
329 kmem_cache_destroy(_rq_cache
);
330 out_free_rq_tio_cache
:
331 kmem_cache_destroy(_rq_tio_cache
);
333 kmem_cache_destroy(_io_cache
);
338 static void local_exit(void)
340 flush_scheduled_work();
341 destroy_workqueue(deferred_remove_workqueue
);
343 kmem_cache_destroy(_rq_cache
);
344 kmem_cache_destroy(_rq_tio_cache
);
345 kmem_cache_destroy(_io_cache
);
346 unregister_blkdev(_major
, _name
);
351 DMINFO("cleaned up");
354 static int (*_inits
[])(void) __initdata
= {
365 static void (*_exits
[])(void) = {
376 static int __init
dm_init(void)
378 const int count
= ARRAY_SIZE(_inits
);
382 for (i
= 0; i
< count
; i
++) {
397 static void __exit
dm_exit(void)
399 int i
= ARRAY_SIZE(_exits
);
405 * Should be empty by this point.
407 idr_destroy(&_minor_idr
);
411 * Block device functions
413 int dm_deleting_md(struct mapped_device
*md
)
415 return test_bit(DMF_DELETING
, &md
->flags
);
418 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
420 struct mapped_device
*md
;
422 spin_lock(&_minor_lock
);
424 md
= bdev
->bd_disk
->private_data
;
428 if (test_bit(DMF_FREEING
, &md
->flags
) ||
429 dm_deleting_md(md
)) {
435 atomic_inc(&md
->open_count
);
437 spin_unlock(&_minor_lock
);
439 return md
? 0 : -ENXIO
;
442 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
444 struct mapped_device
*md
;
446 spin_lock(&_minor_lock
);
448 md
= disk
->private_data
;
452 if (atomic_dec_and_test(&md
->open_count
) &&
453 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
454 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
458 spin_unlock(&_minor_lock
);
461 int dm_open_count(struct mapped_device
*md
)
463 return atomic_read(&md
->open_count
);
467 * Guarantees nothing is using the device before it's deleted.
469 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
473 spin_lock(&_minor_lock
);
475 if (dm_open_count(md
)) {
478 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
479 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
482 set_bit(DMF_DELETING
, &md
->flags
);
484 spin_unlock(&_minor_lock
);
489 int dm_cancel_deferred_remove(struct mapped_device
*md
)
493 spin_lock(&_minor_lock
);
495 if (test_bit(DMF_DELETING
, &md
->flags
))
498 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
500 spin_unlock(&_minor_lock
);
505 static void do_deferred_remove(struct work_struct
*w
)
507 dm_deferred_remove();
510 sector_t
dm_get_size(struct mapped_device
*md
)
512 return get_capacity(md
->disk
);
515 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
520 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
525 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
527 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
529 return dm_get_geometry(md
, geo
);
532 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
533 unsigned int cmd
, unsigned long arg
)
535 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
537 struct dm_table
*map
;
538 struct dm_target
*tgt
;
542 map
= dm_get_live_table(md
, &srcu_idx
);
544 if (!map
|| !dm_table_get_size(map
))
547 /* We only support devices that have a single target */
548 if (dm_table_get_num_targets(map
) != 1)
551 tgt
= dm_table_get_target(map
, 0);
552 if (!tgt
->type
->ioctl
)
555 if (dm_suspended_md(md
)) {
560 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
563 dm_put_live_table(md
, srcu_idx
);
565 if (r
== -ENOTCONN
) {
573 static struct dm_io
*alloc_io(struct mapped_device
*md
)
575 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
578 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
580 mempool_free(io
, md
->io_pool
);
583 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
585 bio_put(&tio
->clone
);
588 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
591 return mempool_alloc(md
->io_pool
, gfp_mask
);
594 static void free_rq_tio(struct dm_rq_target_io
*tio
)
596 mempool_free(tio
, tio
->md
->io_pool
);
599 static struct request
*alloc_clone_request(struct mapped_device
*md
,
602 return mempool_alloc(md
->rq_pool
, gfp_mask
);
605 static void free_clone_request(struct mapped_device
*md
, struct request
*rq
)
607 mempool_free(rq
, md
->rq_pool
);
610 static int md_in_flight(struct mapped_device
*md
)
612 return atomic_read(&md
->pending
[READ
]) +
613 atomic_read(&md
->pending
[WRITE
]);
616 static void start_io_acct(struct dm_io
*io
)
618 struct mapped_device
*md
= io
->md
;
619 struct bio
*bio
= io
->bio
;
621 int rw
= bio_data_dir(bio
);
623 io
->start_time
= jiffies
;
625 cpu
= part_stat_lock();
626 part_round_stats(cpu
, &dm_disk(md
)->part0
);
628 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
629 atomic_inc_return(&md
->pending
[rw
]));
631 if (unlikely(dm_stats_used(&md
->stats
)))
632 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
633 bio_sectors(bio
), false, 0, &io
->stats_aux
);
636 static void end_io_acct(struct dm_io
*io
)
638 struct mapped_device
*md
= io
->md
;
639 struct bio
*bio
= io
->bio
;
640 unsigned long duration
= jiffies
- io
->start_time
;
642 int rw
= bio_data_dir(bio
);
644 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
646 if (unlikely(dm_stats_used(&md
->stats
)))
647 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
648 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
651 * After this is decremented the bio must not be touched if it is
654 pending
= atomic_dec_return(&md
->pending
[rw
]);
655 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
656 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
658 /* nudge anyone waiting on suspend queue */
664 * Add the bio to the list of deferred io.
666 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
670 spin_lock_irqsave(&md
->deferred_lock
, flags
);
671 bio_list_add(&md
->deferred
, bio
);
672 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
673 queue_work(md
->wq
, &md
->work
);
677 * Everyone (including functions in this file), should use this
678 * function to access the md->map field, and make sure they call
679 * dm_put_live_table() when finished.
681 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
683 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
685 return srcu_dereference(md
->map
, &md
->io_barrier
);
688 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
690 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
693 void dm_sync_table(struct mapped_device
*md
)
695 synchronize_srcu(&md
->io_barrier
);
696 synchronize_rcu_expedited();
700 * A fast alternative to dm_get_live_table/dm_put_live_table.
701 * The caller must not block between these two functions.
703 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
706 return rcu_dereference(md
->map
);
709 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
715 * Open a table device so we can use it as a map destination.
717 static int open_table_device(struct table_device
*td
, dev_t dev
,
718 struct mapped_device
*md
)
720 static char *_claim_ptr
= "I belong to device-mapper";
721 struct block_device
*bdev
;
725 BUG_ON(td
->dm_dev
.bdev
);
727 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
729 return PTR_ERR(bdev
);
731 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
733 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
737 td
->dm_dev
.bdev
= bdev
;
742 * Close a table device that we've been using.
744 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
746 if (!td
->dm_dev
.bdev
)
749 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
750 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
751 td
->dm_dev
.bdev
= NULL
;
754 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
756 struct table_device
*td
;
758 list_for_each_entry(td
, l
, list
)
759 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
765 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
766 struct dm_dev
**result
) {
768 struct table_device
*td
;
770 mutex_lock(&md
->table_devices_lock
);
771 td
= find_table_device(&md
->table_devices
, dev
, mode
);
773 td
= kmalloc(sizeof(*td
), GFP_KERNEL
);
775 mutex_unlock(&md
->table_devices_lock
);
779 td
->dm_dev
.mode
= mode
;
780 td
->dm_dev
.bdev
= NULL
;
782 if ((r
= open_table_device(td
, dev
, md
))) {
783 mutex_unlock(&md
->table_devices_lock
);
788 format_dev_t(td
->dm_dev
.name
, dev
);
790 atomic_set(&td
->count
, 0);
791 list_add(&td
->list
, &md
->table_devices
);
793 atomic_inc(&td
->count
);
794 mutex_unlock(&md
->table_devices_lock
);
796 *result
= &td
->dm_dev
;
799 EXPORT_SYMBOL_GPL(dm_get_table_device
);
801 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
803 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
805 mutex_lock(&md
->table_devices_lock
);
806 if (atomic_dec_and_test(&td
->count
)) {
807 close_table_device(td
, md
);
811 mutex_unlock(&md
->table_devices_lock
);
813 EXPORT_SYMBOL(dm_put_table_device
);
815 static void free_table_devices(struct list_head
*devices
)
817 struct list_head
*tmp
, *next
;
819 list_for_each_safe(tmp
, next
, devices
) {
820 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
822 DMWARN("dm_destroy: %s still exists with %d references",
823 td
->dm_dev
.name
, atomic_read(&td
->count
));
829 * Get the geometry associated with a dm device
831 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
839 * Set the geometry of a device.
841 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
843 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
845 if (geo
->start
> sz
) {
846 DMWARN("Start sector is beyond the geometry limits.");
855 /*-----------------------------------------------------------------
857 * A more elegant soln is in the works that uses the queue
858 * merge fn, unfortunately there are a couple of changes to
859 * the block layer that I want to make for this. So in the
860 * interests of getting something for people to use I give
861 * you this clearly demarcated crap.
862 *---------------------------------------------------------------*/
864 static int __noflush_suspending(struct mapped_device
*md
)
866 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
870 * Decrements the number of outstanding ios that a bio has been
871 * cloned into, completing the original io if necc.
873 static void dec_pending(struct dm_io
*io
, int error
)
878 struct mapped_device
*md
= io
->md
;
880 /* Push-back supersedes any I/O errors */
881 if (unlikely(error
)) {
882 spin_lock_irqsave(&io
->endio_lock
, flags
);
883 if (!(io
->error
> 0 && __noflush_suspending(md
)))
885 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
888 if (atomic_dec_and_test(&io
->io_count
)) {
889 if (io
->error
== DM_ENDIO_REQUEUE
) {
891 * Target requested pushing back the I/O.
893 spin_lock_irqsave(&md
->deferred_lock
, flags
);
894 if (__noflush_suspending(md
))
895 bio_list_add_head(&md
->deferred
, io
->bio
);
897 /* noflush suspend was interrupted. */
899 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
902 io_error
= io
->error
;
907 if (io_error
== DM_ENDIO_REQUEUE
)
910 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
912 * Preflush done for flush with data, reissue
915 bio
->bi_rw
&= ~REQ_FLUSH
;
918 /* done with normal IO or empty flush */
919 trace_block_bio_complete(md
->queue
, bio
, io_error
);
920 bio_endio(bio
, io_error
);
925 static void disable_write_same(struct mapped_device
*md
)
927 struct queue_limits
*limits
= dm_get_queue_limits(md
);
929 /* device doesn't really support WRITE SAME, disable it */
930 limits
->max_write_same_sectors
= 0;
933 static void clone_endio(struct bio
*bio
, int error
)
936 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
937 struct dm_io
*io
= tio
->io
;
938 struct mapped_device
*md
= tio
->io
->md
;
939 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
941 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
945 r
= endio(tio
->ti
, bio
, error
);
946 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
948 * error and requeue request are handled
952 else if (r
== DM_ENDIO_INCOMPLETE
)
953 /* The target will handle the io */
956 DMWARN("unimplemented target endio return value: %d", r
);
961 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
962 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
963 disable_write_same(md
);
966 dec_pending(io
, error
);
970 * Partial completion handling for request-based dm
972 static void end_clone_bio(struct bio
*clone
, int error
)
974 struct dm_rq_clone_bio_info
*info
=
975 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
976 struct dm_rq_target_io
*tio
= info
->tio
;
977 struct bio
*bio
= info
->orig
;
978 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
984 * An error has already been detected on the request.
985 * Once error occurred, just let clone->end_io() handle
991 * Don't notice the error to the upper layer yet.
992 * The error handling decision is made by the target driver,
993 * when the request is completed.
1000 * I/O for the bio successfully completed.
1001 * Notice the data completion to the upper layer.
1005 * bios are processed from the head of the list.
1006 * So the completing bio should always be rq->bio.
1007 * If it's not, something wrong is happening.
1009 if (tio
->orig
->bio
!= bio
)
1010 DMERR("bio completion is going in the middle of the request");
1013 * Update the original request.
1014 * Do not use blk_end_request() here, because it may complete
1015 * the original request before the clone, and break the ordering.
1017 blk_update_request(tio
->orig
, 0, nr_bytes
);
1021 * Don't touch any member of the md after calling this function because
1022 * the md may be freed in dm_put() at the end of this function.
1023 * Or do dm_get() before calling this function and dm_put() later.
1025 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1027 atomic_dec(&md
->pending
[rw
]);
1029 /* nudge anyone waiting on suspend queue */
1030 if (!md_in_flight(md
))
1034 * Run this off this callpath, as drivers could invoke end_io while
1035 * inside their request_fn (and holding the queue lock). Calling
1036 * back into ->request_fn() could deadlock attempting to grab the
1040 blk_run_queue_async(md
->queue
);
1043 * dm_put() must be at the end of this function. See the comment above
1048 static void free_rq_clone(struct request
*clone
)
1050 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1052 blk_rq_unprep_clone(clone
);
1053 if (clone
->q
&& clone
->q
->mq_ops
)
1054 tio
->ti
->type
->release_clone_rq(clone
);
1056 free_clone_request(tio
->md
, clone
);
1061 * Complete the clone and the original request.
1062 * Must be called without clone's queue lock held,
1063 * see end_clone_request() for more details.
1065 static void dm_end_request(struct request
*clone
, int error
)
1067 int rw
= rq_data_dir(clone
);
1068 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1069 struct mapped_device
*md
= tio
->md
;
1070 struct request
*rq
= tio
->orig
;
1072 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1073 rq
->errors
= clone
->errors
;
1074 rq
->resid_len
= clone
->resid_len
;
1078 * We are using the sense buffer of the original
1080 * So setting the length of the sense data is enough.
1082 rq
->sense_len
= clone
->sense_len
;
1085 free_rq_clone(clone
);
1086 blk_end_request_all(rq
, error
);
1087 rq_completed(md
, rw
, true);
1090 static void dm_unprep_request(struct request
*rq
)
1092 struct dm_rq_target_io
*tio
= rq
->special
;
1093 struct request
*clone
= tio
->clone
;
1096 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1099 free_rq_clone(clone
);
1103 * Requeue the original request of a clone.
1105 static void dm_requeue_unmapped_original_request(struct mapped_device
*md
,
1108 int rw
= rq_data_dir(rq
);
1109 struct request_queue
*q
= rq
->q
;
1110 unsigned long flags
;
1112 dm_unprep_request(rq
);
1114 spin_lock_irqsave(q
->queue_lock
, flags
);
1115 blk_requeue_request(q
, rq
);
1116 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1118 rq_completed(md
, rw
, false);
1121 static void dm_requeue_unmapped_request(struct request
*clone
)
1123 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1125 dm_requeue_unmapped_original_request(tio
->md
, tio
->orig
);
1128 static void __stop_queue(struct request_queue
*q
)
1133 static void stop_queue(struct request_queue
*q
)
1135 unsigned long flags
;
1137 spin_lock_irqsave(q
->queue_lock
, flags
);
1139 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1142 static void __start_queue(struct request_queue
*q
)
1144 if (blk_queue_stopped(q
))
1148 static void start_queue(struct request_queue
*q
)
1150 unsigned long flags
;
1152 spin_lock_irqsave(q
->queue_lock
, flags
);
1154 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1157 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1160 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1161 dm_request_endio_fn rq_end_io
= NULL
;
1164 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1166 if (mapped
&& rq_end_io
)
1167 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1170 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1171 !clone
->q
->limits
.max_write_same_sectors
))
1172 disable_write_same(tio
->md
);
1175 /* The target wants to complete the I/O */
1176 dm_end_request(clone
, r
);
1177 else if (r
== DM_ENDIO_INCOMPLETE
)
1178 /* The target will handle the I/O */
1180 else if (r
== DM_ENDIO_REQUEUE
)
1181 /* The target wants to requeue the I/O */
1182 dm_requeue_unmapped_request(clone
);
1184 DMWARN("unimplemented target endio return value: %d", r
);
1190 * Request completion handler for request-based dm
1192 static void dm_softirq_done(struct request
*rq
)
1195 struct dm_rq_target_io
*tio
= rq
->special
;
1196 struct request
*clone
= tio
->clone
;
1199 blk_end_request_all(rq
, tio
->error
);
1200 rq_completed(tio
->md
, rq_data_dir(rq
), false);
1205 if (rq
->cmd_flags
& REQ_FAILED
)
1208 dm_done(clone
, tio
->error
, mapped
);
1212 * Complete the clone and the original request with the error status
1213 * through softirq context.
1215 static void dm_complete_request(struct request
*rq
, int error
)
1217 struct dm_rq_target_io
*tio
= rq
->special
;
1220 blk_complete_request(rq
);
1224 * Complete the not-mapped clone and the original request with the error status
1225 * through softirq context.
1226 * Target's rq_end_io() function isn't called.
1227 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1229 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1231 rq
->cmd_flags
|= REQ_FAILED
;
1232 dm_complete_request(rq
, error
);
1236 * Called with the clone's queue lock held
1238 static void end_clone_request(struct request
*clone
, int error
)
1240 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1242 if (!clone
->q
->mq_ops
) {
1244 * For just cleaning up the information of the queue in which
1245 * the clone was dispatched.
1246 * The clone is *NOT* freed actually here because it is alloced
1247 * from dm own mempool (REQ_ALLOCED isn't set).
1249 __blk_put_request(clone
->q
, clone
);
1253 * Actual request completion is done in a softirq context which doesn't
1254 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1255 * - another request may be submitted by the upper level driver
1256 * of the stacking during the completion
1257 * - the submission which requires queue lock may be done
1258 * against this clone's queue
1260 dm_complete_request(tio
->orig
, error
);
1264 * Return maximum size of I/O possible at the supplied sector up to the current
1267 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1269 sector_t target_offset
= dm_target_offset(ti
, sector
);
1271 return ti
->len
- target_offset
;
1274 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1276 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1277 sector_t offset
, max_len
;
1280 * Does the target need to split even further?
1282 if (ti
->max_io_len
) {
1283 offset
= dm_target_offset(ti
, sector
);
1284 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1285 max_len
= sector_div(offset
, ti
->max_io_len
);
1287 max_len
= offset
& (ti
->max_io_len
- 1);
1288 max_len
= ti
->max_io_len
- max_len
;
1297 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1299 if (len
> UINT_MAX
) {
1300 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1301 (unsigned long long)len
, UINT_MAX
);
1302 ti
->error
= "Maximum size of target IO is too large";
1306 ti
->max_io_len
= (uint32_t) len
;
1310 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1313 * A target may call dm_accept_partial_bio only from the map routine. It is
1314 * allowed for all bio types except REQ_FLUSH.
1316 * dm_accept_partial_bio informs the dm that the target only wants to process
1317 * additional n_sectors sectors of the bio and the rest of the data should be
1318 * sent in a next bio.
1320 * A diagram that explains the arithmetics:
1321 * +--------------------+---------------+-------+
1323 * +--------------------+---------------+-------+
1325 * <-------------- *tio->len_ptr --------------->
1326 * <------- bi_size ------->
1329 * Region 1 was already iterated over with bio_advance or similar function.
1330 * (it may be empty if the target doesn't use bio_advance)
1331 * Region 2 is the remaining bio size that the target wants to process.
1332 * (it may be empty if region 1 is non-empty, although there is no reason
1334 * The target requires that region 3 is to be sent in the next bio.
1336 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1337 * the partially processed part (the sum of regions 1+2) must be the same for all
1338 * copies of the bio.
1340 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1342 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1343 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1344 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1345 BUG_ON(bi_size
> *tio
->len_ptr
);
1346 BUG_ON(n_sectors
> bi_size
);
1347 *tio
->len_ptr
-= bi_size
- n_sectors
;
1348 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1350 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1352 static void __map_bio(struct dm_target_io
*tio
)
1356 struct mapped_device
*md
;
1357 struct bio
*clone
= &tio
->clone
;
1358 struct dm_target
*ti
= tio
->ti
;
1360 clone
->bi_end_io
= clone_endio
;
1363 * Map the clone. If r == 0 we don't need to do
1364 * anything, the target has assumed ownership of
1367 atomic_inc(&tio
->io
->io_count
);
1368 sector
= clone
->bi_iter
.bi_sector
;
1369 r
= ti
->type
->map(ti
, clone
);
1370 if (r
== DM_MAPIO_REMAPPED
) {
1371 /* the bio has been remapped so dispatch it */
1373 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1374 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1376 generic_make_request(clone
);
1377 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1378 /* error the io and bail out, or requeue it if needed */
1380 dec_pending(tio
->io
, r
);
1383 DMWARN("unimplemented target map return value: %d", r
);
1389 struct mapped_device
*md
;
1390 struct dm_table
*map
;
1394 unsigned sector_count
;
1397 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1399 bio
->bi_iter
.bi_sector
= sector
;
1400 bio
->bi_iter
.bi_size
= to_bytes(len
);
1404 * Creates a bio that consists of range of complete bvecs.
1406 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1407 sector_t sector
, unsigned len
)
1409 struct bio
*clone
= &tio
->clone
;
1411 __bio_clone_fast(clone
, bio
);
1413 if (bio_integrity(bio
))
1414 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1416 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1417 clone
->bi_iter
.bi_size
= to_bytes(len
);
1419 if (bio_integrity(bio
))
1420 bio_integrity_trim(clone
, 0, len
);
1423 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1424 struct dm_target
*ti
,
1425 unsigned target_bio_nr
)
1427 struct dm_target_io
*tio
;
1430 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1431 tio
= container_of(clone
, struct dm_target_io
, clone
);
1435 tio
->target_bio_nr
= target_bio_nr
;
1440 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1441 struct dm_target
*ti
,
1442 unsigned target_bio_nr
, unsigned *len
)
1444 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1445 struct bio
*clone
= &tio
->clone
;
1449 __bio_clone_fast(clone
, ci
->bio
);
1451 bio_setup_sector(clone
, ci
->sector
, *len
);
1456 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1457 unsigned num_bios
, unsigned *len
)
1459 unsigned target_bio_nr
;
1461 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1462 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1465 static int __send_empty_flush(struct clone_info
*ci
)
1467 unsigned target_nr
= 0;
1468 struct dm_target
*ti
;
1470 BUG_ON(bio_has_data(ci
->bio
));
1471 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1472 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1477 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1478 sector_t sector
, unsigned *len
)
1480 struct bio
*bio
= ci
->bio
;
1481 struct dm_target_io
*tio
;
1482 unsigned target_bio_nr
;
1483 unsigned num_target_bios
= 1;
1486 * Does the target want to receive duplicate copies of the bio?
1488 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1489 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1491 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1492 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1494 clone_bio(tio
, bio
, sector
, *len
);
1499 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1501 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1503 return ti
->num_discard_bios
;
1506 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1508 return ti
->num_write_same_bios
;
1511 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1513 static bool is_split_required_for_discard(struct dm_target
*ti
)
1515 return ti
->split_discard_bios
;
1518 static int __send_changing_extent_only(struct clone_info
*ci
,
1519 get_num_bios_fn get_num_bios
,
1520 is_split_required_fn is_split_required
)
1522 struct dm_target
*ti
;
1527 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1528 if (!dm_target_is_valid(ti
))
1532 * Even though the device advertised support for this type of
1533 * request, that does not mean every target supports it, and
1534 * reconfiguration might also have changed that since the
1535 * check was performed.
1537 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1541 if (is_split_required
&& !is_split_required(ti
))
1542 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1544 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1546 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1549 } while (ci
->sector_count
-= len
);
1554 static int __send_discard(struct clone_info
*ci
)
1556 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1557 is_split_required_for_discard
);
1560 static int __send_write_same(struct clone_info
*ci
)
1562 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1566 * Select the correct strategy for processing a non-flush bio.
1568 static int __split_and_process_non_flush(struct clone_info
*ci
)
1570 struct bio
*bio
= ci
->bio
;
1571 struct dm_target
*ti
;
1574 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1575 return __send_discard(ci
);
1576 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1577 return __send_write_same(ci
);
1579 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1580 if (!dm_target_is_valid(ti
))
1583 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1585 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1588 ci
->sector_count
-= len
;
1594 * Entry point to split a bio into clones and submit them to the targets.
1596 static void __split_and_process_bio(struct mapped_device
*md
,
1597 struct dm_table
*map
, struct bio
*bio
)
1599 struct clone_info ci
;
1602 if (unlikely(!map
)) {
1609 ci
.io
= alloc_io(md
);
1611 atomic_set(&ci
.io
->io_count
, 1);
1614 spin_lock_init(&ci
.io
->endio_lock
);
1615 ci
.sector
= bio
->bi_iter
.bi_sector
;
1617 start_io_acct(ci
.io
);
1619 if (bio
->bi_rw
& REQ_FLUSH
) {
1620 ci
.bio
= &ci
.md
->flush_bio
;
1621 ci
.sector_count
= 0;
1622 error
= __send_empty_flush(&ci
);
1623 /* dec_pending submits any data associated with flush */
1626 ci
.sector_count
= bio_sectors(bio
);
1627 while (ci
.sector_count
&& !error
)
1628 error
= __split_and_process_non_flush(&ci
);
1631 /* drop the extra reference count */
1632 dec_pending(ci
.io
, error
);
1634 /*-----------------------------------------------------------------
1636 *---------------------------------------------------------------*/
1638 static int dm_merge_bvec(struct request_queue
*q
,
1639 struct bvec_merge_data
*bvm
,
1640 struct bio_vec
*biovec
)
1642 struct mapped_device
*md
= q
->queuedata
;
1643 struct dm_table
*map
= dm_get_live_table_fast(md
);
1644 struct dm_target
*ti
;
1645 sector_t max_sectors
;
1651 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1652 if (!dm_target_is_valid(ti
))
1656 * Find maximum amount of I/O that won't need splitting
1658 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1659 (sector_t
) queue_max_sectors(q
));
1660 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1661 if (unlikely(max_size
< 0)) /* this shouldn't _ever_ happen */
1665 * merge_bvec_fn() returns number of bytes
1666 * it can accept at this offset
1667 * max is precomputed maximal io size
1669 if (max_size
&& ti
->type
->merge
)
1670 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1672 * If the target doesn't support merge method and some of the devices
1673 * provided their merge_bvec method (we know this by looking for the
1674 * max_hw_sectors that dm_set_device_limits may set), then we can't
1675 * allow bios with multiple vector entries. So always set max_size
1676 * to 0, and the code below allows just one page.
1678 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1682 dm_put_live_table_fast(md
);
1684 * Always allow an entire first page
1686 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1687 max_size
= biovec
->bv_len
;
1693 * The request function that just remaps the bio built up by
1696 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1698 int rw
= bio_data_dir(bio
);
1699 struct mapped_device
*md
= q
->queuedata
;
1701 struct dm_table
*map
;
1703 map
= dm_get_live_table(md
, &srcu_idx
);
1705 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1707 /* if we're suspended, we have to queue this io for later */
1708 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1709 dm_put_live_table(md
, srcu_idx
);
1711 if (bio_rw(bio
) != READA
)
1718 __split_and_process_bio(md
, map
, bio
);
1719 dm_put_live_table(md
, srcu_idx
);
1723 int dm_request_based(struct mapped_device
*md
)
1725 return blk_queue_stackable(md
->queue
);
1728 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1730 struct mapped_device
*md
= q
->queuedata
;
1732 if (dm_request_based(md
))
1733 blk_queue_bio(q
, bio
);
1735 _dm_request(q
, bio
);
1738 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1742 if (blk_queue_io_stat(clone
->q
))
1743 clone
->cmd_flags
|= REQ_IO_STAT
;
1745 clone
->start_time
= jiffies
;
1746 r
= blk_insert_cloned_request(clone
->q
, clone
);
1748 /* must complete clone in terms of original request */
1749 dm_complete_request(rq
, r
);
1752 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1755 struct dm_rq_target_io
*tio
= data
;
1756 struct dm_rq_clone_bio_info
*info
=
1757 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1759 info
->orig
= bio_orig
;
1761 bio
->bi_end_io
= end_clone_bio
;
1766 static int setup_clone(struct request
*clone
, struct request
*rq
,
1767 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1771 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, gfp_mask
,
1772 dm_rq_bio_constructor
, tio
);
1776 clone
->cmd
= rq
->cmd
;
1777 clone
->cmd_len
= rq
->cmd_len
;
1778 clone
->sense
= rq
->sense
;
1779 clone
->end_io
= end_clone_request
;
1780 clone
->end_io_data
= tio
;
1787 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1788 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1790 struct request
*clone
= alloc_clone_request(md
, gfp_mask
);
1795 blk_rq_init(NULL
, clone
);
1796 if (setup_clone(clone
, rq
, tio
, gfp_mask
)) {
1798 free_clone_request(md
, clone
);
1805 static void map_tio_request(struct kthread_work
*work
);
1807 static struct dm_rq_target_io
*prep_tio(struct request
*rq
,
1808 struct mapped_device
*md
, gfp_t gfp_mask
)
1810 struct dm_rq_target_io
*tio
;
1812 struct dm_table
*table
;
1814 tio
= alloc_rq_tio(md
, gfp_mask
);
1823 memset(&tio
->info
, 0, sizeof(tio
->info
));
1824 init_kthread_work(&tio
->work
, map_tio_request
);
1826 table
= dm_get_live_table(md
, &srcu_idx
);
1827 if (!dm_table_mq_request_based(table
)) {
1828 if (!clone_rq(rq
, md
, tio
, gfp_mask
)) {
1829 dm_put_live_table(md
, srcu_idx
);
1834 dm_put_live_table(md
, srcu_idx
);
1840 * Called with the queue lock held.
1842 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1844 struct mapped_device
*md
= q
->queuedata
;
1845 struct dm_rq_target_io
*tio
;
1847 if (unlikely(rq
->special
)) {
1848 DMWARN("Already has something in rq->special.");
1849 return BLKPREP_KILL
;
1852 tio
= prep_tio(rq
, md
, GFP_ATOMIC
);
1854 return BLKPREP_DEFER
;
1857 rq
->cmd_flags
|= REQ_DONTPREP
;
1864 * 0 : the request has been processed
1865 * DM_MAPIO_REQUEUE : the original request needs to be requeued
1866 * < 0 : the request was completed due to failure
1868 static int map_request(struct dm_target
*ti
, struct request
*rq
,
1869 struct mapped_device
*md
)
1872 struct dm_rq_target_io
*tio
= rq
->special
;
1873 struct request
*clone
= NULL
;
1877 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1879 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
1881 /* The target wants to complete the I/O */
1882 dm_kill_unmapped_request(rq
, r
);
1886 return DM_MAPIO_REQUEUE
;
1887 if (setup_clone(clone
, rq
, tio
, GFP_KERNEL
)) {
1889 ti
->type
->release_clone_rq(clone
);
1890 return DM_MAPIO_REQUEUE
;
1895 case DM_MAPIO_SUBMITTED
:
1896 /* The target has taken the I/O to submit by itself later */
1898 case DM_MAPIO_REMAPPED
:
1899 /* The target has remapped the I/O so dispatch it */
1900 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1902 dm_dispatch_clone_request(clone
, rq
);
1904 case DM_MAPIO_REQUEUE
:
1905 /* The target wants to requeue the I/O */
1906 dm_requeue_unmapped_request(clone
);
1910 DMWARN("unimplemented target map return value: %d", r
);
1914 /* The target wants to complete the I/O */
1915 dm_kill_unmapped_request(rq
, r
);
1922 static void map_tio_request(struct kthread_work
*work
)
1924 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
1925 struct request
*rq
= tio
->orig
;
1926 struct mapped_device
*md
= tio
->md
;
1928 if (map_request(tio
->ti
, rq
, md
) == DM_MAPIO_REQUEUE
)
1929 dm_requeue_unmapped_original_request(md
, rq
);
1932 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1934 blk_start_request(orig
);
1935 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
1938 * Hold the md reference here for the in-flight I/O.
1939 * We can't rely on the reference count by device opener,
1940 * because the device may be closed during the request completion
1941 * when all bios are completed.
1942 * See the comment in rq_completed() too.
1948 * q->request_fn for request-based dm.
1949 * Called with the queue lock held.
1951 static void dm_request_fn(struct request_queue
*q
)
1953 struct mapped_device
*md
= q
->queuedata
;
1955 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
1956 struct dm_target
*ti
;
1958 struct dm_rq_target_io
*tio
;
1962 * For suspend, check blk_queue_stopped() and increment
1963 * ->pending within a single queue_lock not to increment the
1964 * number of in-flight I/Os after the queue is stopped in
1967 while (!blk_queue_stopped(q
)) {
1968 rq
= blk_peek_request(q
);
1972 /* always use block 0 to find the target for flushes for now */
1974 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1975 pos
= blk_rq_pos(rq
);
1977 ti
= dm_table_find_target(map
, pos
);
1978 if (!dm_target_is_valid(ti
)) {
1980 * Must perform setup, that rq_completed() requires,
1981 * before calling dm_kill_unmapped_request
1983 DMERR_LIMIT("request attempted access beyond the end of device");
1984 dm_start_request(md
, rq
);
1985 dm_kill_unmapped_request(rq
, -EIO
);
1989 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1992 dm_start_request(md
, rq
);
1995 /* Establish tio->ti before queuing work (map_tio_request) */
1997 queue_kthread_work(&md
->kworker
, &tio
->work
);
1998 BUG_ON(!irqs_disabled());
2004 blk_delay_queue(q
, HZ
/ 10);
2006 dm_put_live_table(md
, srcu_idx
);
2009 int dm_underlying_device_busy(struct request_queue
*q
)
2011 return blk_lld_busy(q
);
2013 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
2015 static int dm_lld_busy(struct request_queue
*q
)
2018 struct mapped_device
*md
= q
->queuedata
;
2019 struct dm_table
*map
= dm_get_live_table_fast(md
);
2021 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
2024 r
= dm_table_any_busy_target(map
);
2026 dm_put_live_table_fast(md
);
2031 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2034 struct mapped_device
*md
= congested_data
;
2035 struct dm_table
*map
;
2037 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2038 map
= dm_get_live_table_fast(md
);
2041 * Request-based dm cares about only own queue for
2042 * the query about congestion status of request_queue
2044 if (dm_request_based(md
))
2045 r
= md
->queue
->backing_dev_info
.state
&
2048 r
= dm_table_any_congested(map
, bdi_bits
);
2050 dm_put_live_table_fast(md
);
2056 /*-----------------------------------------------------------------
2057 * An IDR is used to keep track of allocated minor numbers.
2058 *---------------------------------------------------------------*/
2059 static void free_minor(int minor
)
2061 spin_lock(&_minor_lock
);
2062 idr_remove(&_minor_idr
, minor
);
2063 spin_unlock(&_minor_lock
);
2067 * See if the device with a specific minor # is free.
2069 static int specific_minor(int minor
)
2073 if (minor
>= (1 << MINORBITS
))
2076 idr_preload(GFP_KERNEL
);
2077 spin_lock(&_minor_lock
);
2079 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2081 spin_unlock(&_minor_lock
);
2084 return r
== -ENOSPC
? -EBUSY
: r
;
2088 static int next_free_minor(int *minor
)
2092 idr_preload(GFP_KERNEL
);
2093 spin_lock(&_minor_lock
);
2095 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2097 spin_unlock(&_minor_lock
);
2105 static const struct block_device_operations dm_blk_dops
;
2107 static void dm_wq_work(struct work_struct
*work
);
2109 static void dm_init_md_queue(struct mapped_device
*md
)
2112 * Request-based dm devices cannot be stacked on top of bio-based dm
2113 * devices. The type of this dm device has not been decided yet.
2114 * The type is decided at the first table loading time.
2115 * To prevent problematic device stacking, clear the queue flag
2116 * for request stacking support until then.
2118 * This queue is new, so no concurrency on the queue_flags.
2120 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2122 md
->queue
->queuedata
= md
;
2123 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2124 md
->queue
->backing_dev_info
.congested_data
= md
;
2125 blk_queue_make_request(md
->queue
, dm_request
);
2126 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2127 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
2131 * Allocate and initialise a blank device with a given minor.
2133 static struct mapped_device
*alloc_dev(int minor
)
2136 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
2140 DMWARN("unable to allocate device, out of memory.");
2144 if (!try_module_get(THIS_MODULE
))
2145 goto bad_module_get
;
2147 /* get a minor number for the dev */
2148 if (minor
== DM_ANY_MINOR
)
2149 r
= next_free_minor(&minor
);
2151 r
= specific_minor(minor
);
2155 r
= init_srcu_struct(&md
->io_barrier
);
2157 goto bad_io_barrier
;
2159 md
->type
= DM_TYPE_NONE
;
2160 mutex_init(&md
->suspend_lock
);
2161 mutex_init(&md
->type_lock
);
2162 mutex_init(&md
->table_devices_lock
);
2163 spin_lock_init(&md
->deferred_lock
);
2164 atomic_set(&md
->holders
, 1);
2165 atomic_set(&md
->open_count
, 0);
2166 atomic_set(&md
->event_nr
, 0);
2167 atomic_set(&md
->uevent_seq
, 0);
2168 INIT_LIST_HEAD(&md
->uevent_list
);
2169 INIT_LIST_HEAD(&md
->table_devices
);
2170 spin_lock_init(&md
->uevent_lock
);
2172 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
2176 dm_init_md_queue(md
);
2178 md
->disk
= alloc_disk(1);
2182 atomic_set(&md
->pending
[0], 0);
2183 atomic_set(&md
->pending
[1], 0);
2184 init_waitqueue_head(&md
->wait
);
2185 INIT_WORK(&md
->work
, dm_wq_work
);
2186 init_waitqueue_head(&md
->eventq
);
2187 init_completion(&md
->kobj_holder
.completion
);
2188 md
->kworker_task
= NULL
;
2190 md
->disk
->major
= _major
;
2191 md
->disk
->first_minor
= minor
;
2192 md
->disk
->fops
= &dm_blk_dops
;
2193 md
->disk
->queue
= md
->queue
;
2194 md
->disk
->private_data
= md
;
2195 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2197 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2199 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2203 md
->bdev
= bdget_disk(md
->disk
, 0);
2207 bio_init(&md
->flush_bio
);
2208 md
->flush_bio
.bi_bdev
= md
->bdev
;
2209 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2211 dm_stats_init(&md
->stats
);
2213 /* Populate the mapping, nobody knows we exist yet */
2214 spin_lock(&_minor_lock
);
2215 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2216 spin_unlock(&_minor_lock
);
2218 BUG_ON(old_md
!= MINOR_ALLOCED
);
2223 destroy_workqueue(md
->wq
);
2225 del_gendisk(md
->disk
);
2228 blk_cleanup_queue(md
->queue
);
2230 cleanup_srcu_struct(&md
->io_barrier
);
2234 module_put(THIS_MODULE
);
2240 static void unlock_fs(struct mapped_device
*md
);
2242 static void free_dev(struct mapped_device
*md
)
2244 int minor
= MINOR(disk_devt(md
->disk
));
2247 destroy_workqueue(md
->wq
);
2249 if (md
->kworker_task
)
2250 kthread_stop(md
->kworker_task
);
2252 mempool_destroy(md
->io_pool
);
2254 mempool_destroy(md
->rq_pool
);
2256 bioset_free(md
->bs
);
2258 cleanup_srcu_struct(&md
->io_barrier
);
2259 free_table_devices(&md
->table_devices
);
2260 dm_stats_cleanup(&md
->stats
);
2262 spin_lock(&_minor_lock
);
2263 md
->disk
->private_data
= NULL
;
2264 spin_unlock(&_minor_lock
);
2265 if (blk_get_integrity(md
->disk
))
2266 blk_integrity_unregister(md
->disk
);
2267 del_gendisk(md
->disk
);
2269 blk_cleanup_queue(md
->queue
);
2273 module_put(THIS_MODULE
);
2277 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2279 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2281 if (md
->io_pool
&& md
->bs
) {
2282 /* The md already has necessary mempools. */
2283 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2285 * Reload bioset because front_pad may have changed
2286 * because a different table was loaded.
2288 bioset_free(md
->bs
);
2293 * There's no need to reload with request-based dm
2294 * because the size of front_pad doesn't change.
2295 * Note for future: If you are to reload bioset,
2296 * prep-ed requests in the queue may refer
2297 * to bio from the old bioset, so you must walk
2298 * through the queue to unprep.
2303 BUG_ON(!p
|| md
->io_pool
|| md
->rq_pool
|| md
->bs
);
2305 md
->io_pool
= p
->io_pool
;
2307 md
->rq_pool
= p
->rq_pool
;
2313 /* mempool bind completed, now no need any mempools in the table */
2314 dm_table_free_md_mempools(t
);
2318 * Bind a table to the device.
2320 static void event_callback(void *context
)
2322 unsigned long flags
;
2324 struct mapped_device
*md
= (struct mapped_device
*) context
;
2326 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2327 list_splice_init(&md
->uevent_list
, &uevents
);
2328 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2330 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2332 atomic_inc(&md
->event_nr
);
2333 wake_up(&md
->eventq
);
2337 * Protected by md->suspend_lock obtained by dm_swap_table().
2339 static void __set_size(struct mapped_device
*md
, sector_t size
)
2341 set_capacity(md
->disk
, size
);
2343 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2347 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2349 * If this function returns 0, then the device is either a non-dm
2350 * device without a merge_bvec_fn, or it is a dm device that is
2351 * able to split any bios it receives that are too big.
2353 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2355 struct mapped_device
*dev_md
;
2357 if (!q
->merge_bvec_fn
)
2360 if (q
->make_request_fn
== dm_request
) {
2361 dev_md
= q
->queuedata
;
2362 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2369 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2370 struct dm_dev
*dev
, sector_t start
,
2371 sector_t len
, void *data
)
2373 struct block_device
*bdev
= dev
->bdev
;
2374 struct request_queue
*q
= bdev_get_queue(bdev
);
2376 return dm_queue_merge_is_compulsory(q
);
2380 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2381 * on the properties of the underlying devices.
2383 static int dm_table_merge_is_optional(struct dm_table
*table
)
2386 struct dm_target
*ti
;
2388 while (i
< dm_table_get_num_targets(table
)) {
2389 ti
= dm_table_get_target(table
, i
++);
2391 if (ti
->type
->iterate_devices
&&
2392 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2400 * Returns old map, which caller must destroy.
2402 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2403 struct queue_limits
*limits
)
2405 struct dm_table
*old_map
;
2406 struct request_queue
*q
= md
->queue
;
2408 int merge_is_optional
;
2410 size
= dm_table_get_size(t
);
2413 * Wipe any geometry if the size of the table changed.
2415 if (size
!= dm_get_size(md
))
2416 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2418 __set_size(md
, size
);
2420 dm_table_event_callback(t
, event_callback
, md
);
2423 * The queue hasn't been stopped yet, if the old table type wasn't
2424 * for request-based during suspension. So stop it to prevent
2425 * I/O mapping before resume.
2426 * This must be done before setting the queue restrictions,
2427 * because request-based dm may be run just after the setting.
2429 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2432 __bind_mempools(md
, t
);
2434 merge_is_optional
= dm_table_merge_is_optional(t
);
2436 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2437 rcu_assign_pointer(md
->map
, t
);
2438 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2440 dm_table_set_restrictions(t
, q
, limits
);
2441 if (merge_is_optional
)
2442 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2444 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2452 * Returns unbound table for the caller to free.
2454 static struct dm_table
*__unbind(struct mapped_device
*md
)
2456 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2461 dm_table_event_callback(map
, NULL
, NULL
);
2462 RCU_INIT_POINTER(md
->map
, NULL
);
2469 * Constructor for a new device.
2471 int dm_create(int minor
, struct mapped_device
**result
)
2473 struct mapped_device
*md
;
2475 md
= alloc_dev(minor
);
2486 * Functions to manage md->type.
2487 * All are required to hold md->type_lock.
2489 void dm_lock_md_type(struct mapped_device
*md
)
2491 mutex_lock(&md
->type_lock
);
2494 void dm_unlock_md_type(struct mapped_device
*md
)
2496 mutex_unlock(&md
->type_lock
);
2499 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2501 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2505 unsigned dm_get_md_type(struct mapped_device
*md
)
2507 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2511 static bool dm_md_type_request_based(struct mapped_device
*md
)
2513 unsigned table_type
= dm_get_md_type(md
);
2515 return (table_type
== DM_TYPE_REQUEST_BASED
||
2516 table_type
== DM_TYPE_MQ_REQUEST_BASED
);
2519 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2521 return md
->immutable_target_type
;
2525 * The queue_limits are only valid as long as you have a reference
2528 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2530 BUG_ON(!atomic_read(&md
->holders
));
2531 return &md
->queue
->limits
;
2533 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2536 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2538 static int dm_init_request_based_queue(struct mapped_device
*md
)
2540 struct request_queue
*q
= NULL
;
2542 if (md
->queue
->elevator
)
2545 /* Fully initialize the queue */
2546 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2551 dm_init_md_queue(md
);
2552 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2553 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2554 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2556 /* Also initialize the request-based DM worker thread */
2557 init_kthread_worker(&md
->kworker
);
2558 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2559 "kdmwork-%s", dm_device_name(md
));
2561 elv_register_queue(md
->queue
);
2567 * Setup the DM device's queue based on md's type
2569 int dm_setup_md_queue(struct mapped_device
*md
)
2571 if (dm_md_type_request_based(md
) && !dm_init_request_based_queue(md
)) {
2572 DMWARN("Cannot initialize queue for request-based mapped device");
2579 struct mapped_device
*dm_get_md(dev_t dev
)
2581 struct mapped_device
*md
;
2582 unsigned minor
= MINOR(dev
);
2584 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2587 spin_lock(&_minor_lock
);
2589 md
= idr_find(&_minor_idr
, minor
);
2591 if ((md
== MINOR_ALLOCED
||
2592 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2593 dm_deleting_md(md
) ||
2594 test_bit(DMF_FREEING
, &md
->flags
))) {
2602 spin_unlock(&_minor_lock
);
2606 EXPORT_SYMBOL_GPL(dm_get_md
);
2608 void *dm_get_mdptr(struct mapped_device
*md
)
2610 return md
->interface_ptr
;
2613 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2615 md
->interface_ptr
= ptr
;
2618 void dm_get(struct mapped_device
*md
)
2620 atomic_inc(&md
->holders
);
2621 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2624 int dm_hold(struct mapped_device
*md
)
2626 spin_lock(&_minor_lock
);
2627 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2628 spin_unlock(&_minor_lock
);
2632 spin_unlock(&_minor_lock
);
2635 EXPORT_SYMBOL_GPL(dm_hold
);
2637 const char *dm_device_name(struct mapped_device
*md
)
2641 EXPORT_SYMBOL_GPL(dm_device_name
);
2643 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2645 struct dm_table
*map
;
2650 map
= dm_get_live_table(md
, &srcu_idx
);
2652 spin_lock(&_minor_lock
);
2653 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2654 set_bit(DMF_FREEING
, &md
->flags
);
2655 spin_unlock(&_minor_lock
);
2657 if (dm_request_based(md
))
2658 flush_kthread_worker(&md
->kworker
);
2661 * Take suspend_lock so that presuspend and postsuspend methods
2662 * do not race with internal suspend.
2664 mutex_lock(&md
->suspend_lock
);
2665 if (!dm_suspended_md(md
)) {
2666 dm_table_presuspend_targets(map
);
2667 dm_table_postsuspend_targets(map
);
2669 mutex_unlock(&md
->suspend_lock
);
2671 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2672 dm_put_live_table(md
, srcu_idx
);
2675 * Rare, but there may be I/O requests still going to complete,
2676 * for example. Wait for all references to disappear.
2677 * No one should increment the reference count of the mapped_device,
2678 * after the mapped_device state becomes DMF_FREEING.
2681 while (atomic_read(&md
->holders
))
2683 else if (atomic_read(&md
->holders
))
2684 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2685 dm_device_name(md
), atomic_read(&md
->holders
));
2688 dm_table_destroy(__unbind(md
));
2692 void dm_destroy(struct mapped_device
*md
)
2694 __dm_destroy(md
, true);
2697 void dm_destroy_immediate(struct mapped_device
*md
)
2699 __dm_destroy(md
, false);
2702 void dm_put(struct mapped_device
*md
)
2704 atomic_dec(&md
->holders
);
2706 EXPORT_SYMBOL_GPL(dm_put
);
2708 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2711 DECLARE_WAITQUEUE(wait
, current
);
2713 add_wait_queue(&md
->wait
, &wait
);
2716 set_current_state(interruptible
);
2718 if (!md_in_flight(md
))
2721 if (interruptible
== TASK_INTERRUPTIBLE
&&
2722 signal_pending(current
)) {
2729 set_current_state(TASK_RUNNING
);
2731 remove_wait_queue(&md
->wait
, &wait
);
2737 * Process the deferred bios
2739 static void dm_wq_work(struct work_struct
*work
)
2741 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2745 struct dm_table
*map
;
2747 map
= dm_get_live_table(md
, &srcu_idx
);
2749 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2750 spin_lock_irq(&md
->deferred_lock
);
2751 c
= bio_list_pop(&md
->deferred
);
2752 spin_unlock_irq(&md
->deferred_lock
);
2757 if (dm_request_based(md
))
2758 generic_make_request(c
);
2760 __split_and_process_bio(md
, map
, c
);
2763 dm_put_live_table(md
, srcu_idx
);
2766 static void dm_queue_flush(struct mapped_device
*md
)
2768 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2769 smp_mb__after_atomic();
2770 queue_work(md
->wq
, &md
->work
);
2774 * Swap in a new table, returning the old one for the caller to destroy.
2776 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2778 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2779 struct queue_limits limits
;
2782 mutex_lock(&md
->suspend_lock
);
2784 /* device must be suspended */
2785 if (!dm_suspended_md(md
))
2789 * If the new table has no data devices, retain the existing limits.
2790 * This helps multipath with queue_if_no_path if all paths disappear,
2791 * then new I/O is queued based on these limits, and then some paths
2794 if (dm_table_has_no_data_devices(table
)) {
2795 live_map
= dm_get_live_table_fast(md
);
2797 limits
= md
->queue
->limits
;
2798 dm_put_live_table_fast(md
);
2802 r
= dm_calculate_queue_limits(table
, &limits
);
2809 map
= __bind(md
, table
, &limits
);
2812 mutex_unlock(&md
->suspend_lock
);
2817 * Functions to lock and unlock any filesystem running on the
2820 static int lock_fs(struct mapped_device
*md
)
2824 WARN_ON(md
->frozen_sb
);
2826 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2827 if (IS_ERR(md
->frozen_sb
)) {
2828 r
= PTR_ERR(md
->frozen_sb
);
2829 md
->frozen_sb
= NULL
;
2833 set_bit(DMF_FROZEN
, &md
->flags
);
2838 static void unlock_fs(struct mapped_device
*md
)
2840 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2843 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2844 md
->frozen_sb
= NULL
;
2845 clear_bit(DMF_FROZEN
, &md
->flags
);
2849 * If __dm_suspend returns 0, the device is completely quiescent
2850 * now. There is no request-processing activity. All new requests
2851 * are being added to md->deferred list.
2853 * Caller must hold md->suspend_lock
2855 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2856 unsigned suspend_flags
, int interruptible
)
2858 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2859 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2863 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2864 * This flag is cleared before dm_suspend returns.
2867 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2870 * This gets reverted if there's an error later and the targets
2871 * provide the .presuspend_undo hook.
2873 dm_table_presuspend_targets(map
);
2876 * Flush I/O to the device.
2877 * Any I/O submitted after lock_fs() may not be flushed.
2878 * noflush takes precedence over do_lockfs.
2879 * (lock_fs() flushes I/Os and waits for them to complete.)
2881 if (!noflush
&& do_lockfs
) {
2884 dm_table_presuspend_undo_targets(map
);
2890 * Here we must make sure that no processes are submitting requests
2891 * to target drivers i.e. no one may be executing
2892 * __split_and_process_bio. This is called from dm_request and
2895 * To get all processes out of __split_and_process_bio in dm_request,
2896 * we take the write lock. To prevent any process from reentering
2897 * __split_and_process_bio from dm_request and quiesce the thread
2898 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2899 * flush_workqueue(md->wq).
2901 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2903 synchronize_srcu(&md
->io_barrier
);
2906 * Stop md->queue before flushing md->wq in case request-based
2907 * dm defers requests to md->wq from md->queue.
2909 if (dm_request_based(md
)) {
2910 stop_queue(md
->queue
);
2911 flush_kthread_worker(&md
->kworker
);
2914 flush_workqueue(md
->wq
);
2917 * At this point no more requests are entering target request routines.
2918 * We call dm_wait_for_completion to wait for all existing requests
2921 r
= dm_wait_for_completion(md
, interruptible
);
2924 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2926 synchronize_srcu(&md
->io_barrier
);
2928 /* were we interrupted ? */
2932 if (dm_request_based(md
))
2933 start_queue(md
->queue
);
2936 dm_table_presuspend_undo_targets(map
);
2937 /* pushback list is already flushed, so skip flush */
2944 * We need to be able to change a mapping table under a mounted
2945 * filesystem. For example we might want to move some data in
2946 * the background. Before the table can be swapped with
2947 * dm_bind_table, dm_suspend must be called to flush any in
2948 * flight bios and ensure that any further io gets deferred.
2951 * Suspend mechanism in request-based dm.
2953 * 1. Flush all I/Os by lock_fs() if needed.
2954 * 2. Stop dispatching any I/O by stopping the request_queue.
2955 * 3. Wait for all in-flight I/Os to be completed or requeued.
2957 * To abort suspend, start the request_queue.
2959 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2961 struct dm_table
*map
= NULL
;
2965 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2967 if (dm_suspended_md(md
)) {
2972 if (dm_suspended_internally_md(md
)) {
2973 /* already internally suspended, wait for internal resume */
2974 mutex_unlock(&md
->suspend_lock
);
2975 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2981 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2983 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
2987 set_bit(DMF_SUSPENDED
, &md
->flags
);
2989 dm_table_postsuspend_targets(map
);
2992 mutex_unlock(&md
->suspend_lock
);
2996 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2999 int r
= dm_table_resume_targets(map
);
3007 * Flushing deferred I/Os must be done after targets are resumed
3008 * so that mapping of targets can work correctly.
3009 * Request-based dm is queueing the deferred I/Os in its request_queue.
3011 if (dm_request_based(md
))
3012 start_queue(md
->queue
);
3019 int dm_resume(struct mapped_device
*md
)
3022 struct dm_table
*map
= NULL
;
3025 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3027 if (!dm_suspended_md(md
))
3030 if (dm_suspended_internally_md(md
)) {
3031 /* already internally suspended, wait for internal resume */
3032 mutex_unlock(&md
->suspend_lock
);
3033 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3039 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3040 if (!map
|| !dm_table_get_size(map
))
3043 r
= __dm_resume(md
, map
);
3047 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3051 mutex_unlock(&md
->suspend_lock
);
3057 * Internal suspend/resume works like userspace-driven suspend. It waits
3058 * until all bios finish and prevents issuing new bios to the target drivers.
3059 * It may be used only from the kernel.
3062 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3064 struct dm_table
*map
= NULL
;
3066 if (md
->internal_suspend_count
++)
3067 return; /* nested internal suspend */
3069 if (dm_suspended_md(md
)) {
3070 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3071 return; /* nest suspend */
3074 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3077 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3078 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3079 * would require changing .presuspend to return an error -- avoid this
3080 * until there is a need for more elaborate variants of internal suspend.
3082 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3084 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3086 dm_table_postsuspend_targets(map
);
3089 static void __dm_internal_resume(struct mapped_device
*md
)
3091 BUG_ON(!md
->internal_suspend_count
);
3093 if (--md
->internal_suspend_count
)
3094 return; /* resume from nested internal suspend */
3096 if (dm_suspended_md(md
))
3097 goto done
; /* resume from nested suspend */
3100 * NOTE: existing callers don't need to call dm_table_resume_targets
3101 * (which may fail -- so best to avoid it for now by passing NULL map)
3103 (void) __dm_resume(md
, NULL
);
3106 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3107 smp_mb__after_atomic();
3108 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3111 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3113 mutex_lock(&md
->suspend_lock
);
3114 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3115 mutex_unlock(&md
->suspend_lock
);
3117 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3119 void dm_internal_resume(struct mapped_device
*md
)
3121 mutex_lock(&md
->suspend_lock
);
3122 __dm_internal_resume(md
);
3123 mutex_unlock(&md
->suspend_lock
);
3125 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3128 * Fast variants of internal suspend/resume hold md->suspend_lock,
3129 * which prevents interaction with userspace-driven suspend.
3132 void dm_internal_suspend_fast(struct mapped_device
*md
)
3134 mutex_lock(&md
->suspend_lock
);
3135 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3138 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3139 synchronize_srcu(&md
->io_barrier
);
3140 flush_workqueue(md
->wq
);
3141 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3143 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
3145 void dm_internal_resume_fast(struct mapped_device
*md
)
3147 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3153 mutex_unlock(&md
->suspend_lock
);
3155 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
3157 /*-----------------------------------------------------------------
3158 * Event notification.
3159 *---------------------------------------------------------------*/
3160 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3163 char udev_cookie
[DM_COOKIE_LENGTH
];
3164 char *envp
[] = { udev_cookie
, NULL
};
3167 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3169 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3170 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3171 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3176 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3178 return atomic_add_return(1, &md
->uevent_seq
);
3181 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3183 return atomic_read(&md
->event_nr
);
3186 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3188 return wait_event_interruptible(md
->eventq
,
3189 (event_nr
!= atomic_read(&md
->event_nr
)));
3192 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3194 unsigned long flags
;
3196 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3197 list_add(elist
, &md
->uevent_list
);
3198 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3202 * The gendisk is only valid as long as you have a reference
3205 struct gendisk
*dm_disk(struct mapped_device
*md
)
3210 struct kobject
*dm_kobject(struct mapped_device
*md
)
3212 return &md
->kobj_holder
.kobj
;
3215 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3217 struct mapped_device
*md
;
3219 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3221 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3229 int dm_suspended_md(struct mapped_device
*md
)
3231 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3234 int dm_suspended_internally_md(struct mapped_device
*md
)
3236 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3239 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3241 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3244 int dm_suspended(struct dm_target
*ti
)
3246 return dm_suspended_md(dm_table_get_md(ti
->table
));
3248 EXPORT_SYMBOL_GPL(dm_suspended
);
3250 int dm_noflush_suspending(struct dm_target
*ti
)
3252 return __noflush_suspending(dm_table_get_md(ti
->table
));
3254 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3256 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
3258 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
3259 struct kmem_cache
*cachep
;
3260 unsigned int pool_size
= 0;
3261 unsigned int front_pad
;
3267 case DM_TYPE_BIO_BASED
:
3269 pool_size
= dm_get_reserved_bio_based_ios();
3270 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3272 case DM_TYPE_REQUEST_BASED
:
3273 pool_size
= dm_get_reserved_rq_based_ios();
3274 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3275 if (!pools
->rq_pool
)
3277 /* fall through to setup remaining rq-based pools */
3278 case DM_TYPE_MQ_REQUEST_BASED
:
3279 cachep
= _rq_tio_cache
;
3281 pool_size
= dm_get_reserved_rq_based_ios();
3282 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3283 /* per_bio_data_size is not used. See __bind_mempools(). */
3284 WARN_ON(per_bio_data_size
!= 0);
3290 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3291 if (!pools
->io_pool
)
3294 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3298 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3304 dm_free_md_mempools(pools
);
3309 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3315 mempool_destroy(pools
->io_pool
);
3318 mempool_destroy(pools
->rq_pool
);
3321 bioset_free(pools
->bs
);
3326 static const struct block_device_operations dm_blk_dops
= {
3327 .open
= dm_blk_open
,
3328 .release
= dm_blk_close
,
3329 .ioctl
= dm_blk_ioctl
,
3330 .getgeo
= dm_blk_getgeo
,
3331 .owner
= THIS_MODULE
3337 module_init(dm_init
);
3338 module_exit(dm_exit
);
3340 module_param(major
, uint
, 0);
3341 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3343 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3344 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3346 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3347 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3349 MODULE_DESCRIPTION(DM_NAME
" driver");
3350 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3351 MODULE_LICENSE("GPL");