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/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
22 #include <linux/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
29 * Cookies are numeric values sent with CHANGE and REMOVE
30 * uevents while resuming, removing or renaming the device.
32 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
33 #define DM_COOKIE_LENGTH 24
35 static const char *_name
= DM_NAME
;
37 static unsigned int major
= 0;
38 static unsigned int _major
= 0;
40 static DEFINE_SPINLOCK(_minor_lock
);
43 * One of these is allocated per bio.
46 struct mapped_device
*md
;
50 unsigned long start_time
;
51 spinlock_t endio_lock
;
56 * One of these is allocated per target within a bio. Hopefully
57 * this will be simplified out one day.
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io
{
70 struct mapped_device
*md
;
72 struct request
*orig
, clone
;
78 * For request-based dm.
79 * One of these is allocated per bio.
81 struct dm_rq_clone_bio_info
{
83 struct dm_rq_target_io
*tio
;
86 union map_info
*dm_get_mapinfo(struct bio
*bio
)
88 if (bio
&& bio
->bi_private
)
89 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
93 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
95 if (rq
&& rq
->end_io_data
)
96 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
99 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
101 #define MINOR_ALLOCED ((void *)-1)
104 * Bits for the md->flags field.
106 #define DMF_BLOCK_IO_FOR_SUSPEND 0
107 #define DMF_SUSPENDED 1
109 #define DMF_FREEING 3
110 #define DMF_DELETING 4
111 #define DMF_NOFLUSH_SUSPENDING 5
114 * Work processed by per-device workqueue.
116 struct mapped_device
{
117 struct rw_semaphore io_lock
;
118 struct mutex suspend_lock
;
125 struct request_queue
*queue
;
127 /* Protect queue and type against concurrent access. */
128 struct mutex type_lock
;
130 struct gendisk
*disk
;
136 * A list of ios that arrived while we were suspended.
139 wait_queue_head_t wait
;
140 struct work_struct work
;
141 struct bio_list deferred
;
142 spinlock_t deferred_lock
;
145 * Processing queue (flush)
147 struct workqueue_struct
*wq
;
150 * The current mapping.
152 struct dm_table
*map
;
155 * io objects are allocated from here.
166 wait_queue_head_t eventq
;
168 struct list_head uevent_list
;
169 spinlock_t uevent_lock
; /* Protect access to uevent_list */
172 * freeze/thaw support require holding onto a super block
174 struct super_block
*frozen_sb
;
175 struct block_device
*bdev
;
177 /* forced geometry settings */
178 struct hd_geometry geometry
;
180 /* For saving the address of __make_request for request based dm */
181 make_request_fn
*saved_make_request_fn
;
186 /* zero-length flush that will be cloned and submitted to targets */
187 struct bio flush_bio
;
191 * For mempools pre-allocation at the table loading time.
193 struct dm_md_mempools
{
200 static struct kmem_cache
*_io_cache
;
201 static struct kmem_cache
*_tio_cache
;
202 static struct kmem_cache
*_rq_tio_cache
;
203 static struct kmem_cache
*_rq_bio_info_cache
;
205 static int __init
local_init(void)
209 /* allocate a slab for the dm_ios */
210 _io_cache
= KMEM_CACHE(dm_io
, 0);
214 /* allocate a slab for the target ios */
215 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
217 goto out_free_io_cache
;
219 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
221 goto out_free_tio_cache
;
223 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
224 if (!_rq_bio_info_cache
)
225 goto out_free_rq_tio_cache
;
227 r
= dm_uevent_init();
229 goto out_free_rq_bio_info_cache
;
232 r
= register_blkdev(_major
, _name
);
234 goto out_uevent_exit
;
243 out_free_rq_bio_info_cache
:
244 kmem_cache_destroy(_rq_bio_info_cache
);
245 out_free_rq_tio_cache
:
246 kmem_cache_destroy(_rq_tio_cache
);
248 kmem_cache_destroy(_tio_cache
);
250 kmem_cache_destroy(_io_cache
);
255 static void local_exit(void)
257 kmem_cache_destroy(_rq_bio_info_cache
);
258 kmem_cache_destroy(_rq_tio_cache
);
259 kmem_cache_destroy(_tio_cache
);
260 kmem_cache_destroy(_io_cache
);
261 unregister_blkdev(_major
, _name
);
266 DMINFO("cleaned up");
269 static int (*_inits
[])(void) __initdata
= {
279 static void (*_exits
[])(void) = {
289 static int __init
dm_init(void)
291 const int count
= ARRAY_SIZE(_inits
);
295 for (i
= 0; i
< count
; i
++) {
310 static void __exit
dm_exit(void)
312 int i
= ARRAY_SIZE(_exits
);
319 * Block device functions
321 int dm_deleting_md(struct mapped_device
*md
)
323 return test_bit(DMF_DELETING
, &md
->flags
);
326 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
328 struct mapped_device
*md
;
330 spin_lock(&_minor_lock
);
332 md
= bdev
->bd_disk
->private_data
;
336 if (test_bit(DMF_FREEING
, &md
->flags
) ||
337 dm_deleting_md(md
)) {
343 atomic_inc(&md
->open_count
);
346 spin_unlock(&_minor_lock
);
348 return md
? 0 : -ENXIO
;
351 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
353 struct mapped_device
*md
= disk
->private_data
;
355 spin_lock(&_minor_lock
);
357 atomic_dec(&md
->open_count
);
360 spin_unlock(&_minor_lock
);
365 int dm_open_count(struct mapped_device
*md
)
367 return atomic_read(&md
->open_count
);
371 * Guarantees nothing is using the device before it's deleted.
373 int dm_lock_for_deletion(struct mapped_device
*md
)
377 spin_lock(&_minor_lock
);
379 if (dm_open_count(md
))
382 set_bit(DMF_DELETING
, &md
->flags
);
384 spin_unlock(&_minor_lock
);
389 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
391 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
393 return dm_get_geometry(md
, geo
);
396 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
397 unsigned int cmd
, unsigned long arg
)
399 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
400 struct dm_table
*map
= dm_get_live_table(md
);
401 struct dm_target
*tgt
;
404 if (!map
|| !dm_table_get_size(map
))
407 /* We only support devices that have a single target */
408 if (dm_table_get_num_targets(map
) != 1)
411 tgt
= dm_table_get_target(map
, 0);
413 if (dm_suspended_md(md
)) {
418 if (tgt
->type
->ioctl
)
419 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
427 static struct dm_io
*alloc_io(struct mapped_device
*md
)
429 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
432 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
434 mempool_free(io
, md
->io_pool
);
437 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
439 mempool_free(tio
, md
->tio_pool
);
442 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
445 return mempool_alloc(md
->tio_pool
, gfp_mask
);
448 static void free_rq_tio(struct dm_rq_target_io
*tio
)
450 mempool_free(tio
, tio
->md
->tio_pool
);
453 static struct dm_rq_clone_bio_info
*alloc_bio_info(struct mapped_device
*md
)
455 return mempool_alloc(md
->io_pool
, GFP_ATOMIC
);
458 static void free_bio_info(struct dm_rq_clone_bio_info
*info
)
460 mempool_free(info
, info
->tio
->md
->io_pool
);
463 static int md_in_flight(struct mapped_device
*md
)
465 return atomic_read(&md
->pending
[READ
]) +
466 atomic_read(&md
->pending
[WRITE
]);
469 static void start_io_acct(struct dm_io
*io
)
471 struct mapped_device
*md
= io
->md
;
473 int rw
= bio_data_dir(io
->bio
);
475 io
->start_time
= jiffies
;
477 cpu
= part_stat_lock();
478 part_round_stats(cpu
, &dm_disk(md
)->part0
);
480 dm_disk(md
)->part0
.in_flight
[rw
] = atomic_inc_return(&md
->pending
[rw
]);
483 static void end_io_acct(struct dm_io
*io
)
485 struct mapped_device
*md
= io
->md
;
486 struct bio
*bio
= io
->bio
;
487 unsigned long duration
= jiffies
- io
->start_time
;
489 int rw
= bio_data_dir(bio
);
491 cpu
= part_stat_lock();
492 part_round_stats(cpu
, &dm_disk(md
)->part0
);
493 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
497 * After this is decremented the bio must not be touched if it is
500 dm_disk(md
)->part0
.in_flight
[rw
] = pending
=
501 atomic_dec_return(&md
->pending
[rw
]);
502 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
504 /* nudge anyone waiting on suspend queue */
510 * Add the bio to the list of deferred io.
512 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
516 spin_lock_irqsave(&md
->deferred_lock
, flags
);
517 bio_list_add(&md
->deferred
, bio
);
518 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
519 queue_work(md
->wq
, &md
->work
);
523 * Everyone (including functions in this file), should use this
524 * function to access the md->map field, and make sure they call
525 * dm_table_put() when finished.
527 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
532 read_lock_irqsave(&md
->map_lock
, flags
);
536 read_unlock_irqrestore(&md
->map_lock
, flags
);
542 * Get the geometry associated with a dm device
544 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
552 * Set the geometry of a device.
554 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
556 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
558 if (geo
->start
> sz
) {
559 DMWARN("Start sector is beyond the geometry limits.");
568 /*-----------------------------------------------------------------
570 * A more elegant soln is in the works that uses the queue
571 * merge fn, unfortunately there are a couple of changes to
572 * the block layer that I want to make for this. So in the
573 * interests of getting something for people to use I give
574 * you this clearly demarcated crap.
575 *---------------------------------------------------------------*/
577 static int __noflush_suspending(struct mapped_device
*md
)
579 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
583 * Decrements the number of outstanding ios that a bio has been
584 * cloned into, completing the original io if necc.
586 static void dec_pending(struct dm_io
*io
, int error
)
591 struct mapped_device
*md
= io
->md
;
593 /* Push-back supersedes any I/O errors */
594 if (unlikely(error
)) {
595 spin_lock_irqsave(&io
->endio_lock
, flags
);
596 if (!(io
->error
> 0 && __noflush_suspending(md
)))
598 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
601 if (atomic_dec_and_test(&io
->io_count
)) {
602 if (io
->error
== DM_ENDIO_REQUEUE
) {
604 * Target requested pushing back the I/O.
606 spin_lock_irqsave(&md
->deferred_lock
, flags
);
607 if (__noflush_suspending(md
))
608 bio_list_add_head(&md
->deferred
, io
->bio
);
610 /* noflush suspend was interrupted. */
612 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
615 io_error
= io
->error
;
620 if (io_error
== DM_ENDIO_REQUEUE
)
623 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
625 * Preflush done for flush with data, reissue
628 bio
->bi_rw
&= ~REQ_FLUSH
;
631 /* done with normal IO or empty flush */
632 trace_block_bio_complete(md
->queue
, bio
, io_error
);
633 bio_endio(bio
, io_error
);
638 static void clone_endio(struct bio
*bio
, int error
)
641 struct dm_target_io
*tio
= bio
->bi_private
;
642 struct dm_io
*io
= tio
->io
;
643 struct mapped_device
*md
= tio
->io
->md
;
644 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
646 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
650 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
651 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
653 * error and requeue request are handled
657 else if (r
== DM_ENDIO_INCOMPLETE
)
658 /* The target will handle the io */
661 DMWARN("unimplemented target endio return value: %d", r
);
667 * Store md for cleanup instead of tio which is about to get freed.
669 bio
->bi_private
= md
->bs
;
673 dec_pending(io
, error
);
677 * Partial completion handling for request-based dm
679 static void end_clone_bio(struct bio
*clone
, int error
)
681 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
682 struct dm_rq_target_io
*tio
= info
->tio
;
683 struct bio
*bio
= info
->orig
;
684 unsigned int nr_bytes
= info
->orig
->bi_size
;
690 * An error has already been detected on the request.
691 * Once error occurred, just let clone->end_io() handle
697 * Don't notice the error to the upper layer yet.
698 * The error handling decision is made by the target driver,
699 * when the request is completed.
706 * I/O for the bio successfully completed.
707 * Notice the data completion to the upper layer.
711 * bios are processed from the head of the list.
712 * So the completing bio should always be rq->bio.
713 * If it's not, something wrong is happening.
715 if (tio
->orig
->bio
!= bio
)
716 DMERR("bio completion is going in the middle of the request");
719 * Update the original request.
720 * Do not use blk_end_request() here, because it may complete
721 * the original request before the clone, and break the ordering.
723 blk_update_request(tio
->orig
, 0, nr_bytes
);
727 * Don't touch any member of the md after calling this function because
728 * the md may be freed in dm_put() at the end of this function.
729 * Or do dm_get() before calling this function and dm_put() later.
731 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
733 atomic_dec(&md
->pending
[rw
]);
735 /* nudge anyone waiting on suspend queue */
736 if (!md_in_flight(md
))
740 blk_run_queue(md
->queue
);
743 * dm_put() must be at the end of this function. See the comment above
748 static void free_rq_clone(struct request
*clone
)
750 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
752 blk_rq_unprep_clone(clone
);
757 * Complete the clone and the original request.
758 * Must be called without queue lock.
760 static void dm_end_request(struct request
*clone
, int error
)
762 int rw
= rq_data_dir(clone
);
763 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
764 struct mapped_device
*md
= tio
->md
;
765 struct request
*rq
= tio
->orig
;
767 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
768 rq
->errors
= clone
->errors
;
769 rq
->resid_len
= clone
->resid_len
;
773 * We are using the sense buffer of the original
775 * So setting the length of the sense data is enough.
777 rq
->sense_len
= clone
->sense_len
;
780 free_rq_clone(clone
);
781 blk_end_request_all(rq
, error
);
782 rq_completed(md
, rw
, true);
785 static void dm_unprep_request(struct request
*rq
)
787 struct request
*clone
= rq
->special
;
790 rq
->cmd_flags
&= ~REQ_DONTPREP
;
792 free_rq_clone(clone
);
796 * Requeue the original request of a clone.
798 void dm_requeue_unmapped_request(struct request
*clone
)
800 int rw
= rq_data_dir(clone
);
801 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
802 struct mapped_device
*md
= tio
->md
;
803 struct request
*rq
= tio
->orig
;
804 struct request_queue
*q
= rq
->q
;
807 dm_unprep_request(rq
);
809 spin_lock_irqsave(q
->queue_lock
, flags
);
810 blk_requeue_request(q
, rq
);
811 spin_unlock_irqrestore(q
->queue_lock
, flags
);
813 rq_completed(md
, rw
, 0);
815 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
817 static void __stop_queue(struct request_queue
*q
)
822 static void stop_queue(struct request_queue
*q
)
826 spin_lock_irqsave(q
->queue_lock
, flags
);
828 spin_unlock_irqrestore(q
->queue_lock
, flags
);
831 static void __start_queue(struct request_queue
*q
)
833 if (blk_queue_stopped(q
))
837 static void start_queue(struct request_queue
*q
)
841 spin_lock_irqsave(q
->queue_lock
, flags
);
843 spin_unlock_irqrestore(q
->queue_lock
, flags
);
846 static void dm_done(struct request
*clone
, int error
, bool mapped
)
849 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
850 dm_request_endio_fn rq_end_io
= tio
->ti
->type
->rq_end_io
;
852 if (mapped
&& rq_end_io
)
853 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
856 /* The target wants to complete the I/O */
857 dm_end_request(clone
, r
);
858 else if (r
== DM_ENDIO_INCOMPLETE
)
859 /* The target will handle the I/O */
861 else if (r
== DM_ENDIO_REQUEUE
)
862 /* The target wants to requeue the I/O */
863 dm_requeue_unmapped_request(clone
);
865 DMWARN("unimplemented target endio return value: %d", r
);
871 * Request completion handler for request-based dm
873 static void dm_softirq_done(struct request
*rq
)
876 struct request
*clone
= rq
->completion_data
;
877 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
879 if (rq
->cmd_flags
& REQ_FAILED
)
882 dm_done(clone
, tio
->error
, mapped
);
886 * Complete the clone and the original request with the error status
887 * through softirq context.
889 static void dm_complete_request(struct request
*clone
, int error
)
891 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
892 struct request
*rq
= tio
->orig
;
895 rq
->completion_data
= clone
;
896 blk_complete_request(rq
);
900 * Complete the not-mapped clone and the original request with the error status
901 * through softirq context.
902 * Target's rq_end_io() function isn't called.
903 * This may be used when the target's map_rq() function fails.
905 void dm_kill_unmapped_request(struct request
*clone
, int error
)
907 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
908 struct request
*rq
= tio
->orig
;
910 rq
->cmd_flags
|= REQ_FAILED
;
911 dm_complete_request(clone
, error
);
913 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
916 * Called with the queue lock held
918 static void end_clone_request(struct request
*clone
, int error
)
921 * For just cleaning up the information of the queue in which
922 * the clone was dispatched.
923 * The clone is *NOT* freed actually here because it is alloced from
924 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
926 __blk_put_request(clone
->q
, clone
);
929 * Actual request completion is done in a softirq context which doesn't
930 * hold the queue lock. Otherwise, deadlock could occur because:
931 * - another request may be submitted by the upper level driver
932 * of the stacking during the completion
933 * - the submission which requires queue lock may be done
936 dm_complete_request(clone
, error
);
940 * Return maximum size of I/O possible at the supplied sector up to the current
943 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
945 sector_t target_offset
= dm_target_offset(ti
, sector
);
947 return ti
->len
- target_offset
;
950 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
952 sector_t len
= max_io_len_target_boundary(sector
, ti
);
955 * Does the target need to split even further ?
959 sector_t offset
= dm_target_offset(ti
, sector
);
960 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
969 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
970 struct dm_target_io
*tio
)
974 struct mapped_device
*md
;
976 clone
->bi_end_io
= clone_endio
;
977 clone
->bi_private
= tio
;
980 * Map the clone. If r == 0 we don't need to do
981 * anything, the target has assumed ownership of
984 atomic_inc(&tio
->io
->io_count
);
985 sector
= clone
->bi_sector
;
986 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
987 if (r
== DM_MAPIO_REMAPPED
) {
988 /* the bio has been remapped so dispatch it */
990 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
991 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
993 generic_make_request(clone
);
994 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
995 /* error the io and bail out, or requeue it if needed */
997 dec_pending(tio
->io
, r
);
999 * Store bio_set for cleanup.
1001 clone
->bi_private
= md
->bs
;
1005 DMWARN("unimplemented target map return value: %d", r
);
1011 struct mapped_device
*md
;
1012 struct dm_table
*map
;
1016 sector_t sector_count
;
1020 static void dm_bio_destructor(struct bio
*bio
)
1022 struct bio_set
*bs
= bio
->bi_private
;
1028 * Creates a little bio that just does part of a bvec.
1030 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
1031 unsigned short idx
, unsigned int offset
,
1032 unsigned int len
, struct bio_set
*bs
)
1035 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1037 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
1038 clone
->bi_destructor
= dm_bio_destructor
;
1039 *clone
->bi_io_vec
= *bv
;
1041 clone
->bi_sector
= sector
;
1042 clone
->bi_bdev
= bio
->bi_bdev
;
1043 clone
->bi_rw
= bio
->bi_rw
;
1045 clone
->bi_size
= to_bytes(len
);
1046 clone
->bi_io_vec
->bv_offset
= offset
;
1047 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1048 clone
->bi_flags
|= 1 << BIO_CLONED
;
1050 if (bio_integrity(bio
)) {
1051 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1052 bio_integrity_trim(clone
,
1053 bio_sector_offset(bio
, idx
, offset
), len
);
1060 * Creates a bio that consists of range of complete bvecs.
1062 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
1063 unsigned short idx
, unsigned short bv_count
,
1064 unsigned int len
, struct bio_set
*bs
)
1068 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
1069 __bio_clone(clone
, bio
);
1070 clone
->bi_destructor
= dm_bio_destructor
;
1071 clone
->bi_sector
= sector
;
1072 clone
->bi_idx
= idx
;
1073 clone
->bi_vcnt
= idx
+ bv_count
;
1074 clone
->bi_size
= to_bytes(len
);
1075 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1077 if (bio_integrity(bio
)) {
1078 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1080 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1081 bio_integrity_trim(clone
,
1082 bio_sector_offset(bio
, idx
, 0), len
);
1088 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1089 struct dm_target
*ti
)
1091 struct dm_target_io
*tio
= mempool_alloc(ci
->md
->tio_pool
, GFP_NOIO
);
1095 memset(&tio
->info
, 0, sizeof(tio
->info
));
1100 static void __issue_target_request(struct clone_info
*ci
, struct dm_target
*ti
,
1101 unsigned request_nr
, sector_t len
)
1103 struct dm_target_io
*tio
= alloc_tio(ci
, ti
);
1106 tio
->info
.target_request_nr
= request_nr
;
1109 * Discard requests require the bio's inline iovecs be initialized.
1110 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1111 * and discard, so no need for concern about wasted bvec allocations.
1113 clone
= bio_alloc_bioset(GFP_NOIO
, ci
->bio
->bi_max_vecs
, ci
->md
->bs
);
1114 __bio_clone(clone
, ci
->bio
);
1115 clone
->bi_destructor
= dm_bio_destructor
;
1117 clone
->bi_sector
= ci
->sector
;
1118 clone
->bi_size
= to_bytes(len
);
1121 __map_bio(ti
, clone
, tio
);
1124 static void __issue_target_requests(struct clone_info
*ci
, struct dm_target
*ti
,
1125 unsigned num_requests
, sector_t len
)
1127 unsigned request_nr
;
1129 for (request_nr
= 0; request_nr
< num_requests
; request_nr
++)
1130 __issue_target_request(ci
, ti
, request_nr
, len
);
1133 static int __clone_and_map_empty_flush(struct clone_info
*ci
)
1135 unsigned target_nr
= 0;
1136 struct dm_target
*ti
;
1138 BUG_ON(bio_has_data(ci
->bio
));
1139 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1140 __issue_target_requests(ci
, ti
, ti
->num_flush_requests
, 0);
1146 * Perform all io with a single clone.
1148 static void __clone_and_map_simple(struct clone_info
*ci
, struct dm_target
*ti
)
1150 struct bio
*clone
, *bio
= ci
->bio
;
1151 struct dm_target_io
*tio
;
1153 tio
= alloc_tio(ci
, ti
);
1154 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
1155 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
1157 __map_bio(ti
, clone
, tio
);
1158 ci
->sector_count
= 0;
1161 static int __clone_and_map_discard(struct clone_info
*ci
)
1163 struct dm_target
*ti
;
1167 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1168 if (!dm_target_is_valid(ti
))
1172 * Even though the device advertised discard support,
1173 * reconfiguration might have changed that since the
1174 * check was performed.
1176 if (!ti
->num_discard_requests
)
1179 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1181 __issue_target_requests(ci
, ti
, ti
->num_discard_requests
, len
);
1184 } while (ci
->sector_count
-= len
);
1189 static int __clone_and_map(struct clone_info
*ci
)
1191 struct bio
*clone
, *bio
= ci
->bio
;
1192 struct dm_target
*ti
;
1193 sector_t len
= 0, max
;
1194 struct dm_target_io
*tio
;
1196 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1197 return __clone_and_map_discard(ci
);
1199 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1200 if (!dm_target_is_valid(ti
))
1203 max
= max_io_len(ci
->sector
, ti
);
1205 if (ci
->sector_count
<= max
) {
1207 * Optimise for the simple case where we can do all of
1208 * the remaining io with a single clone.
1210 __clone_and_map_simple(ci
, ti
);
1212 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1214 * There are some bvecs that don't span targets.
1215 * Do as many of these as possible.
1218 sector_t remaining
= max
;
1221 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1222 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1224 if (bv_len
> remaining
)
1227 remaining
-= bv_len
;
1231 tio
= alloc_tio(ci
, ti
);
1232 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1234 __map_bio(ti
, clone
, tio
);
1237 ci
->sector_count
-= len
;
1242 * Handle a bvec that must be split between two or more targets.
1244 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1245 sector_t remaining
= to_sector(bv
->bv_len
);
1246 unsigned int offset
= 0;
1250 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1251 if (!dm_target_is_valid(ti
))
1254 max
= max_io_len(ci
->sector
, ti
);
1257 len
= min(remaining
, max
);
1259 tio
= alloc_tio(ci
, ti
);
1260 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
1261 bv
->bv_offset
+ offset
, len
,
1264 __map_bio(ti
, clone
, tio
);
1267 ci
->sector_count
-= len
;
1268 offset
+= to_bytes(len
);
1269 } while (remaining
-= len
);
1278 * Split the bio into several clones and submit it to targets.
1280 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1282 struct clone_info ci
;
1285 ci
.map
= dm_get_live_table(md
);
1286 if (unlikely(!ci
.map
)) {
1292 ci
.io
= alloc_io(md
);
1294 atomic_set(&ci
.io
->io_count
, 1);
1297 spin_lock_init(&ci
.io
->endio_lock
);
1298 ci
.sector
= bio
->bi_sector
;
1299 ci
.idx
= bio
->bi_idx
;
1301 start_io_acct(ci
.io
);
1302 if (bio
->bi_rw
& REQ_FLUSH
) {
1303 ci
.bio
= &ci
.md
->flush_bio
;
1304 ci
.sector_count
= 0;
1305 error
= __clone_and_map_empty_flush(&ci
);
1306 /* dec_pending submits any data associated with flush */
1309 ci
.sector_count
= bio_sectors(bio
);
1310 while (ci
.sector_count
&& !error
)
1311 error
= __clone_and_map(&ci
);
1314 /* drop the extra reference count */
1315 dec_pending(ci
.io
, error
);
1316 dm_table_put(ci
.map
);
1318 /*-----------------------------------------------------------------
1320 *---------------------------------------------------------------*/
1322 static int dm_merge_bvec(struct request_queue
*q
,
1323 struct bvec_merge_data
*bvm
,
1324 struct bio_vec
*biovec
)
1326 struct mapped_device
*md
= q
->queuedata
;
1327 struct dm_table
*map
= dm_get_live_table(md
);
1328 struct dm_target
*ti
;
1329 sector_t max_sectors
;
1335 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1336 if (!dm_target_is_valid(ti
))
1340 * Find maximum amount of I/O that won't need splitting
1342 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1343 (sector_t
) BIO_MAX_SECTORS
);
1344 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1349 * merge_bvec_fn() returns number of bytes
1350 * it can accept at this offset
1351 * max is precomputed maximal io size
1353 if (max_size
&& ti
->type
->merge
)
1354 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1356 * If the target doesn't support merge method and some of the devices
1357 * provided their merge_bvec method (we know this by looking at
1358 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1359 * entries. So always set max_size to 0, and the code below allows
1362 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1371 * Always allow an entire first page
1373 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1374 max_size
= biovec
->bv_len
;
1380 * The request function that just remaps the bio built up by
1383 static int _dm_request(struct request_queue
*q
, struct bio
*bio
)
1385 int rw
= bio_data_dir(bio
);
1386 struct mapped_device
*md
= q
->queuedata
;
1389 down_read(&md
->io_lock
);
1391 cpu
= part_stat_lock();
1392 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1393 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1396 /* if we're suspended, we have to queue this io for later */
1397 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1398 up_read(&md
->io_lock
);
1400 if (bio_rw(bio
) != READA
)
1407 __split_and_process_bio(md
, bio
);
1408 up_read(&md
->io_lock
);
1412 static int dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1414 struct mapped_device
*md
= q
->queuedata
;
1416 return md
->saved_make_request_fn(q
, bio
); /* call __make_request() */
1419 static int dm_request_based(struct mapped_device
*md
)
1421 return blk_queue_stackable(md
->queue
);
1424 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
1426 struct mapped_device
*md
= q
->queuedata
;
1428 if (dm_request_based(md
))
1429 return dm_make_request(q
, bio
);
1431 return _dm_request(q
, bio
);
1434 void dm_dispatch_request(struct request
*rq
)
1438 if (blk_queue_io_stat(rq
->q
))
1439 rq
->cmd_flags
|= REQ_IO_STAT
;
1441 rq
->start_time
= jiffies
;
1442 r
= blk_insert_cloned_request(rq
->q
, rq
);
1444 dm_complete_request(rq
, r
);
1446 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1448 static void dm_rq_bio_destructor(struct bio
*bio
)
1450 struct dm_rq_clone_bio_info
*info
= bio
->bi_private
;
1451 struct mapped_device
*md
= info
->tio
->md
;
1453 free_bio_info(info
);
1454 bio_free(bio
, md
->bs
);
1457 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1460 struct dm_rq_target_io
*tio
= data
;
1461 struct mapped_device
*md
= tio
->md
;
1462 struct dm_rq_clone_bio_info
*info
= alloc_bio_info(md
);
1467 info
->orig
= bio_orig
;
1469 bio
->bi_end_io
= end_clone_bio
;
1470 bio
->bi_private
= info
;
1471 bio
->bi_destructor
= dm_rq_bio_destructor
;
1476 static int setup_clone(struct request
*clone
, struct request
*rq
,
1477 struct dm_rq_target_io
*tio
)
1481 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1482 dm_rq_bio_constructor
, tio
);
1486 clone
->cmd
= rq
->cmd
;
1487 clone
->cmd_len
= rq
->cmd_len
;
1488 clone
->sense
= rq
->sense
;
1489 clone
->buffer
= rq
->buffer
;
1490 clone
->end_io
= end_clone_request
;
1491 clone
->end_io_data
= tio
;
1496 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1499 struct request
*clone
;
1500 struct dm_rq_target_io
*tio
;
1502 tio
= alloc_rq_tio(md
, gfp_mask
);
1510 memset(&tio
->info
, 0, sizeof(tio
->info
));
1512 clone
= &tio
->clone
;
1513 if (setup_clone(clone
, rq
, tio
)) {
1523 * Called with the queue lock held.
1525 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1527 struct mapped_device
*md
= q
->queuedata
;
1528 struct request
*clone
;
1530 if (unlikely(rq
->special
)) {
1531 DMWARN("Already has something in rq->special.");
1532 return BLKPREP_KILL
;
1535 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1537 return BLKPREP_DEFER
;
1539 rq
->special
= clone
;
1540 rq
->cmd_flags
|= REQ_DONTPREP
;
1547 * 0 : the request has been processed (not requeued)
1548 * !0 : the request has been requeued
1550 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1551 struct mapped_device
*md
)
1553 int r
, requeued
= 0;
1554 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1557 * Hold the md reference here for the in-flight I/O.
1558 * We can't rely on the reference count by device opener,
1559 * because the device may be closed during the request completion
1560 * when all bios are completed.
1561 * See the comment in rq_completed() too.
1566 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1568 case DM_MAPIO_SUBMITTED
:
1569 /* The target has taken the I/O to submit by itself later */
1571 case DM_MAPIO_REMAPPED
:
1572 /* The target has remapped the I/O so dispatch it */
1573 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1574 blk_rq_pos(tio
->orig
));
1575 dm_dispatch_request(clone
);
1577 case DM_MAPIO_REQUEUE
:
1578 /* The target wants to requeue the I/O */
1579 dm_requeue_unmapped_request(clone
);
1584 DMWARN("unimplemented target map return value: %d", r
);
1588 /* The target wants to complete the I/O */
1589 dm_kill_unmapped_request(clone
, r
);
1597 * q->request_fn for request-based dm.
1598 * Called with the queue lock held.
1600 static void dm_request_fn(struct request_queue
*q
)
1602 struct mapped_device
*md
= q
->queuedata
;
1603 struct dm_table
*map
= dm_get_live_table(md
);
1604 struct dm_target
*ti
;
1605 struct request
*rq
, *clone
;
1609 * For suspend, check blk_queue_stopped() and increment
1610 * ->pending within a single queue_lock not to increment the
1611 * number of in-flight I/Os after the queue is stopped in
1614 while (!blk_queue_stopped(q
)) {
1615 rq
= blk_peek_request(q
);
1619 /* always use block 0 to find the target for flushes for now */
1621 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1622 pos
= blk_rq_pos(rq
);
1624 ti
= dm_table_find_target(map
, pos
);
1625 BUG_ON(!dm_target_is_valid(ti
));
1627 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1630 blk_start_request(rq
);
1631 clone
= rq
->special
;
1632 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1634 spin_unlock(q
->queue_lock
);
1635 if (map_request(ti
, clone
, md
))
1638 BUG_ON(!irqs_disabled());
1639 spin_lock(q
->queue_lock
);
1645 BUG_ON(!irqs_disabled());
1646 spin_lock(q
->queue_lock
);
1649 blk_delay_queue(q
, HZ
/ 10);
1656 int dm_underlying_device_busy(struct request_queue
*q
)
1658 return blk_lld_busy(q
);
1660 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1662 static int dm_lld_busy(struct request_queue
*q
)
1665 struct mapped_device
*md
= q
->queuedata
;
1666 struct dm_table
*map
= dm_get_live_table(md
);
1668 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1671 r
= dm_table_any_busy_target(map
);
1678 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1681 struct mapped_device
*md
= congested_data
;
1682 struct dm_table
*map
;
1684 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1685 map
= dm_get_live_table(md
);
1688 * Request-based dm cares about only own queue for
1689 * the query about congestion status of request_queue
1691 if (dm_request_based(md
))
1692 r
= md
->queue
->backing_dev_info
.state
&
1695 r
= dm_table_any_congested(map
, bdi_bits
);
1704 /*-----------------------------------------------------------------
1705 * An IDR is used to keep track of allocated minor numbers.
1706 *---------------------------------------------------------------*/
1707 static DEFINE_IDR(_minor_idr
);
1709 static void free_minor(int minor
)
1711 spin_lock(&_minor_lock
);
1712 idr_remove(&_minor_idr
, minor
);
1713 spin_unlock(&_minor_lock
);
1717 * See if the device with a specific minor # is free.
1719 static int specific_minor(int minor
)
1723 if (minor
>= (1 << MINORBITS
))
1726 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1730 spin_lock(&_minor_lock
);
1732 if (idr_find(&_minor_idr
, minor
)) {
1737 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1742 idr_remove(&_minor_idr
, m
);
1748 spin_unlock(&_minor_lock
);
1752 static int next_free_minor(int *minor
)
1756 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1760 spin_lock(&_minor_lock
);
1762 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1766 if (m
>= (1 << MINORBITS
)) {
1767 idr_remove(&_minor_idr
, m
);
1775 spin_unlock(&_minor_lock
);
1779 static const struct block_device_operations dm_blk_dops
;
1781 static void dm_wq_work(struct work_struct
*work
);
1783 static void dm_init_md_queue(struct mapped_device
*md
)
1786 * Request-based dm devices cannot be stacked on top of bio-based dm
1787 * devices. The type of this dm device has not been decided yet.
1788 * The type is decided at the first table loading time.
1789 * To prevent problematic device stacking, clear the queue flag
1790 * for request stacking support until then.
1792 * This queue is new, so no concurrency on the queue_flags.
1794 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1796 md
->queue
->queuedata
= md
;
1797 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1798 md
->queue
->backing_dev_info
.congested_data
= md
;
1799 blk_queue_make_request(md
->queue
, dm_request
);
1800 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1801 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1802 blk_queue_flush(md
->queue
, REQ_FLUSH
| REQ_FUA
);
1806 * Allocate and initialise a blank device with a given minor.
1808 static struct mapped_device
*alloc_dev(int minor
)
1811 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1815 DMWARN("unable to allocate device, out of memory.");
1819 if (!try_module_get(THIS_MODULE
))
1820 goto bad_module_get
;
1822 /* get a minor number for the dev */
1823 if (minor
== DM_ANY_MINOR
)
1824 r
= next_free_minor(&minor
);
1826 r
= specific_minor(minor
);
1830 md
->type
= DM_TYPE_NONE
;
1831 init_rwsem(&md
->io_lock
);
1832 mutex_init(&md
->suspend_lock
);
1833 mutex_init(&md
->type_lock
);
1834 spin_lock_init(&md
->deferred_lock
);
1835 rwlock_init(&md
->map_lock
);
1836 atomic_set(&md
->holders
, 1);
1837 atomic_set(&md
->open_count
, 0);
1838 atomic_set(&md
->event_nr
, 0);
1839 atomic_set(&md
->uevent_seq
, 0);
1840 INIT_LIST_HEAD(&md
->uevent_list
);
1841 spin_lock_init(&md
->uevent_lock
);
1843 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1847 dm_init_md_queue(md
);
1849 md
->disk
= alloc_disk(1);
1853 atomic_set(&md
->pending
[0], 0);
1854 atomic_set(&md
->pending
[1], 0);
1855 init_waitqueue_head(&md
->wait
);
1856 INIT_WORK(&md
->work
, dm_wq_work
);
1857 init_waitqueue_head(&md
->eventq
);
1859 md
->disk
->major
= _major
;
1860 md
->disk
->first_minor
= minor
;
1861 md
->disk
->fops
= &dm_blk_dops
;
1862 md
->disk
->queue
= md
->queue
;
1863 md
->disk
->private_data
= md
;
1864 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1866 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1868 md
->wq
= alloc_workqueue("kdmflush",
1869 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1873 md
->bdev
= bdget_disk(md
->disk
, 0);
1877 bio_init(&md
->flush_bio
);
1878 md
->flush_bio
.bi_bdev
= md
->bdev
;
1879 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1881 /* Populate the mapping, nobody knows we exist yet */
1882 spin_lock(&_minor_lock
);
1883 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1884 spin_unlock(&_minor_lock
);
1886 BUG_ON(old_md
!= MINOR_ALLOCED
);
1891 destroy_workqueue(md
->wq
);
1893 del_gendisk(md
->disk
);
1896 blk_cleanup_queue(md
->queue
);
1900 module_put(THIS_MODULE
);
1906 static void unlock_fs(struct mapped_device
*md
);
1908 static void free_dev(struct mapped_device
*md
)
1910 int minor
= MINOR(disk_devt(md
->disk
));
1914 destroy_workqueue(md
->wq
);
1916 mempool_destroy(md
->tio_pool
);
1918 mempool_destroy(md
->io_pool
);
1920 bioset_free(md
->bs
);
1921 blk_integrity_unregister(md
->disk
);
1922 del_gendisk(md
->disk
);
1925 spin_lock(&_minor_lock
);
1926 md
->disk
->private_data
= NULL
;
1927 spin_unlock(&_minor_lock
);
1930 blk_cleanup_queue(md
->queue
);
1931 module_put(THIS_MODULE
);
1935 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1937 struct dm_md_mempools
*p
;
1939 if (md
->io_pool
&& md
->tio_pool
&& md
->bs
)
1940 /* the md already has necessary mempools */
1943 p
= dm_table_get_md_mempools(t
);
1944 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1946 md
->io_pool
= p
->io_pool
;
1948 md
->tio_pool
= p
->tio_pool
;
1954 /* mempool bind completed, now no need any mempools in the table */
1955 dm_table_free_md_mempools(t
);
1959 * Bind a table to the device.
1961 static void event_callback(void *context
)
1963 unsigned long flags
;
1965 struct mapped_device
*md
= (struct mapped_device
*) context
;
1967 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1968 list_splice_init(&md
->uevent_list
, &uevents
);
1969 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1971 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1973 atomic_inc(&md
->event_nr
);
1974 wake_up(&md
->eventq
);
1978 * Protected by md->suspend_lock obtained by dm_swap_table().
1980 static void __set_size(struct mapped_device
*md
, sector_t size
)
1982 set_capacity(md
->disk
, size
);
1984 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1988 * Returns old map, which caller must destroy.
1990 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1991 struct queue_limits
*limits
)
1993 struct dm_table
*old_map
;
1994 struct request_queue
*q
= md
->queue
;
1996 unsigned long flags
;
1998 size
= dm_table_get_size(t
);
2001 * Wipe any geometry if the size of the table changed.
2003 if (size
!= get_capacity(md
->disk
))
2004 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2006 __set_size(md
, size
);
2008 dm_table_event_callback(t
, event_callback
, md
);
2011 * The queue hasn't been stopped yet, if the old table type wasn't
2012 * for request-based during suspension. So stop it to prevent
2013 * I/O mapping before resume.
2014 * This must be done before setting the queue restrictions,
2015 * because request-based dm may be run just after the setting.
2017 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2020 __bind_mempools(md
, t
);
2022 write_lock_irqsave(&md
->map_lock
, flags
);
2025 dm_table_set_restrictions(t
, q
, limits
);
2026 write_unlock_irqrestore(&md
->map_lock
, flags
);
2032 * Returns unbound table for the caller to free.
2034 static struct dm_table
*__unbind(struct mapped_device
*md
)
2036 struct dm_table
*map
= md
->map
;
2037 unsigned long flags
;
2042 dm_table_event_callback(map
, NULL
, NULL
);
2043 write_lock_irqsave(&md
->map_lock
, flags
);
2045 write_unlock_irqrestore(&md
->map_lock
, flags
);
2051 * Constructor for a new device.
2053 int dm_create(int minor
, struct mapped_device
**result
)
2055 struct mapped_device
*md
;
2057 md
= alloc_dev(minor
);
2068 * Functions to manage md->type.
2069 * All are required to hold md->type_lock.
2071 void dm_lock_md_type(struct mapped_device
*md
)
2073 mutex_lock(&md
->type_lock
);
2076 void dm_unlock_md_type(struct mapped_device
*md
)
2078 mutex_unlock(&md
->type_lock
);
2081 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2086 unsigned dm_get_md_type(struct mapped_device
*md
)
2092 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2094 static int dm_init_request_based_queue(struct mapped_device
*md
)
2096 struct request_queue
*q
= NULL
;
2098 if (md
->queue
->elevator
)
2101 /* Fully initialize the queue */
2102 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2107 md
->saved_make_request_fn
= md
->queue
->make_request_fn
;
2108 dm_init_md_queue(md
);
2109 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2110 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2111 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2113 elv_register_queue(md
->queue
);
2119 * Setup the DM device's queue based on md's type
2121 int dm_setup_md_queue(struct mapped_device
*md
)
2123 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2124 !dm_init_request_based_queue(md
)) {
2125 DMWARN("Cannot initialize queue for request-based mapped device");
2132 static struct mapped_device
*dm_find_md(dev_t dev
)
2134 struct mapped_device
*md
;
2135 unsigned minor
= MINOR(dev
);
2137 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2140 spin_lock(&_minor_lock
);
2142 md
= idr_find(&_minor_idr
, minor
);
2143 if (md
&& (md
== MINOR_ALLOCED
||
2144 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2145 dm_deleting_md(md
) ||
2146 test_bit(DMF_FREEING
, &md
->flags
))) {
2152 spin_unlock(&_minor_lock
);
2157 struct mapped_device
*dm_get_md(dev_t dev
)
2159 struct mapped_device
*md
= dm_find_md(dev
);
2167 void *dm_get_mdptr(struct mapped_device
*md
)
2169 return md
->interface_ptr
;
2172 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2174 md
->interface_ptr
= ptr
;
2177 void dm_get(struct mapped_device
*md
)
2179 atomic_inc(&md
->holders
);
2180 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2183 const char *dm_device_name(struct mapped_device
*md
)
2187 EXPORT_SYMBOL_GPL(dm_device_name
);
2189 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2191 struct dm_table
*map
;
2195 spin_lock(&_minor_lock
);
2196 map
= dm_get_live_table(md
);
2197 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2198 set_bit(DMF_FREEING
, &md
->flags
);
2199 spin_unlock(&_minor_lock
);
2201 if (!dm_suspended_md(md
)) {
2202 dm_table_presuspend_targets(map
);
2203 dm_table_postsuspend_targets(map
);
2207 * Rare, but there may be I/O requests still going to complete,
2208 * for example. Wait for all references to disappear.
2209 * No one should increment the reference count of the mapped_device,
2210 * after the mapped_device state becomes DMF_FREEING.
2213 while (atomic_read(&md
->holders
))
2215 else if (atomic_read(&md
->holders
))
2216 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2217 dm_device_name(md
), atomic_read(&md
->holders
));
2221 dm_table_destroy(__unbind(md
));
2225 void dm_destroy(struct mapped_device
*md
)
2227 __dm_destroy(md
, true);
2230 void dm_destroy_immediate(struct mapped_device
*md
)
2232 __dm_destroy(md
, false);
2235 void dm_put(struct mapped_device
*md
)
2237 atomic_dec(&md
->holders
);
2239 EXPORT_SYMBOL_GPL(dm_put
);
2241 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2244 DECLARE_WAITQUEUE(wait
, current
);
2246 add_wait_queue(&md
->wait
, &wait
);
2249 set_current_state(interruptible
);
2252 if (!md_in_flight(md
))
2255 if (interruptible
== TASK_INTERRUPTIBLE
&&
2256 signal_pending(current
)) {
2263 set_current_state(TASK_RUNNING
);
2265 remove_wait_queue(&md
->wait
, &wait
);
2271 * Process the deferred bios
2273 static void dm_wq_work(struct work_struct
*work
)
2275 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2279 down_read(&md
->io_lock
);
2281 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2282 spin_lock_irq(&md
->deferred_lock
);
2283 c
= bio_list_pop(&md
->deferred
);
2284 spin_unlock_irq(&md
->deferred_lock
);
2289 up_read(&md
->io_lock
);
2291 if (dm_request_based(md
))
2292 generic_make_request(c
);
2294 __split_and_process_bio(md
, c
);
2296 down_read(&md
->io_lock
);
2299 up_read(&md
->io_lock
);
2302 static void dm_queue_flush(struct mapped_device
*md
)
2304 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2305 smp_mb__after_clear_bit();
2306 queue_work(md
->wq
, &md
->work
);
2310 * Swap in a new table, returning the old one for the caller to destroy.
2312 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2314 struct dm_table
*map
= ERR_PTR(-EINVAL
);
2315 struct queue_limits limits
;
2318 mutex_lock(&md
->suspend_lock
);
2320 /* device must be suspended */
2321 if (!dm_suspended_md(md
))
2324 r
= dm_calculate_queue_limits(table
, &limits
);
2330 map
= __bind(md
, table
, &limits
);
2333 mutex_unlock(&md
->suspend_lock
);
2338 * Functions to lock and unlock any filesystem running on the
2341 static int lock_fs(struct mapped_device
*md
)
2345 WARN_ON(md
->frozen_sb
);
2347 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2348 if (IS_ERR(md
->frozen_sb
)) {
2349 r
= PTR_ERR(md
->frozen_sb
);
2350 md
->frozen_sb
= NULL
;
2354 set_bit(DMF_FROZEN
, &md
->flags
);
2359 static void unlock_fs(struct mapped_device
*md
)
2361 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2364 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2365 md
->frozen_sb
= NULL
;
2366 clear_bit(DMF_FROZEN
, &md
->flags
);
2370 * We need to be able to change a mapping table under a mounted
2371 * filesystem. For example we might want to move some data in
2372 * the background. Before the table can be swapped with
2373 * dm_bind_table, dm_suspend must be called to flush any in
2374 * flight bios and ensure that any further io gets deferred.
2377 * Suspend mechanism in request-based dm.
2379 * 1. Flush all I/Os by lock_fs() if needed.
2380 * 2. Stop dispatching any I/O by stopping the request_queue.
2381 * 3. Wait for all in-flight I/Os to be completed or requeued.
2383 * To abort suspend, start the request_queue.
2385 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2387 struct dm_table
*map
= NULL
;
2389 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2390 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2392 mutex_lock(&md
->suspend_lock
);
2394 if (dm_suspended_md(md
)) {
2399 map
= dm_get_live_table(md
);
2402 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2403 * This flag is cleared before dm_suspend returns.
2406 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2408 /* This does not get reverted if there's an error later. */
2409 dm_table_presuspend_targets(map
);
2412 * Flush I/O to the device.
2413 * Any I/O submitted after lock_fs() may not be flushed.
2414 * noflush takes precedence over do_lockfs.
2415 * (lock_fs() flushes I/Os and waits for them to complete.)
2417 if (!noflush
&& do_lockfs
) {
2424 * Here we must make sure that no processes are submitting requests
2425 * to target drivers i.e. no one may be executing
2426 * __split_and_process_bio. This is called from dm_request and
2429 * To get all processes out of __split_and_process_bio in dm_request,
2430 * we take the write lock. To prevent any process from reentering
2431 * __split_and_process_bio from dm_request and quiesce the thread
2432 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2433 * flush_workqueue(md->wq).
2435 down_write(&md
->io_lock
);
2436 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2437 up_write(&md
->io_lock
);
2440 * Stop md->queue before flushing md->wq in case request-based
2441 * dm defers requests to md->wq from md->queue.
2443 if (dm_request_based(md
))
2444 stop_queue(md
->queue
);
2446 flush_workqueue(md
->wq
);
2449 * At this point no more requests are entering target request routines.
2450 * We call dm_wait_for_completion to wait for all existing requests
2453 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2455 down_write(&md
->io_lock
);
2457 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2458 up_write(&md
->io_lock
);
2460 /* were we interrupted ? */
2464 if (dm_request_based(md
))
2465 start_queue(md
->queue
);
2468 goto out
; /* pushback list is already flushed, so skip flush */
2472 * If dm_wait_for_completion returned 0, the device is completely
2473 * quiescent now. There is no request-processing activity. All new
2474 * requests are being added to md->deferred list.
2477 set_bit(DMF_SUSPENDED
, &md
->flags
);
2479 dm_table_postsuspend_targets(map
);
2485 mutex_unlock(&md
->suspend_lock
);
2489 int dm_resume(struct mapped_device
*md
)
2492 struct dm_table
*map
= NULL
;
2494 mutex_lock(&md
->suspend_lock
);
2495 if (!dm_suspended_md(md
))
2498 map
= dm_get_live_table(md
);
2499 if (!map
|| !dm_table_get_size(map
))
2502 r
= dm_table_resume_targets(map
);
2509 * Flushing deferred I/Os must be done after targets are resumed
2510 * so that mapping of targets can work correctly.
2511 * Request-based dm is queueing the deferred I/Os in its request_queue.
2513 if (dm_request_based(md
))
2514 start_queue(md
->queue
);
2518 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2523 mutex_unlock(&md
->suspend_lock
);
2528 /*-----------------------------------------------------------------
2529 * Event notification.
2530 *---------------------------------------------------------------*/
2531 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2534 char udev_cookie
[DM_COOKIE_LENGTH
];
2535 char *envp
[] = { udev_cookie
, NULL
};
2538 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2540 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2541 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2542 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2547 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2549 return atomic_add_return(1, &md
->uevent_seq
);
2552 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2554 return atomic_read(&md
->event_nr
);
2557 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2559 return wait_event_interruptible(md
->eventq
,
2560 (event_nr
!= atomic_read(&md
->event_nr
)));
2563 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2565 unsigned long flags
;
2567 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2568 list_add(elist
, &md
->uevent_list
);
2569 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2573 * The gendisk is only valid as long as you have a reference
2576 struct gendisk
*dm_disk(struct mapped_device
*md
)
2581 struct kobject
*dm_kobject(struct mapped_device
*md
)
2587 * struct mapped_device should not be exported outside of dm.c
2588 * so use this check to verify that kobj is part of md structure
2590 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2592 struct mapped_device
*md
;
2594 md
= container_of(kobj
, struct mapped_device
, kobj
);
2595 if (&md
->kobj
!= kobj
)
2598 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2606 int dm_suspended_md(struct mapped_device
*md
)
2608 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2611 int dm_suspended(struct dm_target
*ti
)
2613 return dm_suspended_md(dm_table_get_md(ti
->table
));
2615 EXPORT_SYMBOL_GPL(dm_suspended
);
2617 int dm_noflush_suspending(struct dm_target
*ti
)
2619 return __noflush_suspending(dm_table_get_md(ti
->table
));
2621 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2623 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
)
2625 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2630 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2631 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2632 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2633 if (!pools
->io_pool
)
2634 goto free_pools_and_out
;
2636 pools
->tio_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2637 mempool_create_slab_pool(MIN_IOS
, _tio_cache
) :
2638 mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2639 if (!pools
->tio_pool
)
2640 goto free_io_pool_and_out
;
2642 pools
->bs
= (type
== DM_TYPE_BIO_BASED
) ?
2643 bioset_create(16, 0) : bioset_create(MIN_IOS
, 0);
2645 goto free_tio_pool_and_out
;
2649 free_tio_pool_and_out
:
2650 mempool_destroy(pools
->tio_pool
);
2652 free_io_pool_and_out
:
2653 mempool_destroy(pools
->io_pool
);
2661 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2667 mempool_destroy(pools
->io_pool
);
2669 if (pools
->tio_pool
)
2670 mempool_destroy(pools
->tio_pool
);
2673 bioset_free(pools
->bs
);
2678 static const struct block_device_operations dm_blk_dops
= {
2679 .open
= dm_blk_open
,
2680 .release
= dm_blk_close
,
2681 .ioctl
= dm_blk_ioctl
,
2682 .getgeo
= dm_blk_getgeo
,
2683 .owner
= THIS_MODULE
2686 EXPORT_SYMBOL(dm_get_mapinfo
);
2691 module_init(dm_init
);
2692 module_exit(dm_exit
);
2694 module_param(major
, uint
, 0);
2695 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2696 MODULE_DESCRIPTION(DM_NAME
" driver");
2697 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2698 MODULE_LICENSE("GPL");