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_IDR(_minor_idr
);
42 static DEFINE_SPINLOCK(_minor_lock
);
45 * One of these is allocated per bio.
48 struct mapped_device
*md
;
52 unsigned long start_time
;
53 spinlock_t endio_lock
;
58 * One of these is allocated per target within a bio. Hopefully
59 * this will be simplified out one day.
68 * For request-based dm.
69 * One of these is allocated per request.
71 struct dm_rq_target_io
{
72 struct mapped_device
*md
;
74 struct request
*orig
, clone
;
80 * For request-based dm.
81 * One of these is allocated per bio.
83 struct dm_rq_clone_bio_info
{
85 struct dm_rq_target_io
*tio
;
88 union map_info
*dm_get_mapinfo(struct bio
*bio
)
90 if (bio
&& bio
->bi_private
)
91 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
95 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
97 if (rq
&& rq
->end_io_data
)
98 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
101 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
103 #define MINOR_ALLOCED ((void *)-1)
106 * Bits for the md->flags field.
108 #define DMF_BLOCK_IO_FOR_SUSPEND 0
109 #define DMF_SUSPENDED 1
111 #define DMF_FREEING 3
112 #define DMF_DELETING 4
113 #define DMF_NOFLUSH_SUSPENDING 5
116 * Work processed by per-device workqueue.
118 struct mapped_device
{
119 struct rw_semaphore io_lock
;
120 struct mutex suspend_lock
;
127 struct request_queue
*queue
;
129 /* Protect queue and type against concurrent access. */
130 struct mutex type_lock
;
132 struct gendisk
*disk
;
138 * A list of ios that arrived while we were suspended.
141 wait_queue_head_t wait
;
142 struct work_struct work
;
143 struct bio_list deferred
;
144 spinlock_t deferred_lock
;
147 * Processing queue (flush)
149 struct workqueue_struct
*wq
;
152 * The current mapping.
154 struct dm_table
*map
;
157 * io objects are allocated from here.
168 wait_queue_head_t eventq
;
170 struct list_head uevent_list
;
171 spinlock_t uevent_lock
; /* Protect access to uevent_list */
174 * freeze/thaw support require holding onto a super block
176 struct super_block
*frozen_sb
;
177 struct block_device
*bdev
;
179 /* forced geometry settings */
180 struct hd_geometry geometry
;
182 /* For saving the address of __make_request for request based dm */
183 make_request_fn
*saved_make_request_fn
;
188 /* zero-length flush that will be cloned and submitted to targets */
189 struct bio flush_bio
;
193 * For mempools pre-allocation at the table loading time.
195 struct dm_md_mempools
{
202 static struct kmem_cache
*_io_cache
;
203 static struct kmem_cache
*_tio_cache
;
204 static struct kmem_cache
*_rq_tio_cache
;
205 static struct kmem_cache
*_rq_bio_info_cache
;
207 static int __init
local_init(void)
211 /* allocate a slab for the dm_ios */
212 _io_cache
= KMEM_CACHE(dm_io
, 0);
216 /* allocate a slab for the target ios */
217 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
219 goto out_free_io_cache
;
221 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
223 goto out_free_tio_cache
;
225 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
226 if (!_rq_bio_info_cache
)
227 goto out_free_rq_tio_cache
;
229 r
= dm_uevent_init();
231 goto out_free_rq_bio_info_cache
;
234 r
= register_blkdev(_major
, _name
);
236 goto out_uevent_exit
;
245 out_free_rq_bio_info_cache
:
246 kmem_cache_destroy(_rq_bio_info_cache
);
247 out_free_rq_tio_cache
:
248 kmem_cache_destroy(_rq_tio_cache
);
250 kmem_cache_destroy(_tio_cache
);
252 kmem_cache_destroy(_io_cache
);
257 static void local_exit(void)
259 kmem_cache_destroy(_rq_bio_info_cache
);
260 kmem_cache_destroy(_rq_tio_cache
);
261 kmem_cache_destroy(_tio_cache
);
262 kmem_cache_destroy(_io_cache
);
263 unregister_blkdev(_major
, _name
);
268 DMINFO("cleaned up");
271 static int (*_inits
[])(void) __initdata
= {
281 static void (*_exits
[])(void) = {
291 static int __init
dm_init(void)
293 const int count
= ARRAY_SIZE(_inits
);
297 for (i
= 0; i
< count
; i
++) {
312 static void __exit
dm_exit(void)
314 int i
= ARRAY_SIZE(_exits
);
320 * Should be empty by this point.
322 idr_remove_all(&_minor_idr
);
323 idr_destroy(&_minor_idr
);
327 * Block device functions
329 int dm_deleting_md(struct mapped_device
*md
)
331 return test_bit(DMF_DELETING
, &md
->flags
);
334 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
336 struct mapped_device
*md
;
338 spin_lock(&_minor_lock
);
340 md
= bdev
->bd_disk
->private_data
;
344 if (test_bit(DMF_FREEING
, &md
->flags
) ||
345 dm_deleting_md(md
)) {
351 atomic_inc(&md
->open_count
);
354 spin_unlock(&_minor_lock
);
356 return md
? 0 : -ENXIO
;
359 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
361 struct mapped_device
*md
= disk
->private_data
;
363 spin_lock(&_minor_lock
);
365 atomic_dec(&md
->open_count
);
368 spin_unlock(&_minor_lock
);
373 int dm_open_count(struct mapped_device
*md
)
375 return atomic_read(&md
->open_count
);
379 * Guarantees nothing is using the device before it's deleted.
381 int dm_lock_for_deletion(struct mapped_device
*md
)
385 spin_lock(&_minor_lock
);
387 if (dm_open_count(md
))
390 set_bit(DMF_DELETING
, &md
->flags
);
392 spin_unlock(&_minor_lock
);
397 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
399 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
401 return dm_get_geometry(md
, geo
);
404 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
405 unsigned int cmd
, unsigned long arg
)
407 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
408 struct dm_table
*map
= dm_get_live_table(md
);
409 struct dm_target
*tgt
;
412 if (!map
|| !dm_table_get_size(map
))
415 /* We only support devices that have a single target */
416 if (dm_table_get_num_targets(map
) != 1)
419 tgt
= dm_table_get_target(map
, 0);
421 if (dm_suspended_md(md
)) {
426 if (tgt
->type
->ioctl
)
427 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
435 static struct dm_io
*alloc_io(struct mapped_device
*md
)
437 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
440 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
442 mempool_free(io
, md
->io_pool
);
445 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
447 mempool_free(tio
, md
->tio_pool
);
450 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
453 return mempool_alloc(md
->tio_pool
, gfp_mask
);
456 static void free_rq_tio(struct dm_rq_target_io
*tio
)
458 mempool_free(tio
, tio
->md
->tio_pool
);
461 static struct dm_rq_clone_bio_info
*alloc_bio_info(struct mapped_device
*md
)
463 return mempool_alloc(md
->io_pool
, GFP_ATOMIC
);
466 static void free_bio_info(struct dm_rq_clone_bio_info
*info
)
468 mempool_free(info
, info
->tio
->md
->io_pool
);
471 static int md_in_flight(struct mapped_device
*md
)
473 return atomic_read(&md
->pending
[READ
]) +
474 atomic_read(&md
->pending
[WRITE
]);
477 static void start_io_acct(struct dm_io
*io
)
479 struct mapped_device
*md
= io
->md
;
481 int rw
= bio_data_dir(io
->bio
);
483 io
->start_time
= jiffies
;
485 cpu
= part_stat_lock();
486 part_round_stats(cpu
, &dm_disk(md
)->part0
);
488 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
489 atomic_inc_return(&md
->pending
[rw
]));
492 static void end_io_acct(struct dm_io
*io
)
494 struct mapped_device
*md
= io
->md
;
495 struct bio
*bio
= io
->bio
;
496 unsigned long duration
= jiffies
- io
->start_time
;
498 int rw
= bio_data_dir(bio
);
500 cpu
= part_stat_lock();
501 part_round_stats(cpu
, &dm_disk(md
)->part0
);
502 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
506 * After this is decremented the bio must not be touched if it is
509 pending
= atomic_dec_return(&md
->pending
[rw
]);
510 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
511 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
513 /* nudge anyone waiting on suspend queue */
519 * Add the bio to the list of deferred io.
521 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
525 spin_lock_irqsave(&md
->deferred_lock
, flags
);
526 bio_list_add(&md
->deferred
, bio
);
527 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
528 queue_work(md
->wq
, &md
->work
);
532 * Everyone (including functions in this file), should use this
533 * function to access the md->map field, and make sure they call
534 * dm_table_put() when finished.
536 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
541 read_lock_irqsave(&md
->map_lock
, flags
);
545 read_unlock_irqrestore(&md
->map_lock
, flags
);
551 * Get the geometry associated with a dm device
553 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
561 * Set the geometry of a device.
563 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
565 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
567 if (geo
->start
> sz
) {
568 DMWARN("Start sector is beyond the geometry limits.");
577 /*-----------------------------------------------------------------
579 * A more elegant soln is in the works that uses the queue
580 * merge fn, unfortunately there are a couple of changes to
581 * the block layer that I want to make for this. So in the
582 * interests of getting something for people to use I give
583 * you this clearly demarcated crap.
584 *---------------------------------------------------------------*/
586 static int __noflush_suspending(struct mapped_device
*md
)
588 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
592 * Decrements the number of outstanding ios that a bio has been
593 * cloned into, completing the original io if necc.
595 static void dec_pending(struct dm_io
*io
, int error
)
600 struct mapped_device
*md
= io
->md
;
602 /* Push-back supersedes any I/O errors */
603 if (unlikely(error
)) {
604 spin_lock_irqsave(&io
->endio_lock
, flags
);
605 if (!(io
->error
> 0 && __noflush_suspending(md
)))
607 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
610 if (atomic_dec_and_test(&io
->io_count
)) {
611 if (io
->error
== DM_ENDIO_REQUEUE
) {
613 * Target requested pushing back the I/O.
615 spin_lock_irqsave(&md
->deferred_lock
, flags
);
616 if (__noflush_suspending(md
))
617 bio_list_add_head(&md
->deferred
, io
->bio
);
619 /* noflush suspend was interrupted. */
621 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
624 io_error
= io
->error
;
629 if (io_error
== DM_ENDIO_REQUEUE
)
632 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
634 * Preflush done for flush with data, reissue
637 bio
->bi_rw
&= ~REQ_FLUSH
;
640 /* done with normal IO or empty flush */
641 trace_block_bio_complete(md
->queue
, bio
, io_error
);
642 bio_endio(bio
, io_error
);
647 static void clone_endio(struct bio
*bio
, int error
)
650 struct dm_target_io
*tio
= bio
->bi_private
;
651 struct dm_io
*io
= tio
->io
;
652 struct mapped_device
*md
= tio
->io
->md
;
653 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
655 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
659 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
660 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
662 * error and requeue request are handled
666 else if (r
== DM_ENDIO_INCOMPLETE
)
667 /* The target will handle the io */
670 DMWARN("unimplemented target endio return value: %d", r
);
676 * Store md for cleanup instead of tio which is about to get freed.
678 bio
->bi_private
= md
->bs
;
682 dec_pending(io
, error
);
686 * Partial completion handling for request-based dm
688 static void end_clone_bio(struct bio
*clone
, int error
)
690 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
691 struct dm_rq_target_io
*tio
= info
->tio
;
692 struct bio
*bio
= info
->orig
;
693 unsigned int nr_bytes
= info
->orig
->bi_size
;
699 * An error has already been detected on the request.
700 * Once error occurred, just let clone->end_io() handle
706 * Don't notice the error to the upper layer yet.
707 * The error handling decision is made by the target driver,
708 * when the request is completed.
715 * I/O for the bio successfully completed.
716 * Notice the data completion to the upper layer.
720 * bios are processed from the head of the list.
721 * So the completing bio should always be rq->bio.
722 * If it's not, something wrong is happening.
724 if (tio
->orig
->bio
!= bio
)
725 DMERR("bio completion is going in the middle of the request");
728 * Update the original request.
729 * Do not use blk_end_request() here, because it may complete
730 * the original request before the clone, and break the ordering.
732 blk_update_request(tio
->orig
, 0, nr_bytes
);
736 * Don't touch any member of the md after calling this function because
737 * the md may be freed in dm_put() at the end of this function.
738 * Or do dm_get() before calling this function and dm_put() later.
740 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
742 atomic_dec(&md
->pending
[rw
]);
744 /* nudge anyone waiting on suspend queue */
745 if (!md_in_flight(md
))
749 blk_run_queue(md
->queue
);
752 * dm_put() must be at the end of this function. See the comment above
757 static void free_rq_clone(struct request
*clone
)
759 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
761 blk_rq_unprep_clone(clone
);
766 * Complete the clone and the original request.
767 * Must be called without queue lock.
769 static void dm_end_request(struct request
*clone
, int error
)
771 int rw
= rq_data_dir(clone
);
772 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
773 struct mapped_device
*md
= tio
->md
;
774 struct request
*rq
= tio
->orig
;
776 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
777 rq
->errors
= clone
->errors
;
778 rq
->resid_len
= clone
->resid_len
;
782 * We are using the sense buffer of the original
784 * So setting the length of the sense data is enough.
786 rq
->sense_len
= clone
->sense_len
;
789 free_rq_clone(clone
);
790 blk_end_request_all(rq
, error
);
791 rq_completed(md
, rw
, true);
794 static void dm_unprep_request(struct request
*rq
)
796 struct request
*clone
= rq
->special
;
799 rq
->cmd_flags
&= ~REQ_DONTPREP
;
801 free_rq_clone(clone
);
805 * Requeue the original request of a clone.
807 void dm_requeue_unmapped_request(struct request
*clone
)
809 int rw
= rq_data_dir(clone
);
810 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
811 struct mapped_device
*md
= tio
->md
;
812 struct request
*rq
= tio
->orig
;
813 struct request_queue
*q
= rq
->q
;
816 dm_unprep_request(rq
);
818 spin_lock_irqsave(q
->queue_lock
, flags
);
819 blk_requeue_request(q
, rq
);
820 spin_unlock_irqrestore(q
->queue_lock
, flags
);
822 rq_completed(md
, rw
, 0);
824 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
826 static void __stop_queue(struct request_queue
*q
)
831 static void stop_queue(struct request_queue
*q
)
835 spin_lock_irqsave(q
->queue_lock
, flags
);
837 spin_unlock_irqrestore(q
->queue_lock
, flags
);
840 static void __start_queue(struct request_queue
*q
)
842 if (blk_queue_stopped(q
))
846 static void start_queue(struct request_queue
*q
)
850 spin_lock_irqsave(q
->queue_lock
, flags
);
852 spin_unlock_irqrestore(q
->queue_lock
, flags
);
855 static void dm_done(struct request
*clone
, int error
, bool mapped
)
858 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
859 dm_request_endio_fn rq_end_io
= tio
->ti
->type
->rq_end_io
;
861 if (mapped
&& rq_end_io
)
862 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
865 /* The target wants to complete the I/O */
866 dm_end_request(clone
, r
);
867 else if (r
== DM_ENDIO_INCOMPLETE
)
868 /* The target will handle the I/O */
870 else if (r
== DM_ENDIO_REQUEUE
)
871 /* The target wants to requeue the I/O */
872 dm_requeue_unmapped_request(clone
);
874 DMWARN("unimplemented target endio return value: %d", r
);
880 * Request completion handler for request-based dm
882 static void dm_softirq_done(struct request
*rq
)
885 struct request
*clone
= rq
->completion_data
;
886 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
888 if (rq
->cmd_flags
& REQ_FAILED
)
891 dm_done(clone
, tio
->error
, mapped
);
895 * Complete the clone and the original request with the error status
896 * through softirq context.
898 static void dm_complete_request(struct request
*clone
, int error
)
900 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
901 struct request
*rq
= tio
->orig
;
904 rq
->completion_data
= clone
;
905 blk_complete_request(rq
);
909 * Complete the not-mapped clone and the original request with the error status
910 * through softirq context.
911 * Target's rq_end_io() function isn't called.
912 * This may be used when the target's map_rq() function fails.
914 void dm_kill_unmapped_request(struct request
*clone
, int error
)
916 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
917 struct request
*rq
= tio
->orig
;
919 rq
->cmd_flags
|= REQ_FAILED
;
920 dm_complete_request(clone
, error
);
922 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
925 * Called with the queue lock held
927 static void end_clone_request(struct request
*clone
, int error
)
930 * For just cleaning up the information of the queue in which
931 * the clone was dispatched.
932 * The clone is *NOT* freed actually here because it is alloced from
933 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
935 __blk_put_request(clone
->q
, clone
);
938 * Actual request completion is done in a softirq context which doesn't
939 * hold the queue lock. Otherwise, deadlock could occur because:
940 * - another request may be submitted by the upper level driver
941 * of the stacking during the completion
942 * - the submission which requires queue lock may be done
945 dm_complete_request(clone
, error
);
949 * Return maximum size of I/O possible at the supplied sector up to the current
952 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
954 sector_t target_offset
= dm_target_offset(ti
, sector
);
956 return ti
->len
- target_offset
;
959 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
961 sector_t len
= max_io_len_target_boundary(sector
, ti
);
964 * Does the target need to split even further ?
968 sector_t offset
= dm_target_offset(ti
, sector
);
969 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
978 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
979 struct dm_target_io
*tio
)
983 struct mapped_device
*md
;
985 clone
->bi_end_io
= clone_endio
;
986 clone
->bi_private
= tio
;
989 * Map the clone. If r == 0 we don't need to do
990 * anything, the target has assumed ownership of
993 atomic_inc(&tio
->io
->io_count
);
994 sector
= clone
->bi_sector
;
995 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
996 if (r
== DM_MAPIO_REMAPPED
) {
997 /* the bio has been remapped so dispatch it */
999 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1000 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1002 generic_make_request(clone
);
1003 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1004 /* error the io and bail out, or requeue it if needed */
1006 dec_pending(tio
->io
, r
);
1008 * Store bio_set for cleanup.
1010 clone
->bi_private
= md
->bs
;
1014 DMWARN("unimplemented target map return value: %d", r
);
1020 struct mapped_device
*md
;
1021 struct dm_table
*map
;
1025 sector_t sector_count
;
1029 static void dm_bio_destructor(struct bio
*bio
)
1031 struct bio_set
*bs
= bio
->bi_private
;
1037 * Creates a little bio that just does part of a bvec.
1039 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
1040 unsigned short idx
, unsigned int offset
,
1041 unsigned int len
, struct bio_set
*bs
)
1044 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1046 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
1047 clone
->bi_destructor
= dm_bio_destructor
;
1048 *clone
->bi_io_vec
= *bv
;
1050 clone
->bi_sector
= sector
;
1051 clone
->bi_bdev
= bio
->bi_bdev
;
1052 clone
->bi_rw
= bio
->bi_rw
;
1054 clone
->bi_size
= to_bytes(len
);
1055 clone
->bi_io_vec
->bv_offset
= offset
;
1056 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1057 clone
->bi_flags
|= 1 << BIO_CLONED
;
1059 if (bio_integrity(bio
)) {
1060 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1061 bio_integrity_trim(clone
,
1062 bio_sector_offset(bio
, idx
, offset
), len
);
1069 * Creates a bio that consists of range of complete bvecs.
1071 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
1072 unsigned short idx
, unsigned short bv_count
,
1073 unsigned int len
, struct bio_set
*bs
)
1077 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
1078 __bio_clone(clone
, bio
);
1079 clone
->bi_destructor
= dm_bio_destructor
;
1080 clone
->bi_sector
= sector
;
1081 clone
->bi_idx
= idx
;
1082 clone
->bi_vcnt
= idx
+ bv_count
;
1083 clone
->bi_size
= to_bytes(len
);
1084 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1086 if (bio_integrity(bio
)) {
1087 bio_integrity_clone(clone
, bio
, GFP_NOIO
, bs
);
1089 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1090 bio_integrity_trim(clone
,
1091 bio_sector_offset(bio
, idx
, 0), len
);
1097 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1098 struct dm_target
*ti
)
1100 struct dm_target_io
*tio
= mempool_alloc(ci
->md
->tio_pool
, GFP_NOIO
);
1104 memset(&tio
->info
, 0, sizeof(tio
->info
));
1109 static void __issue_target_request(struct clone_info
*ci
, struct dm_target
*ti
,
1110 unsigned request_nr
, sector_t len
)
1112 struct dm_target_io
*tio
= alloc_tio(ci
, ti
);
1115 tio
->info
.target_request_nr
= request_nr
;
1118 * Discard requests require the bio's inline iovecs be initialized.
1119 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1120 * and discard, so no need for concern about wasted bvec allocations.
1122 clone
= bio_alloc_bioset(GFP_NOIO
, ci
->bio
->bi_max_vecs
, ci
->md
->bs
);
1123 __bio_clone(clone
, ci
->bio
);
1124 clone
->bi_destructor
= dm_bio_destructor
;
1126 clone
->bi_sector
= ci
->sector
;
1127 clone
->bi_size
= to_bytes(len
);
1130 __map_bio(ti
, clone
, tio
);
1133 static void __issue_target_requests(struct clone_info
*ci
, struct dm_target
*ti
,
1134 unsigned num_requests
, sector_t len
)
1136 unsigned request_nr
;
1138 for (request_nr
= 0; request_nr
< num_requests
; request_nr
++)
1139 __issue_target_request(ci
, ti
, request_nr
, len
);
1142 static int __clone_and_map_empty_flush(struct clone_info
*ci
)
1144 unsigned target_nr
= 0;
1145 struct dm_target
*ti
;
1147 BUG_ON(bio_has_data(ci
->bio
));
1148 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1149 __issue_target_requests(ci
, ti
, ti
->num_flush_requests
, 0);
1155 * Perform all io with a single clone.
1157 static void __clone_and_map_simple(struct clone_info
*ci
, struct dm_target
*ti
)
1159 struct bio
*clone
, *bio
= ci
->bio
;
1160 struct dm_target_io
*tio
;
1162 tio
= alloc_tio(ci
, ti
);
1163 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
1164 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
1166 __map_bio(ti
, clone
, tio
);
1167 ci
->sector_count
= 0;
1170 static int __clone_and_map_discard(struct clone_info
*ci
)
1172 struct dm_target
*ti
;
1176 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1177 if (!dm_target_is_valid(ti
))
1181 * Even though the device advertised discard support,
1182 * reconfiguration might have changed that since the
1183 * check was performed.
1185 if (!ti
->num_discard_requests
)
1188 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1190 __issue_target_requests(ci
, ti
, ti
->num_discard_requests
, len
);
1193 } while (ci
->sector_count
-= len
);
1198 static int __clone_and_map(struct clone_info
*ci
)
1200 struct bio
*clone
, *bio
= ci
->bio
;
1201 struct dm_target
*ti
;
1202 sector_t len
= 0, max
;
1203 struct dm_target_io
*tio
;
1205 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1206 return __clone_and_map_discard(ci
);
1208 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1209 if (!dm_target_is_valid(ti
))
1212 max
= max_io_len(ci
->sector
, ti
);
1214 if (ci
->sector_count
<= max
) {
1216 * Optimise for the simple case where we can do all of
1217 * the remaining io with a single clone.
1219 __clone_and_map_simple(ci
, ti
);
1221 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1223 * There are some bvecs that don't span targets.
1224 * Do as many of these as possible.
1227 sector_t remaining
= max
;
1230 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1231 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1233 if (bv_len
> remaining
)
1236 remaining
-= bv_len
;
1240 tio
= alloc_tio(ci
, ti
);
1241 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1243 __map_bio(ti
, clone
, tio
);
1246 ci
->sector_count
-= len
;
1251 * Handle a bvec that must be split between two or more targets.
1253 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1254 sector_t remaining
= to_sector(bv
->bv_len
);
1255 unsigned int offset
= 0;
1259 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1260 if (!dm_target_is_valid(ti
))
1263 max
= max_io_len(ci
->sector
, ti
);
1266 len
= min(remaining
, max
);
1268 tio
= alloc_tio(ci
, ti
);
1269 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
1270 bv
->bv_offset
+ offset
, len
,
1273 __map_bio(ti
, clone
, tio
);
1276 ci
->sector_count
-= len
;
1277 offset
+= to_bytes(len
);
1278 } while (remaining
-= len
);
1287 * Split the bio into several clones and submit it to targets.
1289 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1291 struct clone_info ci
;
1294 ci
.map
= dm_get_live_table(md
);
1295 if (unlikely(!ci
.map
)) {
1301 ci
.io
= alloc_io(md
);
1303 atomic_set(&ci
.io
->io_count
, 1);
1306 spin_lock_init(&ci
.io
->endio_lock
);
1307 ci
.sector
= bio
->bi_sector
;
1308 ci
.idx
= bio
->bi_idx
;
1310 start_io_acct(ci
.io
);
1311 if (bio
->bi_rw
& REQ_FLUSH
) {
1312 ci
.bio
= &ci
.md
->flush_bio
;
1313 ci
.sector_count
= 0;
1314 error
= __clone_and_map_empty_flush(&ci
);
1315 /* dec_pending submits any data associated with flush */
1318 ci
.sector_count
= bio_sectors(bio
);
1319 while (ci
.sector_count
&& !error
)
1320 error
= __clone_and_map(&ci
);
1323 /* drop the extra reference count */
1324 dec_pending(ci
.io
, error
);
1325 dm_table_put(ci
.map
);
1327 /*-----------------------------------------------------------------
1329 *---------------------------------------------------------------*/
1331 static int dm_merge_bvec(struct request_queue
*q
,
1332 struct bvec_merge_data
*bvm
,
1333 struct bio_vec
*biovec
)
1335 struct mapped_device
*md
= q
->queuedata
;
1336 struct dm_table
*map
= dm_get_live_table(md
);
1337 struct dm_target
*ti
;
1338 sector_t max_sectors
;
1344 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1345 if (!dm_target_is_valid(ti
))
1349 * Find maximum amount of I/O that won't need splitting
1351 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1352 (sector_t
) BIO_MAX_SECTORS
);
1353 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1358 * merge_bvec_fn() returns number of bytes
1359 * it can accept at this offset
1360 * max is precomputed maximal io size
1362 if (max_size
&& ti
->type
->merge
)
1363 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1365 * If the target doesn't support merge method and some of the devices
1366 * provided their merge_bvec method (we know this by looking at
1367 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1368 * entries. So always set max_size to 0, and the code below allows
1371 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1380 * Always allow an entire first page
1382 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1383 max_size
= biovec
->bv_len
;
1389 * The request function that just remaps the bio built up by
1392 static int _dm_request(struct request_queue
*q
, struct bio
*bio
)
1394 int rw
= bio_data_dir(bio
);
1395 struct mapped_device
*md
= q
->queuedata
;
1398 down_read(&md
->io_lock
);
1400 cpu
= part_stat_lock();
1401 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1402 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1405 /* if we're suspended, we have to queue this io for later */
1406 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1407 up_read(&md
->io_lock
);
1409 if (bio_rw(bio
) != READA
)
1416 __split_and_process_bio(md
, bio
);
1417 up_read(&md
->io_lock
);
1421 static int dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1423 struct mapped_device
*md
= q
->queuedata
;
1425 return md
->saved_make_request_fn(q
, bio
); /* call __make_request() */
1428 static int dm_request_based(struct mapped_device
*md
)
1430 return blk_queue_stackable(md
->queue
);
1433 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
1435 struct mapped_device
*md
= q
->queuedata
;
1437 if (dm_request_based(md
))
1438 return dm_make_request(q
, bio
);
1440 return _dm_request(q
, bio
);
1443 void dm_dispatch_request(struct request
*rq
)
1447 if (blk_queue_io_stat(rq
->q
))
1448 rq
->cmd_flags
|= REQ_IO_STAT
;
1450 rq
->start_time
= jiffies
;
1451 r
= blk_insert_cloned_request(rq
->q
, rq
);
1453 dm_complete_request(rq
, r
);
1455 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1457 static void dm_rq_bio_destructor(struct bio
*bio
)
1459 struct dm_rq_clone_bio_info
*info
= bio
->bi_private
;
1460 struct mapped_device
*md
= info
->tio
->md
;
1462 free_bio_info(info
);
1463 bio_free(bio
, md
->bs
);
1466 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1469 struct dm_rq_target_io
*tio
= data
;
1470 struct mapped_device
*md
= tio
->md
;
1471 struct dm_rq_clone_bio_info
*info
= alloc_bio_info(md
);
1476 info
->orig
= bio_orig
;
1478 bio
->bi_end_io
= end_clone_bio
;
1479 bio
->bi_private
= info
;
1480 bio
->bi_destructor
= dm_rq_bio_destructor
;
1485 static int setup_clone(struct request
*clone
, struct request
*rq
,
1486 struct dm_rq_target_io
*tio
)
1490 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1491 dm_rq_bio_constructor
, tio
);
1495 clone
->cmd
= rq
->cmd
;
1496 clone
->cmd_len
= rq
->cmd_len
;
1497 clone
->sense
= rq
->sense
;
1498 clone
->buffer
= rq
->buffer
;
1499 clone
->end_io
= end_clone_request
;
1500 clone
->end_io_data
= tio
;
1505 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1508 struct request
*clone
;
1509 struct dm_rq_target_io
*tio
;
1511 tio
= alloc_rq_tio(md
, gfp_mask
);
1519 memset(&tio
->info
, 0, sizeof(tio
->info
));
1521 clone
= &tio
->clone
;
1522 if (setup_clone(clone
, rq
, tio
)) {
1532 * Called with the queue lock held.
1534 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1536 struct mapped_device
*md
= q
->queuedata
;
1537 struct request
*clone
;
1539 if (unlikely(rq
->special
)) {
1540 DMWARN("Already has something in rq->special.");
1541 return BLKPREP_KILL
;
1544 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1546 return BLKPREP_DEFER
;
1548 rq
->special
= clone
;
1549 rq
->cmd_flags
|= REQ_DONTPREP
;
1556 * 0 : the request has been processed (not requeued)
1557 * !0 : the request has been requeued
1559 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1560 struct mapped_device
*md
)
1562 int r
, requeued
= 0;
1563 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1566 * Hold the md reference here for the in-flight I/O.
1567 * We can't rely on the reference count by device opener,
1568 * because the device may be closed during the request completion
1569 * when all bios are completed.
1570 * See the comment in rq_completed() too.
1575 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1577 case DM_MAPIO_SUBMITTED
:
1578 /* The target has taken the I/O to submit by itself later */
1580 case DM_MAPIO_REMAPPED
:
1581 /* The target has remapped the I/O so dispatch it */
1582 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1583 blk_rq_pos(tio
->orig
));
1584 dm_dispatch_request(clone
);
1586 case DM_MAPIO_REQUEUE
:
1587 /* The target wants to requeue the I/O */
1588 dm_requeue_unmapped_request(clone
);
1593 DMWARN("unimplemented target map return value: %d", r
);
1597 /* The target wants to complete the I/O */
1598 dm_kill_unmapped_request(clone
, r
);
1606 * q->request_fn for request-based dm.
1607 * Called with the queue lock held.
1609 static void dm_request_fn(struct request_queue
*q
)
1611 struct mapped_device
*md
= q
->queuedata
;
1612 struct dm_table
*map
= dm_get_live_table(md
);
1613 struct dm_target
*ti
;
1614 struct request
*rq
, *clone
;
1618 * For suspend, check blk_queue_stopped() and increment
1619 * ->pending within a single queue_lock not to increment the
1620 * number of in-flight I/Os after the queue is stopped in
1623 while (!blk_queue_stopped(q
)) {
1624 rq
= blk_peek_request(q
);
1628 /* always use block 0 to find the target for flushes for now */
1630 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1631 pos
= blk_rq_pos(rq
);
1633 ti
= dm_table_find_target(map
, pos
);
1634 BUG_ON(!dm_target_is_valid(ti
));
1636 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1639 blk_start_request(rq
);
1640 clone
= rq
->special
;
1641 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1643 spin_unlock(q
->queue_lock
);
1644 if (map_request(ti
, clone
, md
))
1647 BUG_ON(!irqs_disabled());
1648 spin_lock(q
->queue_lock
);
1654 BUG_ON(!irqs_disabled());
1655 spin_lock(q
->queue_lock
);
1658 blk_delay_queue(q
, HZ
/ 10);
1665 int dm_underlying_device_busy(struct request_queue
*q
)
1667 return blk_lld_busy(q
);
1669 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1671 static int dm_lld_busy(struct request_queue
*q
)
1674 struct mapped_device
*md
= q
->queuedata
;
1675 struct dm_table
*map
= dm_get_live_table(md
);
1677 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1680 r
= dm_table_any_busy_target(map
);
1687 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1690 struct mapped_device
*md
= congested_data
;
1691 struct dm_table
*map
;
1693 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1694 map
= dm_get_live_table(md
);
1697 * Request-based dm cares about only own queue for
1698 * the query about congestion status of request_queue
1700 if (dm_request_based(md
))
1701 r
= md
->queue
->backing_dev_info
.state
&
1704 r
= dm_table_any_congested(map
, bdi_bits
);
1713 /*-----------------------------------------------------------------
1714 * An IDR is used to keep track of allocated minor numbers.
1715 *---------------------------------------------------------------*/
1716 static void free_minor(int minor
)
1718 spin_lock(&_minor_lock
);
1719 idr_remove(&_minor_idr
, minor
);
1720 spin_unlock(&_minor_lock
);
1724 * See if the device with a specific minor # is free.
1726 static int specific_minor(int minor
)
1730 if (minor
>= (1 << MINORBITS
))
1733 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1737 spin_lock(&_minor_lock
);
1739 if (idr_find(&_minor_idr
, minor
)) {
1744 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1749 idr_remove(&_minor_idr
, m
);
1755 spin_unlock(&_minor_lock
);
1759 static int next_free_minor(int *minor
)
1763 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1767 spin_lock(&_minor_lock
);
1769 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1773 if (m
>= (1 << MINORBITS
)) {
1774 idr_remove(&_minor_idr
, m
);
1782 spin_unlock(&_minor_lock
);
1786 static const struct block_device_operations dm_blk_dops
;
1788 static void dm_wq_work(struct work_struct
*work
);
1790 static void dm_init_md_queue(struct mapped_device
*md
)
1793 * Request-based dm devices cannot be stacked on top of bio-based dm
1794 * devices. The type of this dm device has not been decided yet.
1795 * The type is decided at the first table loading time.
1796 * To prevent problematic device stacking, clear the queue flag
1797 * for request stacking support until then.
1799 * This queue is new, so no concurrency on the queue_flags.
1801 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1803 md
->queue
->queuedata
= md
;
1804 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1805 md
->queue
->backing_dev_info
.congested_data
= md
;
1806 blk_queue_make_request(md
->queue
, dm_request
);
1807 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1808 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1809 blk_queue_flush(md
->queue
, REQ_FLUSH
| REQ_FUA
);
1813 * Allocate and initialise a blank device with a given minor.
1815 static struct mapped_device
*alloc_dev(int minor
)
1818 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1822 DMWARN("unable to allocate device, out of memory.");
1826 if (!try_module_get(THIS_MODULE
))
1827 goto bad_module_get
;
1829 /* get a minor number for the dev */
1830 if (minor
== DM_ANY_MINOR
)
1831 r
= next_free_minor(&minor
);
1833 r
= specific_minor(minor
);
1837 md
->type
= DM_TYPE_NONE
;
1838 init_rwsem(&md
->io_lock
);
1839 mutex_init(&md
->suspend_lock
);
1840 mutex_init(&md
->type_lock
);
1841 spin_lock_init(&md
->deferred_lock
);
1842 rwlock_init(&md
->map_lock
);
1843 atomic_set(&md
->holders
, 1);
1844 atomic_set(&md
->open_count
, 0);
1845 atomic_set(&md
->event_nr
, 0);
1846 atomic_set(&md
->uevent_seq
, 0);
1847 INIT_LIST_HEAD(&md
->uevent_list
);
1848 spin_lock_init(&md
->uevent_lock
);
1850 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1854 dm_init_md_queue(md
);
1856 md
->disk
= alloc_disk(1);
1860 atomic_set(&md
->pending
[0], 0);
1861 atomic_set(&md
->pending
[1], 0);
1862 init_waitqueue_head(&md
->wait
);
1863 INIT_WORK(&md
->work
, dm_wq_work
);
1864 init_waitqueue_head(&md
->eventq
);
1866 md
->disk
->major
= _major
;
1867 md
->disk
->first_minor
= minor
;
1868 md
->disk
->fops
= &dm_blk_dops
;
1869 md
->disk
->queue
= md
->queue
;
1870 md
->disk
->private_data
= md
;
1871 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1873 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1875 md
->wq
= alloc_workqueue("kdmflush",
1876 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1880 md
->bdev
= bdget_disk(md
->disk
, 0);
1884 bio_init(&md
->flush_bio
);
1885 md
->flush_bio
.bi_bdev
= md
->bdev
;
1886 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1888 /* Populate the mapping, nobody knows we exist yet */
1889 spin_lock(&_minor_lock
);
1890 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1891 spin_unlock(&_minor_lock
);
1893 BUG_ON(old_md
!= MINOR_ALLOCED
);
1898 destroy_workqueue(md
->wq
);
1900 del_gendisk(md
->disk
);
1903 blk_cleanup_queue(md
->queue
);
1907 module_put(THIS_MODULE
);
1913 static void unlock_fs(struct mapped_device
*md
);
1915 static void free_dev(struct mapped_device
*md
)
1917 int minor
= MINOR(disk_devt(md
->disk
));
1921 destroy_workqueue(md
->wq
);
1923 mempool_destroy(md
->tio_pool
);
1925 mempool_destroy(md
->io_pool
);
1927 bioset_free(md
->bs
);
1928 blk_integrity_unregister(md
->disk
);
1929 del_gendisk(md
->disk
);
1932 spin_lock(&_minor_lock
);
1933 md
->disk
->private_data
= NULL
;
1934 spin_unlock(&_minor_lock
);
1937 blk_cleanup_queue(md
->queue
);
1938 module_put(THIS_MODULE
);
1942 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1944 struct dm_md_mempools
*p
;
1946 if (md
->io_pool
&& md
->tio_pool
&& md
->bs
)
1947 /* the md already has necessary mempools */
1950 p
= dm_table_get_md_mempools(t
);
1951 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1953 md
->io_pool
= p
->io_pool
;
1955 md
->tio_pool
= p
->tio_pool
;
1961 /* mempool bind completed, now no need any mempools in the table */
1962 dm_table_free_md_mempools(t
);
1966 * Bind a table to the device.
1968 static void event_callback(void *context
)
1970 unsigned long flags
;
1972 struct mapped_device
*md
= (struct mapped_device
*) context
;
1974 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1975 list_splice_init(&md
->uevent_list
, &uevents
);
1976 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1978 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1980 atomic_inc(&md
->event_nr
);
1981 wake_up(&md
->eventq
);
1985 * Protected by md->suspend_lock obtained by dm_swap_table().
1987 static void __set_size(struct mapped_device
*md
, sector_t size
)
1989 set_capacity(md
->disk
, size
);
1991 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1995 * Returns old map, which caller must destroy.
1997 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1998 struct queue_limits
*limits
)
2000 struct dm_table
*old_map
;
2001 struct request_queue
*q
= md
->queue
;
2003 unsigned long flags
;
2005 size
= dm_table_get_size(t
);
2008 * Wipe any geometry if the size of the table changed.
2010 if (size
!= get_capacity(md
->disk
))
2011 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2013 __set_size(md
, size
);
2015 dm_table_event_callback(t
, event_callback
, md
);
2018 * The queue hasn't been stopped yet, if the old table type wasn't
2019 * for request-based during suspension. So stop it to prevent
2020 * I/O mapping before resume.
2021 * This must be done before setting the queue restrictions,
2022 * because request-based dm may be run just after the setting.
2024 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2027 __bind_mempools(md
, t
);
2029 write_lock_irqsave(&md
->map_lock
, flags
);
2032 dm_table_set_restrictions(t
, q
, limits
);
2033 write_unlock_irqrestore(&md
->map_lock
, flags
);
2039 * Returns unbound table for the caller to free.
2041 static struct dm_table
*__unbind(struct mapped_device
*md
)
2043 struct dm_table
*map
= md
->map
;
2044 unsigned long flags
;
2049 dm_table_event_callback(map
, NULL
, NULL
);
2050 write_lock_irqsave(&md
->map_lock
, flags
);
2052 write_unlock_irqrestore(&md
->map_lock
, flags
);
2058 * Constructor for a new device.
2060 int dm_create(int minor
, struct mapped_device
**result
)
2062 struct mapped_device
*md
;
2064 md
= alloc_dev(minor
);
2075 * Functions to manage md->type.
2076 * All are required to hold md->type_lock.
2078 void dm_lock_md_type(struct mapped_device
*md
)
2080 mutex_lock(&md
->type_lock
);
2083 void dm_unlock_md_type(struct mapped_device
*md
)
2085 mutex_unlock(&md
->type_lock
);
2088 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2093 unsigned dm_get_md_type(struct mapped_device
*md
)
2099 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2101 static int dm_init_request_based_queue(struct mapped_device
*md
)
2103 struct request_queue
*q
= NULL
;
2105 if (md
->queue
->elevator
)
2108 /* Fully initialize the queue */
2109 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2114 md
->saved_make_request_fn
= md
->queue
->make_request_fn
;
2115 dm_init_md_queue(md
);
2116 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2117 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2118 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2120 elv_register_queue(md
->queue
);
2126 * Setup the DM device's queue based on md's type
2128 int dm_setup_md_queue(struct mapped_device
*md
)
2130 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2131 !dm_init_request_based_queue(md
)) {
2132 DMWARN("Cannot initialize queue for request-based mapped device");
2139 static struct mapped_device
*dm_find_md(dev_t dev
)
2141 struct mapped_device
*md
;
2142 unsigned minor
= MINOR(dev
);
2144 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2147 spin_lock(&_minor_lock
);
2149 md
= idr_find(&_minor_idr
, minor
);
2150 if (md
&& (md
== MINOR_ALLOCED
||
2151 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2152 dm_deleting_md(md
) ||
2153 test_bit(DMF_FREEING
, &md
->flags
))) {
2159 spin_unlock(&_minor_lock
);
2164 struct mapped_device
*dm_get_md(dev_t dev
)
2166 struct mapped_device
*md
= dm_find_md(dev
);
2174 void *dm_get_mdptr(struct mapped_device
*md
)
2176 return md
->interface_ptr
;
2179 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2181 md
->interface_ptr
= ptr
;
2184 void dm_get(struct mapped_device
*md
)
2186 atomic_inc(&md
->holders
);
2187 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2190 const char *dm_device_name(struct mapped_device
*md
)
2194 EXPORT_SYMBOL_GPL(dm_device_name
);
2196 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2198 struct dm_table
*map
;
2202 spin_lock(&_minor_lock
);
2203 map
= dm_get_live_table(md
);
2204 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2205 set_bit(DMF_FREEING
, &md
->flags
);
2206 spin_unlock(&_minor_lock
);
2208 if (!dm_suspended_md(md
)) {
2209 dm_table_presuspend_targets(map
);
2210 dm_table_postsuspend_targets(map
);
2214 * Rare, but there may be I/O requests still going to complete,
2215 * for example. Wait for all references to disappear.
2216 * No one should increment the reference count of the mapped_device,
2217 * after the mapped_device state becomes DMF_FREEING.
2220 while (atomic_read(&md
->holders
))
2222 else if (atomic_read(&md
->holders
))
2223 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2224 dm_device_name(md
), atomic_read(&md
->holders
));
2228 dm_table_destroy(__unbind(md
));
2232 void dm_destroy(struct mapped_device
*md
)
2234 __dm_destroy(md
, true);
2237 void dm_destroy_immediate(struct mapped_device
*md
)
2239 __dm_destroy(md
, false);
2242 void dm_put(struct mapped_device
*md
)
2244 atomic_dec(&md
->holders
);
2246 EXPORT_SYMBOL_GPL(dm_put
);
2248 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2251 DECLARE_WAITQUEUE(wait
, current
);
2253 add_wait_queue(&md
->wait
, &wait
);
2256 set_current_state(interruptible
);
2259 if (!md_in_flight(md
))
2262 if (interruptible
== TASK_INTERRUPTIBLE
&&
2263 signal_pending(current
)) {
2270 set_current_state(TASK_RUNNING
);
2272 remove_wait_queue(&md
->wait
, &wait
);
2278 * Process the deferred bios
2280 static void dm_wq_work(struct work_struct
*work
)
2282 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2286 down_read(&md
->io_lock
);
2288 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2289 spin_lock_irq(&md
->deferred_lock
);
2290 c
= bio_list_pop(&md
->deferred
);
2291 spin_unlock_irq(&md
->deferred_lock
);
2296 up_read(&md
->io_lock
);
2298 if (dm_request_based(md
))
2299 generic_make_request(c
);
2301 __split_and_process_bio(md
, c
);
2303 down_read(&md
->io_lock
);
2306 up_read(&md
->io_lock
);
2309 static void dm_queue_flush(struct mapped_device
*md
)
2311 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2312 smp_mb__after_clear_bit();
2313 queue_work(md
->wq
, &md
->work
);
2317 * Swap in a new table, returning the old one for the caller to destroy.
2319 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2321 struct dm_table
*map
= ERR_PTR(-EINVAL
);
2322 struct queue_limits limits
;
2325 mutex_lock(&md
->suspend_lock
);
2327 /* device must be suspended */
2328 if (!dm_suspended_md(md
))
2331 r
= dm_calculate_queue_limits(table
, &limits
);
2337 map
= __bind(md
, table
, &limits
);
2340 mutex_unlock(&md
->suspend_lock
);
2345 * Functions to lock and unlock any filesystem running on the
2348 static int lock_fs(struct mapped_device
*md
)
2352 WARN_ON(md
->frozen_sb
);
2354 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2355 if (IS_ERR(md
->frozen_sb
)) {
2356 r
= PTR_ERR(md
->frozen_sb
);
2357 md
->frozen_sb
= NULL
;
2361 set_bit(DMF_FROZEN
, &md
->flags
);
2366 static void unlock_fs(struct mapped_device
*md
)
2368 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2371 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2372 md
->frozen_sb
= NULL
;
2373 clear_bit(DMF_FROZEN
, &md
->flags
);
2377 * We need to be able to change a mapping table under a mounted
2378 * filesystem. For example we might want to move some data in
2379 * the background. Before the table can be swapped with
2380 * dm_bind_table, dm_suspend must be called to flush any in
2381 * flight bios and ensure that any further io gets deferred.
2384 * Suspend mechanism in request-based dm.
2386 * 1. Flush all I/Os by lock_fs() if needed.
2387 * 2. Stop dispatching any I/O by stopping the request_queue.
2388 * 3. Wait for all in-flight I/Os to be completed or requeued.
2390 * To abort suspend, start the request_queue.
2392 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2394 struct dm_table
*map
= NULL
;
2396 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2397 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2399 mutex_lock(&md
->suspend_lock
);
2401 if (dm_suspended_md(md
)) {
2406 map
= dm_get_live_table(md
);
2409 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2410 * This flag is cleared before dm_suspend returns.
2413 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2415 /* This does not get reverted if there's an error later. */
2416 dm_table_presuspend_targets(map
);
2419 * Flush I/O to the device.
2420 * Any I/O submitted after lock_fs() may not be flushed.
2421 * noflush takes precedence over do_lockfs.
2422 * (lock_fs() flushes I/Os and waits for them to complete.)
2424 if (!noflush
&& do_lockfs
) {
2431 * Here we must make sure that no processes are submitting requests
2432 * to target drivers i.e. no one may be executing
2433 * __split_and_process_bio. This is called from dm_request and
2436 * To get all processes out of __split_and_process_bio in dm_request,
2437 * we take the write lock. To prevent any process from reentering
2438 * __split_and_process_bio from dm_request and quiesce the thread
2439 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2440 * flush_workqueue(md->wq).
2442 down_write(&md
->io_lock
);
2443 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2444 up_write(&md
->io_lock
);
2447 * Stop md->queue before flushing md->wq in case request-based
2448 * dm defers requests to md->wq from md->queue.
2450 if (dm_request_based(md
))
2451 stop_queue(md
->queue
);
2453 flush_workqueue(md
->wq
);
2456 * At this point no more requests are entering target request routines.
2457 * We call dm_wait_for_completion to wait for all existing requests
2460 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2462 down_write(&md
->io_lock
);
2464 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2465 up_write(&md
->io_lock
);
2467 /* were we interrupted ? */
2471 if (dm_request_based(md
))
2472 start_queue(md
->queue
);
2475 goto out
; /* pushback list is already flushed, so skip flush */
2479 * If dm_wait_for_completion returned 0, the device is completely
2480 * quiescent now. There is no request-processing activity. All new
2481 * requests are being added to md->deferred list.
2484 set_bit(DMF_SUSPENDED
, &md
->flags
);
2486 dm_table_postsuspend_targets(map
);
2492 mutex_unlock(&md
->suspend_lock
);
2496 int dm_resume(struct mapped_device
*md
)
2499 struct dm_table
*map
= NULL
;
2501 mutex_lock(&md
->suspend_lock
);
2502 if (!dm_suspended_md(md
))
2505 map
= dm_get_live_table(md
);
2506 if (!map
|| !dm_table_get_size(map
))
2509 r
= dm_table_resume_targets(map
);
2516 * Flushing deferred I/Os must be done after targets are resumed
2517 * so that mapping of targets can work correctly.
2518 * Request-based dm is queueing the deferred I/Os in its request_queue.
2520 if (dm_request_based(md
))
2521 start_queue(md
->queue
);
2525 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2530 mutex_unlock(&md
->suspend_lock
);
2535 /*-----------------------------------------------------------------
2536 * Event notification.
2537 *---------------------------------------------------------------*/
2538 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2541 char udev_cookie
[DM_COOKIE_LENGTH
];
2542 char *envp
[] = { udev_cookie
, NULL
};
2545 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2547 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2548 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2549 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2554 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2556 return atomic_add_return(1, &md
->uevent_seq
);
2559 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2561 return atomic_read(&md
->event_nr
);
2564 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2566 return wait_event_interruptible(md
->eventq
,
2567 (event_nr
!= atomic_read(&md
->event_nr
)));
2570 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2572 unsigned long flags
;
2574 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2575 list_add(elist
, &md
->uevent_list
);
2576 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2580 * The gendisk is only valid as long as you have a reference
2583 struct gendisk
*dm_disk(struct mapped_device
*md
)
2588 struct kobject
*dm_kobject(struct mapped_device
*md
)
2594 * struct mapped_device should not be exported outside of dm.c
2595 * so use this check to verify that kobj is part of md structure
2597 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2599 struct mapped_device
*md
;
2601 md
= container_of(kobj
, struct mapped_device
, kobj
);
2602 if (&md
->kobj
!= kobj
)
2605 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2613 int dm_suspended_md(struct mapped_device
*md
)
2615 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2618 int dm_suspended(struct dm_target
*ti
)
2620 return dm_suspended_md(dm_table_get_md(ti
->table
));
2622 EXPORT_SYMBOL_GPL(dm_suspended
);
2624 int dm_noflush_suspending(struct dm_target
*ti
)
2626 return __noflush_suspending(dm_table_get_md(ti
->table
));
2628 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2630 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
)
2632 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2633 unsigned int pool_size
= (type
== DM_TYPE_BIO_BASED
) ? 16 : MIN_IOS
;
2638 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2639 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2640 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2641 if (!pools
->io_pool
)
2642 goto free_pools_and_out
;
2644 pools
->tio_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2645 mempool_create_slab_pool(MIN_IOS
, _tio_cache
) :
2646 mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2647 if (!pools
->tio_pool
)
2648 goto free_io_pool_and_out
;
2650 pools
->bs
= bioset_create(pool_size
, 0);
2652 goto free_tio_pool_and_out
;
2654 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2655 goto free_bioset_and_out
;
2659 free_bioset_and_out
:
2660 bioset_free(pools
->bs
);
2662 free_tio_pool_and_out
:
2663 mempool_destroy(pools
->tio_pool
);
2665 free_io_pool_and_out
:
2666 mempool_destroy(pools
->io_pool
);
2674 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2680 mempool_destroy(pools
->io_pool
);
2682 if (pools
->tio_pool
)
2683 mempool_destroy(pools
->tio_pool
);
2686 bioset_free(pools
->bs
);
2691 static const struct block_device_operations dm_blk_dops
= {
2692 .open
= dm_blk_open
,
2693 .release
= dm_blk_close
,
2694 .ioctl
= dm_blk_ioctl
,
2695 .getgeo
= dm_blk_getgeo
,
2696 .owner
= THIS_MODULE
2699 EXPORT_SYMBOL(dm_get_mapinfo
);
2704 module_init(dm_init
);
2705 module_exit(dm_exit
);
2707 module_param(major
, uint
, 0);
2708 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2709 MODULE_DESCRIPTION(DM_NAME
" driver");
2710 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2711 MODULE_LICENSE("GPL");