2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
32 DEFAULT_RATELIMIT_INTERVAL
,
33 DEFAULT_RATELIMIT_BURST
);
34 EXPORT_SYMBOL(dm_ratelimit_state
);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name
= DM_NAME
;
46 static unsigned int major
= 0;
47 static unsigned int _major
= 0;
49 static DEFINE_IDR(_minor_idr
);
51 static DEFINE_SPINLOCK(_minor_lock
);
54 * One of these is allocated per bio.
57 struct mapped_device
*md
;
61 unsigned long start_time
;
62 spinlock_t endio_lock
;
67 * One of these is allocated per target within a bio. Hopefully
68 * this will be simplified out one day.
77 * For request-based dm.
78 * One of these is allocated per request.
80 struct dm_rq_target_io
{
81 struct mapped_device
*md
;
83 struct request
*orig
, clone
;
89 * For request-based dm - the bio clones we allocate are embedded in these
92 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
93 * the bioset is created - this means the bio has to come at the end of the
96 struct dm_rq_clone_bio_info
{
98 struct dm_rq_target_io
*tio
;
102 union map_info
*dm_get_mapinfo(struct bio
*bio
)
104 if (bio
&& bio
->bi_private
)
105 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
109 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
111 if (rq
&& rq
->end_io_data
)
112 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
115 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
117 #define MINOR_ALLOCED ((void *)-1)
120 * Bits for the md->flags field.
122 #define DMF_BLOCK_IO_FOR_SUSPEND 0
123 #define DMF_SUSPENDED 1
125 #define DMF_FREEING 3
126 #define DMF_DELETING 4
127 #define DMF_NOFLUSH_SUSPENDING 5
128 #define DMF_MERGE_IS_OPTIONAL 6
131 * Work processed by per-device workqueue.
133 struct mapped_device
{
134 struct rw_semaphore io_lock
;
135 struct mutex suspend_lock
;
142 struct request_queue
*queue
;
144 /* Protect queue and type against concurrent access. */
145 struct mutex type_lock
;
147 struct target_type
*immutable_target_type
;
149 struct gendisk
*disk
;
155 * A list of ios that arrived while we were suspended.
158 wait_queue_head_t wait
;
159 struct work_struct work
;
160 struct bio_list deferred
;
161 spinlock_t deferred_lock
;
164 * Processing queue (flush)
166 struct workqueue_struct
*wq
;
169 * The current mapping.
171 struct dm_table
*map
;
174 * io objects are allocated from here.
185 wait_queue_head_t eventq
;
187 struct list_head uevent_list
;
188 spinlock_t uevent_lock
; /* Protect access to uevent_list */
191 * freeze/thaw support require holding onto a super block
193 struct super_block
*frozen_sb
;
194 struct block_device
*bdev
;
196 /* forced geometry settings */
197 struct hd_geometry geometry
;
202 /* zero-length flush that will be cloned and submitted to targets */
203 struct bio flush_bio
;
207 * For mempools pre-allocation at the table loading time.
209 struct dm_md_mempools
{
216 static struct kmem_cache
*_io_cache
;
217 static struct kmem_cache
*_tio_cache
;
218 static struct kmem_cache
*_rq_tio_cache
;
221 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
222 * still used for _io_cache, I'm leaving this for a later cleanup
224 static struct kmem_cache
*_rq_bio_info_cache
;
226 static int __init
local_init(void)
230 /* allocate a slab for the dm_ios */
231 _io_cache
= KMEM_CACHE(dm_io
, 0);
235 /* allocate a slab for the target ios */
236 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
238 goto out_free_io_cache
;
240 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
242 goto out_free_tio_cache
;
244 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
245 if (!_rq_bio_info_cache
)
246 goto out_free_rq_tio_cache
;
248 r
= dm_uevent_init();
250 goto out_free_rq_bio_info_cache
;
253 r
= register_blkdev(_major
, _name
);
255 goto out_uevent_exit
;
264 out_free_rq_bio_info_cache
:
265 kmem_cache_destroy(_rq_bio_info_cache
);
266 out_free_rq_tio_cache
:
267 kmem_cache_destroy(_rq_tio_cache
);
269 kmem_cache_destroy(_tio_cache
);
271 kmem_cache_destroy(_io_cache
);
276 static void local_exit(void)
278 kmem_cache_destroy(_rq_bio_info_cache
);
279 kmem_cache_destroy(_rq_tio_cache
);
280 kmem_cache_destroy(_tio_cache
);
281 kmem_cache_destroy(_io_cache
);
282 unregister_blkdev(_major
, _name
);
287 DMINFO("cleaned up");
290 static int (*_inits
[])(void) __initdata
= {
300 static void (*_exits
[])(void) = {
310 static int __init
dm_init(void)
312 const int count
= ARRAY_SIZE(_inits
);
316 for (i
= 0; i
< count
; i
++) {
331 static void __exit
dm_exit(void)
333 int i
= ARRAY_SIZE(_exits
);
339 * Should be empty by this point.
341 idr_remove_all(&_minor_idr
);
342 idr_destroy(&_minor_idr
);
346 * Block device functions
348 int dm_deleting_md(struct mapped_device
*md
)
350 return test_bit(DMF_DELETING
, &md
->flags
);
353 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
355 struct mapped_device
*md
;
357 spin_lock(&_minor_lock
);
359 md
= bdev
->bd_disk
->private_data
;
363 if (test_bit(DMF_FREEING
, &md
->flags
) ||
364 dm_deleting_md(md
)) {
370 atomic_inc(&md
->open_count
);
373 spin_unlock(&_minor_lock
);
375 return md
? 0 : -ENXIO
;
378 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
380 struct mapped_device
*md
= disk
->private_data
;
382 spin_lock(&_minor_lock
);
384 atomic_dec(&md
->open_count
);
387 spin_unlock(&_minor_lock
);
392 int dm_open_count(struct mapped_device
*md
)
394 return atomic_read(&md
->open_count
);
398 * Guarantees nothing is using the device before it's deleted.
400 int dm_lock_for_deletion(struct mapped_device
*md
)
404 spin_lock(&_minor_lock
);
406 if (dm_open_count(md
))
409 set_bit(DMF_DELETING
, &md
->flags
);
411 spin_unlock(&_minor_lock
);
416 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
418 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
420 return dm_get_geometry(md
, geo
);
423 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
424 unsigned int cmd
, unsigned long arg
)
426 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
427 struct dm_table
*map
= dm_get_live_table(md
);
428 struct dm_target
*tgt
;
431 if (!map
|| !dm_table_get_size(map
))
434 /* We only support devices that have a single target */
435 if (dm_table_get_num_targets(map
) != 1)
438 tgt
= dm_table_get_target(map
, 0);
440 if (dm_suspended_md(md
)) {
445 if (tgt
->type
->ioctl
)
446 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
454 static struct dm_io
*alloc_io(struct mapped_device
*md
)
456 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
459 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
461 mempool_free(io
, md
->io_pool
);
464 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
466 mempool_free(tio
, md
->tio_pool
);
469 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
472 return mempool_alloc(md
->tio_pool
, gfp_mask
);
475 static void free_rq_tio(struct dm_rq_target_io
*tio
)
477 mempool_free(tio
, tio
->md
->tio_pool
);
480 static int md_in_flight(struct mapped_device
*md
)
482 return atomic_read(&md
->pending
[READ
]) +
483 atomic_read(&md
->pending
[WRITE
]);
486 static void start_io_acct(struct dm_io
*io
)
488 struct mapped_device
*md
= io
->md
;
490 int rw
= bio_data_dir(io
->bio
);
492 io
->start_time
= jiffies
;
494 cpu
= part_stat_lock();
495 part_round_stats(cpu
, &dm_disk(md
)->part0
);
497 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
498 atomic_inc_return(&md
->pending
[rw
]));
501 static void end_io_acct(struct dm_io
*io
)
503 struct mapped_device
*md
= io
->md
;
504 struct bio
*bio
= io
->bio
;
505 unsigned long duration
= jiffies
- io
->start_time
;
507 int rw
= bio_data_dir(bio
);
509 cpu
= part_stat_lock();
510 part_round_stats(cpu
, &dm_disk(md
)->part0
);
511 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
515 * After this is decremented the bio must not be touched if it is
518 pending
= atomic_dec_return(&md
->pending
[rw
]);
519 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
520 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
522 /* nudge anyone waiting on suspend queue */
528 * Add the bio to the list of deferred io.
530 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
534 spin_lock_irqsave(&md
->deferred_lock
, flags
);
535 bio_list_add(&md
->deferred
, bio
);
536 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
537 queue_work(md
->wq
, &md
->work
);
541 * Everyone (including functions in this file), should use this
542 * function to access the md->map field, and make sure they call
543 * dm_table_put() when finished.
545 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
550 read_lock_irqsave(&md
->map_lock
, flags
);
554 read_unlock_irqrestore(&md
->map_lock
, flags
);
560 * Get the geometry associated with a dm device
562 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
570 * Set the geometry of a device.
572 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
574 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
576 if (geo
->start
> sz
) {
577 DMWARN("Start sector is beyond the geometry limits.");
586 /*-----------------------------------------------------------------
588 * A more elegant soln is in the works that uses the queue
589 * merge fn, unfortunately there are a couple of changes to
590 * the block layer that I want to make for this. So in the
591 * interests of getting something for people to use I give
592 * you this clearly demarcated crap.
593 *---------------------------------------------------------------*/
595 static int __noflush_suspending(struct mapped_device
*md
)
597 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
601 * Decrements the number of outstanding ios that a bio has been
602 * cloned into, completing the original io if necc.
604 static void dec_pending(struct dm_io
*io
, int error
)
609 struct mapped_device
*md
= io
->md
;
611 /* Push-back supersedes any I/O errors */
612 if (unlikely(error
)) {
613 spin_lock_irqsave(&io
->endio_lock
, flags
);
614 if (!(io
->error
> 0 && __noflush_suspending(md
)))
616 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
619 if (atomic_dec_and_test(&io
->io_count
)) {
620 if (io
->error
== DM_ENDIO_REQUEUE
) {
622 * Target requested pushing back the I/O.
624 spin_lock_irqsave(&md
->deferred_lock
, flags
);
625 if (__noflush_suspending(md
))
626 bio_list_add_head(&md
->deferred
, io
->bio
);
628 /* noflush suspend was interrupted. */
630 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
633 io_error
= io
->error
;
638 if (io_error
== DM_ENDIO_REQUEUE
)
641 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
643 * Preflush done for flush with data, reissue
646 bio
->bi_rw
&= ~REQ_FLUSH
;
649 /* done with normal IO or empty flush */
650 trace_block_bio_complete(md
->queue
, bio
, io_error
);
651 bio_endio(bio
, io_error
);
656 static void clone_endio(struct bio
*bio
, int error
)
659 struct dm_target_io
*tio
= bio
->bi_private
;
660 struct dm_io
*io
= tio
->io
;
661 struct mapped_device
*md
= tio
->io
->md
;
662 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
664 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
668 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
669 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
671 * error and requeue request are handled
675 else if (r
== DM_ENDIO_INCOMPLETE
)
676 /* The target will handle the io */
679 DMWARN("unimplemented target endio return value: %d", r
);
686 dec_pending(io
, error
);
690 * Partial completion handling for request-based dm
692 static void end_clone_bio(struct bio
*clone
, int error
)
694 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
695 struct dm_rq_target_io
*tio
= info
->tio
;
696 struct bio
*bio
= info
->orig
;
697 unsigned int nr_bytes
= info
->orig
->bi_size
;
703 * An error has already been detected on the request.
704 * Once error occurred, just let clone->end_io() handle
710 * Don't notice the error to the upper layer yet.
711 * The error handling decision is made by the target driver,
712 * when the request is completed.
719 * I/O for the bio successfully completed.
720 * Notice the data completion to the upper layer.
724 * bios are processed from the head of the list.
725 * So the completing bio should always be rq->bio.
726 * If it's not, something wrong is happening.
728 if (tio
->orig
->bio
!= bio
)
729 DMERR("bio completion is going in the middle of the request");
732 * Update the original request.
733 * Do not use blk_end_request() here, because it may complete
734 * the original request before the clone, and break the ordering.
736 blk_update_request(tio
->orig
, 0, nr_bytes
);
740 * Don't touch any member of the md after calling this function because
741 * the md may be freed in dm_put() at the end of this function.
742 * Or do dm_get() before calling this function and dm_put() later.
744 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
746 atomic_dec(&md
->pending
[rw
]);
748 /* nudge anyone waiting on suspend queue */
749 if (!md_in_flight(md
))
753 blk_run_queue(md
->queue
);
756 * dm_put() must be at the end of this function. See the comment above
761 static void free_rq_clone(struct request
*clone
)
763 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
765 blk_rq_unprep_clone(clone
);
770 * Complete the clone and the original request.
771 * Must be called without queue lock.
773 static void dm_end_request(struct request
*clone
, int error
)
775 int rw
= rq_data_dir(clone
);
776 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
777 struct mapped_device
*md
= tio
->md
;
778 struct request
*rq
= tio
->orig
;
780 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
781 rq
->errors
= clone
->errors
;
782 rq
->resid_len
= clone
->resid_len
;
786 * We are using the sense buffer of the original
788 * So setting the length of the sense data is enough.
790 rq
->sense_len
= clone
->sense_len
;
793 free_rq_clone(clone
);
794 blk_end_request_all(rq
, error
);
795 rq_completed(md
, rw
, true);
798 static void dm_unprep_request(struct request
*rq
)
800 struct request
*clone
= rq
->special
;
803 rq
->cmd_flags
&= ~REQ_DONTPREP
;
805 free_rq_clone(clone
);
809 * Requeue the original request of a clone.
811 void dm_requeue_unmapped_request(struct request
*clone
)
813 int rw
= rq_data_dir(clone
);
814 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
815 struct mapped_device
*md
= tio
->md
;
816 struct request
*rq
= tio
->orig
;
817 struct request_queue
*q
= rq
->q
;
820 dm_unprep_request(rq
);
822 spin_lock_irqsave(q
->queue_lock
, flags
);
823 blk_requeue_request(q
, rq
);
824 spin_unlock_irqrestore(q
->queue_lock
, flags
);
826 rq_completed(md
, rw
, 0);
828 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
830 static void __stop_queue(struct request_queue
*q
)
835 static void stop_queue(struct request_queue
*q
)
839 spin_lock_irqsave(q
->queue_lock
, flags
);
841 spin_unlock_irqrestore(q
->queue_lock
, flags
);
844 static void __start_queue(struct request_queue
*q
)
846 if (blk_queue_stopped(q
))
850 static void start_queue(struct request_queue
*q
)
854 spin_lock_irqsave(q
->queue_lock
, flags
);
856 spin_unlock_irqrestore(q
->queue_lock
, flags
);
859 static void dm_done(struct request
*clone
, int error
, bool mapped
)
862 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
863 dm_request_endio_fn rq_end_io
= tio
->ti
->type
->rq_end_io
;
865 if (mapped
&& rq_end_io
)
866 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
869 /* The target wants to complete the I/O */
870 dm_end_request(clone
, r
);
871 else if (r
== DM_ENDIO_INCOMPLETE
)
872 /* The target will handle the I/O */
874 else if (r
== DM_ENDIO_REQUEUE
)
875 /* The target wants to requeue the I/O */
876 dm_requeue_unmapped_request(clone
);
878 DMWARN("unimplemented target endio return value: %d", r
);
884 * Request completion handler for request-based dm
886 static void dm_softirq_done(struct request
*rq
)
889 struct request
*clone
= rq
->completion_data
;
890 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
892 if (rq
->cmd_flags
& REQ_FAILED
)
895 dm_done(clone
, tio
->error
, mapped
);
899 * Complete the clone and the original request with the error status
900 * through softirq context.
902 static void dm_complete_request(struct request
*clone
, int error
)
904 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
905 struct request
*rq
= tio
->orig
;
908 rq
->completion_data
= clone
;
909 blk_complete_request(rq
);
913 * Complete the not-mapped clone and the original request with the error status
914 * through softirq context.
915 * Target's rq_end_io() function isn't called.
916 * This may be used when the target's map_rq() function fails.
918 void dm_kill_unmapped_request(struct request
*clone
, int error
)
920 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
921 struct request
*rq
= tio
->orig
;
923 rq
->cmd_flags
|= REQ_FAILED
;
924 dm_complete_request(clone
, error
);
926 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
929 * Called with the queue lock held
931 static void end_clone_request(struct request
*clone
, int error
)
934 * For just cleaning up the information of the queue in which
935 * the clone was dispatched.
936 * The clone is *NOT* freed actually here because it is alloced from
937 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
939 __blk_put_request(clone
->q
, clone
);
942 * Actual request completion is done in a softirq context which doesn't
943 * hold the queue lock. Otherwise, deadlock could occur because:
944 * - another request may be submitted by the upper level driver
945 * of the stacking during the completion
946 * - the submission which requires queue lock may be done
949 dm_complete_request(clone
, error
);
953 * Return maximum size of I/O possible at the supplied sector up to the current
956 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
958 sector_t target_offset
= dm_target_offset(ti
, sector
);
960 return ti
->len
- target_offset
;
963 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
965 sector_t len
= max_io_len_target_boundary(sector
, ti
);
966 sector_t offset
, max_len
;
969 * Does the target need to split even further?
971 if (ti
->max_io_len
) {
972 offset
= dm_target_offset(ti
, sector
);
973 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
974 max_len
= sector_div(offset
, ti
->max_io_len
);
976 max_len
= offset
& (ti
->max_io_len
- 1);
977 max_len
= ti
->max_io_len
- max_len
;
986 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
988 if (len
> UINT_MAX
) {
989 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
990 (unsigned long long)len
, UINT_MAX
);
991 ti
->error
= "Maximum size of target IO is too large";
995 ti
->max_io_len
= (uint32_t) len
;
999 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1001 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
1002 struct dm_target_io
*tio
)
1006 struct mapped_device
*md
;
1008 clone
->bi_end_io
= clone_endio
;
1009 clone
->bi_private
= tio
;
1012 * Map the clone. If r == 0 we don't need to do
1013 * anything, the target has assumed ownership of
1016 atomic_inc(&tio
->io
->io_count
);
1017 sector
= clone
->bi_sector
;
1018 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
1019 if (r
== DM_MAPIO_REMAPPED
) {
1020 /* the bio has been remapped so dispatch it */
1022 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1023 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1025 generic_make_request(clone
);
1026 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1027 /* error the io and bail out, or requeue it if needed */
1029 dec_pending(tio
->io
, r
);
1033 DMWARN("unimplemented target map return value: %d", r
);
1039 struct mapped_device
*md
;
1040 struct dm_table
*map
;
1044 sector_t sector_count
;
1049 * Creates a little bio that just does part of a bvec.
1051 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
1052 unsigned short idx
, unsigned int offset
,
1053 unsigned int len
, struct bio_set
*bs
)
1056 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1058 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
1059 *clone
->bi_io_vec
= *bv
;
1061 clone
->bi_sector
= sector
;
1062 clone
->bi_bdev
= bio
->bi_bdev
;
1063 clone
->bi_rw
= bio
->bi_rw
;
1065 clone
->bi_size
= to_bytes(len
);
1066 clone
->bi_io_vec
->bv_offset
= offset
;
1067 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1068 clone
->bi_flags
|= 1 << BIO_CLONED
;
1070 if (bio_integrity(bio
)) {
1071 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1072 bio_integrity_trim(clone
,
1073 bio_sector_offset(bio
, idx
, offset
), len
);
1080 * Creates a bio that consists of range of complete bvecs.
1082 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
1083 unsigned short idx
, unsigned short bv_count
,
1084 unsigned int len
, struct bio_set
*bs
)
1088 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
1089 __bio_clone(clone
, bio
);
1090 clone
->bi_sector
= sector
;
1091 clone
->bi_idx
= idx
;
1092 clone
->bi_vcnt
= idx
+ bv_count
;
1093 clone
->bi_size
= to_bytes(len
);
1094 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1096 if (bio_integrity(bio
)) {
1097 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1099 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1100 bio_integrity_trim(clone
,
1101 bio_sector_offset(bio
, idx
, 0), len
);
1107 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1108 struct dm_target
*ti
)
1110 struct dm_target_io
*tio
= mempool_alloc(ci
->md
->tio_pool
, GFP_NOIO
);
1114 memset(&tio
->info
, 0, sizeof(tio
->info
));
1119 static void __issue_target_request(struct clone_info
*ci
, struct dm_target
*ti
,
1120 unsigned request_nr
, sector_t len
)
1122 struct dm_target_io
*tio
= alloc_tio(ci
, ti
);
1125 tio
->info
.target_request_nr
= request_nr
;
1128 * Discard requests require the bio's inline iovecs be initialized.
1129 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1130 * and discard, so no need for concern about wasted bvec allocations.
1132 clone
= bio_clone_bioset(ci
->bio
, GFP_NOIO
, ci
->md
->bs
);
1135 clone
->bi_sector
= ci
->sector
;
1136 clone
->bi_size
= to_bytes(len
);
1139 __map_bio(ti
, clone
, tio
);
1142 static void __issue_target_requests(struct clone_info
*ci
, struct dm_target
*ti
,
1143 unsigned num_requests
, sector_t len
)
1145 unsigned request_nr
;
1147 for (request_nr
= 0; request_nr
< num_requests
; request_nr
++)
1148 __issue_target_request(ci
, ti
, request_nr
, len
);
1151 static int __clone_and_map_empty_flush(struct clone_info
*ci
)
1153 unsigned target_nr
= 0;
1154 struct dm_target
*ti
;
1156 BUG_ON(bio_has_data(ci
->bio
));
1157 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1158 __issue_target_requests(ci
, ti
, ti
->num_flush_requests
, 0);
1164 * Perform all io with a single clone.
1166 static void __clone_and_map_simple(struct clone_info
*ci
, struct dm_target
*ti
)
1168 struct bio
*clone
, *bio
= ci
->bio
;
1169 struct dm_target_io
*tio
;
1171 tio
= alloc_tio(ci
, ti
);
1172 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
1173 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
1175 __map_bio(ti
, clone
, tio
);
1176 ci
->sector_count
= 0;
1179 static int __clone_and_map_discard(struct clone_info
*ci
)
1181 struct dm_target
*ti
;
1185 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1186 if (!dm_target_is_valid(ti
))
1190 * Even though the device advertised discard support,
1191 * that does not mean every target supports it, and
1192 * reconfiguration might also have changed that since the
1193 * check was performed.
1195 if (!ti
->num_discard_requests
)
1198 if (!ti
->split_discard_requests
)
1199 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1201 len
= min(ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1203 __issue_target_requests(ci
, ti
, ti
->num_discard_requests
, len
);
1206 } while (ci
->sector_count
-= len
);
1211 static int __clone_and_map(struct clone_info
*ci
)
1213 struct bio
*clone
, *bio
= ci
->bio
;
1214 struct dm_target
*ti
;
1215 sector_t len
= 0, max
;
1216 struct dm_target_io
*tio
;
1218 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1219 return __clone_and_map_discard(ci
);
1221 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1222 if (!dm_target_is_valid(ti
))
1225 max
= max_io_len(ci
->sector
, ti
);
1227 if (ci
->sector_count
<= max
) {
1229 * Optimise for the simple case where we can do all of
1230 * the remaining io with a single clone.
1232 __clone_and_map_simple(ci
, ti
);
1234 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1236 * There are some bvecs that don't span targets.
1237 * Do as many of these as possible.
1240 sector_t remaining
= max
;
1243 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1244 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1246 if (bv_len
> remaining
)
1249 remaining
-= bv_len
;
1253 tio
= alloc_tio(ci
, ti
);
1254 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1256 __map_bio(ti
, clone
, tio
);
1259 ci
->sector_count
-= len
;
1264 * Handle a bvec that must be split between two or more targets.
1266 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1267 sector_t remaining
= to_sector(bv
->bv_len
);
1268 unsigned int offset
= 0;
1272 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1273 if (!dm_target_is_valid(ti
))
1276 max
= max_io_len(ci
->sector
, ti
);
1279 len
= min(remaining
, max
);
1281 tio
= alloc_tio(ci
, ti
);
1282 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
1283 bv
->bv_offset
+ offset
, len
,
1286 __map_bio(ti
, clone
, tio
);
1289 ci
->sector_count
-= len
;
1290 offset
+= to_bytes(len
);
1291 } while (remaining
-= len
);
1300 * Split the bio into several clones and submit it to targets.
1302 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1304 struct clone_info ci
;
1307 ci
.map
= dm_get_live_table(md
);
1308 if (unlikely(!ci
.map
)) {
1314 ci
.io
= alloc_io(md
);
1316 atomic_set(&ci
.io
->io_count
, 1);
1319 spin_lock_init(&ci
.io
->endio_lock
);
1320 ci
.sector
= bio
->bi_sector
;
1321 ci
.idx
= bio
->bi_idx
;
1323 start_io_acct(ci
.io
);
1324 if (bio
->bi_rw
& REQ_FLUSH
) {
1325 ci
.bio
= &ci
.md
->flush_bio
;
1326 ci
.sector_count
= 0;
1327 error
= __clone_and_map_empty_flush(&ci
);
1328 /* dec_pending submits any data associated with flush */
1331 ci
.sector_count
= bio_sectors(bio
);
1332 while (ci
.sector_count
&& !error
)
1333 error
= __clone_and_map(&ci
);
1336 /* drop the extra reference count */
1337 dec_pending(ci
.io
, error
);
1338 dm_table_put(ci
.map
);
1340 /*-----------------------------------------------------------------
1342 *---------------------------------------------------------------*/
1344 static int dm_merge_bvec(struct request_queue
*q
,
1345 struct bvec_merge_data
*bvm
,
1346 struct bio_vec
*biovec
)
1348 struct mapped_device
*md
= q
->queuedata
;
1349 struct dm_table
*map
= dm_get_live_table(md
);
1350 struct dm_target
*ti
;
1351 sector_t max_sectors
;
1357 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1358 if (!dm_target_is_valid(ti
))
1362 * Find maximum amount of I/O that won't need splitting
1364 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1365 (sector_t
) BIO_MAX_SECTORS
);
1366 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1371 * merge_bvec_fn() returns number of bytes
1372 * it can accept at this offset
1373 * max is precomputed maximal io size
1375 if (max_size
&& ti
->type
->merge
)
1376 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1378 * If the target doesn't support merge method and some of the devices
1379 * provided their merge_bvec method (we know this by looking at
1380 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1381 * entries. So always set max_size to 0, and the code below allows
1384 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1393 * Always allow an entire first page
1395 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1396 max_size
= biovec
->bv_len
;
1402 * The request function that just remaps the bio built up by
1405 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1407 int rw
= bio_data_dir(bio
);
1408 struct mapped_device
*md
= q
->queuedata
;
1411 down_read(&md
->io_lock
);
1413 cpu
= part_stat_lock();
1414 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1415 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1418 /* if we're suspended, we have to queue this io for later */
1419 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1420 up_read(&md
->io_lock
);
1422 if (bio_rw(bio
) != READA
)
1429 __split_and_process_bio(md
, bio
);
1430 up_read(&md
->io_lock
);
1434 static int dm_request_based(struct mapped_device
*md
)
1436 return blk_queue_stackable(md
->queue
);
1439 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1441 struct mapped_device
*md
= q
->queuedata
;
1443 if (dm_request_based(md
))
1444 blk_queue_bio(q
, bio
);
1446 _dm_request(q
, bio
);
1449 void dm_dispatch_request(struct request
*rq
)
1453 if (blk_queue_io_stat(rq
->q
))
1454 rq
->cmd_flags
|= REQ_IO_STAT
;
1456 rq
->start_time
= jiffies
;
1457 r
= blk_insert_cloned_request(rq
->q
, rq
);
1459 dm_complete_request(rq
, r
);
1461 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1463 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1466 struct dm_rq_target_io
*tio
= data
;
1467 struct dm_rq_clone_bio_info
*info
=
1468 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1470 info
->orig
= bio_orig
;
1472 bio
->bi_end_io
= end_clone_bio
;
1473 bio
->bi_private
= info
;
1478 static int setup_clone(struct request
*clone
, struct request
*rq
,
1479 struct dm_rq_target_io
*tio
)
1483 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1484 dm_rq_bio_constructor
, tio
);
1488 clone
->cmd
= rq
->cmd
;
1489 clone
->cmd_len
= rq
->cmd_len
;
1490 clone
->sense
= rq
->sense
;
1491 clone
->buffer
= rq
->buffer
;
1492 clone
->end_io
= end_clone_request
;
1493 clone
->end_io_data
= tio
;
1498 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1501 struct request
*clone
;
1502 struct dm_rq_target_io
*tio
;
1504 tio
= alloc_rq_tio(md
, gfp_mask
);
1512 memset(&tio
->info
, 0, sizeof(tio
->info
));
1514 clone
= &tio
->clone
;
1515 if (setup_clone(clone
, rq
, tio
)) {
1525 * Called with the queue lock held.
1527 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1529 struct mapped_device
*md
= q
->queuedata
;
1530 struct request
*clone
;
1532 if (unlikely(rq
->special
)) {
1533 DMWARN("Already has something in rq->special.");
1534 return BLKPREP_KILL
;
1537 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1539 return BLKPREP_DEFER
;
1541 rq
->special
= clone
;
1542 rq
->cmd_flags
|= REQ_DONTPREP
;
1549 * 0 : the request has been processed (not requeued)
1550 * !0 : the request has been requeued
1552 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1553 struct mapped_device
*md
)
1555 int r
, requeued
= 0;
1556 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1559 * Hold the md reference here for the in-flight I/O.
1560 * We can't rely on the reference count by device opener,
1561 * because the device may be closed during the request completion
1562 * when all bios are completed.
1563 * See the comment in rq_completed() too.
1568 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1570 case DM_MAPIO_SUBMITTED
:
1571 /* The target has taken the I/O to submit by itself later */
1573 case DM_MAPIO_REMAPPED
:
1574 /* The target has remapped the I/O so dispatch it */
1575 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1576 blk_rq_pos(tio
->orig
));
1577 dm_dispatch_request(clone
);
1579 case DM_MAPIO_REQUEUE
:
1580 /* The target wants to requeue the I/O */
1581 dm_requeue_unmapped_request(clone
);
1586 DMWARN("unimplemented target map return value: %d", r
);
1590 /* The target wants to complete the I/O */
1591 dm_kill_unmapped_request(clone
, r
);
1599 * q->request_fn for request-based dm.
1600 * Called with the queue lock held.
1602 static void dm_request_fn(struct request_queue
*q
)
1604 struct mapped_device
*md
= q
->queuedata
;
1605 struct dm_table
*map
= dm_get_live_table(md
);
1606 struct dm_target
*ti
;
1607 struct request
*rq
, *clone
;
1611 * For suspend, check blk_queue_stopped() and increment
1612 * ->pending within a single queue_lock not to increment the
1613 * number of in-flight I/Os after the queue is stopped in
1616 while (!blk_queue_stopped(q
)) {
1617 rq
= blk_peek_request(q
);
1621 /* always use block 0 to find the target for flushes for now */
1623 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1624 pos
= blk_rq_pos(rq
);
1626 ti
= dm_table_find_target(map
, pos
);
1627 BUG_ON(!dm_target_is_valid(ti
));
1629 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1632 blk_start_request(rq
);
1633 clone
= rq
->special
;
1634 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1636 spin_unlock(q
->queue_lock
);
1637 if (map_request(ti
, clone
, md
))
1640 BUG_ON(!irqs_disabled());
1641 spin_lock(q
->queue_lock
);
1647 BUG_ON(!irqs_disabled());
1648 spin_lock(q
->queue_lock
);
1651 blk_delay_queue(q
, HZ
/ 10);
1658 int dm_underlying_device_busy(struct request_queue
*q
)
1660 return blk_lld_busy(q
);
1662 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1664 static int dm_lld_busy(struct request_queue
*q
)
1667 struct mapped_device
*md
= q
->queuedata
;
1668 struct dm_table
*map
= dm_get_live_table(md
);
1670 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1673 r
= dm_table_any_busy_target(map
);
1680 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1683 struct mapped_device
*md
= congested_data
;
1684 struct dm_table
*map
;
1686 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1687 map
= dm_get_live_table(md
);
1690 * Request-based dm cares about only own queue for
1691 * the query about congestion status of request_queue
1693 if (dm_request_based(md
))
1694 r
= md
->queue
->backing_dev_info
.state
&
1697 r
= dm_table_any_congested(map
, bdi_bits
);
1706 /*-----------------------------------------------------------------
1707 * An IDR is used to keep track of allocated minor numbers.
1708 *---------------------------------------------------------------*/
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
);
1805 * Allocate and initialise a blank device with a given minor.
1807 static struct mapped_device
*alloc_dev(int minor
)
1810 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1814 DMWARN("unable to allocate device, out of memory.");
1818 if (!try_module_get(THIS_MODULE
))
1819 goto bad_module_get
;
1821 /* get a minor number for the dev */
1822 if (minor
== DM_ANY_MINOR
)
1823 r
= next_free_minor(&minor
);
1825 r
= specific_minor(minor
);
1829 md
->type
= DM_TYPE_NONE
;
1830 init_rwsem(&md
->io_lock
);
1831 mutex_init(&md
->suspend_lock
);
1832 mutex_init(&md
->type_lock
);
1833 spin_lock_init(&md
->deferred_lock
);
1834 rwlock_init(&md
->map_lock
);
1835 atomic_set(&md
->holders
, 1);
1836 atomic_set(&md
->open_count
, 0);
1837 atomic_set(&md
->event_nr
, 0);
1838 atomic_set(&md
->uevent_seq
, 0);
1839 INIT_LIST_HEAD(&md
->uevent_list
);
1840 spin_lock_init(&md
->uevent_lock
);
1842 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1846 dm_init_md_queue(md
);
1848 md
->disk
= alloc_disk(1);
1852 atomic_set(&md
->pending
[0], 0);
1853 atomic_set(&md
->pending
[1], 0);
1854 init_waitqueue_head(&md
->wait
);
1855 INIT_WORK(&md
->work
, dm_wq_work
);
1856 init_waitqueue_head(&md
->eventq
);
1858 md
->disk
->major
= _major
;
1859 md
->disk
->first_minor
= minor
;
1860 md
->disk
->fops
= &dm_blk_dops
;
1861 md
->disk
->queue
= md
->queue
;
1862 md
->disk
->private_data
= md
;
1863 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1865 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1867 md
->wq
= alloc_workqueue("kdmflush",
1868 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1872 md
->bdev
= bdget_disk(md
->disk
, 0);
1876 bio_init(&md
->flush_bio
);
1877 md
->flush_bio
.bi_bdev
= md
->bdev
;
1878 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1880 /* Populate the mapping, nobody knows we exist yet */
1881 spin_lock(&_minor_lock
);
1882 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1883 spin_unlock(&_minor_lock
);
1885 BUG_ON(old_md
!= MINOR_ALLOCED
);
1890 destroy_workqueue(md
->wq
);
1892 del_gendisk(md
->disk
);
1895 blk_cleanup_queue(md
->queue
);
1899 module_put(THIS_MODULE
);
1905 static void unlock_fs(struct mapped_device
*md
);
1907 static void free_dev(struct mapped_device
*md
)
1909 int minor
= MINOR(disk_devt(md
->disk
));
1913 destroy_workqueue(md
->wq
);
1915 mempool_destroy(md
->tio_pool
);
1917 mempool_destroy(md
->io_pool
);
1919 bioset_free(md
->bs
);
1920 blk_integrity_unregister(md
->disk
);
1921 del_gendisk(md
->disk
);
1924 spin_lock(&_minor_lock
);
1925 md
->disk
->private_data
= NULL
;
1926 spin_unlock(&_minor_lock
);
1929 blk_cleanup_queue(md
->queue
);
1930 module_put(THIS_MODULE
);
1934 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1936 struct dm_md_mempools
*p
;
1938 if (md
->io_pool
&& md
->tio_pool
&& md
->bs
)
1939 /* the md already has necessary mempools */
1942 p
= dm_table_get_md_mempools(t
);
1943 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1945 md
->io_pool
= p
->io_pool
;
1947 md
->tio_pool
= p
->tio_pool
;
1953 /* mempool bind completed, now no need any mempools in the table */
1954 dm_table_free_md_mempools(t
);
1958 * Bind a table to the device.
1960 static void event_callback(void *context
)
1962 unsigned long flags
;
1964 struct mapped_device
*md
= (struct mapped_device
*) context
;
1966 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1967 list_splice_init(&md
->uevent_list
, &uevents
);
1968 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1970 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1972 atomic_inc(&md
->event_nr
);
1973 wake_up(&md
->eventq
);
1977 * Protected by md->suspend_lock obtained by dm_swap_table().
1979 static void __set_size(struct mapped_device
*md
, sector_t size
)
1981 set_capacity(md
->disk
, size
);
1983 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1987 * Return 1 if the queue has a compulsory merge_bvec_fn function.
1989 * If this function returns 0, then the device is either a non-dm
1990 * device without a merge_bvec_fn, or it is a dm device that is
1991 * able to split any bios it receives that are too big.
1993 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
1995 struct mapped_device
*dev_md
;
1997 if (!q
->merge_bvec_fn
)
2000 if (q
->make_request_fn
== dm_request
) {
2001 dev_md
= q
->queuedata
;
2002 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2009 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2010 struct dm_dev
*dev
, sector_t start
,
2011 sector_t len
, void *data
)
2013 struct block_device
*bdev
= dev
->bdev
;
2014 struct request_queue
*q
= bdev_get_queue(bdev
);
2016 return dm_queue_merge_is_compulsory(q
);
2020 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2021 * on the properties of the underlying devices.
2023 static int dm_table_merge_is_optional(struct dm_table
*table
)
2026 struct dm_target
*ti
;
2028 while (i
< dm_table_get_num_targets(table
)) {
2029 ti
= dm_table_get_target(table
, i
++);
2031 if (ti
->type
->iterate_devices
&&
2032 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2040 * Returns old map, which caller must destroy.
2042 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2043 struct queue_limits
*limits
)
2045 struct dm_table
*old_map
;
2046 struct request_queue
*q
= md
->queue
;
2048 unsigned long flags
;
2049 int merge_is_optional
;
2051 size
= dm_table_get_size(t
);
2054 * Wipe any geometry if the size of the table changed.
2056 if (size
!= get_capacity(md
->disk
))
2057 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2059 __set_size(md
, size
);
2061 dm_table_event_callback(t
, event_callback
, md
);
2064 * The queue hasn't been stopped yet, if the old table type wasn't
2065 * for request-based during suspension. So stop it to prevent
2066 * I/O mapping before resume.
2067 * This must be done before setting the queue restrictions,
2068 * because request-based dm may be run just after the setting.
2070 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2073 __bind_mempools(md
, t
);
2075 merge_is_optional
= dm_table_merge_is_optional(t
);
2077 write_lock_irqsave(&md
->map_lock
, flags
);
2080 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2082 dm_table_set_restrictions(t
, q
, limits
);
2083 if (merge_is_optional
)
2084 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2086 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2087 write_unlock_irqrestore(&md
->map_lock
, flags
);
2093 * Returns unbound table for the caller to free.
2095 static struct dm_table
*__unbind(struct mapped_device
*md
)
2097 struct dm_table
*map
= md
->map
;
2098 unsigned long flags
;
2103 dm_table_event_callback(map
, NULL
, NULL
);
2104 write_lock_irqsave(&md
->map_lock
, flags
);
2106 write_unlock_irqrestore(&md
->map_lock
, flags
);
2112 * Constructor for a new device.
2114 int dm_create(int minor
, struct mapped_device
**result
)
2116 struct mapped_device
*md
;
2118 md
= alloc_dev(minor
);
2129 * Functions to manage md->type.
2130 * All are required to hold md->type_lock.
2132 void dm_lock_md_type(struct mapped_device
*md
)
2134 mutex_lock(&md
->type_lock
);
2137 void dm_unlock_md_type(struct mapped_device
*md
)
2139 mutex_unlock(&md
->type_lock
);
2142 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2147 unsigned dm_get_md_type(struct mapped_device
*md
)
2152 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2154 return md
->immutable_target_type
;
2158 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2160 static int dm_init_request_based_queue(struct mapped_device
*md
)
2162 struct request_queue
*q
= NULL
;
2164 if (md
->queue
->elevator
)
2167 /* Fully initialize the queue */
2168 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2173 dm_init_md_queue(md
);
2174 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2175 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2176 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2178 elv_register_queue(md
->queue
);
2184 * Setup the DM device's queue based on md's type
2186 int dm_setup_md_queue(struct mapped_device
*md
)
2188 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2189 !dm_init_request_based_queue(md
)) {
2190 DMWARN("Cannot initialize queue for request-based mapped device");
2197 static struct mapped_device
*dm_find_md(dev_t dev
)
2199 struct mapped_device
*md
;
2200 unsigned minor
= MINOR(dev
);
2202 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2205 spin_lock(&_minor_lock
);
2207 md
= idr_find(&_minor_idr
, minor
);
2208 if (md
&& (md
== MINOR_ALLOCED
||
2209 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2210 dm_deleting_md(md
) ||
2211 test_bit(DMF_FREEING
, &md
->flags
))) {
2217 spin_unlock(&_minor_lock
);
2222 struct mapped_device
*dm_get_md(dev_t dev
)
2224 struct mapped_device
*md
= dm_find_md(dev
);
2231 EXPORT_SYMBOL_GPL(dm_get_md
);
2233 void *dm_get_mdptr(struct mapped_device
*md
)
2235 return md
->interface_ptr
;
2238 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2240 md
->interface_ptr
= ptr
;
2243 void dm_get(struct mapped_device
*md
)
2245 atomic_inc(&md
->holders
);
2246 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2249 const char *dm_device_name(struct mapped_device
*md
)
2253 EXPORT_SYMBOL_GPL(dm_device_name
);
2255 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2257 struct dm_table
*map
;
2261 spin_lock(&_minor_lock
);
2262 map
= dm_get_live_table(md
);
2263 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2264 set_bit(DMF_FREEING
, &md
->flags
);
2265 spin_unlock(&_minor_lock
);
2267 if (!dm_suspended_md(md
)) {
2268 dm_table_presuspend_targets(map
);
2269 dm_table_postsuspend_targets(map
);
2273 * Rare, but there may be I/O requests still going to complete,
2274 * for example. Wait for all references to disappear.
2275 * No one should increment the reference count of the mapped_device,
2276 * after the mapped_device state becomes DMF_FREEING.
2279 while (atomic_read(&md
->holders
))
2281 else if (atomic_read(&md
->holders
))
2282 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2283 dm_device_name(md
), atomic_read(&md
->holders
));
2287 dm_table_destroy(__unbind(md
));
2291 void dm_destroy(struct mapped_device
*md
)
2293 __dm_destroy(md
, true);
2296 void dm_destroy_immediate(struct mapped_device
*md
)
2298 __dm_destroy(md
, false);
2301 void dm_put(struct mapped_device
*md
)
2303 atomic_dec(&md
->holders
);
2305 EXPORT_SYMBOL_GPL(dm_put
);
2307 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2310 DECLARE_WAITQUEUE(wait
, current
);
2312 add_wait_queue(&md
->wait
, &wait
);
2315 set_current_state(interruptible
);
2317 if (!md_in_flight(md
))
2320 if (interruptible
== TASK_INTERRUPTIBLE
&&
2321 signal_pending(current
)) {
2328 set_current_state(TASK_RUNNING
);
2330 remove_wait_queue(&md
->wait
, &wait
);
2336 * Process the deferred bios
2338 static void dm_wq_work(struct work_struct
*work
)
2340 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2344 down_read(&md
->io_lock
);
2346 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2347 spin_lock_irq(&md
->deferred_lock
);
2348 c
= bio_list_pop(&md
->deferred
);
2349 spin_unlock_irq(&md
->deferred_lock
);
2354 up_read(&md
->io_lock
);
2356 if (dm_request_based(md
))
2357 generic_make_request(c
);
2359 __split_and_process_bio(md
, c
);
2361 down_read(&md
->io_lock
);
2364 up_read(&md
->io_lock
);
2367 static void dm_queue_flush(struct mapped_device
*md
)
2369 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2370 smp_mb__after_clear_bit();
2371 queue_work(md
->wq
, &md
->work
);
2375 * Swap in a new table, returning the old one for the caller to destroy.
2377 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2379 struct dm_table
*map
= ERR_PTR(-EINVAL
);
2380 struct queue_limits limits
;
2383 mutex_lock(&md
->suspend_lock
);
2385 /* device must be suspended */
2386 if (!dm_suspended_md(md
))
2389 r
= dm_calculate_queue_limits(table
, &limits
);
2395 map
= __bind(md
, table
, &limits
);
2398 mutex_unlock(&md
->suspend_lock
);
2403 * Functions to lock and unlock any filesystem running on the
2406 static int lock_fs(struct mapped_device
*md
)
2410 WARN_ON(md
->frozen_sb
);
2412 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2413 if (IS_ERR(md
->frozen_sb
)) {
2414 r
= PTR_ERR(md
->frozen_sb
);
2415 md
->frozen_sb
= NULL
;
2419 set_bit(DMF_FROZEN
, &md
->flags
);
2424 static void unlock_fs(struct mapped_device
*md
)
2426 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2429 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2430 md
->frozen_sb
= NULL
;
2431 clear_bit(DMF_FROZEN
, &md
->flags
);
2435 * We need to be able to change a mapping table under a mounted
2436 * filesystem. For example we might want to move some data in
2437 * the background. Before the table can be swapped with
2438 * dm_bind_table, dm_suspend must be called to flush any in
2439 * flight bios and ensure that any further io gets deferred.
2442 * Suspend mechanism in request-based dm.
2444 * 1. Flush all I/Os by lock_fs() if needed.
2445 * 2. Stop dispatching any I/O by stopping the request_queue.
2446 * 3. Wait for all in-flight I/Os to be completed or requeued.
2448 * To abort suspend, start the request_queue.
2450 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2452 struct dm_table
*map
= NULL
;
2454 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2455 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2457 mutex_lock(&md
->suspend_lock
);
2459 if (dm_suspended_md(md
)) {
2464 map
= dm_get_live_table(md
);
2467 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2468 * This flag is cleared before dm_suspend returns.
2471 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2473 /* This does not get reverted if there's an error later. */
2474 dm_table_presuspend_targets(map
);
2477 * Flush I/O to the device.
2478 * Any I/O submitted after lock_fs() may not be flushed.
2479 * noflush takes precedence over do_lockfs.
2480 * (lock_fs() flushes I/Os and waits for them to complete.)
2482 if (!noflush
&& do_lockfs
) {
2489 * Here we must make sure that no processes are submitting requests
2490 * to target drivers i.e. no one may be executing
2491 * __split_and_process_bio. This is called from dm_request and
2494 * To get all processes out of __split_and_process_bio in dm_request,
2495 * we take the write lock. To prevent any process from reentering
2496 * __split_and_process_bio from dm_request and quiesce the thread
2497 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2498 * flush_workqueue(md->wq).
2500 down_write(&md
->io_lock
);
2501 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2502 up_write(&md
->io_lock
);
2505 * Stop md->queue before flushing md->wq in case request-based
2506 * dm defers requests to md->wq from md->queue.
2508 if (dm_request_based(md
))
2509 stop_queue(md
->queue
);
2511 flush_workqueue(md
->wq
);
2514 * At this point no more requests are entering target request routines.
2515 * We call dm_wait_for_completion to wait for all existing requests
2518 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2520 down_write(&md
->io_lock
);
2522 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2523 up_write(&md
->io_lock
);
2525 /* were we interrupted ? */
2529 if (dm_request_based(md
))
2530 start_queue(md
->queue
);
2533 goto out
; /* pushback list is already flushed, so skip flush */
2537 * If dm_wait_for_completion returned 0, the device is completely
2538 * quiescent now. There is no request-processing activity. All new
2539 * requests are being added to md->deferred list.
2542 set_bit(DMF_SUSPENDED
, &md
->flags
);
2544 dm_table_postsuspend_targets(map
);
2550 mutex_unlock(&md
->suspend_lock
);
2554 int dm_resume(struct mapped_device
*md
)
2557 struct dm_table
*map
= NULL
;
2559 mutex_lock(&md
->suspend_lock
);
2560 if (!dm_suspended_md(md
))
2563 map
= dm_get_live_table(md
);
2564 if (!map
|| !dm_table_get_size(map
))
2567 r
= dm_table_resume_targets(map
);
2574 * Flushing deferred I/Os must be done after targets are resumed
2575 * so that mapping of targets can work correctly.
2576 * Request-based dm is queueing the deferred I/Os in its request_queue.
2578 if (dm_request_based(md
))
2579 start_queue(md
->queue
);
2583 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2588 mutex_unlock(&md
->suspend_lock
);
2593 /*-----------------------------------------------------------------
2594 * Event notification.
2595 *---------------------------------------------------------------*/
2596 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2599 char udev_cookie
[DM_COOKIE_LENGTH
];
2600 char *envp
[] = { udev_cookie
, NULL
};
2603 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2605 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2606 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2607 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2612 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2614 return atomic_add_return(1, &md
->uevent_seq
);
2617 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2619 return atomic_read(&md
->event_nr
);
2622 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2624 return wait_event_interruptible(md
->eventq
,
2625 (event_nr
!= atomic_read(&md
->event_nr
)));
2628 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2630 unsigned long flags
;
2632 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2633 list_add(elist
, &md
->uevent_list
);
2634 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2638 * The gendisk is only valid as long as you have a reference
2641 struct gendisk
*dm_disk(struct mapped_device
*md
)
2646 struct kobject
*dm_kobject(struct mapped_device
*md
)
2652 * struct mapped_device should not be exported outside of dm.c
2653 * so use this check to verify that kobj is part of md structure
2655 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2657 struct mapped_device
*md
;
2659 md
= container_of(kobj
, struct mapped_device
, kobj
);
2660 if (&md
->kobj
!= kobj
)
2663 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2671 int dm_suspended_md(struct mapped_device
*md
)
2673 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2676 int dm_suspended(struct dm_target
*ti
)
2678 return dm_suspended_md(dm_table_get_md(ti
->table
));
2680 EXPORT_SYMBOL_GPL(dm_suspended
);
2682 int dm_noflush_suspending(struct dm_target
*ti
)
2684 return __noflush_suspending(dm_table_get_md(ti
->table
));
2686 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2688 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
)
2690 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2691 unsigned int pool_size
= (type
== DM_TYPE_BIO_BASED
) ? 16 : MIN_IOS
;
2696 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2697 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2698 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2699 if (!pools
->io_pool
)
2700 goto free_pools_and_out
;
2702 pools
->tio_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2703 mempool_create_slab_pool(MIN_IOS
, _tio_cache
) :
2704 mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2705 if (!pools
->tio_pool
)
2706 goto free_io_pool_and_out
;
2708 pools
->bs
= (type
== DM_TYPE_BIO_BASED
) ?
2709 bioset_create(pool_size
, 0) :
2710 bioset_create(pool_size
,
2711 offsetof(struct dm_rq_clone_bio_info
, clone
));
2713 goto free_tio_pool_and_out
;
2715 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2716 goto free_bioset_and_out
;
2720 free_bioset_and_out
:
2721 bioset_free(pools
->bs
);
2723 free_tio_pool_and_out
:
2724 mempool_destroy(pools
->tio_pool
);
2726 free_io_pool_and_out
:
2727 mempool_destroy(pools
->io_pool
);
2735 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2741 mempool_destroy(pools
->io_pool
);
2743 if (pools
->tio_pool
)
2744 mempool_destroy(pools
->tio_pool
);
2747 bioset_free(pools
->bs
);
2752 static const struct block_device_operations dm_blk_dops
= {
2753 .open
= dm_blk_open
,
2754 .release
= dm_blk_close
,
2755 .ioctl
= dm_blk_ioctl
,
2756 .getgeo
= dm_blk_getgeo
,
2757 .owner
= THIS_MODULE
2760 EXPORT_SYMBOL(dm_get_mapinfo
);
2765 module_init(dm_init
);
2766 module_exit(dm_exit
);
2768 module_param(major
, uint
, 0);
2769 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2770 MODULE_DESCRIPTION(DM_NAME
" driver");
2771 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2772 MODULE_LICENSE("GPL");