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
;
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io
{
70 struct mapped_device
*md
;
72 struct request
*orig
, clone
;
78 * For request-based dm - the bio clones we allocate are embedded in these
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
85 struct dm_rq_clone_bio_info
{
87 struct dm_rq_target_io
*tio
;
91 union map_info
*dm_get_mapinfo(struct bio
*bio
)
93 if (bio
&& bio
->bi_private
)
94 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
98 union map_info
*dm_get_rq_mapinfo(struct request
*rq
)
100 if (rq
&& rq
->end_io_data
)
101 return &((struct dm_rq_target_io
*)rq
->end_io_data
)->info
;
104 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo
);
106 #define MINOR_ALLOCED ((void *)-1)
109 * Bits for the md->flags field.
111 #define DMF_BLOCK_IO_FOR_SUSPEND 0
112 #define DMF_SUSPENDED 1
114 #define DMF_FREEING 3
115 #define DMF_DELETING 4
116 #define DMF_NOFLUSH_SUSPENDING 5
117 #define DMF_MERGE_IS_OPTIONAL 6
120 * Work processed by per-device workqueue.
122 struct mapped_device
{
123 struct rw_semaphore io_lock
;
124 struct mutex suspend_lock
;
131 struct request_queue
*queue
;
133 /* Protect queue and type against concurrent access. */
134 struct mutex type_lock
;
136 struct target_type
*immutable_target_type
;
138 struct gendisk
*disk
;
144 * A list of ios that arrived while we were suspended.
147 wait_queue_head_t wait
;
148 struct work_struct work
;
149 struct bio_list deferred
;
150 spinlock_t deferred_lock
;
153 * Processing queue (flush)
155 struct workqueue_struct
*wq
;
158 * The current mapping.
160 struct dm_table
*map
;
163 * io objects are allocated from here.
174 wait_queue_head_t eventq
;
176 struct list_head uevent_list
;
177 spinlock_t uevent_lock
; /* Protect access to uevent_list */
180 * freeze/thaw support require holding onto a super block
182 struct super_block
*frozen_sb
;
183 struct block_device
*bdev
;
185 /* forced geometry settings */
186 struct hd_geometry geometry
;
191 /* zero-length flush that will be cloned and submitted to targets */
192 struct bio flush_bio
;
196 * For mempools pre-allocation at the table loading time.
198 struct dm_md_mempools
{
205 static struct kmem_cache
*_io_cache
;
206 static struct kmem_cache
*_rq_tio_cache
;
209 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
210 * still used for _io_cache, I'm leaving this for a later cleanup
212 static struct kmem_cache
*_rq_bio_info_cache
;
214 static int __init
local_init(void)
218 /* allocate a slab for the dm_ios */
219 _io_cache
= KMEM_CACHE(dm_io
, 0);
223 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
225 goto out_free_io_cache
;
227 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
228 if (!_rq_bio_info_cache
)
229 goto out_free_rq_tio_cache
;
231 r
= dm_uevent_init();
233 goto out_free_rq_bio_info_cache
;
236 r
= register_blkdev(_major
, _name
);
238 goto out_uevent_exit
;
247 out_free_rq_bio_info_cache
:
248 kmem_cache_destroy(_rq_bio_info_cache
);
249 out_free_rq_tio_cache
:
250 kmem_cache_destroy(_rq_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(_io_cache
);
262 unregister_blkdev(_major
, _name
);
267 DMINFO("cleaned up");
270 static int (*_inits
[])(void) __initdata
= {
280 static void (*_exits
[])(void) = {
290 static int __init
dm_init(void)
292 const int count
= ARRAY_SIZE(_inits
);
296 for (i
= 0; i
< count
; i
++) {
311 static void __exit
dm_exit(void)
313 int i
= ARRAY_SIZE(_exits
);
319 * Should be empty by this point.
321 idr_remove_all(&_minor_idr
);
322 idr_destroy(&_minor_idr
);
326 * Block device functions
328 int dm_deleting_md(struct mapped_device
*md
)
330 return test_bit(DMF_DELETING
, &md
->flags
);
333 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
335 struct mapped_device
*md
;
337 spin_lock(&_minor_lock
);
339 md
= bdev
->bd_disk
->private_data
;
343 if (test_bit(DMF_FREEING
, &md
->flags
) ||
344 dm_deleting_md(md
)) {
350 atomic_inc(&md
->open_count
);
353 spin_unlock(&_minor_lock
);
355 return md
? 0 : -ENXIO
;
358 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
360 struct mapped_device
*md
= disk
->private_data
;
362 spin_lock(&_minor_lock
);
364 atomic_dec(&md
->open_count
);
367 spin_unlock(&_minor_lock
);
372 int dm_open_count(struct mapped_device
*md
)
374 return atomic_read(&md
->open_count
);
378 * Guarantees nothing is using the device before it's deleted.
380 int dm_lock_for_deletion(struct mapped_device
*md
)
384 spin_lock(&_minor_lock
);
386 if (dm_open_count(md
))
389 set_bit(DMF_DELETING
, &md
->flags
);
391 spin_unlock(&_minor_lock
);
396 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
398 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
400 return dm_get_geometry(md
, geo
);
403 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
404 unsigned int cmd
, unsigned long arg
)
406 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
407 struct dm_table
*map
= dm_get_live_table(md
);
408 struct dm_target
*tgt
;
411 if (!map
|| !dm_table_get_size(map
))
414 /* We only support devices that have a single target */
415 if (dm_table_get_num_targets(map
) != 1)
418 tgt
= dm_table_get_target(map
, 0);
420 if (dm_suspended_md(md
)) {
425 if (tgt
->type
->ioctl
)
426 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
434 static struct dm_io
*alloc_io(struct mapped_device
*md
)
436 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
439 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
441 mempool_free(io
, md
->io_pool
);
444 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
446 bio_put(&tio
->clone
);
449 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
452 return mempool_alloc(md
->tio_pool
, gfp_mask
);
455 static void free_rq_tio(struct dm_rq_target_io
*tio
)
457 mempool_free(tio
, tio
->md
->tio_pool
);
460 static int md_in_flight(struct mapped_device
*md
)
462 return atomic_read(&md
->pending
[READ
]) +
463 atomic_read(&md
->pending
[WRITE
]);
466 static void start_io_acct(struct dm_io
*io
)
468 struct mapped_device
*md
= io
->md
;
470 int rw
= bio_data_dir(io
->bio
);
472 io
->start_time
= jiffies
;
474 cpu
= part_stat_lock();
475 part_round_stats(cpu
, &dm_disk(md
)->part0
);
477 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
478 atomic_inc_return(&md
->pending
[rw
]));
481 static void end_io_acct(struct dm_io
*io
)
483 struct mapped_device
*md
= io
->md
;
484 struct bio
*bio
= io
->bio
;
485 unsigned long duration
= jiffies
- io
->start_time
;
487 int rw
= bio_data_dir(bio
);
489 cpu
= part_stat_lock();
490 part_round_stats(cpu
, &dm_disk(md
)->part0
);
491 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
495 * After this is decremented the bio must not be touched if it is
498 pending
= atomic_dec_return(&md
->pending
[rw
]);
499 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
500 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
502 /* nudge anyone waiting on suspend queue */
508 * Add the bio to the list of deferred io.
510 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
514 spin_lock_irqsave(&md
->deferred_lock
, flags
);
515 bio_list_add(&md
->deferred
, bio
);
516 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
517 queue_work(md
->wq
, &md
->work
);
521 * Everyone (including functions in this file), should use this
522 * function to access the md->map field, and make sure they call
523 * dm_table_put() when finished.
525 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
530 read_lock_irqsave(&md
->map_lock
, flags
);
534 read_unlock_irqrestore(&md
->map_lock
, flags
);
540 * Get the geometry associated with a dm device
542 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
550 * Set the geometry of a device.
552 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
554 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
556 if (geo
->start
> sz
) {
557 DMWARN("Start sector is beyond the geometry limits.");
566 /*-----------------------------------------------------------------
568 * A more elegant soln is in the works that uses the queue
569 * merge fn, unfortunately there are a couple of changes to
570 * the block layer that I want to make for this. So in the
571 * interests of getting something for people to use I give
572 * you this clearly demarcated crap.
573 *---------------------------------------------------------------*/
575 static int __noflush_suspending(struct mapped_device
*md
)
577 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
581 * Decrements the number of outstanding ios that a bio has been
582 * cloned into, completing the original io if necc.
584 static void dec_pending(struct dm_io
*io
, int error
)
589 struct mapped_device
*md
= io
->md
;
591 /* Push-back supersedes any I/O errors */
592 if (unlikely(error
)) {
593 spin_lock_irqsave(&io
->endio_lock
, flags
);
594 if (!(io
->error
> 0 && __noflush_suspending(md
)))
596 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
599 if (atomic_dec_and_test(&io
->io_count
)) {
600 if (io
->error
== DM_ENDIO_REQUEUE
) {
602 * Target requested pushing back the I/O.
604 spin_lock_irqsave(&md
->deferred_lock
, flags
);
605 if (__noflush_suspending(md
))
606 bio_list_add_head(&md
->deferred
, io
->bio
);
608 /* noflush suspend was interrupted. */
610 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
613 io_error
= io
->error
;
618 if (io_error
== DM_ENDIO_REQUEUE
)
621 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
623 * Preflush done for flush with data, reissue
626 bio
->bi_rw
&= ~REQ_FLUSH
;
629 /* done with normal IO or empty flush */
630 trace_block_bio_complete(md
->queue
, bio
, io_error
);
631 bio_endio(bio
, io_error
);
636 static void clone_endio(struct bio
*bio
, int error
)
639 struct dm_target_io
*tio
= bio
->bi_private
;
640 struct dm_io
*io
= tio
->io
;
641 struct mapped_device
*md
= tio
->io
->md
;
642 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
644 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
648 r
= endio(tio
->ti
, bio
, error
);
649 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
651 * error and requeue request are handled
655 else if (r
== DM_ENDIO_INCOMPLETE
)
656 /* The target will handle the io */
659 DMWARN("unimplemented target endio return value: %d", r
);
665 dec_pending(io
, error
);
669 * Partial completion handling for request-based dm
671 static void end_clone_bio(struct bio
*clone
, int error
)
673 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
674 struct dm_rq_target_io
*tio
= info
->tio
;
675 struct bio
*bio
= info
->orig
;
676 unsigned int nr_bytes
= info
->orig
->bi_size
;
682 * An error has already been detected on the request.
683 * Once error occurred, just let clone->end_io() handle
689 * Don't notice the error to the upper layer yet.
690 * The error handling decision is made by the target driver,
691 * when the request is completed.
698 * I/O for the bio successfully completed.
699 * Notice the data completion to the upper layer.
703 * bios are processed from the head of the list.
704 * So the completing bio should always be rq->bio.
705 * If it's not, something wrong is happening.
707 if (tio
->orig
->bio
!= bio
)
708 DMERR("bio completion is going in the middle of the request");
711 * Update the original request.
712 * Do not use blk_end_request() here, because it may complete
713 * the original request before the clone, and break the ordering.
715 blk_update_request(tio
->orig
, 0, nr_bytes
);
719 * Don't touch any member of the md after calling this function because
720 * the md may be freed in dm_put() at the end of this function.
721 * Or do dm_get() before calling this function and dm_put() later.
723 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
725 atomic_dec(&md
->pending
[rw
]);
727 /* nudge anyone waiting on suspend queue */
728 if (!md_in_flight(md
))
732 * Run this off this callpath, as drivers could invoke end_io while
733 * inside their request_fn (and holding the queue lock). Calling
734 * back into ->request_fn() could deadlock attempting to grab the
738 blk_run_queue_async(md
->queue
);
741 * dm_put() must be at the end of this function. See the comment above
746 static void free_rq_clone(struct request
*clone
)
748 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
750 blk_rq_unprep_clone(clone
);
755 * Complete the clone and the original request.
756 * Must be called without queue lock.
758 static void dm_end_request(struct request
*clone
, int error
)
760 int rw
= rq_data_dir(clone
);
761 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
762 struct mapped_device
*md
= tio
->md
;
763 struct request
*rq
= tio
->orig
;
765 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
766 rq
->errors
= clone
->errors
;
767 rq
->resid_len
= clone
->resid_len
;
771 * We are using the sense buffer of the original
773 * So setting the length of the sense data is enough.
775 rq
->sense_len
= clone
->sense_len
;
778 free_rq_clone(clone
);
779 blk_end_request_all(rq
, error
);
780 rq_completed(md
, rw
, true);
783 static void dm_unprep_request(struct request
*rq
)
785 struct request
*clone
= rq
->special
;
788 rq
->cmd_flags
&= ~REQ_DONTPREP
;
790 free_rq_clone(clone
);
794 * Requeue the original request of a clone.
796 void dm_requeue_unmapped_request(struct request
*clone
)
798 int rw
= rq_data_dir(clone
);
799 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
800 struct mapped_device
*md
= tio
->md
;
801 struct request
*rq
= tio
->orig
;
802 struct request_queue
*q
= rq
->q
;
805 dm_unprep_request(rq
);
807 spin_lock_irqsave(q
->queue_lock
, flags
);
808 blk_requeue_request(q
, rq
);
809 spin_unlock_irqrestore(q
->queue_lock
, flags
);
811 rq_completed(md
, rw
, 0);
813 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
815 static void __stop_queue(struct request_queue
*q
)
820 static void stop_queue(struct request_queue
*q
)
824 spin_lock_irqsave(q
->queue_lock
, flags
);
826 spin_unlock_irqrestore(q
->queue_lock
, flags
);
829 static void __start_queue(struct request_queue
*q
)
831 if (blk_queue_stopped(q
))
835 static void start_queue(struct request_queue
*q
)
839 spin_lock_irqsave(q
->queue_lock
, flags
);
841 spin_unlock_irqrestore(q
->queue_lock
, flags
);
844 static void dm_done(struct request
*clone
, int error
, bool mapped
)
847 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
848 dm_request_endio_fn rq_end_io
= NULL
;
851 rq_end_io
= tio
->ti
->type
->rq_end_io
;
853 if (mapped
&& rq_end_io
)
854 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
858 /* The target wants to complete the I/O */
859 dm_end_request(clone
, r
);
860 else if (r
== DM_ENDIO_INCOMPLETE
)
861 /* The target will handle the I/O */
863 else if (r
== DM_ENDIO_REQUEUE
)
864 /* The target wants to requeue the I/O */
865 dm_requeue_unmapped_request(clone
);
867 DMWARN("unimplemented target endio return value: %d", r
);
873 * Request completion handler for request-based dm
875 static void dm_softirq_done(struct request
*rq
)
878 struct request
*clone
= rq
->completion_data
;
879 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
881 if (rq
->cmd_flags
& REQ_FAILED
)
884 dm_done(clone
, tio
->error
, mapped
);
888 * Complete the clone and the original request with the error status
889 * through softirq context.
891 static void dm_complete_request(struct request
*clone
, int error
)
893 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
894 struct request
*rq
= tio
->orig
;
897 rq
->completion_data
= clone
;
898 blk_complete_request(rq
);
902 * Complete the not-mapped clone and the original request with the error status
903 * through softirq context.
904 * Target's rq_end_io() function isn't called.
905 * This may be used when the target's map_rq() function fails.
907 void dm_kill_unmapped_request(struct request
*clone
, int error
)
909 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
910 struct request
*rq
= tio
->orig
;
912 rq
->cmd_flags
|= REQ_FAILED
;
913 dm_complete_request(clone
, error
);
915 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
918 * Called with the queue lock held
920 static void end_clone_request(struct request
*clone
, int error
)
923 * For just cleaning up the information of the queue in which
924 * the clone was dispatched.
925 * The clone is *NOT* freed actually here because it is alloced from
926 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
928 __blk_put_request(clone
->q
, clone
);
931 * Actual request completion is done in a softirq context which doesn't
932 * hold the queue lock. Otherwise, deadlock could occur because:
933 * - another request may be submitted by the upper level driver
934 * of the stacking during the completion
935 * - the submission which requires queue lock may be done
938 dm_complete_request(clone
, error
);
942 * Return maximum size of I/O possible at the supplied sector up to the current
945 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
947 sector_t target_offset
= dm_target_offset(ti
, sector
);
949 return ti
->len
- target_offset
;
952 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
954 sector_t len
= max_io_len_target_boundary(sector
, ti
);
955 sector_t offset
, max_len
;
958 * Does the target need to split even further?
960 if (ti
->max_io_len
) {
961 offset
= dm_target_offset(ti
, sector
);
962 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
963 max_len
= sector_div(offset
, ti
->max_io_len
);
965 max_len
= offset
& (ti
->max_io_len
- 1);
966 max_len
= ti
->max_io_len
- max_len
;
975 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
977 if (len
> UINT_MAX
) {
978 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
979 (unsigned long long)len
, UINT_MAX
);
980 ti
->error
= "Maximum size of target IO is too large";
984 ti
->max_io_len
= (uint32_t) len
;
988 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
990 static void __map_bio(struct dm_target
*ti
, struct dm_target_io
*tio
)
994 struct mapped_device
*md
;
995 struct bio
*clone
= &tio
->clone
;
997 clone
->bi_end_io
= clone_endio
;
998 clone
->bi_private
= tio
;
1001 * Map the clone. If r == 0 we don't need to do
1002 * anything, the target has assumed ownership of
1005 atomic_inc(&tio
->io
->io_count
);
1006 sector
= clone
->bi_sector
;
1007 r
= ti
->type
->map(ti
, clone
);
1008 if (r
== DM_MAPIO_REMAPPED
) {
1009 /* the bio has been remapped so dispatch it */
1011 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1012 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1014 generic_make_request(clone
);
1015 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1016 /* error the io and bail out, or requeue it if needed */
1018 dec_pending(tio
->io
, r
);
1021 DMWARN("unimplemented target map return value: %d", r
);
1027 struct mapped_device
*md
;
1028 struct dm_table
*map
;
1032 sector_t sector_count
;
1037 * Creates a little bio that just does part of a bvec.
1039 static void split_bvec(struct dm_target_io
*tio
, struct bio
*bio
,
1040 sector_t sector
, unsigned short idx
, unsigned int offset
,
1041 unsigned int len
, struct bio_set
*bs
)
1043 struct bio
*clone
= &tio
->clone
;
1044 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1046 *clone
->bi_io_vec
= *bv
;
1048 clone
->bi_sector
= sector
;
1049 clone
->bi_bdev
= bio
->bi_bdev
;
1050 clone
->bi_rw
= bio
->bi_rw
;
1052 clone
->bi_size
= to_bytes(len
);
1053 clone
->bi_io_vec
->bv_offset
= offset
;
1054 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1055 clone
->bi_flags
|= 1 << BIO_CLONED
;
1057 if (bio_integrity(bio
)) {
1058 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1059 bio_integrity_trim(clone
,
1060 bio_sector_offset(bio
, idx
, offset
), len
);
1065 * Creates a bio that consists of range of complete bvecs.
1067 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1068 sector_t sector
, unsigned short idx
,
1069 unsigned short bv_count
, unsigned int len
,
1072 struct bio
*clone
= &tio
->clone
;
1074 __bio_clone(clone
, bio
);
1075 clone
->bi_sector
= sector
;
1076 clone
->bi_idx
= idx
;
1077 clone
->bi_vcnt
= idx
+ bv_count
;
1078 clone
->bi_size
= to_bytes(len
);
1079 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1081 if (bio_integrity(bio
)) {
1082 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1084 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1085 bio_integrity_trim(clone
,
1086 bio_sector_offset(bio
, idx
, 0), len
);
1090 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1091 struct dm_target
*ti
, int nr_iovecs
)
1093 struct dm_target_io
*tio
;
1096 clone
= bio_alloc_bioset(GFP_NOIO
, nr_iovecs
, ci
->md
->bs
);
1097 tio
= container_of(clone
, struct dm_target_io
, clone
);
1101 memset(&tio
->info
, 0, sizeof(tio
->info
));
1102 tio
->target_request_nr
= 0;
1107 static void __issue_target_request(struct clone_info
*ci
, struct dm_target
*ti
,
1108 unsigned request_nr
, sector_t len
)
1110 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, ci
->bio
->bi_max_vecs
);
1111 struct bio
*clone
= &tio
->clone
;
1113 tio
->target_request_nr
= request_nr
;
1116 * Discard requests require the bio's inline iovecs be initialized.
1117 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1118 * and discard, so no need for concern about wasted bvec allocations.
1121 __bio_clone(clone
, ci
->bio
);
1123 clone
->bi_sector
= ci
->sector
;
1124 clone
->bi_size
= to_bytes(len
);
1130 static void __issue_target_requests(struct clone_info
*ci
, struct dm_target
*ti
,
1131 unsigned num_requests
, sector_t len
)
1133 unsigned request_nr
;
1135 for (request_nr
= 0; request_nr
< num_requests
; request_nr
++)
1136 __issue_target_request(ci
, ti
, request_nr
, len
);
1139 static int __clone_and_map_empty_flush(struct clone_info
*ci
)
1141 unsigned target_nr
= 0;
1142 struct dm_target
*ti
;
1144 BUG_ON(bio_has_data(ci
->bio
));
1145 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1146 __issue_target_requests(ci
, ti
, ti
->num_flush_requests
, 0);
1152 * Perform all io with a single clone.
1154 static void __clone_and_map_simple(struct clone_info
*ci
, struct dm_target
*ti
)
1156 struct bio
*bio
= ci
->bio
;
1157 struct dm_target_io
*tio
;
1159 tio
= alloc_tio(ci
, ti
, bio
->bi_max_vecs
);
1160 clone_bio(tio
, bio
, ci
->sector
, ci
->idx
, bio
->bi_vcnt
- ci
->idx
,
1161 ci
->sector_count
, ci
->md
->bs
);
1163 ci
->sector_count
= 0;
1166 typedef unsigned (*get_num_requests_fn
)(struct dm_target
*ti
);
1168 static unsigned get_num_discard_requests(struct dm_target
*ti
)
1170 return ti
->num_discard_requests
;
1173 static unsigned get_num_write_same_requests(struct dm_target
*ti
)
1175 return ti
->num_write_same_requests
;
1178 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1180 static bool is_split_required_for_discard(struct dm_target
*ti
)
1182 return ti
->split_discard_requests
;
1185 static int __clone_and_map_changing_extent_only(struct clone_info
*ci
,
1186 get_num_requests_fn get_num_requests
,
1187 is_split_required_fn is_split_required
)
1189 struct dm_target
*ti
;
1191 unsigned num_requests
;
1194 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1195 if (!dm_target_is_valid(ti
))
1199 * Even though the device advertised support for this type of
1200 * request, that does not mean every target supports it, and
1201 * reconfiguration might also have changed that since the
1202 * check was performed.
1204 num_requests
= get_num_requests
? get_num_requests(ti
) : 0;
1208 if (is_split_required
&& !is_split_required(ti
))
1209 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1211 len
= min(ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1213 __issue_target_requests(ci
, ti
, num_requests
, len
);
1216 } while (ci
->sector_count
-= len
);
1221 static int __clone_and_map_discard(struct clone_info
*ci
)
1223 return __clone_and_map_changing_extent_only(ci
, get_num_discard_requests
,
1224 is_split_required_for_discard
);
1227 static int __clone_and_map_write_same(struct clone_info
*ci
)
1229 return __clone_and_map_changing_extent_only(ci
, get_num_write_same_requests
, NULL
);
1232 static int __clone_and_map(struct clone_info
*ci
)
1234 struct bio
*bio
= ci
->bio
;
1235 struct dm_target
*ti
;
1236 sector_t len
= 0, max
;
1237 struct dm_target_io
*tio
;
1239 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1240 return __clone_and_map_discard(ci
);
1241 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1242 return __clone_and_map_write_same(ci
);
1244 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1245 if (!dm_target_is_valid(ti
))
1248 max
= max_io_len(ci
->sector
, ti
);
1250 if (ci
->sector_count
<= max
) {
1252 * Optimise for the simple case where we can do all of
1253 * the remaining io with a single clone.
1255 __clone_and_map_simple(ci
, ti
);
1257 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1259 * There are some bvecs that don't span targets.
1260 * Do as many of these as possible.
1263 sector_t remaining
= max
;
1266 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
1267 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
1269 if (bv_len
> remaining
)
1272 remaining
-= bv_len
;
1276 tio
= alloc_tio(ci
, ti
, bio
->bi_max_vecs
);
1277 clone_bio(tio
, bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
1282 ci
->sector_count
-= len
;
1287 * Handle a bvec that must be split between two or more targets.
1289 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1290 sector_t remaining
= to_sector(bv
->bv_len
);
1291 unsigned int offset
= 0;
1295 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1296 if (!dm_target_is_valid(ti
))
1299 max
= max_io_len(ci
->sector
, ti
);
1302 len
= min(remaining
, max
);
1304 tio
= alloc_tio(ci
, ti
, 1);
1305 split_bvec(tio
, bio
, ci
->sector
, ci
->idx
,
1306 bv
->bv_offset
+ offset
, len
, ci
->md
->bs
);
1311 ci
->sector_count
-= len
;
1312 offset
+= to_bytes(len
);
1313 } while (remaining
-= len
);
1322 * Split the bio into several clones and submit it to targets.
1324 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1326 struct clone_info ci
;
1329 ci
.map
= dm_get_live_table(md
);
1330 if (unlikely(!ci
.map
)) {
1336 ci
.io
= alloc_io(md
);
1338 atomic_set(&ci
.io
->io_count
, 1);
1341 spin_lock_init(&ci
.io
->endio_lock
);
1342 ci
.sector
= bio
->bi_sector
;
1343 ci
.idx
= bio
->bi_idx
;
1345 start_io_acct(ci
.io
);
1346 if (bio
->bi_rw
& REQ_FLUSH
) {
1347 ci
.bio
= &ci
.md
->flush_bio
;
1348 ci
.sector_count
= 0;
1349 error
= __clone_and_map_empty_flush(&ci
);
1350 /* dec_pending submits any data associated with flush */
1353 ci
.sector_count
= bio_sectors(bio
);
1354 while (ci
.sector_count
&& !error
)
1355 error
= __clone_and_map(&ci
);
1358 /* drop the extra reference count */
1359 dec_pending(ci
.io
, error
);
1360 dm_table_put(ci
.map
);
1362 /*-----------------------------------------------------------------
1364 *---------------------------------------------------------------*/
1366 static int dm_merge_bvec(struct request_queue
*q
,
1367 struct bvec_merge_data
*bvm
,
1368 struct bio_vec
*biovec
)
1370 struct mapped_device
*md
= q
->queuedata
;
1371 struct dm_table
*map
= dm_get_live_table(md
);
1372 struct dm_target
*ti
;
1373 sector_t max_sectors
;
1379 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1380 if (!dm_target_is_valid(ti
))
1384 * Find maximum amount of I/O that won't need splitting
1386 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1387 (sector_t
) BIO_MAX_SECTORS
);
1388 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1393 * merge_bvec_fn() returns number of bytes
1394 * it can accept at this offset
1395 * max is precomputed maximal io size
1397 if (max_size
&& ti
->type
->merge
)
1398 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1400 * If the target doesn't support merge method and some of the devices
1401 * provided their merge_bvec method (we know this by looking at
1402 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1403 * entries. So always set max_size to 0, and the code below allows
1406 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1415 * Always allow an entire first page
1417 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1418 max_size
= biovec
->bv_len
;
1424 * The request function that just remaps the bio built up by
1427 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1429 int rw
= bio_data_dir(bio
);
1430 struct mapped_device
*md
= q
->queuedata
;
1433 down_read(&md
->io_lock
);
1435 cpu
= part_stat_lock();
1436 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1437 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1440 /* if we're suspended, we have to queue this io for later */
1441 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1442 up_read(&md
->io_lock
);
1444 if (bio_rw(bio
) != READA
)
1451 __split_and_process_bio(md
, bio
);
1452 up_read(&md
->io_lock
);
1456 static int dm_request_based(struct mapped_device
*md
)
1458 return blk_queue_stackable(md
->queue
);
1461 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1463 struct mapped_device
*md
= q
->queuedata
;
1465 if (dm_request_based(md
))
1466 blk_queue_bio(q
, bio
);
1468 _dm_request(q
, bio
);
1471 void dm_dispatch_request(struct request
*rq
)
1475 if (blk_queue_io_stat(rq
->q
))
1476 rq
->cmd_flags
|= REQ_IO_STAT
;
1478 rq
->start_time
= jiffies
;
1479 r
= blk_insert_cloned_request(rq
->q
, rq
);
1481 dm_complete_request(rq
, r
);
1483 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1485 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1488 struct dm_rq_target_io
*tio
= data
;
1489 struct dm_rq_clone_bio_info
*info
=
1490 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1492 info
->orig
= bio_orig
;
1494 bio
->bi_end_io
= end_clone_bio
;
1495 bio
->bi_private
= info
;
1500 static int setup_clone(struct request
*clone
, struct request
*rq
,
1501 struct dm_rq_target_io
*tio
)
1505 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1506 dm_rq_bio_constructor
, tio
);
1510 clone
->cmd
= rq
->cmd
;
1511 clone
->cmd_len
= rq
->cmd_len
;
1512 clone
->sense
= rq
->sense
;
1513 clone
->buffer
= rq
->buffer
;
1514 clone
->end_io
= end_clone_request
;
1515 clone
->end_io_data
= tio
;
1520 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1523 struct request
*clone
;
1524 struct dm_rq_target_io
*tio
;
1526 tio
= alloc_rq_tio(md
, gfp_mask
);
1534 memset(&tio
->info
, 0, sizeof(tio
->info
));
1536 clone
= &tio
->clone
;
1537 if (setup_clone(clone
, rq
, tio
)) {
1547 * Called with the queue lock held.
1549 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1551 struct mapped_device
*md
= q
->queuedata
;
1552 struct request
*clone
;
1554 if (unlikely(rq
->special
)) {
1555 DMWARN("Already has something in rq->special.");
1556 return BLKPREP_KILL
;
1559 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1561 return BLKPREP_DEFER
;
1563 rq
->special
= clone
;
1564 rq
->cmd_flags
|= REQ_DONTPREP
;
1571 * 0 : the request has been processed (not requeued)
1572 * !0 : the request has been requeued
1574 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1575 struct mapped_device
*md
)
1577 int r
, requeued
= 0;
1578 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1581 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1583 case DM_MAPIO_SUBMITTED
:
1584 /* The target has taken the I/O to submit by itself later */
1586 case DM_MAPIO_REMAPPED
:
1587 /* The target has remapped the I/O so dispatch it */
1588 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1589 blk_rq_pos(tio
->orig
));
1590 dm_dispatch_request(clone
);
1592 case DM_MAPIO_REQUEUE
:
1593 /* The target wants to requeue the I/O */
1594 dm_requeue_unmapped_request(clone
);
1599 DMWARN("unimplemented target map return value: %d", r
);
1603 /* The target wants to complete the I/O */
1604 dm_kill_unmapped_request(clone
, r
);
1611 static struct request
*dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1613 struct request
*clone
;
1615 blk_start_request(orig
);
1616 clone
= orig
->special
;
1617 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1620 * Hold the md reference here for the in-flight I/O.
1621 * We can't rely on the reference count by device opener,
1622 * because the device may be closed during the request completion
1623 * when all bios are completed.
1624 * See the comment in rq_completed() too.
1632 * q->request_fn for request-based dm.
1633 * Called with the queue lock held.
1635 static void dm_request_fn(struct request_queue
*q
)
1637 struct mapped_device
*md
= q
->queuedata
;
1638 struct dm_table
*map
= dm_get_live_table(md
);
1639 struct dm_target
*ti
;
1640 struct request
*rq
, *clone
;
1644 * For suspend, check blk_queue_stopped() and increment
1645 * ->pending within a single queue_lock not to increment the
1646 * number of in-flight I/Os after the queue is stopped in
1649 while (!blk_queue_stopped(q
)) {
1650 rq
= blk_peek_request(q
);
1654 /* always use block 0 to find the target for flushes for now */
1656 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1657 pos
= blk_rq_pos(rq
);
1659 ti
= dm_table_find_target(map
, pos
);
1660 if (!dm_target_is_valid(ti
)) {
1662 * Must perform setup, that dm_done() requires,
1663 * before calling dm_kill_unmapped_request
1665 DMERR_LIMIT("request attempted access beyond the end of device");
1666 clone
= dm_start_request(md
, rq
);
1667 dm_kill_unmapped_request(clone
, -EIO
);
1671 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1674 clone
= dm_start_request(md
, rq
);
1676 spin_unlock(q
->queue_lock
);
1677 if (map_request(ti
, clone
, md
))
1680 BUG_ON(!irqs_disabled());
1681 spin_lock(q
->queue_lock
);
1687 BUG_ON(!irqs_disabled());
1688 spin_lock(q
->queue_lock
);
1691 blk_delay_queue(q
, HZ
/ 10);
1696 int dm_underlying_device_busy(struct request_queue
*q
)
1698 return blk_lld_busy(q
);
1700 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1702 static int dm_lld_busy(struct request_queue
*q
)
1705 struct mapped_device
*md
= q
->queuedata
;
1706 struct dm_table
*map
= dm_get_live_table(md
);
1708 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1711 r
= dm_table_any_busy_target(map
);
1718 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1721 struct mapped_device
*md
= congested_data
;
1722 struct dm_table
*map
;
1724 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1725 map
= dm_get_live_table(md
);
1728 * Request-based dm cares about only own queue for
1729 * the query about congestion status of request_queue
1731 if (dm_request_based(md
))
1732 r
= md
->queue
->backing_dev_info
.state
&
1735 r
= dm_table_any_congested(map
, bdi_bits
);
1744 /*-----------------------------------------------------------------
1745 * An IDR is used to keep track of allocated minor numbers.
1746 *---------------------------------------------------------------*/
1747 static void free_minor(int minor
)
1749 spin_lock(&_minor_lock
);
1750 idr_remove(&_minor_idr
, minor
);
1751 spin_unlock(&_minor_lock
);
1755 * See if the device with a specific minor # is free.
1757 static int specific_minor(int minor
)
1761 if (minor
>= (1 << MINORBITS
))
1764 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1768 spin_lock(&_minor_lock
);
1770 if (idr_find(&_minor_idr
, minor
)) {
1775 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1780 idr_remove(&_minor_idr
, m
);
1786 spin_unlock(&_minor_lock
);
1790 static int next_free_minor(int *minor
)
1794 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1798 spin_lock(&_minor_lock
);
1800 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1804 if (m
>= (1 << MINORBITS
)) {
1805 idr_remove(&_minor_idr
, m
);
1813 spin_unlock(&_minor_lock
);
1817 static const struct block_device_operations dm_blk_dops
;
1819 static void dm_wq_work(struct work_struct
*work
);
1821 static void dm_init_md_queue(struct mapped_device
*md
)
1824 * Request-based dm devices cannot be stacked on top of bio-based dm
1825 * devices. The type of this dm device has not been decided yet.
1826 * The type is decided at the first table loading time.
1827 * To prevent problematic device stacking, clear the queue flag
1828 * for request stacking support until then.
1830 * This queue is new, so no concurrency on the queue_flags.
1832 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1834 md
->queue
->queuedata
= md
;
1835 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1836 md
->queue
->backing_dev_info
.congested_data
= md
;
1837 blk_queue_make_request(md
->queue
, dm_request
);
1838 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1839 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1843 * Allocate and initialise a blank device with a given minor.
1845 static struct mapped_device
*alloc_dev(int minor
)
1848 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1852 DMWARN("unable to allocate device, out of memory.");
1856 if (!try_module_get(THIS_MODULE
))
1857 goto bad_module_get
;
1859 /* get a minor number for the dev */
1860 if (minor
== DM_ANY_MINOR
)
1861 r
= next_free_minor(&minor
);
1863 r
= specific_minor(minor
);
1867 md
->type
= DM_TYPE_NONE
;
1868 init_rwsem(&md
->io_lock
);
1869 mutex_init(&md
->suspend_lock
);
1870 mutex_init(&md
->type_lock
);
1871 spin_lock_init(&md
->deferred_lock
);
1872 rwlock_init(&md
->map_lock
);
1873 atomic_set(&md
->holders
, 1);
1874 atomic_set(&md
->open_count
, 0);
1875 atomic_set(&md
->event_nr
, 0);
1876 atomic_set(&md
->uevent_seq
, 0);
1877 INIT_LIST_HEAD(&md
->uevent_list
);
1878 spin_lock_init(&md
->uevent_lock
);
1880 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1884 dm_init_md_queue(md
);
1886 md
->disk
= alloc_disk(1);
1890 atomic_set(&md
->pending
[0], 0);
1891 atomic_set(&md
->pending
[1], 0);
1892 init_waitqueue_head(&md
->wait
);
1893 INIT_WORK(&md
->work
, dm_wq_work
);
1894 init_waitqueue_head(&md
->eventq
);
1896 md
->disk
->major
= _major
;
1897 md
->disk
->first_minor
= minor
;
1898 md
->disk
->fops
= &dm_blk_dops
;
1899 md
->disk
->queue
= md
->queue
;
1900 md
->disk
->private_data
= md
;
1901 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1903 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1905 md
->wq
= alloc_workqueue("kdmflush",
1906 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1910 md
->bdev
= bdget_disk(md
->disk
, 0);
1914 bio_init(&md
->flush_bio
);
1915 md
->flush_bio
.bi_bdev
= md
->bdev
;
1916 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1918 /* Populate the mapping, nobody knows we exist yet */
1919 spin_lock(&_minor_lock
);
1920 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1921 spin_unlock(&_minor_lock
);
1923 BUG_ON(old_md
!= MINOR_ALLOCED
);
1928 destroy_workqueue(md
->wq
);
1930 del_gendisk(md
->disk
);
1933 blk_cleanup_queue(md
->queue
);
1937 module_put(THIS_MODULE
);
1943 static void unlock_fs(struct mapped_device
*md
);
1945 static void free_dev(struct mapped_device
*md
)
1947 int minor
= MINOR(disk_devt(md
->disk
));
1951 destroy_workqueue(md
->wq
);
1953 mempool_destroy(md
->tio_pool
);
1955 mempool_destroy(md
->io_pool
);
1957 bioset_free(md
->bs
);
1958 blk_integrity_unregister(md
->disk
);
1959 del_gendisk(md
->disk
);
1962 spin_lock(&_minor_lock
);
1963 md
->disk
->private_data
= NULL
;
1964 spin_unlock(&_minor_lock
);
1967 blk_cleanup_queue(md
->queue
);
1968 module_put(THIS_MODULE
);
1972 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1974 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1976 if (md
->io_pool
&& (md
->tio_pool
|| dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) && md
->bs
) {
1978 * The md already has necessary mempools. Reload just the
1979 * bioset because front_pad may have changed because
1980 * a different table was loaded.
1982 bioset_free(md
->bs
);
1988 BUG_ON(!p
|| md
->io_pool
|| md
->tio_pool
|| md
->bs
);
1990 md
->io_pool
= p
->io_pool
;
1992 md
->tio_pool
= p
->tio_pool
;
1998 /* mempool bind completed, now no need any mempools in the table */
1999 dm_table_free_md_mempools(t
);
2003 * Bind a table to the device.
2005 static void event_callback(void *context
)
2007 unsigned long flags
;
2009 struct mapped_device
*md
= (struct mapped_device
*) context
;
2011 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2012 list_splice_init(&md
->uevent_list
, &uevents
);
2013 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2015 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2017 atomic_inc(&md
->event_nr
);
2018 wake_up(&md
->eventq
);
2022 * Protected by md->suspend_lock obtained by dm_swap_table().
2024 static void __set_size(struct mapped_device
*md
, sector_t size
)
2026 set_capacity(md
->disk
, size
);
2028 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2032 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2034 * If this function returns 0, then the device is either a non-dm
2035 * device without a merge_bvec_fn, or it is a dm device that is
2036 * able to split any bios it receives that are too big.
2038 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2040 struct mapped_device
*dev_md
;
2042 if (!q
->merge_bvec_fn
)
2045 if (q
->make_request_fn
== dm_request
) {
2046 dev_md
= q
->queuedata
;
2047 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2054 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2055 struct dm_dev
*dev
, sector_t start
,
2056 sector_t len
, void *data
)
2058 struct block_device
*bdev
= dev
->bdev
;
2059 struct request_queue
*q
= bdev_get_queue(bdev
);
2061 return dm_queue_merge_is_compulsory(q
);
2065 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2066 * on the properties of the underlying devices.
2068 static int dm_table_merge_is_optional(struct dm_table
*table
)
2071 struct dm_target
*ti
;
2073 while (i
< dm_table_get_num_targets(table
)) {
2074 ti
= dm_table_get_target(table
, i
++);
2076 if (ti
->type
->iterate_devices
&&
2077 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2085 * Returns old map, which caller must destroy.
2087 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2088 struct queue_limits
*limits
)
2090 struct dm_table
*old_map
;
2091 struct request_queue
*q
= md
->queue
;
2093 unsigned long flags
;
2094 int merge_is_optional
;
2096 size
= dm_table_get_size(t
);
2099 * Wipe any geometry if the size of the table changed.
2101 if (size
!= get_capacity(md
->disk
))
2102 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2104 __set_size(md
, size
);
2106 dm_table_event_callback(t
, event_callback
, md
);
2109 * The queue hasn't been stopped yet, if the old table type wasn't
2110 * for request-based during suspension. So stop it to prevent
2111 * I/O mapping before resume.
2112 * This must be done before setting the queue restrictions,
2113 * because request-based dm may be run just after the setting.
2115 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2118 __bind_mempools(md
, t
);
2120 merge_is_optional
= dm_table_merge_is_optional(t
);
2122 write_lock_irqsave(&md
->map_lock
, flags
);
2125 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2127 dm_table_set_restrictions(t
, q
, limits
);
2128 if (merge_is_optional
)
2129 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2131 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2132 write_unlock_irqrestore(&md
->map_lock
, flags
);
2138 * Returns unbound table for the caller to free.
2140 static struct dm_table
*__unbind(struct mapped_device
*md
)
2142 struct dm_table
*map
= md
->map
;
2143 unsigned long flags
;
2148 dm_table_event_callback(map
, NULL
, NULL
);
2149 write_lock_irqsave(&md
->map_lock
, flags
);
2151 write_unlock_irqrestore(&md
->map_lock
, flags
);
2157 * Constructor for a new device.
2159 int dm_create(int minor
, struct mapped_device
**result
)
2161 struct mapped_device
*md
;
2163 md
= alloc_dev(minor
);
2174 * Functions to manage md->type.
2175 * All are required to hold md->type_lock.
2177 void dm_lock_md_type(struct mapped_device
*md
)
2179 mutex_lock(&md
->type_lock
);
2182 void dm_unlock_md_type(struct mapped_device
*md
)
2184 mutex_unlock(&md
->type_lock
);
2187 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2192 unsigned dm_get_md_type(struct mapped_device
*md
)
2197 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2199 return md
->immutable_target_type
;
2203 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2205 static int dm_init_request_based_queue(struct mapped_device
*md
)
2207 struct request_queue
*q
= NULL
;
2209 if (md
->queue
->elevator
)
2212 /* Fully initialize the queue */
2213 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2218 dm_init_md_queue(md
);
2219 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2220 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2221 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2223 elv_register_queue(md
->queue
);
2229 * Setup the DM device's queue based on md's type
2231 int dm_setup_md_queue(struct mapped_device
*md
)
2233 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2234 !dm_init_request_based_queue(md
)) {
2235 DMWARN("Cannot initialize queue for request-based mapped device");
2242 static struct mapped_device
*dm_find_md(dev_t dev
)
2244 struct mapped_device
*md
;
2245 unsigned minor
= MINOR(dev
);
2247 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2250 spin_lock(&_minor_lock
);
2252 md
= idr_find(&_minor_idr
, minor
);
2253 if (md
&& (md
== MINOR_ALLOCED
||
2254 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2255 dm_deleting_md(md
) ||
2256 test_bit(DMF_FREEING
, &md
->flags
))) {
2262 spin_unlock(&_minor_lock
);
2267 struct mapped_device
*dm_get_md(dev_t dev
)
2269 struct mapped_device
*md
= dm_find_md(dev
);
2276 EXPORT_SYMBOL_GPL(dm_get_md
);
2278 void *dm_get_mdptr(struct mapped_device
*md
)
2280 return md
->interface_ptr
;
2283 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2285 md
->interface_ptr
= ptr
;
2288 void dm_get(struct mapped_device
*md
)
2290 atomic_inc(&md
->holders
);
2291 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2294 const char *dm_device_name(struct mapped_device
*md
)
2298 EXPORT_SYMBOL_GPL(dm_device_name
);
2300 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2302 struct dm_table
*map
;
2306 spin_lock(&_minor_lock
);
2307 map
= dm_get_live_table(md
);
2308 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2309 set_bit(DMF_FREEING
, &md
->flags
);
2310 spin_unlock(&_minor_lock
);
2312 if (!dm_suspended_md(md
)) {
2313 dm_table_presuspend_targets(map
);
2314 dm_table_postsuspend_targets(map
);
2318 * Rare, but there may be I/O requests still going to complete,
2319 * for example. Wait for all references to disappear.
2320 * No one should increment the reference count of the mapped_device,
2321 * after the mapped_device state becomes DMF_FREEING.
2324 while (atomic_read(&md
->holders
))
2326 else if (atomic_read(&md
->holders
))
2327 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2328 dm_device_name(md
), atomic_read(&md
->holders
));
2332 dm_table_destroy(__unbind(md
));
2336 void dm_destroy(struct mapped_device
*md
)
2338 __dm_destroy(md
, true);
2341 void dm_destroy_immediate(struct mapped_device
*md
)
2343 __dm_destroy(md
, false);
2346 void dm_put(struct mapped_device
*md
)
2348 atomic_dec(&md
->holders
);
2350 EXPORT_SYMBOL_GPL(dm_put
);
2352 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2355 DECLARE_WAITQUEUE(wait
, current
);
2357 add_wait_queue(&md
->wait
, &wait
);
2360 set_current_state(interruptible
);
2362 if (!md_in_flight(md
))
2365 if (interruptible
== TASK_INTERRUPTIBLE
&&
2366 signal_pending(current
)) {
2373 set_current_state(TASK_RUNNING
);
2375 remove_wait_queue(&md
->wait
, &wait
);
2381 * Process the deferred bios
2383 static void dm_wq_work(struct work_struct
*work
)
2385 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2389 down_read(&md
->io_lock
);
2391 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2392 spin_lock_irq(&md
->deferred_lock
);
2393 c
= bio_list_pop(&md
->deferred
);
2394 spin_unlock_irq(&md
->deferred_lock
);
2399 up_read(&md
->io_lock
);
2401 if (dm_request_based(md
))
2402 generic_make_request(c
);
2404 __split_and_process_bio(md
, c
);
2406 down_read(&md
->io_lock
);
2409 up_read(&md
->io_lock
);
2412 static void dm_queue_flush(struct mapped_device
*md
)
2414 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2415 smp_mb__after_clear_bit();
2416 queue_work(md
->wq
, &md
->work
);
2420 * Swap in a new table, returning the old one for the caller to destroy.
2422 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2424 struct dm_table
*live_map
, *map
= ERR_PTR(-EINVAL
);
2425 struct queue_limits limits
;
2428 mutex_lock(&md
->suspend_lock
);
2430 /* device must be suspended */
2431 if (!dm_suspended_md(md
))
2435 * If the new table has no data devices, retain the existing limits.
2436 * This helps multipath with queue_if_no_path if all paths disappear,
2437 * then new I/O is queued based on these limits, and then some paths
2440 if (dm_table_has_no_data_devices(table
)) {
2441 live_map
= dm_get_live_table(md
);
2443 limits
= md
->queue
->limits
;
2444 dm_table_put(live_map
);
2447 r
= dm_calculate_queue_limits(table
, &limits
);
2453 map
= __bind(md
, table
, &limits
);
2456 mutex_unlock(&md
->suspend_lock
);
2461 * Functions to lock and unlock any filesystem running on the
2464 static int lock_fs(struct mapped_device
*md
)
2468 WARN_ON(md
->frozen_sb
);
2470 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2471 if (IS_ERR(md
->frozen_sb
)) {
2472 r
= PTR_ERR(md
->frozen_sb
);
2473 md
->frozen_sb
= NULL
;
2477 set_bit(DMF_FROZEN
, &md
->flags
);
2482 static void unlock_fs(struct mapped_device
*md
)
2484 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2487 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2488 md
->frozen_sb
= NULL
;
2489 clear_bit(DMF_FROZEN
, &md
->flags
);
2493 * We need to be able to change a mapping table under a mounted
2494 * filesystem. For example we might want to move some data in
2495 * the background. Before the table can be swapped with
2496 * dm_bind_table, dm_suspend must be called to flush any in
2497 * flight bios and ensure that any further io gets deferred.
2500 * Suspend mechanism in request-based dm.
2502 * 1. Flush all I/Os by lock_fs() if needed.
2503 * 2. Stop dispatching any I/O by stopping the request_queue.
2504 * 3. Wait for all in-flight I/Os to be completed or requeued.
2506 * To abort suspend, start the request_queue.
2508 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2510 struct dm_table
*map
= NULL
;
2512 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2513 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2515 mutex_lock(&md
->suspend_lock
);
2517 if (dm_suspended_md(md
)) {
2522 map
= dm_get_live_table(md
);
2525 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2526 * This flag is cleared before dm_suspend returns.
2529 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2531 /* This does not get reverted if there's an error later. */
2532 dm_table_presuspend_targets(map
);
2535 * Flush I/O to the device.
2536 * Any I/O submitted after lock_fs() may not be flushed.
2537 * noflush takes precedence over do_lockfs.
2538 * (lock_fs() flushes I/Os and waits for them to complete.)
2540 if (!noflush
&& do_lockfs
) {
2547 * Here we must make sure that no processes are submitting requests
2548 * to target drivers i.e. no one may be executing
2549 * __split_and_process_bio. This is called from dm_request and
2552 * To get all processes out of __split_and_process_bio in dm_request,
2553 * we take the write lock. To prevent any process from reentering
2554 * __split_and_process_bio from dm_request and quiesce the thread
2555 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2556 * flush_workqueue(md->wq).
2558 down_write(&md
->io_lock
);
2559 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2560 up_write(&md
->io_lock
);
2563 * Stop md->queue before flushing md->wq in case request-based
2564 * dm defers requests to md->wq from md->queue.
2566 if (dm_request_based(md
))
2567 stop_queue(md
->queue
);
2569 flush_workqueue(md
->wq
);
2572 * At this point no more requests are entering target request routines.
2573 * We call dm_wait_for_completion to wait for all existing requests
2576 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2578 down_write(&md
->io_lock
);
2580 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2581 up_write(&md
->io_lock
);
2583 /* were we interrupted ? */
2587 if (dm_request_based(md
))
2588 start_queue(md
->queue
);
2591 goto out
; /* pushback list is already flushed, so skip flush */
2595 * If dm_wait_for_completion returned 0, the device is completely
2596 * quiescent now. There is no request-processing activity. All new
2597 * requests are being added to md->deferred list.
2600 set_bit(DMF_SUSPENDED
, &md
->flags
);
2602 dm_table_postsuspend_targets(map
);
2608 mutex_unlock(&md
->suspend_lock
);
2612 int dm_resume(struct mapped_device
*md
)
2615 struct dm_table
*map
= NULL
;
2617 mutex_lock(&md
->suspend_lock
);
2618 if (!dm_suspended_md(md
))
2621 map
= dm_get_live_table(md
);
2622 if (!map
|| !dm_table_get_size(map
))
2625 r
= dm_table_resume_targets(map
);
2632 * Flushing deferred I/Os must be done after targets are resumed
2633 * so that mapping of targets can work correctly.
2634 * Request-based dm is queueing the deferred I/Os in its request_queue.
2636 if (dm_request_based(md
))
2637 start_queue(md
->queue
);
2641 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2646 mutex_unlock(&md
->suspend_lock
);
2651 /*-----------------------------------------------------------------
2652 * Event notification.
2653 *---------------------------------------------------------------*/
2654 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2657 char udev_cookie
[DM_COOKIE_LENGTH
];
2658 char *envp
[] = { udev_cookie
, NULL
};
2661 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2663 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2664 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2665 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2670 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2672 return atomic_add_return(1, &md
->uevent_seq
);
2675 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2677 return atomic_read(&md
->event_nr
);
2680 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2682 return wait_event_interruptible(md
->eventq
,
2683 (event_nr
!= atomic_read(&md
->event_nr
)));
2686 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2688 unsigned long flags
;
2690 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2691 list_add(elist
, &md
->uevent_list
);
2692 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2696 * The gendisk is only valid as long as you have a reference
2699 struct gendisk
*dm_disk(struct mapped_device
*md
)
2704 struct kobject
*dm_kobject(struct mapped_device
*md
)
2710 * struct mapped_device should not be exported outside of dm.c
2711 * so use this check to verify that kobj is part of md structure
2713 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2715 struct mapped_device
*md
;
2717 md
= container_of(kobj
, struct mapped_device
, kobj
);
2718 if (&md
->kobj
!= kobj
)
2721 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2729 int dm_suspended_md(struct mapped_device
*md
)
2731 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2734 int dm_suspended(struct dm_target
*ti
)
2736 return dm_suspended_md(dm_table_get_md(ti
->table
));
2738 EXPORT_SYMBOL_GPL(dm_suspended
);
2740 int dm_noflush_suspending(struct dm_target
*ti
)
2742 return __noflush_suspending(dm_table_get_md(ti
->table
));
2744 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2746 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
2748 struct dm_md_mempools
*pools
= kmalloc(sizeof(*pools
), GFP_KERNEL
);
2749 unsigned int pool_size
= (type
== DM_TYPE_BIO_BASED
) ? 16 : MIN_IOS
;
2754 per_bio_data_size
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
));
2756 pools
->io_pool
= (type
== DM_TYPE_BIO_BASED
) ?
2757 mempool_create_slab_pool(MIN_IOS
, _io_cache
) :
2758 mempool_create_slab_pool(MIN_IOS
, _rq_bio_info_cache
);
2759 if (!pools
->io_pool
)
2760 goto free_pools_and_out
;
2762 pools
->tio_pool
= NULL
;
2763 if (type
== DM_TYPE_REQUEST_BASED
) {
2764 pools
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _rq_tio_cache
);
2765 if (!pools
->tio_pool
)
2766 goto free_io_pool_and_out
;
2769 pools
->bs
= (type
== DM_TYPE_BIO_BASED
) ?
2770 bioset_create(pool_size
,
2771 per_bio_data_size
+ offsetof(struct dm_target_io
, clone
)) :
2772 bioset_create(pool_size
,
2773 offsetof(struct dm_rq_clone_bio_info
, clone
));
2775 goto free_tio_pool_and_out
;
2777 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2778 goto free_bioset_and_out
;
2782 free_bioset_and_out
:
2783 bioset_free(pools
->bs
);
2785 free_tio_pool_and_out
:
2786 if (pools
->tio_pool
)
2787 mempool_destroy(pools
->tio_pool
);
2789 free_io_pool_and_out
:
2790 mempool_destroy(pools
->io_pool
);
2798 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2804 mempool_destroy(pools
->io_pool
);
2806 if (pools
->tio_pool
)
2807 mempool_destroy(pools
->tio_pool
);
2810 bioset_free(pools
->bs
);
2815 static const struct block_device_operations dm_blk_dops
= {
2816 .open
= dm_blk_open
,
2817 .release
= dm_blk_close
,
2818 .ioctl
= dm_blk_ioctl
,
2819 .getgeo
= dm_blk_getgeo
,
2820 .owner
= THIS_MODULE
2823 EXPORT_SYMBOL(dm_get_mapinfo
);
2828 module_init(dm_init
);
2829 module_exit(dm_exit
);
2831 module_param(major
, uint
, 0);
2832 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2833 MODULE_DESCRIPTION(DM_NAME
" driver");
2834 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2835 MODULE_LICENSE("GPL");