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.
173 wait_queue_head_t eventq
;
175 struct list_head uevent_list
;
176 spinlock_t uevent_lock
; /* Protect access to uevent_list */
179 * freeze/thaw support require holding onto a super block
181 struct super_block
*frozen_sb
;
182 struct block_device
*bdev
;
184 /* forced geometry settings */
185 struct hd_geometry geometry
;
190 /* zero-length flush that will be cloned and submitted to targets */
191 struct bio flush_bio
;
195 * For mempools pre-allocation at the table loading time.
197 struct dm_md_mempools
{
203 static struct kmem_cache
*_io_cache
;
204 static struct kmem_cache
*_rq_tio_cache
;
206 static int __init
local_init(void)
210 /* allocate a slab for the dm_ios */
211 _io_cache
= KMEM_CACHE(dm_io
, 0);
215 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
217 goto out_free_io_cache
;
219 r
= dm_uevent_init();
221 goto out_free_rq_tio_cache
;
224 r
= register_blkdev(_major
, _name
);
226 goto out_uevent_exit
;
235 out_free_rq_tio_cache
:
236 kmem_cache_destroy(_rq_tio_cache
);
238 kmem_cache_destroy(_io_cache
);
243 static void local_exit(void)
245 kmem_cache_destroy(_rq_tio_cache
);
246 kmem_cache_destroy(_io_cache
);
247 unregister_blkdev(_major
, _name
);
252 DMINFO("cleaned up");
255 static int (*_inits
[])(void) __initdata
= {
265 static void (*_exits
[])(void) = {
275 static int __init
dm_init(void)
277 const int count
= ARRAY_SIZE(_inits
);
281 for (i
= 0; i
< count
; i
++) {
296 static void __exit
dm_exit(void)
298 int i
= ARRAY_SIZE(_exits
);
304 * Should be empty by this point.
306 idr_destroy(&_minor_idr
);
310 * Block device functions
312 int dm_deleting_md(struct mapped_device
*md
)
314 return test_bit(DMF_DELETING
, &md
->flags
);
317 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
319 struct mapped_device
*md
;
321 spin_lock(&_minor_lock
);
323 md
= bdev
->bd_disk
->private_data
;
327 if (test_bit(DMF_FREEING
, &md
->flags
) ||
328 dm_deleting_md(md
)) {
334 atomic_inc(&md
->open_count
);
337 spin_unlock(&_minor_lock
);
339 return md
? 0 : -ENXIO
;
342 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
344 struct mapped_device
*md
= disk
->private_data
;
346 spin_lock(&_minor_lock
);
348 atomic_dec(&md
->open_count
);
351 spin_unlock(&_minor_lock
);
356 int dm_open_count(struct mapped_device
*md
)
358 return atomic_read(&md
->open_count
);
362 * Guarantees nothing is using the device before it's deleted.
364 int dm_lock_for_deletion(struct mapped_device
*md
)
368 spin_lock(&_minor_lock
);
370 if (dm_open_count(md
))
373 set_bit(DMF_DELETING
, &md
->flags
);
375 spin_unlock(&_minor_lock
);
380 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
382 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
384 return dm_get_geometry(md
, geo
);
387 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
388 unsigned int cmd
, unsigned long arg
)
390 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
391 struct dm_table
*map
= dm_get_live_table(md
);
392 struct dm_target
*tgt
;
395 if (!map
|| !dm_table_get_size(map
))
398 /* We only support devices that have a single target */
399 if (dm_table_get_num_targets(map
) != 1)
402 tgt
= dm_table_get_target(map
, 0);
404 if (dm_suspended_md(md
)) {
409 if (tgt
->type
->ioctl
)
410 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
418 static struct dm_io
*alloc_io(struct mapped_device
*md
)
420 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
423 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
425 mempool_free(io
, md
->io_pool
);
428 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
430 bio_put(&tio
->clone
);
433 static struct dm_rq_target_io
*alloc_rq_tio(struct mapped_device
*md
,
436 return mempool_alloc(md
->io_pool
, gfp_mask
);
439 static void free_rq_tio(struct dm_rq_target_io
*tio
)
441 mempool_free(tio
, tio
->md
->io_pool
);
444 static int md_in_flight(struct mapped_device
*md
)
446 return atomic_read(&md
->pending
[READ
]) +
447 atomic_read(&md
->pending
[WRITE
]);
450 static void start_io_acct(struct dm_io
*io
)
452 struct mapped_device
*md
= io
->md
;
454 int rw
= bio_data_dir(io
->bio
);
456 io
->start_time
= jiffies
;
458 cpu
= part_stat_lock();
459 part_round_stats(cpu
, &dm_disk(md
)->part0
);
461 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
462 atomic_inc_return(&md
->pending
[rw
]));
465 static void end_io_acct(struct dm_io
*io
)
467 struct mapped_device
*md
= io
->md
;
468 struct bio
*bio
= io
->bio
;
469 unsigned long duration
= jiffies
- io
->start_time
;
471 int rw
= bio_data_dir(bio
);
473 cpu
= part_stat_lock();
474 part_round_stats(cpu
, &dm_disk(md
)->part0
);
475 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
479 * After this is decremented the bio must not be touched if it is
482 pending
= atomic_dec_return(&md
->pending
[rw
]);
483 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
484 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
486 /* nudge anyone waiting on suspend queue */
492 * Add the bio to the list of deferred io.
494 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
498 spin_lock_irqsave(&md
->deferred_lock
, flags
);
499 bio_list_add(&md
->deferred
, bio
);
500 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
501 queue_work(md
->wq
, &md
->work
);
505 * Everyone (including functions in this file), should use this
506 * function to access the md->map field, and make sure they call
507 * dm_table_put() when finished.
509 struct dm_table
*dm_get_live_table(struct mapped_device
*md
)
514 read_lock_irqsave(&md
->map_lock
, flags
);
518 read_unlock_irqrestore(&md
->map_lock
, flags
);
524 * Get the geometry associated with a dm device
526 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
534 * Set the geometry of a device.
536 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
538 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
540 if (geo
->start
> sz
) {
541 DMWARN("Start sector is beyond the geometry limits.");
550 /*-----------------------------------------------------------------
552 * A more elegant soln is in the works that uses the queue
553 * merge fn, unfortunately there are a couple of changes to
554 * the block layer that I want to make for this. So in the
555 * interests of getting something for people to use I give
556 * you this clearly demarcated crap.
557 *---------------------------------------------------------------*/
559 static int __noflush_suspending(struct mapped_device
*md
)
561 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
565 * Decrements the number of outstanding ios that a bio has been
566 * cloned into, completing the original io if necc.
568 static void dec_pending(struct dm_io
*io
, int error
)
573 struct mapped_device
*md
= io
->md
;
575 /* Push-back supersedes any I/O errors */
576 if (unlikely(error
)) {
577 spin_lock_irqsave(&io
->endio_lock
, flags
);
578 if (!(io
->error
> 0 && __noflush_suspending(md
)))
580 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
583 if (atomic_dec_and_test(&io
->io_count
)) {
584 if (io
->error
== DM_ENDIO_REQUEUE
) {
586 * Target requested pushing back the I/O.
588 spin_lock_irqsave(&md
->deferred_lock
, flags
);
589 if (__noflush_suspending(md
))
590 bio_list_add_head(&md
->deferred
, io
->bio
);
592 /* noflush suspend was interrupted. */
594 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
597 io_error
= io
->error
;
602 if (io_error
== DM_ENDIO_REQUEUE
)
605 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_size
) {
607 * Preflush done for flush with data, reissue
610 bio
->bi_rw
&= ~REQ_FLUSH
;
613 /* done with normal IO or empty flush */
614 trace_block_bio_complete(md
->queue
, bio
, io_error
);
615 bio_endio(bio
, io_error
);
620 static void clone_endio(struct bio
*bio
, int error
)
623 struct dm_target_io
*tio
= bio
->bi_private
;
624 struct dm_io
*io
= tio
->io
;
625 struct mapped_device
*md
= tio
->io
->md
;
626 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
628 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
632 r
= endio(tio
->ti
, bio
, error
);
633 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
635 * error and requeue request are handled
639 else if (r
== DM_ENDIO_INCOMPLETE
)
640 /* The target will handle the io */
643 DMWARN("unimplemented target endio return value: %d", r
);
649 dec_pending(io
, error
);
653 * Partial completion handling for request-based dm
655 static void end_clone_bio(struct bio
*clone
, int error
)
657 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
658 struct dm_rq_target_io
*tio
= info
->tio
;
659 struct bio
*bio
= info
->orig
;
660 unsigned int nr_bytes
= info
->orig
->bi_size
;
666 * An error has already been detected on the request.
667 * Once error occurred, just let clone->end_io() handle
673 * Don't notice the error to the upper layer yet.
674 * The error handling decision is made by the target driver,
675 * when the request is completed.
682 * I/O for the bio successfully completed.
683 * Notice the data completion to the upper layer.
687 * bios are processed from the head of the list.
688 * So the completing bio should always be rq->bio.
689 * If it's not, something wrong is happening.
691 if (tio
->orig
->bio
!= bio
)
692 DMERR("bio completion is going in the middle of the request");
695 * Update the original request.
696 * Do not use blk_end_request() here, because it may complete
697 * the original request before the clone, and break the ordering.
699 blk_update_request(tio
->orig
, 0, nr_bytes
);
703 * Don't touch any member of the md after calling this function because
704 * the md may be freed in dm_put() at the end of this function.
705 * Or do dm_get() before calling this function and dm_put() later.
707 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
709 atomic_dec(&md
->pending
[rw
]);
711 /* nudge anyone waiting on suspend queue */
712 if (!md_in_flight(md
))
716 * Run this off this callpath, as drivers could invoke end_io while
717 * inside their request_fn (and holding the queue lock). Calling
718 * back into ->request_fn() could deadlock attempting to grab the
722 blk_run_queue_async(md
->queue
);
725 * dm_put() must be at the end of this function. See the comment above
730 static void free_rq_clone(struct request
*clone
)
732 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
734 blk_rq_unprep_clone(clone
);
739 * Complete the clone and the original request.
740 * Must be called without queue lock.
742 static void dm_end_request(struct request
*clone
, int error
)
744 int rw
= rq_data_dir(clone
);
745 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
746 struct mapped_device
*md
= tio
->md
;
747 struct request
*rq
= tio
->orig
;
749 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
750 rq
->errors
= clone
->errors
;
751 rq
->resid_len
= clone
->resid_len
;
755 * We are using the sense buffer of the original
757 * So setting the length of the sense data is enough.
759 rq
->sense_len
= clone
->sense_len
;
762 free_rq_clone(clone
);
763 blk_end_request_all(rq
, error
);
764 rq_completed(md
, rw
, true);
767 static void dm_unprep_request(struct request
*rq
)
769 struct request
*clone
= rq
->special
;
772 rq
->cmd_flags
&= ~REQ_DONTPREP
;
774 free_rq_clone(clone
);
778 * Requeue the original request of a clone.
780 void dm_requeue_unmapped_request(struct request
*clone
)
782 int rw
= rq_data_dir(clone
);
783 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
784 struct mapped_device
*md
= tio
->md
;
785 struct request
*rq
= tio
->orig
;
786 struct request_queue
*q
= rq
->q
;
789 dm_unprep_request(rq
);
791 spin_lock_irqsave(q
->queue_lock
, flags
);
792 blk_requeue_request(q
, rq
);
793 spin_unlock_irqrestore(q
->queue_lock
, flags
);
795 rq_completed(md
, rw
, 0);
797 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
799 static void __stop_queue(struct request_queue
*q
)
804 static void stop_queue(struct request_queue
*q
)
808 spin_lock_irqsave(q
->queue_lock
, flags
);
810 spin_unlock_irqrestore(q
->queue_lock
, flags
);
813 static void __start_queue(struct request_queue
*q
)
815 if (blk_queue_stopped(q
))
819 static void start_queue(struct request_queue
*q
)
823 spin_lock_irqsave(q
->queue_lock
, flags
);
825 spin_unlock_irqrestore(q
->queue_lock
, flags
);
828 static void dm_done(struct request
*clone
, int error
, bool mapped
)
831 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
832 dm_request_endio_fn rq_end_io
= NULL
;
835 rq_end_io
= tio
->ti
->type
->rq_end_io
;
837 if (mapped
&& rq_end_io
)
838 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
842 /* The target wants to complete the I/O */
843 dm_end_request(clone
, r
);
844 else if (r
== DM_ENDIO_INCOMPLETE
)
845 /* The target will handle the I/O */
847 else if (r
== DM_ENDIO_REQUEUE
)
848 /* The target wants to requeue the I/O */
849 dm_requeue_unmapped_request(clone
);
851 DMWARN("unimplemented target endio return value: %d", r
);
857 * Request completion handler for request-based dm
859 static void dm_softirq_done(struct request
*rq
)
862 struct request
*clone
= rq
->completion_data
;
863 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
865 if (rq
->cmd_flags
& REQ_FAILED
)
868 dm_done(clone
, tio
->error
, mapped
);
872 * Complete the clone and the original request with the error status
873 * through softirq context.
875 static void dm_complete_request(struct request
*clone
, int error
)
877 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
878 struct request
*rq
= tio
->orig
;
881 rq
->completion_data
= clone
;
882 blk_complete_request(rq
);
886 * Complete the not-mapped clone and the original request with the error status
887 * through softirq context.
888 * Target's rq_end_io() function isn't called.
889 * This may be used when the target's map_rq() function fails.
891 void dm_kill_unmapped_request(struct request
*clone
, int error
)
893 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
894 struct request
*rq
= tio
->orig
;
896 rq
->cmd_flags
|= REQ_FAILED
;
897 dm_complete_request(clone
, error
);
899 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
902 * Called with the queue lock held
904 static void end_clone_request(struct request
*clone
, int error
)
907 * For just cleaning up the information of the queue in which
908 * the clone was dispatched.
909 * The clone is *NOT* freed actually here because it is alloced from
910 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
912 __blk_put_request(clone
->q
, clone
);
915 * Actual request completion is done in a softirq context which doesn't
916 * hold the queue lock. Otherwise, deadlock could occur because:
917 * - another request may be submitted by the upper level driver
918 * of the stacking during the completion
919 * - the submission which requires queue lock may be done
922 dm_complete_request(clone
, error
);
926 * Return maximum size of I/O possible at the supplied sector up to the current
929 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
931 sector_t target_offset
= dm_target_offset(ti
, sector
);
933 return ti
->len
- target_offset
;
936 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
938 sector_t len
= max_io_len_target_boundary(sector
, ti
);
939 sector_t offset
, max_len
;
942 * Does the target need to split even further?
944 if (ti
->max_io_len
) {
945 offset
= dm_target_offset(ti
, sector
);
946 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
947 max_len
= sector_div(offset
, ti
->max_io_len
);
949 max_len
= offset
& (ti
->max_io_len
- 1);
950 max_len
= ti
->max_io_len
- max_len
;
959 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
961 if (len
> UINT_MAX
) {
962 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
963 (unsigned long long)len
, UINT_MAX
);
964 ti
->error
= "Maximum size of target IO is too large";
968 ti
->max_io_len
= (uint32_t) len
;
972 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
974 static void __map_bio(struct dm_target_io
*tio
)
978 struct mapped_device
*md
;
979 struct bio
*clone
= &tio
->clone
;
980 struct dm_target
*ti
= tio
->ti
;
982 clone
->bi_end_io
= clone_endio
;
983 clone
->bi_private
= tio
;
986 * Map the clone. If r == 0 we don't need to do
987 * anything, the target has assumed ownership of
990 atomic_inc(&tio
->io
->io_count
);
991 sector
= clone
->bi_sector
;
992 r
= ti
->type
->map(ti
, clone
);
993 if (r
== DM_MAPIO_REMAPPED
) {
994 /* the bio has been remapped so dispatch it */
996 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
997 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
999 generic_make_request(clone
);
1000 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1001 /* error the io and bail out, or requeue it if needed */
1003 dec_pending(tio
->io
, r
);
1006 DMWARN("unimplemented target map return value: %d", r
);
1012 struct mapped_device
*md
;
1013 struct dm_table
*map
;
1017 sector_t sector_count
;
1021 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, sector_t len
)
1023 bio
->bi_sector
= sector
;
1024 bio
->bi_size
= to_bytes(len
);
1027 static void bio_setup_bv(struct bio
*bio
, unsigned short idx
, unsigned short bv_count
)
1030 bio
->bi_vcnt
= idx
+ bv_count
;
1031 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1034 static void clone_bio_integrity(struct bio
*bio
, struct bio
*clone
,
1035 unsigned short idx
, unsigned len
, unsigned offset
,
1038 if (!bio_integrity(bio
))
1041 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1044 bio_integrity_trim(clone
, bio_sector_offset(bio
, idx
, offset
), len
);
1048 * Creates a little bio that just does part of a bvec.
1050 static void clone_split_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1051 sector_t sector
, unsigned short idx
,
1052 unsigned offset
, unsigned len
)
1054 struct bio
*clone
= &tio
->clone
;
1055 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1057 *clone
->bi_io_vec
= *bv
;
1059 bio_setup_sector(clone
, sector
, len
);
1061 clone
->bi_bdev
= bio
->bi_bdev
;
1062 clone
->bi_rw
= bio
->bi_rw
;
1064 clone
->bi_io_vec
->bv_offset
= offset
;
1065 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1066 clone
->bi_flags
|= 1 << BIO_CLONED
;
1068 clone_bio_integrity(bio
, clone
, idx
, len
, offset
, 1);
1072 * Creates a bio that consists of range of complete bvecs.
1074 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1075 sector_t sector
, unsigned short idx
,
1076 unsigned short bv_count
, unsigned len
)
1078 struct bio
*clone
= &tio
->clone
;
1081 __bio_clone(clone
, bio
);
1082 bio_setup_sector(clone
, sector
, len
);
1083 bio_setup_bv(clone
, idx
, bv_count
);
1085 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1087 clone_bio_integrity(bio
, clone
, idx
, len
, 0, trim
);
1090 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1091 struct dm_target
*ti
, int nr_iovecs
,
1092 unsigned target_bio_nr
)
1094 struct dm_target_io
*tio
;
1097 clone
= bio_alloc_bioset(GFP_NOIO
, nr_iovecs
, ci
->md
->bs
);
1098 tio
= container_of(clone
, struct dm_target_io
, clone
);
1102 memset(&tio
->info
, 0, sizeof(tio
->info
));
1103 tio
->target_bio_nr
= target_bio_nr
;
1108 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1109 struct dm_target
*ti
,
1110 unsigned target_bio_nr
, sector_t len
)
1112 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, ci
->bio
->bi_max_vecs
, target_bio_nr
);
1113 struct bio
*clone
= &tio
->clone
;
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.
1120 __bio_clone(clone
, ci
->bio
);
1122 bio_setup_sector(clone
, ci
->sector
, len
);
1127 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1128 unsigned num_bios
, sector_t len
)
1130 unsigned target_bio_nr
;
1132 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1133 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1136 static int __send_empty_flush(struct clone_info
*ci
)
1138 unsigned target_nr
= 0;
1139 struct dm_target
*ti
;
1141 BUG_ON(bio_has_data(ci
->bio
));
1142 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1143 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, 0);
1148 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1149 sector_t sector
, int nr_iovecs
,
1150 unsigned short idx
, unsigned short bv_count
,
1151 unsigned offset
, unsigned len
,
1152 unsigned split_bvec
)
1154 struct bio
*bio
= ci
->bio
;
1155 struct dm_target_io
*tio
;
1156 unsigned target_bio_nr
;
1157 unsigned num_target_bios
= 1;
1160 * Does the target want to receive duplicate copies of the bio?
1162 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1163 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1165 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1166 tio
= alloc_tio(ci
, ti
, nr_iovecs
, target_bio_nr
);
1168 clone_split_bio(tio
, bio
, sector
, idx
, offset
, len
);
1170 clone_bio(tio
, bio
, sector
, idx
, bv_count
, len
);
1175 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1177 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1179 return ti
->num_discard_bios
;
1182 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1184 return ti
->num_write_same_bios
;
1187 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1189 static bool is_split_required_for_discard(struct dm_target
*ti
)
1191 return ti
->split_discard_bios
;
1194 static int __send_changing_extent_only(struct clone_info
*ci
,
1195 get_num_bios_fn get_num_bios
,
1196 is_split_required_fn is_split_required
)
1198 struct dm_target
*ti
;
1203 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1204 if (!dm_target_is_valid(ti
))
1208 * Even though the device advertised support for this type of
1209 * request, that does not mean every target supports it, and
1210 * reconfiguration might also have changed that since the
1211 * check was performed.
1213 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1217 if (is_split_required
&& !is_split_required(ti
))
1218 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1220 len
= min(ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1222 __send_duplicate_bios(ci
, ti
, num_bios
, len
);
1225 } while (ci
->sector_count
-= len
);
1230 static int __send_discard(struct clone_info
*ci
)
1232 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1233 is_split_required_for_discard
);
1236 static int __send_write_same(struct clone_info
*ci
)
1238 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1242 * Find maximum number of sectors / bvecs we can process with a single bio.
1244 static sector_t
__len_within_target(struct clone_info
*ci
, sector_t max
, int *idx
)
1246 struct bio
*bio
= ci
->bio
;
1247 sector_t bv_len
, total_len
= 0;
1249 for (*idx
= ci
->idx
; max
&& (*idx
< bio
->bi_vcnt
); (*idx
)++) {
1250 bv_len
= to_sector(bio
->bi_io_vec
[*idx
].bv_len
);
1256 total_len
+= bv_len
;
1262 static int __split_bvec_across_targets(struct clone_info
*ci
,
1263 struct dm_target
*ti
, sector_t max
)
1265 struct bio
*bio
= ci
->bio
;
1266 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1267 sector_t remaining
= to_sector(bv
->bv_len
);
1268 unsigned offset
= 0;
1273 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1274 if (!dm_target_is_valid(ti
))
1277 max
= max_io_len(ci
->sector
, ti
);
1280 len
= min(remaining
, max
);
1282 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, 1, ci
->idx
, 0,
1283 bv
->bv_offset
+ offset
, len
, 1);
1286 ci
->sector_count
-= len
;
1287 offset
+= to_bytes(len
);
1288 } while (remaining
-= len
);
1296 * Select the correct strategy for processing a non-flush bio.
1298 static int __split_and_process_non_flush(struct clone_info
*ci
)
1300 struct bio
*bio
= ci
->bio
;
1301 struct dm_target
*ti
;
1305 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1306 return __send_discard(ci
);
1307 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1308 return __send_write_same(ci
);
1310 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1311 if (!dm_target_is_valid(ti
))
1314 max
= max_io_len(ci
->sector
, ti
);
1317 * Optimise for the simple case where we can do all of
1318 * the remaining io with a single clone.
1320 if (ci
->sector_count
<= max
) {
1321 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, bio
->bi_max_vecs
,
1322 ci
->idx
, bio
->bi_vcnt
- ci
->idx
, 0,
1323 ci
->sector_count
, 0);
1324 ci
->sector_count
= 0;
1329 * There are some bvecs that don't span targets.
1330 * Do as many of these as possible.
1332 if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1333 len
= __len_within_target(ci
, max
, &idx
);
1335 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, bio
->bi_max_vecs
,
1336 ci
->idx
, idx
- ci
->idx
, 0, len
, 0);
1339 ci
->sector_count
-= len
;
1346 * Handle a bvec that must be split between two or more targets.
1348 return __split_bvec_across_targets(ci
, ti
, max
);
1352 * Entry point to split a bio into clones and submit them to the targets.
1354 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1356 struct clone_info ci
;
1359 ci
.map
= dm_get_live_table(md
);
1360 if (unlikely(!ci
.map
)) {
1366 ci
.io
= alloc_io(md
);
1368 atomic_set(&ci
.io
->io_count
, 1);
1371 spin_lock_init(&ci
.io
->endio_lock
);
1372 ci
.sector
= bio
->bi_sector
;
1373 ci
.idx
= bio
->bi_idx
;
1375 start_io_acct(ci
.io
);
1377 if (bio
->bi_rw
& REQ_FLUSH
) {
1378 ci
.bio
= &ci
.md
->flush_bio
;
1379 ci
.sector_count
= 0;
1380 error
= __send_empty_flush(&ci
);
1381 /* dec_pending submits any data associated with flush */
1384 ci
.sector_count
= bio_sectors(bio
);
1385 while (ci
.sector_count
&& !error
)
1386 error
= __split_and_process_non_flush(&ci
);
1389 /* drop the extra reference count */
1390 dec_pending(ci
.io
, error
);
1391 dm_table_put(ci
.map
);
1393 /*-----------------------------------------------------------------
1395 *---------------------------------------------------------------*/
1397 static int dm_merge_bvec(struct request_queue
*q
,
1398 struct bvec_merge_data
*bvm
,
1399 struct bio_vec
*biovec
)
1401 struct mapped_device
*md
= q
->queuedata
;
1402 struct dm_table
*map
= dm_get_live_table(md
);
1403 struct dm_target
*ti
;
1404 sector_t max_sectors
;
1410 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1411 if (!dm_target_is_valid(ti
))
1415 * Find maximum amount of I/O that won't need splitting
1417 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1418 (sector_t
) BIO_MAX_SECTORS
);
1419 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1424 * merge_bvec_fn() returns number of bytes
1425 * it can accept at this offset
1426 * max is precomputed maximal io size
1428 if (max_size
&& ti
->type
->merge
)
1429 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1431 * If the target doesn't support merge method and some of the devices
1432 * provided their merge_bvec method (we know this by looking at
1433 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1434 * entries. So always set max_size to 0, and the code below allows
1437 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1446 * Always allow an entire first page
1448 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1449 max_size
= biovec
->bv_len
;
1455 * The request function that just remaps the bio built up by
1458 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1460 int rw
= bio_data_dir(bio
);
1461 struct mapped_device
*md
= q
->queuedata
;
1464 down_read(&md
->io_lock
);
1466 cpu
= part_stat_lock();
1467 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1468 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1471 /* if we're suspended, we have to queue this io for later */
1472 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1473 up_read(&md
->io_lock
);
1475 if (bio_rw(bio
) != READA
)
1482 __split_and_process_bio(md
, bio
);
1483 up_read(&md
->io_lock
);
1487 static int dm_request_based(struct mapped_device
*md
)
1489 return blk_queue_stackable(md
->queue
);
1492 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1494 struct mapped_device
*md
= q
->queuedata
;
1496 if (dm_request_based(md
))
1497 blk_queue_bio(q
, bio
);
1499 _dm_request(q
, bio
);
1502 void dm_dispatch_request(struct request
*rq
)
1506 if (blk_queue_io_stat(rq
->q
))
1507 rq
->cmd_flags
|= REQ_IO_STAT
;
1509 rq
->start_time
= jiffies
;
1510 r
= blk_insert_cloned_request(rq
->q
, rq
);
1512 dm_complete_request(rq
, r
);
1514 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1516 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1519 struct dm_rq_target_io
*tio
= data
;
1520 struct dm_rq_clone_bio_info
*info
=
1521 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1523 info
->orig
= bio_orig
;
1525 bio
->bi_end_io
= end_clone_bio
;
1526 bio
->bi_private
= info
;
1531 static int setup_clone(struct request
*clone
, struct request
*rq
,
1532 struct dm_rq_target_io
*tio
)
1536 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1537 dm_rq_bio_constructor
, tio
);
1541 clone
->cmd
= rq
->cmd
;
1542 clone
->cmd_len
= rq
->cmd_len
;
1543 clone
->sense
= rq
->sense
;
1544 clone
->buffer
= rq
->buffer
;
1545 clone
->end_io
= end_clone_request
;
1546 clone
->end_io_data
= tio
;
1551 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1554 struct request
*clone
;
1555 struct dm_rq_target_io
*tio
;
1557 tio
= alloc_rq_tio(md
, gfp_mask
);
1565 memset(&tio
->info
, 0, sizeof(tio
->info
));
1567 clone
= &tio
->clone
;
1568 if (setup_clone(clone
, rq
, tio
)) {
1578 * Called with the queue lock held.
1580 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1582 struct mapped_device
*md
= q
->queuedata
;
1583 struct request
*clone
;
1585 if (unlikely(rq
->special
)) {
1586 DMWARN("Already has something in rq->special.");
1587 return BLKPREP_KILL
;
1590 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1592 return BLKPREP_DEFER
;
1594 rq
->special
= clone
;
1595 rq
->cmd_flags
|= REQ_DONTPREP
;
1602 * 0 : the request has been processed (not requeued)
1603 * !0 : the request has been requeued
1605 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1606 struct mapped_device
*md
)
1608 int r
, requeued
= 0;
1609 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1612 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1614 case DM_MAPIO_SUBMITTED
:
1615 /* The target has taken the I/O to submit by itself later */
1617 case DM_MAPIO_REMAPPED
:
1618 /* The target has remapped the I/O so dispatch it */
1619 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1620 blk_rq_pos(tio
->orig
));
1621 dm_dispatch_request(clone
);
1623 case DM_MAPIO_REQUEUE
:
1624 /* The target wants to requeue the I/O */
1625 dm_requeue_unmapped_request(clone
);
1630 DMWARN("unimplemented target map return value: %d", r
);
1634 /* The target wants to complete the I/O */
1635 dm_kill_unmapped_request(clone
, r
);
1642 static struct request
*dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1644 struct request
*clone
;
1646 blk_start_request(orig
);
1647 clone
= orig
->special
;
1648 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1651 * Hold the md reference here for the in-flight I/O.
1652 * We can't rely on the reference count by device opener,
1653 * because the device may be closed during the request completion
1654 * when all bios are completed.
1655 * See the comment in rq_completed() too.
1663 * q->request_fn for request-based dm.
1664 * Called with the queue lock held.
1666 static void dm_request_fn(struct request_queue
*q
)
1668 struct mapped_device
*md
= q
->queuedata
;
1669 struct dm_table
*map
= dm_get_live_table(md
);
1670 struct dm_target
*ti
;
1671 struct request
*rq
, *clone
;
1675 * For suspend, check blk_queue_stopped() and increment
1676 * ->pending within a single queue_lock not to increment the
1677 * number of in-flight I/Os after the queue is stopped in
1680 while (!blk_queue_stopped(q
)) {
1681 rq
= blk_peek_request(q
);
1685 /* always use block 0 to find the target for flushes for now */
1687 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1688 pos
= blk_rq_pos(rq
);
1690 ti
= dm_table_find_target(map
, pos
);
1691 if (!dm_target_is_valid(ti
)) {
1693 * Must perform setup, that dm_done() requires,
1694 * before calling dm_kill_unmapped_request
1696 DMERR_LIMIT("request attempted access beyond the end of device");
1697 clone
= dm_start_request(md
, rq
);
1698 dm_kill_unmapped_request(clone
, -EIO
);
1702 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1705 clone
= dm_start_request(md
, rq
);
1707 spin_unlock(q
->queue_lock
);
1708 if (map_request(ti
, clone
, md
))
1711 BUG_ON(!irqs_disabled());
1712 spin_lock(q
->queue_lock
);
1718 BUG_ON(!irqs_disabled());
1719 spin_lock(q
->queue_lock
);
1722 blk_delay_queue(q
, HZ
/ 10);
1727 int dm_underlying_device_busy(struct request_queue
*q
)
1729 return blk_lld_busy(q
);
1731 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1733 static int dm_lld_busy(struct request_queue
*q
)
1736 struct mapped_device
*md
= q
->queuedata
;
1737 struct dm_table
*map
= dm_get_live_table(md
);
1739 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1742 r
= dm_table_any_busy_target(map
);
1749 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1752 struct mapped_device
*md
= congested_data
;
1753 struct dm_table
*map
;
1755 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1756 map
= dm_get_live_table(md
);
1759 * Request-based dm cares about only own queue for
1760 * the query about congestion status of request_queue
1762 if (dm_request_based(md
))
1763 r
= md
->queue
->backing_dev_info
.state
&
1766 r
= dm_table_any_congested(map
, bdi_bits
);
1775 /*-----------------------------------------------------------------
1776 * An IDR is used to keep track of allocated minor numbers.
1777 *---------------------------------------------------------------*/
1778 static void free_minor(int minor
)
1780 spin_lock(&_minor_lock
);
1781 idr_remove(&_minor_idr
, minor
);
1782 spin_unlock(&_minor_lock
);
1786 * See if the device with a specific minor # is free.
1788 static int specific_minor(int minor
)
1792 if (minor
>= (1 << MINORBITS
))
1795 idr_preload(GFP_KERNEL
);
1796 spin_lock(&_minor_lock
);
1798 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1800 spin_unlock(&_minor_lock
);
1803 return r
== -ENOSPC
? -EBUSY
: r
;
1807 static int next_free_minor(int *minor
)
1811 idr_preload(GFP_KERNEL
);
1812 spin_lock(&_minor_lock
);
1814 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1816 spin_unlock(&_minor_lock
);
1824 static const struct block_device_operations dm_blk_dops
;
1826 static void dm_wq_work(struct work_struct
*work
);
1828 static void dm_init_md_queue(struct mapped_device
*md
)
1831 * Request-based dm devices cannot be stacked on top of bio-based dm
1832 * devices. The type of this dm device has not been decided yet.
1833 * The type is decided at the first table loading time.
1834 * To prevent problematic device stacking, clear the queue flag
1835 * for request stacking support until then.
1837 * This queue is new, so no concurrency on the queue_flags.
1839 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1841 md
->queue
->queuedata
= md
;
1842 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1843 md
->queue
->backing_dev_info
.congested_data
= md
;
1844 blk_queue_make_request(md
->queue
, dm_request
);
1845 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1846 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1850 * Allocate and initialise a blank device with a given minor.
1852 static struct mapped_device
*alloc_dev(int minor
)
1855 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1859 DMWARN("unable to allocate device, out of memory.");
1863 if (!try_module_get(THIS_MODULE
))
1864 goto bad_module_get
;
1866 /* get a minor number for the dev */
1867 if (minor
== DM_ANY_MINOR
)
1868 r
= next_free_minor(&minor
);
1870 r
= specific_minor(minor
);
1874 md
->type
= DM_TYPE_NONE
;
1875 init_rwsem(&md
->io_lock
);
1876 mutex_init(&md
->suspend_lock
);
1877 mutex_init(&md
->type_lock
);
1878 spin_lock_init(&md
->deferred_lock
);
1879 rwlock_init(&md
->map_lock
);
1880 atomic_set(&md
->holders
, 1);
1881 atomic_set(&md
->open_count
, 0);
1882 atomic_set(&md
->event_nr
, 0);
1883 atomic_set(&md
->uevent_seq
, 0);
1884 INIT_LIST_HEAD(&md
->uevent_list
);
1885 spin_lock_init(&md
->uevent_lock
);
1887 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1891 dm_init_md_queue(md
);
1893 md
->disk
= alloc_disk(1);
1897 atomic_set(&md
->pending
[0], 0);
1898 atomic_set(&md
->pending
[1], 0);
1899 init_waitqueue_head(&md
->wait
);
1900 INIT_WORK(&md
->work
, dm_wq_work
);
1901 init_waitqueue_head(&md
->eventq
);
1903 md
->disk
->major
= _major
;
1904 md
->disk
->first_minor
= minor
;
1905 md
->disk
->fops
= &dm_blk_dops
;
1906 md
->disk
->queue
= md
->queue
;
1907 md
->disk
->private_data
= md
;
1908 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1910 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1912 md
->wq
= alloc_workqueue("kdmflush",
1913 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1917 md
->bdev
= bdget_disk(md
->disk
, 0);
1921 bio_init(&md
->flush_bio
);
1922 md
->flush_bio
.bi_bdev
= md
->bdev
;
1923 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1925 /* Populate the mapping, nobody knows we exist yet */
1926 spin_lock(&_minor_lock
);
1927 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1928 spin_unlock(&_minor_lock
);
1930 BUG_ON(old_md
!= MINOR_ALLOCED
);
1935 destroy_workqueue(md
->wq
);
1937 del_gendisk(md
->disk
);
1940 blk_cleanup_queue(md
->queue
);
1944 module_put(THIS_MODULE
);
1950 static void unlock_fs(struct mapped_device
*md
);
1952 static void free_dev(struct mapped_device
*md
)
1954 int minor
= MINOR(disk_devt(md
->disk
));
1958 destroy_workqueue(md
->wq
);
1960 mempool_destroy(md
->io_pool
);
1962 bioset_free(md
->bs
);
1963 blk_integrity_unregister(md
->disk
);
1964 del_gendisk(md
->disk
);
1967 spin_lock(&_minor_lock
);
1968 md
->disk
->private_data
= NULL
;
1969 spin_unlock(&_minor_lock
);
1972 blk_cleanup_queue(md
->queue
);
1973 module_put(THIS_MODULE
);
1977 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1979 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1981 if (md
->io_pool
&& md
->bs
) {
1982 /* The md already has necessary mempools. */
1983 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
1985 * Reload bioset because front_pad may have changed
1986 * because a different table was loaded.
1988 bioset_free(md
->bs
);
1991 } else if (dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
) {
1993 * There's no need to reload with request-based dm
1994 * because the size of front_pad doesn't change.
1995 * Note for future: If you are to reload bioset,
1996 * prep-ed requests in the queue may refer
1997 * to bio from the old bioset, so you must walk
1998 * through the queue to unprep.
2004 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
2006 md
->io_pool
= p
->io_pool
;
2012 /* mempool bind completed, now no need any mempools in the table */
2013 dm_table_free_md_mempools(t
);
2017 * Bind a table to the device.
2019 static void event_callback(void *context
)
2021 unsigned long flags
;
2023 struct mapped_device
*md
= (struct mapped_device
*) context
;
2025 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2026 list_splice_init(&md
->uevent_list
, &uevents
);
2027 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2029 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2031 atomic_inc(&md
->event_nr
);
2032 wake_up(&md
->eventq
);
2036 * Protected by md->suspend_lock obtained by dm_swap_table().
2038 static void __set_size(struct mapped_device
*md
, sector_t size
)
2040 set_capacity(md
->disk
, size
);
2042 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2046 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2048 * If this function returns 0, then the device is either a non-dm
2049 * device without a merge_bvec_fn, or it is a dm device that is
2050 * able to split any bios it receives that are too big.
2052 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2054 struct mapped_device
*dev_md
;
2056 if (!q
->merge_bvec_fn
)
2059 if (q
->make_request_fn
== dm_request
) {
2060 dev_md
= q
->queuedata
;
2061 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2068 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2069 struct dm_dev
*dev
, sector_t start
,
2070 sector_t len
, void *data
)
2072 struct block_device
*bdev
= dev
->bdev
;
2073 struct request_queue
*q
= bdev_get_queue(bdev
);
2075 return dm_queue_merge_is_compulsory(q
);
2079 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2080 * on the properties of the underlying devices.
2082 static int dm_table_merge_is_optional(struct dm_table
*table
)
2085 struct dm_target
*ti
;
2087 while (i
< dm_table_get_num_targets(table
)) {
2088 ti
= dm_table_get_target(table
, i
++);
2090 if (ti
->type
->iterate_devices
&&
2091 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2099 * Returns old map, which caller must destroy.
2101 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2102 struct queue_limits
*limits
)
2104 struct dm_table
*old_map
;
2105 struct request_queue
*q
= md
->queue
;
2107 unsigned long flags
;
2108 int merge_is_optional
;
2110 size
= dm_table_get_size(t
);
2113 * Wipe any geometry if the size of the table changed.
2115 if (size
!= get_capacity(md
->disk
))
2116 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2118 __set_size(md
, size
);
2120 dm_table_event_callback(t
, event_callback
, md
);
2123 * The queue hasn't been stopped yet, if the old table type wasn't
2124 * for request-based during suspension. So stop it to prevent
2125 * I/O mapping before resume.
2126 * This must be done before setting the queue restrictions,
2127 * because request-based dm may be run just after the setting.
2129 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2132 __bind_mempools(md
, t
);
2134 merge_is_optional
= dm_table_merge_is_optional(t
);
2136 write_lock_irqsave(&md
->map_lock
, flags
);
2139 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2141 dm_table_set_restrictions(t
, q
, limits
);
2142 if (merge_is_optional
)
2143 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2145 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2146 write_unlock_irqrestore(&md
->map_lock
, flags
);
2152 * Returns unbound table for the caller to free.
2154 static struct dm_table
*__unbind(struct mapped_device
*md
)
2156 struct dm_table
*map
= md
->map
;
2157 unsigned long flags
;
2162 dm_table_event_callback(map
, NULL
, NULL
);
2163 write_lock_irqsave(&md
->map_lock
, flags
);
2165 write_unlock_irqrestore(&md
->map_lock
, flags
);
2171 * Constructor for a new device.
2173 int dm_create(int minor
, struct mapped_device
**result
)
2175 struct mapped_device
*md
;
2177 md
= alloc_dev(minor
);
2188 * Functions to manage md->type.
2189 * All are required to hold md->type_lock.
2191 void dm_lock_md_type(struct mapped_device
*md
)
2193 mutex_lock(&md
->type_lock
);
2196 void dm_unlock_md_type(struct mapped_device
*md
)
2198 mutex_unlock(&md
->type_lock
);
2201 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2206 unsigned dm_get_md_type(struct mapped_device
*md
)
2211 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2213 return md
->immutable_target_type
;
2217 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2219 static int dm_init_request_based_queue(struct mapped_device
*md
)
2221 struct request_queue
*q
= NULL
;
2223 if (md
->queue
->elevator
)
2226 /* Fully initialize the queue */
2227 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2232 dm_init_md_queue(md
);
2233 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2234 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2235 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2237 elv_register_queue(md
->queue
);
2243 * Setup the DM device's queue based on md's type
2245 int dm_setup_md_queue(struct mapped_device
*md
)
2247 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2248 !dm_init_request_based_queue(md
)) {
2249 DMWARN("Cannot initialize queue for request-based mapped device");
2256 static struct mapped_device
*dm_find_md(dev_t dev
)
2258 struct mapped_device
*md
;
2259 unsigned minor
= MINOR(dev
);
2261 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2264 spin_lock(&_minor_lock
);
2266 md
= idr_find(&_minor_idr
, minor
);
2267 if (md
&& (md
== MINOR_ALLOCED
||
2268 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2269 dm_deleting_md(md
) ||
2270 test_bit(DMF_FREEING
, &md
->flags
))) {
2276 spin_unlock(&_minor_lock
);
2281 struct mapped_device
*dm_get_md(dev_t dev
)
2283 struct mapped_device
*md
= dm_find_md(dev
);
2290 EXPORT_SYMBOL_GPL(dm_get_md
);
2292 void *dm_get_mdptr(struct mapped_device
*md
)
2294 return md
->interface_ptr
;
2297 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2299 md
->interface_ptr
= ptr
;
2302 void dm_get(struct mapped_device
*md
)
2304 atomic_inc(&md
->holders
);
2305 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2308 const char *dm_device_name(struct mapped_device
*md
)
2312 EXPORT_SYMBOL_GPL(dm_device_name
);
2314 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2316 struct dm_table
*map
;
2320 spin_lock(&_minor_lock
);
2321 map
= dm_get_live_table(md
);
2322 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2323 set_bit(DMF_FREEING
, &md
->flags
);
2324 spin_unlock(&_minor_lock
);
2326 if (!dm_suspended_md(md
)) {
2327 dm_table_presuspend_targets(map
);
2328 dm_table_postsuspend_targets(map
);
2332 * Rare, but there may be I/O requests still going to complete,
2333 * for example. Wait for all references to disappear.
2334 * No one should increment the reference count of the mapped_device,
2335 * after the mapped_device state becomes DMF_FREEING.
2338 while (atomic_read(&md
->holders
))
2340 else if (atomic_read(&md
->holders
))
2341 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2342 dm_device_name(md
), atomic_read(&md
->holders
));
2346 dm_table_destroy(__unbind(md
));
2350 void dm_destroy(struct mapped_device
*md
)
2352 __dm_destroy(md
, true);
2355 void dm_destroy_immediate(struct mapped_device
*md
)
2357 __dm_destroy(md
, false);
2360 void dm_put(struct mapped_device
*md
)
2362 atomic_dec(&md
->holders
);
2364 EXPORT_SYMBOL_GPL(dm_put
);
2366 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2369 DECLARE_WAITQUEUE(wait
, current
);
2371 add_wait_queue(&md
->wait
, &wait
);
2374 set_current_state(interruptible
);
2376 if (!md_in_flight(md
))
2379 if (interruptible
== TASK_INTERRUPTIBLE
&&
2380 signal_pending(current
)) {
2387 set_current_state(TASK_RUNNING
);
2389 remove_wait_queue(&md
->wait
, &wait
);
2395 * Process the deferred bios
2397 static void dm_wq_work(struct work_struct
*work
)
2399 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2403 down_read(&md
->io_lock
);
2405 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2406 spin_lock_irq(&md
->deferred_lock
);
2407 c
= bio_list_pop(&md
->deferred
);
2408 spin_unlock_irq(&md
->deferred_lock
);
2413 up_read(&md
->io_lock
);
2415 if (dm_request_based(md
))
2416 generic_make_request(c
);
2418 __split_and_process_bio(md
, c
);
2420 down_read(&md
->io_lock
);
2423 up_read(&md
->io_lock
);
2426 static void dm_queue_flush(struct mapped_device
*md
)
2428 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2429 smp_mb__after_clear_bit();
2430 queue_work(md
->wq
, &md
->work
);
2434 * Swap in a new table, returning the old one for the caller to destroy.
2436 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2438 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2439 struct queue_limits limits
;
2442 mutex_lock(&md
->suspend_lock
);
2444 /* device must be suspended */
2445 if (!dm_suspended_md(md
))
2449 * If the new table has no data devices, retain the existing limits.
2450 * This helps multipath with queue_if_no_path if all paths disappear,
2451 * then new I/O is queued based on these limits, and then some paths
2454 if (dm_table_has_no_data_devices(table
)) {
2455 live_map
= dm_get_live_table(md
);
2457 limits
= md
->queue
->limits
;
2458 dm_table_put(live_map
);
2462 r
= dm_calculate_queue_limits(table
, &limits
);
2469 map
= __bind(md
, table
, &limits
);
2472 mutex_unlock(&md
->suspend_lock
);
2477 * Functions to lock and unlock any filesystem running on the
2480 static int lock_fs(struct mapped_device
*md
)
2484 WARN_ON(md
->frozen_sb
);
2486 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2487 if (IS_ERR(md
->frozen_sb
)) {
2488 r
= PTR_ERR(md
->frozen_sb
);
2489 md
->frozen_sb
= NULL
;
2493 set_bit(DMF_FROZEN
, &md
->flags
);
2498 static void unlock_fs(struct mapped_device
*md
)
2500 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2503 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2504 md
->frozen_sb
= NULL
;
2505 clear_bit(DMF_FROZEN
, &md
->flags
);
2509 * We need to be able to change a mapping table under a mounted
2510 * filesystem. For example we might want to move some data in
2511 * the background. Before the table can be swapped with
2512 * dm_bind_table, dm_suspend must be called to flush any in
2513 * flight bios and ensure that any further io gets deferred.
2516 * Suspend mechanism in request-based dm.
2518 * 1. Flush all I/Os by lock_fs() if needed.
2519 * 2. Stop dispatching any I/O by stopping the request_queue.
2520 * 3. Wait for all in-flight I/Os to be completed or requeued.
2522 * To abort suspend, start the request_queue.
2524 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2526 struct dm_table
*map
= NULL
;
2528 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2529 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2531 mutex_lock(&md
->suspend_lock
);
2533 if (dm_suspended_md(md
)) {
2538 map
= dm_get_live_table(md
);
2541 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2542 * This flag is cleared before dm_suspend returns.
2545 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2547 /* This does not get reverted if there's an error later. */
2548 dm_table_presuspend_targets(map
);
2551 * Flush I/O to the device.
2552 * Any I/O submitted after lock_fs() may not be flushed.
2553 * noflush takes precedence over do_lockfs.
2554 * (lock_fs() flushes I/Os and waits for them to complete.)
2556 if (!noflush
&& do_lockfs
) {
2563 * Here we must make sure that no processes are submitting requests
2564 * to target drivers i.e. no one may be executing
2565 * __split_and_process_bio. This is called from dm_request and
2568 * To get all processes out of __split_and_process_bio in dm_request,
2569 * we take the write lock. To prevent any process from reentering
2570 * __split_and_process_bio from dm_request and quiesce the thread
2571 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2572 * flush_workqueue(md->wq).
2574 down_write(&md
->io_lock
);
2575 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2576 up_write(&md
->io_lock
);
2579 * Stop md->queue before flushing md->wq in case request-based
2580 * dm defers requests to md->wq from md->queue.
2582 if (dm_request_based(md
))
2583 stop_queue(md
->queue
);
2585 flush_workqueue(md
->wq
);
2588 * At this point no more requests are entering target request routines.
2589 * We call dm_wait_for_completion to wait for all existing requests
2592 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2594 down_write(&md
->io_lock
);
2596 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2597 up_write(&md
->io_lock
);
2599 /* were we interrupted ? */
2603 if (dm_request_based(md
))
2604 start_queue(md
->queue
);
2607 goto out
; /* pushback list is already flushed, so skip flush */
2611 * If dm_wait_for_completion returned 0, the device is completely
2612 * quiescent now. There is no request-processing activity. All new
2613 * requests are being added to md->deferred list.
2616 set_bit(DMF_SUSPENDED
, &md
->flags
);
2618 dm_table_postsuspend_targets(map
);
2624 mutex_unlock(&md
->suspend_lock
);
2628 int dm_resume(struct mapped_device
*md
)
2631 struct dm_table
*map
= NULL
;
2633 mutex_lock(&md
->suspend_lock
);
2634 if (!dm_suspended_md(md
))
2637 map
= dm_get_live_table(md
);
2638 if (!map
|| !dm_table_get_size(map
))
2641 r
= dm_table_resume_targets(map
);
2648 * Flushing deferred I/Os must be done after targets are resumed
2649 * so that mapping of targets can work correctly.
2650 * Request-based dm is queueing the deferred I/Os in its request_queue.
2652 if (dm_request_based(md
))
2653 start_queue(md
->queue
);
2657 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2662 mutex_unlock(&md
->suspend_lock
);
2667 /*-----------------------------------------------------------------
2668 * Event notification.
2669 *---------------------------------------------------------------*/
2670 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2673 char udev_cookie
[DM_COOKIE_LENGTH
];
2674 char *envp
[] = { udev_cookie
, NULL
};
2677 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2679 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2680 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2681 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2686 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2688 return atomic_add_return(1, &md
->uevent_seq
);
2691 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2693 return atomic_read(&md
->event_nr
);
2696 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2698 return wait_event_interruptible(md
->eventq
,
2699 (event_nr
!= atomic_read(&md
->event_nr
)));
2702 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2704 unsigned long flags
;
2706 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2707 list_add(elist
, &md
->uevent_list
);
2708 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2712 * The gendisk is only valid as long as you have a reference
2715 struct gendisk
*dm_disk(struct mapped_device
*md
)
2720 struct kobject
*dm_kobject(struct mapped_device
*md
)
2726 * struct mapped_device should not be exported outside of dm.c
2727 * so use this check to verify that kobj is part of md structure
2729 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2731 struct mapped_device
*md
;
2733 md
= container_of(kobj
, struct mapped_device
, kobj
);
2734 if (&md
->kobj
!= kobj
)
2737 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2745 int dm_suspended_md(struct mapped_device
*md
)
2747 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2750 int dm_suspended(struct dm_target
*ti
)
2752 return dm_suspended_md(dm_table_get_md(ti
->table
));
2754 EXPORT_SYMBOL_GPL(dm_suspended
);
2756 int dm_noflush_suspending(struct dm_target
*ti
)
2758 return __noflush_suspending(dm_table_get_md(ti
->table
));
2760 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2762 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
2764 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
2765 struct kmem_cache
*cachep
;
2766 unsigned int pool_size
;
2767 unsigned int front_pad
;
2772 if (type
== DM_TYPE_BIO_BASED
) {
2775 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2776 } else if (type
== DM_TYPE_REQUEST_BASED
) {
2777 cachep
= _rq_tio_cache
;
2778 pool_size
= MIN_IOS
;
2779 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2780 /* per_bio_data_size is not used. See __bind_mempools(). */
2781 WARN_ON(per_bio_data_size
!= 0);
2785 pools
->io_pool
= mempool_create_slab_pool(MIN_IOS
, cachep
);
2786 if (!pools
->io_pool
)
2789 pools
->bs
= bioset_create(pool_size
, front_pad
);
2793 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2799 dm_free_md_mempools(pools
);
2804 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2810 mempool_destroy(pools
->io_pool
);
2813 bioset_free(pools
->bs
);
2818 static const struct block_device_operations dm_blk_dops
= {
2819 .open
= dm_blk_open
,
2820 .release
= dm_blk_close
,
2821 .ioctl
= dm_blk_ioctl
,
2822 .getgeo
= dm_blk_getgeo
,
2823 .owner
= THIS_MODULE
2826 EXPORT_SYMBOL(dm_get_mapinfo
);
2831 module_init(dm_init
);
2832 module_exit(dm_exit
);
2834 module_param(major
, uint
, 0);
2835 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2836 MODULE_DESCRIPTION(DM_NAME
" driver");
2837 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2838 MODULE_LICENSE("GPL");