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 bio_endio(bio
, io_error
);
619 static void clone_endio(struct bio
*bio
, int error
)
622 struct dm_target_io
*tio
= bio
->bi_private
;
623 struct dm_io
*io
= tio
->io
;
624 struct mapped_device
*md
= tio
->io
->md
;
625 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
627 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
631 r
= endio(tio
->ti
, bio
, error
);
632 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
634 * error and requeue request are handled
638 else if (r
== DM_ENDIO_INCOMPLETE
)
639 /* The target will handle the io */
642 DMWARN("unimplemented target endio return value: %d", r
);
648 dec_pending(io
, error
);
652 * Partial completion handling for request-based dm
654 static void end_clone_bio(struct bio
*clone
, int error
)
656 struct dm_rq_clone_bio_info
*info
= clone
->bi_private
;
657 struct dm_rq_target_io
*tio
= info
->tio
;
658 struct bio
*bio
= info
->orig
;
659 unsigned int nr_bytes
= info
->orig
->bi_size
;
665 * An error has already been detected on the request.
666 * Once error occurred, just let clone->end_io() handle
672 * Don't notice the error to the upper layer yet.
673 * The error handling decision is made by the target driver,
674 * when the request is completed.
681 * I/O for the bio successfully completed.
682 * Notice the data completion to the upper layer.
686 * bios are processed from the head of the list.
687 * So the completing bio should always be rq->bio.
688 * If it's not, something wrong is happening.
690 if (tio
->orig
->bio
!= bio
)
691 DMERR("bio completion is going in the middle of the request");
694 * Update the original request.
695 * Do not use blk_end_request() here, because it may complete
696 * the original request before the clone, and break the ordering.
698 blk_update_request(tio
->orig
, 0, nr_bytes
);
702 * Don't touch any member of the md after calling this function because
703 * the md may be freed in dm_put() at the end of this function.
704 * Or do dm_get() before calling this function and dm_put() later.
706 static void rq_completed(struct mapped_device
*md
, int rw
, int run_queue
)
708 atomic_dec(&md
->pending
[rw
]);
710 /* nudge anyone waiting on suspend queue */
711 if (!md_in_flight(md
))
715 * Run this off this callpath, as drivers could invoke end_io while
716 * inside their request_fn (and holding the queue lock). Calling
717 * back into ->request_fn() could deadlock attempting to grab the
721 blk_run_queue_async(md
->queue
);
724 * dm_put() must be at the end of this function. See the comment above
729 static void free_rq_clone(struct request
*clone
)
731 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
733 blk_rq_unprep_clone(clone
);
738 * Complete the clone and the original request.
739 * Must be called without queue lock.
741 static void dm_end_request(struct request
*clone
, int error
)
743 int rw
= rq_data_dir(clone
);
744 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
745 struct mapped_device
*md
= tio
->md
;
746 struct request
*rq
= tio
->orig
;
748 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
749 rq
->errors
= clone
->errors
;
750 rq
->resid_len
= clone
->resid_len
;
754 * We are using the sense buffer of the original
756 * So setting the length of the sense data is enough.
758 rq
->sense_len
= clone
->sense_len
;
761 free_rq_clone(clone
);
762 blk_end_request_all(rq
, error
);
763 rq_completed(md
, rw
, true);
766 static void dm_unprep_request(struct request
*rq
)
768 struct request
*clone
= rq
->special
;
771 rq
->cmd_flags
&= ~REQ_DONTPREP
;
773 free_rq_clone(clone
);
777 * Requeue the original request of a clone.
779 void dm_requeue_unmapped_request(struct request
*clone
)
781 int rw
= rq_data_dir(clone
);
782 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
783 struct mapped_device
*md
= tio
->md
;
784 struct request
*rq
= tio
->orig
;
785 struct request_queue
*q
= rq
->q
;
788 dm_unprep_request(rq
);
790 spin_lock_irqsave(q
->queue_lock
, flags
);
791 blk_requeue_request(q
, rq
);
792 spin_unlock_irqrestore(q
->queue_lock
, flags
);
794 rq_completed(md
, rw
, 0);
796 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request
);
798 static void __stop_queue(struct request_queue
*q
)
803 static void stop_queue(struct request_queue
*q
)
807 spin_lock_irqsave(q
->queue_lock
, flags
);
809 spin_unlock_irqrestore(q
->queue_lock
, flags
);
812 static void __start_queue(struct request_queue
*q
)
814 if (blk_queue_stopped(q
))
818 static void start_queue(struct request_queue
*q
)
822 spin_lock_irqsave(q
->queue_lock
, flags
);
824 spin_unlock_irqrestore(q
->queue_lock
, flags
);
827 static void dm_done(struct request
*clone
, int error
, bool mapped
)
830 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
831 dm_request_endio_fn rq_end_io
= NULL
;
834 rq_end_io
= tio
->ti
->type
->rq_end_io
;
836 if (mapped
&& rq_end_io
)
837 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
841 /* The target wants to complete the I/O */
842 dm_end_request(clone
, r
);
843 else if (r
== DM_ENDIO_INCOMPLETE
)
844 /* The target will handle the I/O */
846 else if (r
== DM_ENDIO_REQUEUE
)
847 /* The target wants to requeue the I/O */
848 dm_requeue_unmapped_request(clone
);
850 DMWARN("unimplemented target endio return value: %d", r
);
856 * Request completion handler for request-based dm
858 static void dm_softirq_done(struct request
*rq
)
861 struct request
*clone
= rq
->completion_data
;
862 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
864 if (rq
->cmd_flags
& REQ_FAILED
)
867 dm_done(clone
, tio
->error
, mapped
);
871 * Complete the clone and the original request with the error status
872 * through softirq context.
874 static void dm_complete_request(struct request
*clone
, int error
)
876 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
877 struct request
*rq
= tio
->orig
;
880 rq
->completion_data
= clone
;
881 blk_complete_request(rq
);
885 * Complete the not-mapped clone and the original request with the error status
886 * through softirq context.
887 * Target's rq_end_io() function isn't called.
888 * This may be used when the target's map_rq() function fails.
890 void dm_kill_unmapped_request(struct request
*clone
, int error
)
892 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
893 struct request
*rq
= tio
->orig
;
895 rq
->cmd_flags
|= REQ_FAILED
;
896 dm_complete_request(clone
, error
);
898 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request
);
901 * Called with the queue lock held
903 static void end_clone_request(struct request
*clone
, int error
)
906 * For just cleaning up the information of the queue in which
907 * the clone was dispatched.
908 * The clone is *NOT* freed actually here because it is alloced from
909 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
911 __blk_put_request(clone
->q
, clone
);
914 * Actual request completion is done in a softirq context which doesn't
915 * hold the queue lock. Otherwise, deadlock could occur because:
916 * - another request may be submitted by the upper level driver
917 * of the stacking during the completion
918 * - the submission which requires queue lock may be done
921 dm_complete_request(clone
, error
);
925 * Return maximum size of I/O possible at the supplied sector up to the current
928 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
930 sector_t target_offset
= dm_target_offset(ti
, sector
);
932 return ti
->len
- target_offset
;
935 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
937 sector_t len
= max_io_len_target_boundary(sector
, ti
);
938 sector_t offset
, max_len
;
941 * Does the target need to split even further?
943 if (ti
->max_io_len
) {
944 offset
= dm_target_offset(ti
, sector
);
945 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
946 max_len
= sector_div(offset
, ti
->max_io_len
);
948 max_len
= offset
& (ti
->max_io_len
- 1);
949 max_len
= ti
->max_io_len
- max_len
;
958 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
960 if (len
> UINT_MAX
) {
961 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
962 (unsigned long long)len
, UINT_MAX
);
963 ti
->error
= "Maximum size of target IO is too large";
967 ti
->max_io_len
= (uint32_t) len
;
971 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
973 static void __map_bio(struct dm_target_io
*tio
)
977 struct mapped_device
*md
;
978 struct bio
*clone
= &tio
->clone
;
979 struct dm_target
*ti
= tio
->ti
;
981 clone
->bi_end_io
= clone_endio
;
982 clone
->bi_private
= tio
;
985 * Map the clone. If r == 0 we don't need to do
986 * anything, the target has assumed ownership of
989 atomic_inc(&tio
->io
->io_count
);
990 sector
= clone
->bi_sector
;
991 r
= ti
->type
->map(ti
, clone
);
992 if (r
== DM_MAPIO_REMAPPED
) {
993 /* the bio has been remapped so dispatch it */
995 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
996 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
998 generic_make_request(clone
);
999 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1000 /* error the io and bail out, or requeue it if needed */
1002 dec_pending(tio
->io
, r
);
1005 DMWARN("unimplemented target map return value: %d", r
);
1011 struct mapped_device
*md
;
1012 struct dm_table
*map
;
1016 sector_t sector_count
;
1020 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, sector_t len
)
1022 bio
->bi_sector
= sector
;
1023 bio
->bi_size
= to_bytes(len
);
1026 static void bio_setup_bv(struct bio
*bio
, unsigned short idx
, unsigned short bv_count
)
1029 bio
->bi_vcnt
= idx
+ bv_count
;
1030 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
1033 static void clone_bio_integrity(struct bio
*bio
, struct bio
*clone
,
1034 unsigned short idx
, unsigned len
, unsigned offset
,
1037 if (!bio_integrity(bio
))
1040 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1043 bio_integrity_trim(clone
, bio_sector_offset(bio
, idx
, offset
), len
);
1047 * Creates a little bio that just does part of a bvec.
1049 static void clone_split_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1050 sector_t sector
, unsigned short idx
,
1051 unsigned offset
, unsigned len
)
1053 struct bio
*clone
= &tio
->clone
;
1054 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
1056 *clone
->bi_io_vec
= *bv
;
1058 bio_setup_sector(clone
, sector
, len
);
1060 clone
->bi_bdev
= bio
->bi_bdev
;
1061 clone
->bi_rw
= bio
->bi_rw
;
1063 clone
->bi_io_vec
->bv_offset
= offset
;
1064 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
1065 clone
->bi_flags
|= 1 << BIO_CLONED
;
1067 clone_bio_integrity(bio
, clone
, idx
, len
, offset
, 1);
1071 * Creates a bio that consists of range of complete bvecs.
1073 static void clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1074 sector_t sector
, unsigned short idx
,
1075 unsigned short bv_count
, unsigned len
)
1077 struct bio
*clone
= &tio
->clone
;
1080 __bio_clone(clone
, bio
);
1081 bio_setup_sector(clone
, sector
, len
);
1082 bio_setup_bv(clone
, idx
, bv_count
);
1084 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
1086 clone_bio_integrity(bio
, clone
, idx
, len
, 0, trim
);
1089 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1090 struct dm_target
*ti
, int nr_iovecs
,
1091 unsigned target_bio_nr
)
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_bio_nr
= target_bio_nr
;
1107 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1108 struct dm_target
*ti
,
1109 unsigned target_bio_nr
, sector_t len
)
1111 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, ci
->bio
->bi_max_vecs
, target_bio_nr
);
1112 struct bio
*clone
= &tio
->clone
;
1115 * Discard requests require the bio's inline iovecs be initialized.
1116 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1117 * and discard, so no need for concern about wasted bvec allocations.
1119 __bio_clone(clone
, ci
->bio
);
1121 bio_setup_sector(clone
, ci
->sector
, len
);
1126 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1127 unsigned num_bios
, sector_t len
)
1129 unsigned target_bio_nr
;
1131 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1132 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1135 static int __send_empty_flush(struct clone_info
*ci
)
1137 unsigned target_nr
= 0;
1138 struct dm_target
*ti
;
1140 BUG_ON(bio_has_data(ci
->bio
));
1141 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1142 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, 0);
1147 static void __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1148 sector_t sector
, int nr_iovecs
,
1149 unsigned short idx
, unsigned short bv_count
,
1150 unsigned offset
, unsigned len
,
1151 unsigned split_bvec
)
1153 struct bio
*bio
= ci
->bio
;
1154 struct dm_target_io
*tio
;
1155 unsigned target_bio_nr
;
1156 unsigned num_target_bios
= 1;
1159 * Does the target want to receive duplicate copies of the bio?
1161 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1162 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1164 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1165 tio
= alloc_tio(ci
, ti
, nr_iovecs
, target_bio_nr
);
1167 clone_split_bio(tio
, bio
, sector
, idx
, offset
, len
);
1169 clone_bio(tio
, bio
, sector
, idx
, bv_count
, len
);
1174 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1176 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1178 return ti
->num_discard_bios
;
1181 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1183 return ti
->num_write_same_bios
;
1186 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1188 static bool is_split_required_for_discard(struct dm_target
*ti
)
1190 return ti
->split_discard_bios
;
1193 static int __send_changing_extent_only(struct clone_info
*ci
,
1194 get_num_bios_fn get_num_bios
,
1195 is_split_required_fn is_split_required
)
1197 struct dm_target
*ti
;
1202 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1203 if (!dm_target_is_valid(ti
))
1207 * Even though the device advertised support for this type of
1208 * request, that does not mean every target supports it, and
1209 * reconfiguration might also have changed that since the
1210 * check was performed.
1212 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1216 if (is_split_required
&& !is_split_required(ti
))
1217 len
= min(ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1219 len
= min(ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1221 __send_duplicate_bios(ci
, ti
, num_bios
, len
);
1224 } while (ci
->sector_count
-= len
);
1229 static int __send_discard(struct clone_info
*ci
)
1231 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1232 is_split_required_for_discard
);
1235 static int __send_write_same(struct clone_info
*ci
)
1237 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1241 * Find maximum number of sectors / bvecs we can process with a single bio.
1243 static sector_t
__len_within_target(struct clone_info
*ci
, sector_t max
, int *idx
)
1245 struct bio
*bio
= ci
->bio
;
1246 sector_t bv_len
, total_len
= 0;
1248 for (*idx
= ci
->idx
; max
&& (*idx
< bio
->bi_vcnt
); (*idx
)++) {
1249 bv_len
= to_sector(bio
->bi_io_vec
[*idx
].bv_len
);
1255 total_len
+= bv_len
;
1261 static int __split_bvec_across_targets(struct clone_info
*ci
,
1262 struct dm_target
*ti
, sector_t max
)
1264 struct bio
*bio
= ci
->bio
;
1265 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
1266 sector_t remaining
= to_sector(bv
->bv_len
);
1267 unsigned offset
= 0;
1272 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1273 if (!dm_target_is_valid(ti
))
1276 max
= max_io_len(ci
->sector
, ti
);
1279 len
= min(remaining
, max
);
1281 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, 1, ci
->idx
, 0,
1282 bv
->bv_offset
+ offset
, len
, 1);
1285 ci
->sector_count
-= len
;
1286 offset
+= to_bytes(len
);
1287 } while (remaining
-= len
);
1295 * Select the correct strategy for processing a non-flush bio.
1297 static int __split_and_process_non_flush(struct clone_info
*ci
)
1299 struct bio
*bio
= ci
->bio
;
1300 struct dm_target
*ti
;
1304 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1305 return __send_discard(ci
);
1306 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1307 return __send_write_same(ci
);
1309 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1310 if (!dm_target_is_valid(ti
))
1313 max
= max_io_len(ci
->sector
, ti
);
1316 * Optimise for the simple case where we can do all of
1317 * the remaining io with a single clone.
1319 if (ci
->sector_count
<= max
) {
1320 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, bio
->bi_max_vecs
,
1321 ci
->idx
, bio
->bi_vcnt
- ci
->idx
, 0,
1322 ci
->sector_count
, 0);
1323 ci
->sector_count
= 0;
1328 * There are some bvecs that don't span targets.
1329 * Do as many of these as possible.
1331 if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
1332 len
= __len_within_target(ci
, max
, &idx
);
1334 __clone_and_map_data_bio(ci
, ti
, ci
->sector
, bio
->bi_max_vecs
,
1335 ci
->idx
, idx
- ci
->idx
, 0, len
, 0);
1338 ci
->sector_count
-= len
;
1345 * Handle a bvec that must be split between two or more targets.
1347 return __split_bvec_across_targets(ci
, ti
, max
);
1351 * Entry point to split a bio into clones and submit them to the targets.
1353 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
1355 struct clone_info ci
;
1358 ci
.map
= dm_get_live_table(md
);
1359 if (unlikely(!ci
.map
)) {
1365 ci
.io
= alloc_io(md
);
1367 atomic_set(&ci
.io
->io_count
, 1);
1370 spin_lock_init(&ci
.io
->endio_lock
);
1371 ci
.sector
= bio
->bi_sector
;
1372 ci
.idx
= bio
->bi_idx
;
1374 start_io_acct(ci
.io
);
1376 if (bio
->bi_rw
& REQ_FLUSH
) {
1377 ci
.bio
= &ci
.md
->flush_bio
;
1378 ci
.sector_count
= 0;
1379 error
= __send_empty_flush(&ci
);
1380 /* dec_pending submits any data associated with flush */
1383 ci
.sector_count
= bio_sectors(bio
);
1384 while (ci
.sector_count
&& !error
)
1385 error
= __split_and_process_non_flush(&ci
);
1388 /* drop the extra reference count */
1389 dec_pending(ci
.io
, error
);
1390 dm_table_put(ci
.map
);
1392 /*-----------------------------------------------------------------
1394 *---------------------------------------------------------------*/
1396 static int dm_merge_bvec(struct request_queue
*q
,
1397 struct bvec_merge_data
*bvm
,
1398 struct bio_vec
*biovec
)
1400 struct mapped_device
*md
= q
->queuedata
;
1401 struct dm_table
*map
= dm_get_live_table(md
);
1402 struct dm_target
*ti
;
1403 sector_t max_sectors
;
1409 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
1410 if (!dm_target_is_valid(ti
))
1414 * Find maximum amount of I/O that won't need splitting
1416 max_sectors
= min(max_io_len(bvm
->bi_sector
, ti
),
1417 (sector_t
) BIO_MAX_SECTORS
);
1418 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
1423 * merge_bvec_fn() returns number of bytes
1424 * it can accept at this offset
1425 * max is precomputed maximal io size
1427 if (max_size
&& ti
->type
->merge
)
1428 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
1430 * If the target doesn't support merge method and some of the devices
1431 * provided their merge_bvec method (we know this by looking at
1432 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1433 * entries. So always set max_size to 0, and the code below allows
1436 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
1445 * Always allow an entire first page
1447 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
1448 max_size
= biovec
->bv_len
;
1454 * The request function that just remaps the bio built up by
1457 static void _dm_request(struct request_queue
*q
, struct bio
*bio
)
1459 int rw
= bio_data_dir(bio
);
1460 struct mapped_device
*md
= q
->queuedata
;
1463 down_read(&md
->io_lock
);
1465 cpu
= part_stat_lock();
1466 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
1467 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
1470 /* if we're suspended, we have to queue this io for later */
1471 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1472 up_read(&md
->io_lock
);
1474 if (bio_rw(bio
) != READA
)
1481 __split_and_process_bio(md
, bio
);
1482 up_read(&md
->io_lock
);
1486 static int dm_request_based(struct mapped_device
*md
)
1488 return blk_queue_stackable(md
->queue
);
1491 static void dm_request(struct request_queue
*q
, struct bio
*bio
)
1493 struct mapped_device
*md
= q
->queuedata
;
1495 if (dm_request_based(md
))
1496 blk_queue_bio(q
, bio
);
1498 _dm_request(q
, bio
);
1501 void dm_dispatch_request(struct request
*rq
)
1505 if (blk_queue_io_stat(rq
->q
))
1506 rq
->cmd_flags
|= REQ_IO_STAT
;
1508 rq
->start_time
= jiffies
;
1509 r
= blk_insert_cloned_request(rq
->q
, rq
);
1511 dm_complete_request(rq
, r
);
1513 EXPORT_SYMBOL_GPL(dm_dispatch_request
);
1515 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1518 struct dm_rq_target_io
*tio
= data
;
1519 struct dm_rq_clone_bio_info
*info
=
1520 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1522 info
->orig
= bio_orig
;
1524 bio
->bi_end_io
= end_clone_bio
;
1525 bio
->bi_private
= info
;
1530 static int setup_clone(struct request
*clone
, struct request
*rq
,
1531 struct dm_rq_target_io
*tio
)
1535 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, GFP_ATOMIC
,
1536 dm_rq_bio_constructor
, tio
);
1540 clone
->cmd
= rq
->cmd
;
1541 clone
->cmd_len
= rq
->cmd_len
;
1542 clone
->sense
= rq
->sense
;
1543 clone
->buffer
= rq
->buffer
;
1544 clone
->end_io
= end_clone_request
;
1545 clone
->end_io_data
= tio
;
1550 static struct request
*clone_rq(struct request
*rq
, struct mapped_device
*md
,
1553 struct request
*clone
;
1554 struct dm_rq_target_io
*tio
;
1556 tio
= alloc_rq_tio(md
, gfp_mask
);
1564 memset(&tio
->info
, 0, sizeof(tio
->info
));
1566 clone
= &tio
->clone
;
1567 if (setup_clone(clone
, rq
, tio
)) {
1577 * Called with the queue lock held.
1579 static int dm_prep_fn(struct request_queue
*q
, struct request
*rq
)
1581 struct mapped_device
*md
= q
->queuedata
;
1582 struct request
*clone
;
1584 if (unlikely(rq
->special
)) {
1585 DMWARN("Already has something in rq->special.");
1586 return BLKPREP_KILL
;
1589 clone
= clone_rq(rq
, md
, GFP_ATOMIC
);
1591 return BLKPREP_DEFER
;
1593 rq
->special
= clone
;
1594 rq
->cmd_flags
|= REQ_DONTPREP
;
1601 * 0 : the request has been processed (not requeued)
1602 * !0 : the request has been requeued
1604 static int map_request(struct dm_target
*ti
, struct request
*clone
,
1605 struct mapped_device
*md
)
1607 int r
, requeued
= 0;
1608 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1611 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
1613 case DM_MAPIO_SUBMITTED
:
1614 /* The target has taken the I/O to submit by itself later */
1616 case DM_MAPIO_REMAPPED
:
1617 /* The target has remapped the I/O so dispatch it */
1618 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
1619 blk_rq_pos(tio
->orig
));
1620 dm_dispatch_request(clone
);
1622 case DM_MAPIO_REQUEUE
:
1623 /* The target wants to requeue the I/O */
1624 dm_requeue_unmapped_request(clone
);
1629 DMWARN("unimplemented target map return value: %d", r
);
1633 /* The target wants to complete the I/O */
1634 dm_kill_unmapped_request(clone
, r
);
1641 static struct request
*dm_start_request(struct mapped_device
*md
, struct request
*orig
)
1643 struct request
*clone
;
1645 blk_start_request(orig
);
1646 clone
= orig
->special
;
1647 atomic_inc(&md
->pending
[rq_data_dir(clone
)]);
1650 * Hold the md reference here for the in-flight I/O.
1651 * We can't rely on the reference count by device opener,
1652 * because the device may be closed during the request completion
1653 * when all bios are completed.
1654 * See the comment in rq_completed() too.
1662 * q->request_fn for request-based dm.
1663 * Called with the queue lock held.
1665 static void dm_request_fn(struct request_queue
*q
)
1667 struct mapped_device
*md
= q
->queuedata
;
1668 struct dm_table
*map
= dm_get_live_table(md
);
1669 struct dm_target
*ti
;
1670 struct request
*rq
, *clone
;
1674 * For suspend, check blk_queue_stopped() and increment
1675 * ->pending within a single queue_lock not to increment the
1676 * number of in-flight I/Os after the queue is stopped in
1679 while (!blk_queue_stopped(q
)) {
1680 rq
= blk_peek_request(q
);
1684 /* always use block 0 to find the target for flushes for now */
1686 if (!(rq
->cmd_flags
& REQ_FLUSH
))
1687 pos
= blk_rq_pos(rq
);
1689 ti
= dm_table_find_target(map
, pos
);
1690 if (!dm_target_is_valid(ti
)) {
1692 * Must perform setup, that dm_done() requires,
1693 * before calling dm_kill_unmapped_request
1695 DMERR_LIMIT("request attempted access beyond the end of device");
1696 clone
= dm_start_request(md
, rq
);
1697 dm_kill_unmapped_request(clone
, -EIO
);
1701 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
1704 clone
= dm_start_request(md
, rq
);
1706 spin_unlock(q
->queue_lock
);
1707 if (map_request(ti
, clone
, md
))
1710 BUG_ON(!irqs_disabled());
1711 spin_lock(q
->queue_lock
);
1717 BUG_ON(!irqs_disabled());
1718 spin_lock(q
->queue_lock
);
1721 blk_delay_queue(q
, HZ
/ 10);
1726 int dm_underlying_device_busy(struct request_queue
*q
)
1728 return blk_lld_busy(q
);
1730 EXPORT_SYMBOL_GPL(dm_underlying_device_busy
);
1732 static int dm_lld_busy(struct request_queue
*q
)
1735 struct mapped_device
*md
= q
->queuedata
;
1736 struct dm_table
*map
= dm_get_live_table(md
);
1738 if (!map
|| test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))
1741 r
= dm_table_any_busy_target(map
);
1748 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1751 struct mapped_device
*md
= congested_data
;
1752 struct dm_table
*map
;
1754 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1755 map
= dm_get_live_table(md
);
1758 * Request-based dm cares about only own queue for
1759 * the query about congestion status of request_queue
1761 if (dm_request_based(md
))
1762 r
= md
->queue
->backing_dev_info
.state
&
1765 r
= dm_table_any_congested(map
, bdi_bits
);
1774 /*-----------------------------------------------------------------
1775 * An IDR is used to keep track of allocated minor numbers.
1776 *---------------------------------------------------------------*/
1777 static void free_minor(int minor
)
1779 spin_lock(&_minor_lock
);
1780 idr_remove(&_minor_idr
, minor
);
1781 spin_unlock(&_minor_lock
);
1785 * See if the device with a specific minor # is free.
1787 static int specific_minor(int minor
)
1791 if (minor
>= (1 << MINORBITS
))
1794 idr_preload(GFP_KERNEL
);
1795 spin_lock(&_minor_lock
);
1797 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1799 spin_unlock(&_minor_lock
);
1802 return r
== -ENOSPC
? -EBUSY
: r
;
1806 static int next_free_minor(int *minor
)
1810 idr_preload(GFP_KERNEL
);
1811 spin_lock(&_minor_lock
);
1813 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1815 spin_unlock(&_minor_lock
);
1823 static const struct block_device_operations dm_blk_dops
;
1825 static void dm_wq_work(struct work_struct
*work
);
1827 static void dm_init_md_queue(struct mapped_device
*md
)
1830 * Request-based dm devices cannot be stacked on top of bio-based dm
1831 * devices. The type of this dm device has not been decided yet.
1832 * The type is decided at the first table loading time.
1833 * To prevent problematic device stacking, clear the queue flag
1834 * for request stacking support until then.
1836 * This queue is new, so no concurrency on the queue_flags.
1838 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
1840 md
->queue
->queuedata
= md
;
1841 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1842 md
->queue
->backing_dev_info
.congested_data
= md
;
1843 blk_queue_make_request(md
->queue
, dm_request
);
1844 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1845 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1849 * Allocate and initialise a blank device with a given minor.
1851 static struct mapped_device
*alloc_dev(int minor
)
1854 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1858 DMWARN("unable to allocate device, out of memory.");
1862 if (!try_module_get(THIS_MODULE
))
1863 goto bad_module_get
;
1865 /* get a minor number for the dev */
1866 if (minor
== DM_ANY_MINOR
)
1867 r
= next_free_minor(&minor
);
1869 r
= specific_minor(minor
);
1873 md
->type
= DM_TYPE_NONE
;
1874 init_rwsem(&md
->io_lock
);
1875 mutex_init(&md
->suspend_lock
);
1876 mutex_init(&md
->type_lock
);
1877 spin_lock_init(&md
->deferred_lock
);
1878 rwlock_init(&md
->map_lock
);
1879 atomic_set(&md
->holders
, 1);
1880 atomic_set(&md
->open_count
, 0);
1881 atomic_set(&md
->event_nr
, 0);
1882 atomic_set(&md
->uevent_seq
, 0);
1883 INIT_LIST_HEAD(&md
->uevent_list
);
1884 spin_lock_init(&md
->uevent_lock
);
1886 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1890 dm_init_md_queue(md
);
1892 md
->disk
= alloc_disk(1);
1896 atomic_set(&md
->pending
[0], 0);
1897 atomic_set(&md
->pending
[1], 0);
1898 init_waitqueue_head(&md
->wait
);
1899 INIT_WORK(&md
->work
, dm_wq_work
);
1900 init_waitqueue_head(&md
->eventq
);
1902 md
->disk
->major
= _major
;
1903 md
->disk
->first_minor
= minor
;
1904 md
->disk
->fops
= &dm_blk_dops
;
1905 md
->disk
->queue
= md
->queue
;
1906 md
->disk
->private_data
= md
;
1907 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1909 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1911 md
->wq
= alloc_workqueue("kdmflush",
1912 WQ_NON_REENTRANT
| WQ_MEM_RECLAIM
, 0);
1916 md
->bdev
= bdget_disk(md
->disk
, 0);
1920 bio_init(&md
->flush_bio
);
1921 md
->flush_bio
.bi_bdev
= md
->bdev
;
1922 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
1924 /* Populate the mapping, nobody knows we exist yet */
1925 spin_lock(&_minor_lock
);
1926 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1927 spin_unlock(&_minor_lock
);
1929 BUG_ON(old_md
!= MINOR_ALLOCED
);
1934 destroy_workqueue(md
->wq
);
1936 del_gendisk(md
->disk
);
1939 blk_cleanup_queue(md
->queue
);
1943 module_put(THIS_MODULE
);
1949 static void unlock_fs(struct mapped_device
*md
);
1951 static void free_dev(struct mapped_device
*md
)
1953 int minor
= MINOR(disk_devt(md
->disk
));
1957 destroy_workqueue(md
->wq
);
1959 mempool_destroy(md
->io_pool
);
1961 bioset_free(md
->bs
);
1962 blk_integrity_unregister(md
->disk
);
1963 del_gendisk(md
->disk
);
1966 spin_lock(&_minor_lock
);
1967 md
->disk
->private_data
= NULL
;
1968 spin_unlock(&_minor_lock
);
1971 blk_cleanup_queue(md
->queue
);
1972 module_put(THIS_MODULE
);
1976 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1978 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1980 if (md
->io_pool
&& md
->bs
) {
1981 /* The md already has necessary mempools. */
1982 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
1984 * Reload bioset because front_pad may have changed
1985 * because a different table was loaded.
1987 bioset_free(md
->bs
);
1990 } else if (dm_table_get_type(t
) == DM_TYPE_REQUEST_BASED
) {
1992 * There's no need to reload with request-based dm
1993 * because the size of front_pad doesn't change.
1994 * Note for future: If you are to reload bioset,
1995 * prep-ed requests in the queue may refer
1996 * to bio from the old bioset, so you must walk
1997 * through the queue to unprep.
2003 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
2005 md
->io_pool
= p
->io_pool
;
2011 /* mempool bind completed, now no need any mempools in the table */
2012 dm_table_free_md_mempools(t
);
2016 * Bind a table to the device.
2018 static void event_callback(void *context
)
2020 unsigned long flags
;
2022 struct mapped_device
*md
= (struct mapped_device
*) context
;
2024 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2025 list_splice_init(&md
->uevent_list
, &uevents
);
2026 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2028 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2030 atomic_inc(&md
->event_nr
);
2031 wake_up(&md
->eventq
);
2035 * Protected by md->suspend_lock obtained by dm_swap_table().
2037 static void __set_size(struct mapped_device
*md
, sector_t size
)
2039 set_capacity(md
->disk
, size
);
2041 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2045 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2047 * If this function returns 0, then the device is either a non-dm
2048 * device without a merge_bvec_fn, or it is a dm device that is
2049 * able to split any bios it receives that are too big.
2051 int dm_queue_merge_is_compulsory(struct request_queue
*q
)
2053 struct mapped_device
*dev_md
;
2055 if (!q
->merge_bvec_fn
)
2058 if (q
->make_request_fn
== dm_request
) {
2059 dev_md
= q
->queuedata
;
2060 if (test_bit(DMF_MERGE_IS_OPTIONAL
, &dev_md
->flags
))
2067 static int dm_device_merge_is_compulsory(struct dm_target
*ti
,
2068 struct dm_dev
*dev
, sector_t start
,
2069 sector_t len
, void *data
)
2071 struct block_device
*bdev
= dev
->bdev
;
2072 struct request_queue
*q
= bdev_get_queue(bdev
);
2074 return dm_queue_merge_is_compulsory(q
);
2078 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2079 * on the properties of the underlying devices.
2081 static int dm_table_merge_is_optional(struct dm_table
*table
)
2084 struct dm_target
*ti
;
2086 while (i
< dm_table_get_num_targets(table
)) {
2087 ti
= dm_table_get_target(table
, i
++);
2089 if (ti
->type
->iterate_devices
&&
2090 ti
->type
->iterate_devices(ti
, dm_device_merge_is_compulsory
, NULL
))
2098 * Returns old map, which caller must destroy.
2100 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2101 struct queue_limits
*limits
)
2103 struct dm_table
*old_map
;
2104 struct request_queue
*q
= md
->queue
;
2106 unsigned long flags
;
2107 int merge_is_optional
;
2109 size
= dm_table_get_size(t
);
2112 * Wipe any geometry if the size of the table changed.
2114 if (size
!= get_capacity(md
->disk
))
2115 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2117 __set_size(md
, size
);
2119 dm_table_event_callback(t
, event_callback
, md
);
2122 * The queue hasn't been stopped yet, if the old table type wasn't
2123 * for request-based during suspension. So stop it to prevent
2124 * I/O mapping before resume.
2125 * This must be done before setting the queue restrictions,
2126 * because request-based dm may be run just after the setting.
2128 if (dm_table_request_based(t
) && !blk_queue_stopped(q
))
2131 __bind_mempools(md
, t
);
2133 merge_is_optional
= dm_table_merge_is_optional(t
);
2135 write_lock_irqsave(&md
->map_lock
, flags
);
2138 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2140 dm_table_set_restrictions(t
, q
, limits
);
2141 if (merge_is_optional
)
2142 set_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2144 clear_bit(DMF_MERGE_IS_OPTIONAL
, &md
->flags
);
2145 write_unlock_irqrestore(&md
->map_lock
, flags
);
2151 * Returns unbound table for the caller to free.
2153 static struct dm_table
*__unbind(struct mapped_device
*md
)
2155 struct dm_table
*map
= md
->map
;
2156 unsigned long flags
;
2161 dm_table_event_callback(map
, NULL
, NULL
);
2162 write_lock_irqsave(&md
->map_lock
, flags
);
2164 write_unlock_irqrestore(&md
->map_lock
, flags
);
2170 * Constructor for a new device.
2172 int dm_create(int minor
, struct mapped_device
**result
)
2174 struct mapped_device
*md
;
2176 md
= alloc_dev(minor
);
2187 * Functions to manage md->type.
2188 * All are required to hold md->type_lock.
2190 void dm_lock_md_type(struct mapped_device
*md
)
2192 mutex_lock(&md
->type_lock
);
2195 void dm_unlock_md_type(struct mapped_device
*md
)
2197 mutex_unlock(&md
->type_lock
);
2200 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2205 unsigned dm_get_md_type(struct mapped_device
*md
)
2210 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2212 return md
->immutable_target_type
;
2216 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2218 static int dm_init_request_based_queue(struct mapped_device
*md
)
2220 struct request_queue
*q
= NULL
;
2222 if (md
->queue
->elevator
)
2225 /* Fully initialize the queue */
2226 q
= blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
);
2231 dm_init_md_queue(md
);
2232 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2233 blk_queue_prep_rq(md
->queue
, dm_prep_fn
);
2234 blk_queue_lld_busy(md
->queue
, dm_lld_busy
);
2236 elv_register_queue(md
->queue
);
2242 * Setup the DM device's queue based on md's type
2244 int dm_setup_md_queue(struct mapped_device
*md
)
2246 if ((dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) &&
2247 !dm_init_request_based_queue(md
)) {
2248 DMWARN("Cannot initialize queue for request-based mapped device");
2255 static struct mapped_device
*dm_find_md(dev_t dev
)
2257 struct mapped_device
*md
;
2258 unsigned minor
= MINOR(dev
);
2260 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2263 spin_lock(&_minor_lock
);
2265 md
= idr_find(&_minor_idr
, minor
);
2266 if (md
&& (md
== MINOR_ALLOCED
||
2267 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2268 dm_deleting_md(md
) ||
2269 test_bit(DMF_FREEING
, &md
->flags
))) {
2275 spin_unlock(&_minor_lock
);
2280 struct mapped_device
*dm_get_md(dev_t dev
)
2282 struct mapped_device
*md
= dm_find_md(dev
);
2289 EXPORT_SYMBOL_GPL(dm_get_md
);
2291 void *dm_get_mdptr(struct mapped_device
*md
)
2293 return md
->interface_ptr
;
2296 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2298 md
->interface_ptr
= ptr
;
2301 void dm_get(struct mapped_device
*md
)
2303 atomic_inc(&md
->holders
);
2304 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2307 const char *dm_device_name(struct mapped_device
*md
)
2311 EXPORT_SYMBOL_GPL(dm_device_name
);
2313 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2315 struct dm_table
*map
;
2319 spin_lock(&_minor_lock
);
2320 map
= dm_get_live_table(md
);
2321 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2322 set_bit(DMF_FREEING
, &md
->flags
);
2323 spin_unlock(&_minor_lock
);
2325 if (!dm_suspended_md(md
)) {
2326 dm_table_presuspend_targets(map
);
2327 dm_table_postsuspend_targets(map
);
2331 * Rare, but there may be I/O requests still going to complete,
2332 * for example. Wait for all references to disappear.
2333 * No one should increment the reference count of the mapped_device,
2334 * after the mapped_device state becomes DMF_FREEING.
2337 while (atomic_read(&md
->holders
))
2339 else if (atomic_read(&md
->holders
))
2340 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2341 dm_device_name(md
), atomic_read(&md
->holders
));
2345 dm_table_destroy(__unbind(md
));
2349 void dm_destroy(struct mapped_device
*md
)
2351 __dm_destroy(md
, true);
2354 void dm_destroy_immediate(struct mapped_device
*md
)
2356 __dm_destroy(md
, false);
2359 void dm_put(struct mapped_device
*md
)
2361 atomic_dec(&md
->holders
);
2363 EXPORT_SYMBOL_GPL(dm_put
);
2365 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2368 DECLARE_WAITQUEUE(wait
, current
);
2370 add_wait_queue(&md
->wait
, &wait
);
2373 set_current_state(interruptible
);
2375 if (!md_in_flight(md
))
2378 if (interruptible
== TASK_INTERRUPTIBLE
&&
2379 signal_pending(current
)) {
2386 set_current_state(TASK_RUNNING
);
2388 remove_wait_queue(&md
->wait
, &wait
);
2394 * Process the deferred bios
2396 static void dm_wq_work(struct work_struct
*work
)
2398 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2402 down_read(&md
->io_lock
);
2404 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2405 spin_lock_irq(&md
->deferred_lock
);
2406 c
= bio_list_pop(&md
->deferred
);
2407 spin_unlock_irq(&md
->deferred_lock
);
2412 up_read(&md
->io_lock
);
2414 if (dm_request_based(md
))
2415 generic_make_request(c
);
2417 __split_and_process_bio(md
, c
);
2419 down_read(&md
->io_lock
);
2422 up_read(&md
->io_lock
);
2425 static void dm_queue_flush(struct mapped_device
*md
)
2427 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2428 smp_mb__after_clear_bit();
2429 queue_work(md
->wq
, &md
->work
);
2433 * Swap in a new table, returning the old one for the caller to destroy.
2435 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2437 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2438 struct queue_limits limits
;
2441 mutex_lock(&md
->suspend_lock
);
2443 /* device must be suspended */
2444 if (!dm_suspended_md(md
))
2448 * If the new table has no data devices, retain the existing limits.
2449 * This helps multipath with queue_if_no_path if all paths disappear,
2450 * then new I/O is queued based on these limits, and then some paths
2453 if (dm_table_has_no_data_devices(table
)) {
2454 live_map
= dm_get_live_table(md
);
2456 limits
= md
->queue
->limits
;
2457 dm_table_put(live_map
);
2461 r
= dm_calculate_queue_limits(table
, &limits
);
2468 map
= __bind(md
, table
, &limits
);
2471 mutex_unlock(&md
->suspend_lock
);
2476 * Functions to lock and unlock any filesystem running on the
2479 static int lock_fs(struct mapped_device
*md
)
2483 WARN_ON(md
->frozen_sb
);
2485 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2486 if (IS_ERR(md
->frozen_sb
)) {
2487 r
= PTR_ERR(md
->frozen_sb
);
2488 md
->frozen_sb
= NULL
;
2492 set_bit(DMF_FROZEN
, &md
->flags
);
2497 static void unlock_fs(struct mapped_device
*md
)
2499 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2502 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2503 md
->frozen_sb
= NULL
;
2504 clear_bit(DMF_FROZEN
, &md
->flags
);
2508 * We need to be able to change a mapping table under a mounted
2509 * filesystem. For example we might want to move some data in
2510 * the background. Before the table can be swapped with
2511 * dm_bind_table, dm_suspend must be called to flush any in
2512 * flight bios and ensure that any further io gets deferred.
2515 * Suspend mechanism in request-based dm.
2517 * 1. Flush all I/Os by lock_fs() if needed.
2518 * 2. Stop dispatching any I/O by stopping the request_queue.
2519 * 3. Wait for all in-flight I/Os to be completed or requeued.
2521 * To abort suspend, start the request_queue.
2523 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2525 struct dm_table
*map
= NULL
;
2527 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
2528 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
2530 mutex_lock(&md
->suspend_lock
);
2532 if (dm_suspended_md(md
)) {
2537 map
= dm_get_live_table(md
);
2540 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2541 * This flag is cleared before dm_suspend returns.
2544 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2546 /* This does not get reverted if there's an error later. */
2547 dm_table_presuspend_targets(map
);
2550 * Flush I/O to the device.
2551 * Any I/O submitted after lock_fs() may not be flushed.
2552 * noflush takes precedence over do_lockfs.
2553 * (lock_fs() flushes I/Os and waits for them to complete.)
2555 if (!noflush
&& do_lockfs
) {
2562 * Here we must make sure that no processes are submitting requests
2563 * to target drivers i.e. no one may be executing
2564 * __split_and_process_bio. This is called from dm_request and
2567 * To get all processes out of __split_and_process_bio in dm_request,
2568 * we take the write lock. To prevent any process from reentering
2569 * __split_and_process_bio from dm_request and quiesce the thread
2570 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2571 * flush_workqueue(md->wq).
2573 down_write(&md
->io_lock
);
2574 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2575 up_write(&md
->io_lock
);
2578 * Stop md->queue before flushing md->wq in case request-based
2579 * dm defers requests to md->wq from md->queue.
2581 if (dm_request_based(md
))
2582 stop_queue(md
->queue
);
2584 flush_workqueue(md
->wq
);
2587 * At this point no more requests are entering target request routines.
2588 * We call dm_wait_for_completion to wait for all existing requests
2591 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
2593 down_write(&md
->io_lock
);
2595 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2596 up_write(&md
->io_lock
);
2598 /* were we interrupted ? */
2602 if (dm_request_based(md
))
2603 start_queue(md
->queue
);
2606 goto out
; /* pushback list is already flushed, so skip flush */
2610 * If dm_wait_for_completion returned 0, the device is completely
2611 * quiescent now. There is no request-processing activity. All new
2612 * requests are being added to md->deferred list.
2615 set_bit(DMF_SUSPENDED
, &md
->flags
);
2617 dm_table_postsuspend_targets(map
);
2623 mutex_unlock(&md
->suspend_lock
);
2627 int dm_resume(struct mapped_device
*md
)
2630 struct dm_table
*map
= NULL
;
2632 mutex_lock(&md
->suspend_lock
);
2633 if (!dm_suspended_md(md
))
2636 map
= dm_get_live_table(md
);
2637 if (!map
|| !dm_table_get_size(map
))
2640 r
= dm_table_resume_targets(map
);
2647 * Flushing deferred I/Os must be done after targets are resumed
2648 * so that mapping of targets can work correctly.
2649 * Request-based dm is queueing the deferred I/Os in its request_queue.
2651 if (dm_request_based(md
))
2652 start_queue(md
->queue
);
2656 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2661 mutex_unlock(&md
->suspend_lock
);
2666 /*-----------------------------------------------------------------
2667 * Event notification.
2668 *---------------------------------------------------------------*/
2669 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2672 char udev_cookie
[DM_COOKIE_LENGTH
];
2673 char *envp
[] = { udev_cookie
, NULL
};
2676 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2678 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2679 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2680 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2685 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2687 return atomic_add_return(1, &md
->uevent_seq
);
2690 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2692 return atomic_read(&md
->event_nr
);
2695 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2697 return wait_event_interruptible(md
->eventq
,
2698 (event_nr
!= atomic_read(&md
->event_nr
)));
2701 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2703 unsigned long flags
;
2705 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2706 list_add(elist
, &md
->uevent_list
);
2707 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2711 * The gendisk is only valid as long as you have a reference
2714 struct gendisk
*dm_disk(struct mapped_device
*md
)
2719 struct kobject
*dm_kobject(struct mapped_device
*md
)
2725 * struct mapped_device should not be exported outside of dm.c
2726 * so use this check to verify that kobj is part of md structure
2728 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2730 struct mapped_device
*md
;
2732 md
= container_of(kobj
, struct mapped_device
, kobj
);
2733 if (&md
->kobj
!= kobj
)
2736 if (test_bit(DMF_FREEING
, &md
->flags
) ||
2744 int dm_suspended_md(struct mapped_device
*md
)
2746 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2749 int dm_suspended(struct dm_target
*ti
)
2751 return dm_suspended_md(dm_table_get_md(ti
->table
));
2753 EXPORT_SYMBOL_GPL(dm_suspended
);
2755 int dm_noflush_suspending(struct dm_target
*ti
)
2757 return __noflush_suspending(dm_table_get_md(ti
->table
));
2759 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2761 struct dm_md_mempools
*dm_alloc_md_mempools(unsigned type
, unsigned integrity
, unsigned per_bio_data_size
)
2763 struct dm_md_mempools
*pools
= kzalloc(sizeof(*pools
), GFP_KERNEL
);
2764 struct kmem_cache
*cachep
;
2765 unsigned int pool_size
;
2766 unsigned int front_pad
;
2771 if (type
== DM_TYPE_BIO_BASED
) {
2774 front_pad
= roundup(per_bio_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2775 } else if (type
== DM_TYPE_REQUEST_BASED
) {
2776 cachep
= _rq_tio_cache
;
2777 pool_size
= MIN_IOS
;
2778 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2779 /* per_bio_data_size is not used. See __bind_mempools(). */
2780 WARN_ON(per_bio_data_size
!= 0);
2784 pools
->io_pool
= mempool_create_slab_pool(MIN_IOS
, cachep
);
2785 if (!pools
->io_pool
)
2788 pools
->bs
= bioset_create(pool_size
, front_pad
);
2792 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2798 dm_free_md_mempools(pools
);
2803 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2809 mempool_destroy(pools
->io_pool
);
2812 bioset_free(pools
->bs
);
2817 static const struct block_device_operations dm_blk_dops
= {
2818 .open
= dm_blk_open
,
2819 .release
= dm_blk_close
,
2820 .ioctl
= dm_blk_ioctl
,
2821 .getgeo
= dm_blk_getgeo
,
2822 .owner
= THIS_MODULE
2825 EXPORT_SYMBOL(dm_get_mapinfo
);
2830 module_init(dm_init
);
2831 module_exit(dm_exit
);
2833 module_param(major
, uint
, 0);
2834 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2835 MODULE_DESCRIPTION(DM_NAME
" driver");
2836 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2837 MODULE_LICENSE("GPL");