2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
9 #include "dm-bio-list.h"
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/moduleparam.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/blktrace_api.h>
24 #include <trace/block.h>
26 #define DM_MSG_PREFIX "core"
28 static const char *_name
= DM_NAME
;
30 static unsigned int major
= 0;
31 static unsigned int _major
= 0;
33 static DEFINE_SPINLOCK(_minor_lock
);
35 * One of these is allocated per bio.
38 struct mapped_device
*md
;
42 unsigned long start_time
;
46 * One of these is allocated per target within a bio. Hopefully
47 * this will be simplified out one day.
55 DEFINE_TRACE(block_bio_complete
);
57 union map_info
*dm_get_mapinfo(struct bio
*bio
)
59 if (bio
&& bio
->bi_private
)
60 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
64 #define MINOR_ALLOCED ((void *)-1)
67 * Bits for the md->flags field.
69 #define DMF_BLOCK_IO 0
70 #define DMF_SUSPENDED 1
73 #define DMF_DELETING 4
74 #define DMF_NOFLUSH_SUSPENDING 5
77 * Work processed by per-device workqueue.
83 struct work_struct work
;
84 struct mapped_device
*md
;
88 struct mapped_device
{
89 struct rw_semaphore io_lock
;
90 struct mutex suspend_lock
;
91 spinlock_t pushback_lock
;
98 struct request_queue
*queue
;
105 * A list of ios that arrived while we were suspended.
108 wait_queue_head_t wait
;
109 struct bio_list deferred
;
110 struct bio_list pushback
;
113 * Processing queue (flush/barriers)
115 struct workqueue_struct
*wq
;
118 * The current mapping.
120 struct dm_table
*map
;
123 * io objects are allocated from here.
134 wait_queue_head_t eventq
;
136 struct list_head uevent_list
;
137 spinlock_t uevent_lock
; /* Protect access to uevent_list */
140 * freeze/thaw support require holding onto a super block
142 struct super_block
*frozen_sb
;
143 struct block_device
*suspended_bdev
;
145 /* forced geometry settings */
146 struct hd_geometry geometry
;
150 static struct kmem_cache
*_io_cache
;
151 static struct kmem_cache
*_tio_cache
;
153 static int __init
local_init(void)
157 /* allocate a slab for the dm_ios */
158 _io_cache
= KMEM_CACHE(dm_io
, 0);
162 /* allocate a slab for the target ios */
163 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
165 goto out_free_io_cache
;
167 r
= dm_uevent_init();
169 goto out_free_tio_cache
;
172 r
= register_blkdev(_major
, _name
);
174 goto out_uevent_exit
;
184 kmem_cache_destroy(_tio_cache
);
186 kmem_cache_destroy(_io_cache
);
191 static void local_exit(void)
193 kmem_cache_destroy(_tio_cache
);
194 kmem_cache_destroy(_io_cache
);
195 unregister_blkdev(_major
, _name
);
200 DMINFO("cleaned up");
203 static int (*_inits
[])(void) __initdata
= {
212 static void (*_exits
[])(void) = {
221 static int __init
dm_init(void)
223 const int count
= ARRAY_SIZE(_inits
);
227 for (i
= 0; i
< count
; i
++) {
242 static void __exit
dm_exit(void)
244 int i
= ARRAY_SIZE(_exits
);
251 * Block device functions
253 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
255 struct mapped_device
*md
;
257 spin_lock(&_minor_lock
);
259 md
= bdev
->bd_disk
->private_data
;
263 if (test_bit(DMF_FREEING
, &md
->flags
) ||
264 test_bit(DMF_DELETING
, &md
->flags
)) {
270 atomic_inc(&md
->open_count
);
273 spin_unlock(&_minor_lock
);
275 return md
? 0 : -ENXIO
;
278 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
280 struct mapped_device
*md
= disk
->private_data
;
281 atomic_dec(&md
->open_count
);
286 int dm_open_count(struct mapped_device
*md
)
288 return atomic_read(&md
->open_count
);
292 * Guarantees nothing is using the device before it's deleted.
294 int dm_lock_for_deletion(struct mapped_device
*md
)
298 spin_lock(&_minor_lock
);
300 if (dm_open_count(md
))
303 set_bit(DMF_DELETING
, &md
->flags
);
305 spin_unlock(&_minor_lock
);
310 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
312 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
314 return dm_get_geometry(md
, geo
);
317 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
318 unsigned int cmd
, unsigned long arg
)
320 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
321 struct dm_table
*map
= dm_get_table(md
);
322 struct dm_target
*tgt
;
325 if (!map
|| !dm_table_get_size(map
))
328 /* We only support devices that have a single target */
329 if (dm_table_get_num_targets(map
) != 1)
332 tgt
= dm_table_get_target(map
, 0);
334 if (dm_suspended(md
)) {
339 if (tgt
->type
->ioctl
)
340 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
348 static struct dm_io
*alloc_io(struct mapped_device
*md
)
350 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
353 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
355 mempool_free(io
, md
->io_pool
);
358 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
360 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
363 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
365 mempool_free(tio
, md
->tio_pool
);
368 static void start_io_acct(struct dm_io
*io
)
370 struct mapped_device
*md
= io
->md
;
373 io
->start_time
= jiffies
;
375 cpu
= part_stat_lock();
376 part_round_stats(cpu
, &dm_disk(md
)->part0
);
378 dm_disk(md
)->part0
.in_flight
= atomic_inc_return(&md
->pending
);
381 static void end_io_acct(struct dm_io
*io
)
383 struct mapped_device
*md
= io
->md
;
384 struct bio
*bio
= io
->bio
;
385 unsigned long duration
= jiffies
- io
->start_time
;
387 int rw
= bio_data_dir(bio
);
389 cpu
= part_stat_lock();
390 part_round_stats(cpu
, &dm_disk(md
)->part0
);
391 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
394 dm_disk(md
)->part0
.in_flight
= pending
=
395 atomic_dec_return(&md
->pending
);
397 /* nudge anyone waiting on suspend queue */
403 * Add the bio to the list of deferred io.
405 static int queue_io(struct mapped_device
*md
, struct bio
*bio
)
407 down_write(&md
->io_lock
);
409 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
410 up_write(&md
->io_lock
);
414 bio_list_add(&md
->deferred
, bio
);
416 up_write(&md
->io_lock
);
417 return 0; /* deferred successfully */
421 * Everyone (including functions in this file), should use this
422 * function to access the md->map field, and make sure they call
423 * dm_table_put() when finished.
425 struct dm_table
*dm_get_table(struct mapped_device
*md
)
429 read_lock(&md
->map_lock
);
433 read_unlock(&md
->map_lock
);
439 * Get the geometry associated with a dm device
441 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
449 * Set the geometry of a device.
451 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
453 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
455 if (geo
->start
> sz
) {
456 DMWARN("Start sector is beyond the geometry limits.");
465 /*-----------------------------------------------------------------
467 * A more elegant soln is in the works that uses the queue
468 * merge fn, unfortunately there are a couple of changes to
469 * the block layer that I want to make for this. So in the
470 * interests of getting something for people to use I give
471 * you this clearly demarcated crap.
472 *---------------------------------------------------------------*/
474 static int __noflush_suspending(struct mapped_device
*md
)
476 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
480 * Decrements the number of outstanding ios that a bio has been
481 * cloned into, completing the original io if necc.
483 static void dec_pending(struct dm_io
*io
, int error
)
487 /* Push-back supersedes any I/O errors */
488 if (error
&& !(io
->error
> 0 && __noflush_suspending(io
->md
)))
491 if (atomic_dec_and_test(&io
->io_count
)) {
492 if (io
->error
== DM_ENDIO_REQUEUE
) {
494 * Target requested pushing back the I/O.
495 * This must be handled before the sleeper on
496 * suspend queue merges the pushback list.
498 spin_lock_irqsave(&io
->md
->pushback_lock
, flags
);
499 if (__noflush_suspending(io
->md
))
500 bio_list_add(&io
->md
->pushback
, io
->bio
);
502 /* noflush suspend was interrupted. */
504 spin_unlock_irqrestore(&io
->md
->pushback_lock
, flags
);
509 if (io
->error
!= DM_ENDIO_REQUEUE
) {
510 trace_block_bio_complete(io
->md
->queue
, io
->bio
);
512 bio_endio(io
->bio
, io
->error
);
519 static void clone_endio(struct bio
*bio
, int error
)
522 struct dm_target_io
*tio
= bio
->bi_private
;
523 struct mapped_device
*md
= tio
->io
->md
;
524 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
526 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
530 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
531 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
533 * error and requeue request are handled
537 else if (r
== DM_ENDIO_INCOMPLETE
)
538 /* The target will handle the io */
541 DMWARN("unimplemented target endio return value: %d", r
);
546 dec_pending(tio
->io
, error
);
549 * Store md for cleanup instead of tio which is about to get freed.
551 bio
->bi_private
= md
->bs
;
557 static sector_t
max_io_len(struct mapped_device
*md
,
558 sector_t sector
, struct dm_target
*ti
)
560 sector_t offset
= sector
- ti
->begin
;
561 sector_t len
= ti
->len
- offset
;
564 * Does the target need to split even further ?
568 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
577 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
578 struct dm_target_io
*tio
)
582 struct mapped_device
*md
;
587 BUG_ON(!clone
->bi_size
);
589 clone
->bi_end_io
= clone_endio
;
590 clone
->bi_private
= tio
;
593 * Map the clone. If r == 0 we don't need to do
594 * anything, the target has assumed ownership of
597 atomic_inc(&tio
->io
->io_count
);
598 sector
= clone
->bi_sector
;
599 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
600 if (r
== DM_MAPIO_REMAPPED
) {
601 /* the bio has been remapped so dispatch it */
603 trace_block_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
604 tio
->io
->bio
->bi_bdev
->bd_dev
,
605 clone
->bi_sector
, sector
);
607 generic_make_request(clone
);
608 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
609 /* error the io and bail out, or requeue it if needed */
611 dec_pending(tio
->io
, r
);
613 * Store bio_set for cleanup.
615 clone
->bi_private
= md
->bs
;
619 DMWARN("unimplemented target map return value: %d", r
);
625 struct mapped_device
*md
;
626 struct dm_table
*map
;
630 sector_t sector_count
;
634 static void dm_bio_destructor(struct bio
*bio
)
636 struct bio_set
*bs
= bio
->bi_private
;
642 * Creates a little bio that is just does part of a bvec.
644 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
645 unsigned short idx
, unsigned int offset
,
646 unsigned int len
, struct bio_set
*bs
)
649 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
651 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
652 clone
->bi_destructor
= dm_bio_destructor
;
653 *clone
->bi_io_vec
= *bv
;
655 clone
->bi_sector
= sector
;
656 clone
->bi_bdev
= bio
->bi_bdev
;
657 clone
->bi_rw
= bio
->bi_rw
;
659 clone
->bi_size
= to_bytes(len
);
660 clone
->bi_io_vec
->bv_offset
= offset
;
661 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
662 clone
->bi_flags
|= 1 << BIO_CLONED
;
668 * Creates a bio that consists of range of complete bvecs.
670 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
671 unsigned short idx
, unsigned short bv_count
,
672 unsigned int len
, struct bio_set
*bs
)
676 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
677 __bio_clone(clone
, bio
);
678 clone
->bi_destructor
= dm_bio_destructor
;
679 clone
->bi_sector
= sector
;
681 clone
->bi_vcnt
= idx
+ bv_count
;
682 clone
->bi_size
= to_bytes(len
);
683 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
688 static int __clone_and_map(struct clone_info
*ci
)
690 struct bio
*clone
, *bio
= ci
->bio
;
691 struct dm_target
*ti
;
692 sector_t len
= 0, max
;
693 struct dm_target_io
*tio
;
695 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
696 if (!dm_target_is_valid(ti
))
699 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
702 * Allocate a target io object.
704 tio
= alloc_tio(ci
->md
);
707 memset(&tio
->info
, 0, sizeof(tio
->info
));
709 if (ci
->sector_count
<= max
) {
711 * Optimise for the simple case where we can do all of
712 * the remaining io with a single clone.
714 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
715 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
717 __map_bio(ti
, clone
, tio
);
718 ci
->sector_count
= 0;
720 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
722 * There are some bvecs that don't span targets.
723 * Do as many of these as possible.
726 sector_t remaining
= max
;
729 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
730 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
732 if (bv_len
> remaining
)
739 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
741 __map_bio(ti
, clone
, tio
);
744 ci
->sector_count
-= len
;
749 * Handle a bvec that must be split between two or more targets.
751 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
752 sector_t remaining
= to_sector(bv
->bv_len
);
753 unsigned int offset
= 0;
757 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
758 if (!dm_target_is_valid(ti
))
761 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
763 tio
= alloc_tio(ci
->md
);
766 memset(&tio
->info
, 0, sizeof(tio
->info
));
769 len
= min(remaining
, max
);
771 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
772 bv
->bv_offset
+ offset
, len
,
775 __map_bio(ti
, clone
, tio
);
778 ci
->sector_count
-= len
;
779 offset
+= to_bytes(len
);
780 } while (remaining
-= len
);
789 * Split the bio into several clones.
791 static int __split_bio(struct mapped_device
*md
, struct bio
*bio
)
793 struct clone_info ci
;
796 ci
.map
= dm_get_table(md
);
797 if (unlikely(!ci
.map
))
802 ci
.io
= alloc_io(md
);
804 atomic_set(&ci
.io
->io_count
, 1);
807 ci
.sector
= bio
->bi_sector
;
808 ci
.sector_count
= bio_sectors(bio
);
809 ci
.idx
= bio
->bi_idx
;
811 start_io_acct(ci
.io
);
812 while (ci
.sector_count
&& !error
)
813 error
= __clone_and_map(&ci
);
815 /* drop the extra reference count */
816 dec_pending(ci
.io
, error
);
817 dm_table_put(ci
.map
);
821 /*-----------------------------------------------------------------
823 *---------------------------------------------------------------*/
825 static int dm_merge_bvec(struct request_queue
*q
,
826 struct bvec_merge_data
*bvm
,
827 struct bio_vec
*biovec
)
829 struct mapped_device
*md
= q
->queuedata
;
830 struct dm_table
*map
= dm_get_table(md
);
831 struct dm_target
*ti
;
832 sector_t max_sectors
;
838 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
839 if (!dm_target_is_valid(ti
))
843 * Find maximum amount of I/O that won't need splitting
845 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
846 (sector_t
) BIO_MAX_SECTORS
);
847 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
852 * merge_bvec_fn() returns number of bytes
853 * it can accept at this offset
854 * max is precomputed maximal io size
856 if (max_size
&& ti
->type
->merge
)
857 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
864 * Always allow an entire first page
866 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
867 max_size
= biovec
->bv_len
;
873 * The request function that just remaps the bio built up by
876 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
879 int rw
= bio_data_dir(bio
);
880 struct mapped_device
*md
= q
->queuedata
;
884 * There is no use in forwarding any barrier request since we can't
885 * guarantee it is (or can be) handled by the targets correctly.
887 if (unlikely(bio_barrier(bio
))) {
888 bio_endio(bio
, -EOPNOTSUPP
);
892 down_read(&md
->io_lock
);
894 cpu
= part_stat_lock();
895 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
896 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
900 * If we're suspended we have to queue
903 while (test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
904 up_read(&md
->io_lock
);
906 if (bio_rw(bio
) != READA
)
907 r
= queue_io(md
, bio
);
913 * We're in a while loop, because someone could suspend
914 * before we get to the following read lock.
916 down_read(&md
->io_lock
);
919 r
= __split_bio(md
, bio
);
920 up_read(&md
->io_lock
);
929 static void dm_unplug_all(struct request_queue
*q
)
931 struct mapped_device
*md
= q
->queuedata
;
932 struct dm_table
*map
= dm_get_table(md
);
935 dm_table_unplug_all(map
);
940 static int dm_any_congested(void *congested_data
, int bdi_bits
)
943 struct mapped_device
*md
= congested_data
;
944 struct dm_table
*map
;
946 atomic_inc(&md
->pending
);
948 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
949 map
= dm_get_table(md
);
951 r
= dm_table_any_congested(map
, bdi_bits
);
956 if (!atomic_dec_return(&md
->pending
))
957 /* nudge anyone waiting on suspend queue */
963 /*-----------------------------------------------------------------
964 * An IDR is used to keep track of allocated minor numbers.
965 *---------------------------------------------------------------*/
966 static DEFINE_IDR(_minor_idr
);
968 static void free_minor(int minor
)
970 spin_lock(&_minor_lock
);
971 idr_remove(&_minor_idr
, minor
);
972 spin_unlock(&_minor_lock
);
976 * See if the device with a specific minor # is free.
978 static int specific_minor(int minor
)
982 if (minor
>= (1 << MINORBITS
))
985 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
989 spin_lock(&_minor_lock
);
991 if (idr_find(&_minor_idr
, minor
)) {
996 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1001 idr_remove(&_minor_idr
, m
);
1007 spin_unlock(&_minor_lock
);
1011 static int next_free_minor(int *minor
)
1015 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1019 spin_lock(&_minor_lock
);
1021 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1025 if (m
>= (1 << MINORBITS
)) {
1026 idr_remove(&_minor_idr
, m
);
1034 spin_unlock(&_minor_lock
);
1038 static struct block_device_operations dm_blk_dops
;
1041 * Allocate and initialise a blank device with a given minor.
1043 static struct mapped_device
*alloc_dev(int minor
)
1046 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1050 DMWARN("unable to allocate device, out of memory.");
1054 if (!try_module_get(THIS_MODULE
))
1055 goto bad_module_get
;
1057 /* get a minor number for the dev */
1058 if (minor
== DM_ANY_MINOR
)
1059 r
= next_free_minor(&minor
);
1061 r
= specific_minor(minor
);
1065 init_rwsem(&md
->io_lock
);
1066 mutex_init(&md
->suspend_lock
);
1067 spin_lock_init(&md
->pushback_lock
);
1068 rwlock_init(&md
->map_lock
);
1069 atomic_set(&md
->holders
, 1);
1070 atomic_set(&md
->open_count
, 0);
1071 atomic_set(&md
->event_nr
, 0);
1072 atomic_set(&md
->uevent_seq
, 0);
1073 INIT_LIST_HEAD(&md
->uevent_list
);
1074 spin_lock_init(&md
->uevent_lock
);
1076 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1080 md
->queue
->queuedata
= md
;
1081 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1082 md
->queue
->backing_dev_info
.congested_data
= md
;
1083 blk_queue_make_request(md
->queue
, dm_request
);
1084 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1085 md
->queue
->unplug_fn
= dm_unplug_all
;
1086 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1088 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1092 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1096 md
->bs
= bioset_create(16, 0);
1100 md
->disk
= alloc_disk(1);
1104 atomic_set(&md
->pending
, 0);
1105 init_waitqueue_head(&md
->wait
);
1106 init_waitqueue_head(&md
->eventq
);
1108 md
->disk
->major
= _major
;
1109 md
->disk
->first_minor
= minor
;
1110 md
->disk
->fops
= &dm_blk_dops
;
1111 md
->disk
->queue
= md
->queue
;
1112 md
->disk
->private_data
= md
;
1113 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1115 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1117 md
->wq
= create_singlethread_workqueue("kdmflush");
1121 /* Populate the mapping, nobody knows we exist yet */
1122 spin_lock(&_minor_lock
);
1123 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1124 spin_unlock(&_minor_lock
);
1126 BUG_ON(old_md
!= MINOR_ALLOCED
);
1133 bioset_free(md
->bs
);
1135 mempool_destroy(md
->tio_pool
);
1137 mempool_destroy(md
->io_pool
);
1139 blk_cleanup_queue(md
->queue
);
1143 module_put(THIS_MODULE
);
1149 static void unlock_fs(struct mapped_device
*md
);
1151 static void free_dev(struct mapped_device
*md
)
1153 int minor
= MINOR(disk_devt(md
->disk
));
1155 if (md
->suspended_bdev
) {
1157 bdput(md
->suspended_bdev
);
1159 destroy_workqueue(md
->wq
);
1160 mempool_destroy(md
->tio_pool
);
1161 mempool_destroy(md
->io_pool
);
1162 bioset_free(md
->bs
);
1163 del_gendisk(md
->disk
);
1166 spin_lock(&_minor_lock
);
1167 md
->disk
->private_data
= NULL
;
1168 spin_unlock(&_minor_lock
);
1171 blk_cleanup_queue(md
->queue
);
1172 module_put(THIS_MODULE
);
1177 * Bind a table to the device.
1179 static void event_callback(void *context
)
1181 unsigned long flags
;
1183 struct mapped_device
*md
= (struct mapped_device
*) context
;
1185 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1186 list_splice_init(&md
->uevent_list
, &uevents
);
1187 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1189 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1191 atomic_inc(&md
->event_nr
);
1192 wake_up(&md
->eventq
);
1195 static void __set_size(struct mapped_device
*md
, sector_t size
)
1197 set_capacity(md
->disk
, size
);
1199 mutex_lock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1200 i_size_write(md
->suspended_bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1201 mutex_unlock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1204 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1206 struct request_queue
*q
= md
->queue
;
1209 size
= dm_table_get_size(t
);
1212 * Wipe any geometry if the size of the table changed.
1214 if (size
!= get_capacity(md
->disk
))
1215 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1217 if (md
->suspended_bdev
)
1218 __set_size(md
, size
);
1223 dm_table_event_callback(t
, event_callback
, md
);
1225 write_lock(&md
->map_lock
);
1227 dm_table_set_restrictions(t
, q
);
1228 write_unlock(&md
->map_lock
);
1233 static void __unbind(struct mapped_device
*md
)
1235 struct dm_table
*map
= md
->map
;
1240 dm_table_event_callback(map
, NULL
, NULL
);
1241 write_lock(&md
->map_lock
);
1243 write_unlock(&md
->map_lock
);
1248 * Constructor for a new device.
1250 int dm_create(int minor
, struct mapped_device
**result
)
1252 struct mapped_device
*md
;
1254 md
= alloc_dev(minor
);
1262 static struct mapped_device
*dm_find_md(dev_t dev
)
1264 struct mapped_device
*md
;
1265 unsigned minor
= MINOR(dev
);
1267 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1270 spin_lock(&_minor_lock
);
1272 md
= idr_find(&_minor_idr
, minor
);
1273 if (md
&& (md
== MINOR_ALLOCED
||
1274 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1275 test_bit(DMF_FREEING
, &md
->flags
))) {
1281 spin_unlock(&_minor_lock
);
1286 struct mapped_device
*dm_get_md(dev_t dev
)
1288 struct mapped_device
*md
= dm_find_md(dev
);
1296 void *dm_get_mdptr(struct mapped_device
*md
)
1298 return md
->interface_ptr
;
1301 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1303 md
->interface_ptr
= ptr
;
1306 void dm_get(struct mapped_device
*md
)
1308 atomic_inc(&md
->holders
);
1311 const char *dm_device_name(struct mapped_device
*md
)
1315 EXPORT_SYMBOL_GPL(dm_device_name
);
1317 void dm_put(struct mapped_device
*md
)
1319 struct dm_table
*map
;
1321 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1323 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1324 map
= dm_get_table(md
);
1325 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
1326 MINOR(disk_devt(dm_disk(md
))));
1327 set_bit(DMF_FREEING
, &md
->flags
);
1328 spin_unlock(&_minor_lock
);
1329 if (!dm_suspended(md
)) {
1330 dm_table_presuspend_targets(map
);
1331 dm_table_postsuspend_targets(map
);
1338 EXPORT_SYMBOL_GPL(dm_put
);
1340 static int dm_wait_for_completion(struct mapped_device
*md
)
1345 set_current_state(TASK_INTERRUPTIBLE
);
1348 if (!atomic_read(&md
->pending
))
1351 if (signal_pending(current
)) {
1358 set_current_state(TASK_RUNNING
);
1364 * Process the deferred bios
1366 static void __flush_deferred_io(struct mapped_device
*md
)
1370 while ((c
= bio_list_pop(&md
->deferred
))) {
1371 if (__split_bio(md
, c
))
1375 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1378 static void __merge_pushback_list(struct mapped_device
*md
)
1380 unsigned long flags
;
1382 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1383 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1384 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1385 bio_list_init(&md
->pushback
);
1386 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1389 static void dm_wq_work(struct work_struct
*work
)
1391 struct dm_wq_req
*req
= container_of(work
, struct dm_wq_req
, work
);
1392 struct mapped_device
*md
= req
->md
;
1394 down_write(&md
->io_lock
);
1395 switch (req
->type
) {
1396 case DM_WQ_FLUSH_DEFERRED
:
1397 __flush_deferred_io(md
);
1400 DMERR("dm_wq_work: unrecognised work type %d", req
->type
);
1403 up_write(&md
->io_lock
);
1406 static void dm_wq_queue(struct mapped_device
*md
, int type
, void *context
,
1407 struct dm_wq_req
*req
)
1411 req
->context
= context
;
1412 INIT_WORK(&req
->work
, dm_wq_work
);
1413 queue_work(md
->wq
, &req
->work
);
1416 static void dm_queue_flush(struct mapped_device
*md
, int type
, void *context
)
1418 struct dm_wq_req req
;
1420 dm_wq_queue(md
, type
, context
, &req
);
1421 flush_workqueue(md
->wq
);
1425 * Swap in a new table (destroying old one).
1427 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1431 mutex_lock(&md
->suspend_lock
);
1433 /* device must be suspended */
1434 if (!dm_suspended(md
))
1437 /* without bdev, the device size cannot be changed */
1438 if (!md
->suspended_bdev
)
1439 if (get_capacity(md
->disk
) != dm_table_get_size(table
))
1443 r
= __bind(md
, table
);
1446 mutex_unlock(&md
->suspend_lock
);
1451 * Functions to lock and unlock any filesystem running on the
1454 static int lock_fs(struct mapped_device
*md
)
1458 WARN_ON(md
->frozen_sb
);
1460 md
->frozen_sb
= freeze_bdev(md
->suspended_bdev
);
1461 if (IS_ERR(md
->frozen_sb
)) {
1462 r
= PTR_ERR(md
->frozen_sb
);
1463 md
->frozen_sb
= NULL
;
1467 set_bit(DMF_FROZEN
, &md
->flags
);
1469 /* don't bdput right now, we don't want the bdev
1470 * to go away while it is locked.
1475 static void unlock_fs(struct mapped_device
*md
)
1477 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1480 thaw_bdev(md
->suspended_bdev
, md
->frozen_sb
);
1481 md
->frozen_sb
= NULL
;
1482 clear_bit(DMF_FROZEN
, &md
->flags
);
1486 * We need to be able to change a mapping table under a mounted
1487 * filesystem. For example we might want to move some data in
1488 * the background. Before the table can be swapped with
1489 * dm_bind_table, dm_suspend must be called to flush any in
1490 * flight bios and ensure that any further io gets deferred.
1492 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1494 struct dm_table
*map
= NULL
;
1495 DECLARE_WAITQUEUE(wait
, current
);
1497 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1498 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1500 mutex_lock(&md
->suspend_lock
);
1502 if (dm_suspended(md
)) {
1507 map
= dm_get_table(md
);
1510 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1511 * This flag is cleared before dm_suspend returns.
1514 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1516 /* This does not get reverted if there's an error later. */
1517 dm_table_presuspend_targets(map
);
1519 /* bdget() can stall if the pending I/Os are not flushed */
1521 md
->suspended_bdev
= bdget_disk(md
->disk
, 0);
1522 if (!md
->suspended_bdev
) {
1523 DMWARN("bdget failed in dm_suspend");
1529 * Flush I/O to the device. noflush supersedes do_lockfs,
1530 * because lock_fs() needs to flush I/Os.
1540 * First we set the BLOCK_IO flag so no more ios will be mapped.
1542 down_write(&md
->io_lock
);
1543 set_bit(DMF_BLOCK_IO
, &md
->flags
);
1545 add_wait_queue(&md
->wait
, &wait
);
1546 up_write(&md
->io_lock
);
1550 dm_table_unplug_all(map
);
1553 * Wait for the already-mapped ios to complete.
1555 r
= dm_wait_for_completion(md
);
1557 down_write(&md
->io_lock
);
1558 remove_wait_queue(&md
->wait
, &wait
);
1561 __merge_pushback_list(md
);
1562 up_write(&md
->io_lock
);
1564 /* were we interrupted ? */
1566 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1569 goto out
; /* pushback list is already flushed, so skip flush */
1572 dm_table_postsuspend_targets(map
);
1574 set_bit(DMF_SUSPENDED
, &md
->flags
);
1577 if (r
&& md
->suspended_bdev
) {
1578 bdput(md
->suspended_bdev
);
1579 md
->suspended_bdev
= NULL
;
1585 mutex_unlock(&md
->suspend_lock
);
1589 int dm_resume(struct mapped_device
*md
)
1592 struct dm_table
*map
= NULL
;
1594 mutex_lock(&md
->suspend_lock
);
1595 if (!dm_suspended(md
))
1598 map
= dm_get_table(md
);
1599 if (!map
|| !dm_table_get_size(map
))
1602 r
= dm_table_resume_targets(map
);
1606 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1610 if (md
->suspended_bdev
) {
1611 bdput(md
->suspended_bdev
);
1612 md
->suspended_bdev
= NULL
;
1615 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1617 dm_table_unplug_all(map
);
1619 dm_kobject_uevent(md
);
1625 mutex_unlock(&md
->suspend_lock
);
1630 /*-----------------------------------------------------------------
1631 * Event notification.
1632 *---------------------------------------------------------------*/
1633 void dm_kobject_uevent(struct mapped_device
*md
)
1635 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, KOBJ_CHANGE
);
1638 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
1640 return atomic_add_return(1, &md
->uevent_seq
);
1643 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1645 return atomic_read(&md
->event_nr
);
1648 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1650 return wait_event_interruptible(md
->eventq
,
1651 (event_nr
!= atomic_read(&md
->event_nr
)));
1654 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
1656 unsigned long flags
;
1658 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1659 list_add(elist
, &md
->uevent_list
);
1660 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1664 * The gendisk is only valid as long as you have a reference
1667 struct gendisk
*dm_disk(struct mapped_device
*md
)
1672 int dm_suspended(struct mapped_device
*md
)
1674 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1677 int dm_noflush_suspending(struct dm_target
*ti
)
1679 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1680 int r
= __noflush_suspending(md
);
1686 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1688 static struct block_device_operations dm_blk_dops
= {
1689 .open
= dm_blk_open
,
1690 .release
= dm_blk_close
,
1691 .ioctl
= dm_blk_ioctl
,
1692 .getgeo
= dm_blk_getgeo
,
1693 .owner
= THIS_MODULE
1696 EXPORT_SYMBOL(dm_get_mapinfo
);
1701 module_init(dm_init
);
1702 module_exit(dm_exit
);
1704 module_param(major
, uint
, 0);
1705 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1706 MODULE_DESCRIPTION(DM_NAME
" driver");
1707 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1708 MODULE_LICENSE("GPL");