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.
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
);
36 * One of these is allocated per bio.
39 struct mapped_device
*md
;
43 unsigned long start_time
;
48 * One of these is allocated per target within a bio. Hopefully
49 * this will be simplified out one day.
57 DEFINE_TRACE(block_bio_complete
);
60 * For request-based dm.
61 * One of these is allocated per request.
63 struct dm_rq_target_io
{
64 struct mapped_device
*md
;
66 struct request
*orig
, clone
;
72 * For request-based dm.
73 * One of these is allocated per bio.
75 struct dm_rq_clone_bio_info
{
80 union map_info
*dm_get_mapinfo(struct bio
*bio
)
82 if (bio
&& bio
->bi_private
)
83 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
87 #define MINOR_ALLOCED ((void *)-1)
90 * Bits for the md->flags field.
92 #define DMF_BLOCK_IO 0
93 #define DMF_SUSPENDED 1
96 #define DMF_DELETING 4
97 #define DMF_NOFLUSH_SUSPENDING 5
100 * Work processed by per-device workqueue.
102 struct mapped_device
{
103 struct rw_semaphore io_lock
;
104 struct mutex suspend_lock
;
105 spinlock_t pushback_lock
;
112 struct request_queue
*queue
;
113 struct gendisk
*disk
;
119 * A list of ios that arrived while we were suspended.
122 wait_queue_head_t wait
;
123 struct work_struct work
;
124 struct bio_list deferred
;
125 struct bio_list pushback
;
128 * Processing queue (flush/barriers)
130 struct workqueue_struct
*wq
;
133 * The current mapping.
135 struct dm_table
*map
;
138 * io objects are allocated from here.
149 wait_queue_head_t eventq
;
151 struct list_head uevent_list
;
152 spinlock_t uevent_lock
; /* Protect access to uevent_list */
155 * freeze/thaw support require holding onto a super block
157 struct super_block
*frozen_sb
;
158 struct block_device
*suspended_bdev
;
160 /* forced geometry settings */
161 struct hd_geometry geometry
;
168 static struct kmem_cache
*_io_cache
;
169 static struct kmem_cache
*_tio_cache
;
170 static struct kmem_cache
*_rq_tio_cache
;
171 static struct kmem_cache
*_rq_bio_info_cache
;
173 static int __init
local_init(void)
177 /* allocate a slab for the dm_ios */
178 _io_cache
= KMEM_CACHE(dm_io
, 0);
182 /* allocate a slab for the target ios */
183 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
185 goto out_free_io_cache
;
187 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
189 goto out_free_tio_cache
;
191 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
192 if (!_rq_bio_info_cache
)
193 goto out_free_rq_tio_cache
;
195 r
= dm_uevent_init();
197 goto out_free_rq_bio_info_cache
;
200 r
= register_blkdev(_major
, _name
);
202 goto out_uevent_exit
;
211 out_free_rq_bio_info_cache
:
212 kmem_cache_destroy(_rq_bio_info_cache
);
213 out_free_rq_tio_cache
:
214 kmem_cache_destroy(_rq_tio_cache
);
216 kmem_cache_destroy(_tio_cache
);
218 kmem_cache_destroy(_io_cache
);
223 static void local_exit(void)
225 kmem_cache_destroy(_rq_bio_info_cache
);
226 kmem_cache_destroy(_rq_tio_cache
);
227 kmem_cache_destroy(_tio_cache
);
228 kmem_cache_destroy(_io_cache
);
229 unregister_blkdev(_major
, _name
);
234 DMINFO("cleaned up");
237 static int (*_inits
[])(void) __initdata
= {
246 static void (*_exits
[])(void) = {
255 static int __init
dm_init(void)
257 const int count
= ARRAY_SIZE(_inits
);
261 for (i
= 0; i
< count
; i
++) {
276 static void __exit
dm_exit(void)
278 int i
= ARRAY_SIZE(_exits
);
285 * Block device functions
287 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
289 struct mapped_device
*md
;
291 spin_lock(&_minor_lock
);
293 md
= bdev
->bd_disk
->private_data
;
297 if (test_bit(DMF_FREEING
, &md
->flags
) ||
298 test_bit(DMF_DELETING
, &md
->flags
)) {
304 atomic_inc(&md
->open_count
);
307 spin_unlock(&_minor_lock
);
309 return md
? 0 : -ENXIO
;
312 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
314 struct mapped_device
*md
= disk
->private_data
;
315 atomic_dec(&md
->open_count
);
320 int dm_open_count(struct mapped_device
*md
)
322 return atomic_read(&md
->open_count
);
326 * Guarantees nothing is using the device before it's deleted.
328 int dm_lock_for_deletion(struct mapped_device
*md
)
332 spin_lock(&_minor_lock
);
334 if (dm_open_count(md
))
337 set_bit(DMF_DELETING
, &md
->flags
);
339 spin_unlock(&_minor_lock
);
344 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
346 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
348 return dm_get_geometry(md
, geo
);
351 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
352 unsigned int cmd
, unsigned long arg
)
354 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
355 struct dm_table
*map
= dm_get_table(md
);
356 struct dm_target
*tgt
;
359 if (!map
|| !dm_table_get_size(map
))
362 /* We only support devices that have a single target */
363 if (dm_table_get_num_targets(map
) != 1)
366 tgt
= dm_table_get_target(map
, 0);
368 if (dm_suspended(md
)) {
373 if (tgt
->type
->ioctl
)
374 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
382 static struct dm_io
*alloc_io(struct mapped_device
*md
)
384 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
387 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
389 mempool_free(io
, md
->io_pool
);
392 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
394 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
397 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
399 mempool_free(tio
, md
->tio_pool
);
402 static void start_io_acct(struct dm_io
*io
)
404 struct mapped_device
*md
= io
->md
;
407 io
->start_time
= jiffies
;
409 cpu
= part_stat_lock();
410 part_round_stats(cpu
, &dm_disk(md
)->part0
);
412 dm_disk(md
)->part0
.in_flight
= atomic_inc_return(&md
->pending
);
415 static void end_io_acct(struct dm_io
*io
)
417 struct mapped_device
*md
= io
->md
;
418 struct bio
*bio
= io
->bio
;
419 unsigned long duration
= jiffies
- io
->start_time
;
421 int rw
= bio_data_dir(bio
);
423 cpu
= part_stat_lock();
424 part_round_stats(cpu
, &dm_disk(md
)->part0
);
425 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
428 dm_disk(md
)->part0
.in_flight
= pending
=
429 atomic_dec_return(&md
->pending
);
431 /* nudge anyone waiting on suspend queue */
437 * Add the bio to the list of deferred io.
439 static int queue_io(struct mapped_device
*md
, struct bio
*bio
)
441 down_write(&md
->io_lock
);
443 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
444 up_write(&md
->io_lock
);
448 bio_list_add(&md
->deferred
, bio
);
450 up_write(&md
->io_lock
);
451 return 0; /* deferred successfully */
455 * Everyone (including functions in this file), should use this
456 * function to access the md->map field, and make sure they call
457 * dm_table_put() when finished.
459 struct dm_table
*dm_get_table(struct mapped_device
*md
)
463 read_lock(&md
->map_lock
);
467 read_unlock(&md
->map_lock
);
473 * Get the geometry associated with a dm device
475 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
483 * Set the geometry of a device.
485 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
487 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
489 if (geo
->start
> sz
) {
490 DMWARN("Start sector is beyond the geometry limits.");
499 /*-----------------------------------------------------------------
501 * A more elegant soln is in the works that uses the queue
502 * merge fn, unfortunately there are a couple of changes to
503 * the block layer that I want to make for this. So in the
504 * interests of getting something for people to use I give
505 * you this clearly demarcated crap.
506 *---------------------------------------------------------------*/
508 static int __noflush_suspending(struct mapped_device
*md
)
510 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
514 * Decrements the number of outstanding ios that a bio has been
515 * cloned into, completing the original io if necc.
517 static void dec_pending(struct dm_io
*io
, int error
)
522 struct mapped_device
*md
= io
->md
;
524 /* Push-back supersedes any I/O errors */
525 if (error
&& !(io
->error
> 0 && __noflush_suspending(md
)))
528 if (atomic_dec_and_test(&io
->io_count
)) {
529 if (io
->error
== DM_ENDIO_REQUEUE
) {
531 * Target requested pushing back the I/O.
532 * This must be handled before the sleeper on
533 * suspend queue merges the pushback list.
535 spin_lock_irqsave(&md
->pushback_lock
, flags
);
536 if (__noflush_suspending(md
))
537 bio_list_add(&md
->pushback
, io
->bio
);
539 /* noflush suspend was interrupted. */
541 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
546 io_error
= io
->error
;
551 if (io_error
!= DM_ENDIO_REQUEUE
) {
552 trace_block_bio_complete(md
->queue
, bio
);
554 bio_endio(bio
, io_error
);
559 static void clone_endio(struct bio
*bio
, int error
)
562 struct dm_target_io
*tio
= bio
->bi_private
;
563 struct dm_io
*io
= tio
->io
;
564 struct mapped_device
*md
= tio
->io
->md
;
565 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
567 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
571 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
572 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
574 * error and requeue request are handled
578 else if (r
== DM_ENDIO_INCOMPLETE
)
579 /* The target will handle the io */
582 DMWARN("unimplemented target endio return value: %d", r
);
588 * Store md for cleanup instead of tio which is about to get freed.
590 bio
->bi_private
= md
->bs
;
594 dec_pending(io
, error
);
597 static sector_t
max_io_len(struct mapped_device
*md
,
598 sector_t sector
, struct dm_target
*ti
)
600 sector_t offset
= sector
- ti
->begin
;
601 sector_t len
= ti
->len
- offset
;
604 * Does the target need to split even further ?
608 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
617 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
618 struct dm_target_io
*tio
)
622 struct mapped_device
*md
;
627 BUG_ON(!clone
->bi_size
);
629 clone
->bi_end_io
= clone_endio
;
630 clone
->bi_private
= tio
;
633 * Map the clone. If r == 0 we don't need to do
634 * anything, the target has assumed ownership of
637 atomic_inc(&tio
->io
->io_count
);
638 sector
= clone
->bi_sector
;
639 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
640 if (r
== DM_MAPIO_REMAPPED
) {
641 /* the bio has been remapped so dispatch it */
643 trace_block_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
644 tio
->io
->bio
->bi_bdev
->bd_dev
,
645 clone
->bi_sector
, sector
);
647 generic_make_request(clone
);
648 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
649 /* error the io and bail out, or requeue it if needed */
651 dec_pending(tio
->io
, r
);
653 * Store bio_set for cleanup.
655 clone
->bi_private
= md
->bs
;
659 DMWARN("unimplemented target map return value: %d", r
);
665 struct mapped_device
*md
;
666 struct dm_table
*map
;
670 sector_t sector_count
;
674 static void dm_bio_destructor(struct bio
*bio
)
676 struct bio_set
*bs
= bio
->bi_private
;
682 * Creates a little bio that is just does part of a bvec.
684 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
685 unsigned short idx
, unsigned int offset
,
686 unsigned int len
, struct bio_set
*bs
)
689 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
691 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
692 clone
->bi_destructor
= dm_bio_destructor
;
693 *clone
->bi_io_vec
= *bv
;
695 clone
->bi_sector
= sector
;
696 clone
->bi_bdev
= bio
->bi_bdev
;
697 clone
->bi_rw
= bio
->bi_rw
;
699 clone
->bi_size
= to_bytes(len
);
700 clone
->bi_io_vec
->bv_offset
= offset
;
701 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
702 clone
->bi_flags
|= 1 << BIO_CLONED
;
708 * Creates a bio that consists of range of complete bvecs.
710 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
711 unsigned short idx
, unsigned short bv_count
,
712 unsigned int len
, struct bio_set
*bs
)
716 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
717 __bio_clone(clone
, bio
);
718 clone
->bi_destructor
= dm_bio_destructor
;
719 clone
->bi_sector
= sector
;
721 clone
->bi_vcnt
= idx
+ bv_count
;
722 clone
->bi_size
= to_bytes(len
);
723 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
728 static int __clone_and_map(struct clone_info
*ci
)
730 struct bio
*clone
, *bio
= ci
->bio
;
731 struct dm_target
*ti
;
732 sector_t len
= 0, max
;
733 struct dm_target_io
*tio
;
735 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
736 if (!dm_target_is_valid(ti
))
739 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
742 * Allocate a target io object.
744 tio
= alloc_tio(ci
->md
);
747 memset(&tio
->info
, 0, sizeof(tio
->info
));
749 if (ci
->sector_count
<= max
) {
751 * Optimise for the simple case where we can do all of
752 * the remaining io with a single clone.
754 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
755 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
757 __map_bio(ti
, clone
, tio
);
758 ci
->sector_count
= 0;
760 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
762 * There are some bvecs that don't span targets.
763 * Do as many of these as possible.
766 sector_t remaining
= max
;
769 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
770 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
772 if (bv_len
> remaining
)
779 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
781 __map_bio(ti
, clone
, tio
);
784 ci
->sector_count
-= len
;
789 * Handle a bvec that must be split between two or more targets.
791 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
792 sector_t remaining
= to_sector(bv
->bv_len
);
793 unsigned int offset
= 0;
797 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
798 if (!dm_target_is_valid(ti
))
801 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
803 tio
= alloc_tio(ci
->md
);
806 memset(&tio
->info
, 0, sizeof(tio
->info
));
809 len
= min(remaining
, max
);
811 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
812 bv
->bv_offset
+ offset
, len
,
815 __map_bio(ti
, clone
, tio
);
818 ci
->sector_count
-= len
;
819 offset
+= to_bytes(len
);
820 } while (remaining
-= len
);
829 * Split the bio into several clones and submit it to targets.
831 static int __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
833 struct clone_info ci
;
836 ci
.map
= dm_get_table(md
);
837 if (unlikely(!ci
.map
))
839 if (unlikely(bio_barrier(bio
) && !dm_table_barrier_ok(ci
.map
))) {
840 dm_table_put(ci
.map
);
841 bio_endio(bio
, -EOPNOTSUPP
);
846 ci
.io
= alloc_io(md
);
848 atomic_set(&ci
.io
->io_count
, 1);
851 ci
.sector
= bio
->bi_sector
;
852 ci
.sector_count
= bio_sectors(bio
);
853 ci
.idx
= bio
->bi_idx
;
855 start_io_acct(ci
.io
);
856 while (ci
.sector_count
&& !error
)
857 error
= __clone_and_map(&ci
);
859 /* drop the extra reference count */
860 dec_pending(ci
.io
, error
);
861 dm_table_put(ci
.map
);
865 /*-----------------------------------------------------------------
867 *---------------------------------------------------------------*/
869 static int dm_merge_bvec(struct request_queue
*q
,
870 struct bvec_merge_data
*bvm
,
871 struct bio_vec
*biovec
)
873 struct mapped_device
*md
= q
->queuedata
;
874 struct dm_table
*map
= dm_get_table(md
);
875 struct dm_target
*ti
;
876 sector_t max_sectors
;
882 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
883 if (!dm_target_is_valid(ti
))
887 * Find maximum amount of I/O that won't need splitting
889 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
890 (sector_t
) BIO_MAX_SECTORS
);
891 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
896 * merge_bvec_fn() returns number of bytes
897 * it can accept at this offset
898 * max is precomputed maximal io size
900 if (max_size
&& ti
->type
->merge
)
901 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
908 * Always allow an entire first page
910 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
911 max_size
= biovec
->bv_len
;
917 * The request function that just remaps the bio built up by
920 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
923 int rw
= bio_data_dir(bio
);
924 struct mapped_device
*md
= q
->queuedata
;
927 down_read(&md
->io_lock
);
929 cpu
= part_stat_lock();
930 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
931 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
935 * If we're suspended we have to queue
938 while (test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
939 up_read(&md
->io_lock
);
941 if (bio_rw(bio
) != READA
)
942 r
= queue_io(md
, bio
);
948 * We're in a while loop, because someone could suspend
949 * before we get to the following read lock.
951 down_read(&md
->io_lock
);
954 r
= __split_and_process_bio(md
, bio
);
955 up_read(&md
->io_lock
);
964 static void dm_unplug_all(struct request_queue
*q
)
966 struct mapped_device
*md
= q
->queuedata
;
967 struct dm_table
*map
= dm_get_table(md
);
970 dm_table_unplug_all(map
);
975 static int dm_any_congested(void *congested_data
, int bdi_bits
)
978 struct mapped_device
*md
= congested_data
;
979 struct dm_table
*map
;
981 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
982 map
= dm_get_table(md
);
984 r
= dm_table_any_congested(map
, bdi_bits
);
992 /*-----------------------------------------------------------------
993 * An IDR is used to keep track of allocated minor numbers.
994 *---------------------------------------------------------------*/
995 static DEFINE_IDR(_minor_idr
);
997 static void free_minor(int minor
)
999 spin_lock(&_minor_lock
);
1000 idr_remove(&_minor_idr
, minor
);
1001 spin_unlock(&_minor_lock
);
1005 * See if the device with a specific minor # is free.
1007 static int specific_minor(int minor
)
1011 if (minor
>= (1 << MINORBITS
))
1014 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1018 spin_lock(&_minor_lock
);
1020 if (idr_find(&_minor_idr
, minor
)) {
1025 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1030 idr_remove(&_minor_idr
, m
);
1036 spin_unlock(&_minor_lock
);
1040 static int next_free_minor(int *minor
)
1044 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1048 spin_lock(&_minor_lock
);
1050 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1054 if (m
>= (1 << MINORBITS
)) {
1055 idr_remove(&_minor_idr
, m
);
1063 spin_unlock(&_minor_lock
);
1067 static struct block_device_operations dm_blk_dops
;
1069 static void dm_wq_work(struct work_struct
*work
);
1072 * Allocate and initialise a blank device with a given minor.
1074 static struct mapped_device
*alloc_dev(int minor
)
1077 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1081 DMWARN("unable to allocate device, out of memory.");
1085 if (!try_module_get(THIS_MODULE
))
1086 goto bad_module_get
;
1088 /* get a minor number for the dev */
1089 if (minor
== DM_ANY_MINOR
)
1090 r
= next_free_minor(&minor
);
1092 r
= specific_minor(minor
);
1096 init_rwsem(&md
->io_lock
);
1097 mutex_init(&md
->suspend_lock
);
1098 spin_lock_init(&md
->pushback_lock
);
1099 rwlock_init(&md
->map_lock
);
1100 atomic_set(&md
->holders
, 1);
1101 atomic_set(&md
->open_count
, 0);
1102 atomic_set(&md
->event_nr
, 0);
1103 atomic_set(&md
->uevent_seq
, 0);
1104 INIT_LIST_HEAD(&md
->uevent_list
);
1105 spin_lock_init(&md
->uevent_lock
);
1107 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1111 md
->queue
->queuedata
= md
;
1112 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1113 md
->queue
->backing_dev_info
.congested_data
= md
;
1114 blk_queue_make_request(md
->queue
, dm_request
);
1115 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1116 md
->queue
->unplug_fn
= dm_unplug_all
;
1117 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1119 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1123 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1127 md
->bs
= bioset_create(16, 0);
1131 md
->disk
= alloc_disk(1);
1135 atomic_set(&md
->pending
, 0);
1136 init_waitqueue_head(&md
->wait
);
1137 INIT_WORK(&md
->work
, dm_wq_work
);
1138 init_waitqueue_head(&md
->eventq
);
1140 md
->disk
->major
= _major
;
1141 md
->disk
->first_minor
= minor
;
1142 md
->disk
->fops
= &dm_blk_dops
;
1143 md
->disk
->queue
= md
->queue
;
1144 md
->disk
->private_data
= md
;
1145 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1147 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1149 md
->wq
= create_singlethread_workqueue("kdmflush");
1153 /* Populate the mapping, nobody knows we exist yet */
1154 spin_lock(&_minor_lock
);
1155 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1156 spin_unlock(&_minor_lock
);
1158 BUG_ON(old_md
!= MINOR_ALLOCED
);
1165 bioset_free(md
->bs
);
1167 mempool_destroy(md
->tio_pool
);
1169 mempool_destroy(md
->io_pool
);
1171 blk_cleanup_queue(md
->queue
);
1175 module_put(THIS_MODULE
);
1181 static void unlock_fs(struct mapped_device
*md
);
1183 static void free_dev(struct mapped_device
*md
)
1185 int minor
= MINOR(disk_devt(md
->disk
));
1187 if (md
->suspended_bdev
) {
1189 bdput(md
->suspended_bdev
);
1191 destroy_workqueue(md
->wq
);
1192 mempool_destroy(md
->tio_pool
);
1193 mempool_destroy(md
->io_pool
);
1194 bioset_free(md
->bs
);
1195 del_gendisk(md
->disk
);
1198 spin_lock(&_minor_lock
);
1199 md
->disk
->private_data
= NULL
;
1200 spin_unlock(&_minor_lock
);
1203 blk_cleanup_queue(md
->queue
);
1204 module_put(THIS_MODULE
);
1209 * Bind a table to the device.
1211 static void event_callback(void *context
)
1213 unsigned long flags
;
1215 struct mapped_device
*md
= (struct mapped_device
*) context
;
1217 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1218 list_splice_init(&md
->uevent_list
, &uevents
);
1219 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1221 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1223 atomic_inc(&md
->event_nr
);
1224 wake_up(&md
->eventq
);
1227 static void __set_size(struct mapped_device
*md
, sector_t size
)
1229 set_capacity(md
->disk
, size
);
1231 mutex_lock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1232 i_size_write(md
->suspended_bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1233 mutex_unlock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1236 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1238 struct request_queue
*q
= md
->queue
;
1241 size
= dm_table_get_size(t
);
1244 * Wipe any geometry if the size of the table changed.
1246 if (size
!= get_capacity(md
->disk
))
1247 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1249 if (md
->suspended_bdev
)
1250 __set_size(md
, size
);
1253 dm_table_destroy(t
);
1257 dm_table_event_callback(t
, event_callback
, md
);
1259 write_lock(&md
->map_lock
);
1261 dm_table_set_restrictions(t
, q
);
1262 write_unlock(&md
->map_lock
);
1267 static void __unbind(struct mapped_device
*md
)
1269 struct dm_table
*map
= md
->map
;
1274 dm_table_event_callback(map
, NULL
, NULL
);
1275 write_lock(&md
->map_lock
);
1277 write_unlock(&md
->map_lock
);
1278 dm_table_destroy(map
);
1282 * Constructor for a new device.
1284 int dm_create(int minor
, struct mapped_device
**result
)
1286 struct mapped_device
*md
;
1288 md
= alloc_dev(minor
);
1298 static struct mapped_device
*dm_find_md(dev_t dev
)
1300 struct mapped_device
*md
;
1301 unsigned minor
= MINOR(dev
);
1303 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1306 spin_lock(&_minor_lock
);
1308 md
= idr_find(&_minor_idr
, minor
);
1309 if (md
&& (md
== MINOR_ALLOCED
||
1310 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1311 test_bit(DMF_FREEING
, &md
->flags
))) {
1317 spin_unlock(&_minor_lock
);
1322 struct mapped_device
*dm_get_md(dev_t dev
)
1324 struct mapped_device
*md
= dm_find_md(dev
);
1332 void *dm_get_mdptr(struct mapped_device
*md
)
1334 return md
->interface_ptr
;
1337 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1339 md
->interface_ptr
= ptr
;
1342 void dm_get(struct mapped_device
*md
)
1344 atomic_inc(&md
->holders
);
1347 const char *dm_device_name(struct mapped_device
*md
)
1351 EXPORT_SYMBOL_GPL(dm_device_name
);
1353 void dm_put(struct mapped_device
*md
)
1355 struct dm_table
*map
;
1357 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1359 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1360 map
= dm_get_table(md
);
1361 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
1362 MINOR(disk_devt(dm_disk(md
))));
1363 set_bit(DMF_FREEING
, &md
->flags
);
1364 spin_unlock(&_minor_lock
);
1365 if (!dm_suspended(md
)) {
1366 dm_table_presuspend_targets(map
);
1367 dm_table_postsuspend_targets(map
);
1375 EXPORT_SYMBOL_GPL(dm_put
);
1377 static int dm_wait_for_completion(struct mapped_device
*md
)
1382 set_current_state(TASK_INTERRUPTIBLE
);
1385 if (!atomic_read(&md
->pending
))
1388 if (signal_pending(current
)) {
1395 set_current_state(TASK_RUNNING
);
1401 * Process the deferred bios
1403 static void __flush_deferred_io(struct mapped_device
*md
)
1407 while ((c
= bio_list_pop(&md
->deferred
))) {
1408 if (__split_and_process_bio(md
, c
))
1412 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1415 static void __merge_pushback_list(struct mapped_device
*md
)
1417 unsigned long flags
;
1419 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1420 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1421 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1422 bio_list_init(&md
->pushback
);
1423 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1426 static void dm_wq_work(struct work_struct
*work
)
1428 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
1431 down_write(&md
->io_lock
);
1432 __flush_deferred_io(md
);
1433 up_write(&md
->io_lock
);
1436 static void dm_queue_flush(struct mapped_device
*md
)
1438 queue_work(md
->wq
, &md
->work
);
1439 flush_workqueue(md
->wq
);
1443 * Swap in a new table (destroying old one).
1445 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1449 mutex_lock(&md
->suspend_lock
);
1451 /* device must be suspended */
1452 if (!dm_suspended(md
))
1455 /* without bdev, the device size cannot be changed */
1456 if (!md
->suspended_bdev
)
1457 if (get_capacity(md
->disk
) != dm_table_get_size(table
))
1461 r
= __bind(md
, table
);
1464 mutex_unlock(&md
->suspend_lock
);
1469 * Functions to lock and unlock any filesystem running on the
1472 static int lock_fs(struct mapped_device
*md
)
1476 WARN_ON(md
->frozen_sb
);
1478 md
->frozen_sb
= freeze_bdev(md
->suspended_bdev
);
1479 if (IS_ERR(md
->frozen_sb
)) {
1480 r
= PTR_ERR(md
->frozen_sb
);
1481 md
->frozen_sb
= NULL
;
1485 set_bit(DMF_FROZEN
, &md
->flags
);
1487 /* don't bdput right now, we don't want the bdev
1488 * to go away while it is locked.
1493 static void unlock_fs(struct mapped_device
*md
)
1495 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1498 thaw_bdev(md
->suspended_bdev
, md
->frozen_sb
);
1499 md
->frozen_sb
= NULL
;
1500 clear_bit(DMF_FROZEN
, &md
->flags
);
1504 * We need to be able to change a mapping table under a mounted
1505 * filesystem. For example we might want to move some data in
1506 * the background. Before the table can be swapped with
1507 * dm_bind_table, dm_suspend must be called to flush any in
1508 * flight bios and ensure that any further io gets deferred.
1510 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1512 struct dm_table
*map
= NULL
;
1513 DECLARE_WAITQUEUE(wait
, current
);
1515 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1516 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1518 mutex_lock(&md
->suspend_lock
);
1520 if (dm_suspended(md
)) {
1525 map
= dm_get_table(md
);
1528 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1529 * This flag is cleared before dm_suspend returns.
1532 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1534 /* This does not get reverted if there's an error later. */
1535 dm_table_presuspend_targets(map
);
1537 /* bdget() can stall if the pending I/Os are not flushed */
1539 md
->suspended_bdev
= bdget_disk(md
->disk
, 0);
1540 if (!md
->suspended_bdev
) {
1541 DMWARN("bdget failed in dm_suspend");
1547 * Flush I/O to the device. noflush supersedes do_lockfs,
1548 * because lock_fs() needs to flush I/Os.
1558 * First we set the BLOCK_IO flag so no more ios will be mapped.
1560 down_write(&md
->io_lock
);
1561 set_bit(DMF_BLOCK_IO
, &md
->flags
);
1563 add_wait_queue(&md
->wait
, &wait
);
1564 up_write(&md
->io_lock
);
1568 dm_table_unplug_all(map
);
1571 * Wait for the already-mapped ios to complete.
1573 r
= dm_wait_for_completion(md
);
1575 down_write(&md
->io_lock
);
1576 remove_wait_queue(&md
->wait
, &wait
);
1579 __merge_pushback_list(md
);
1580 up_write(&md
->io_lock
);
1582 /* were we interrupted ? */
1587 goto out
; /* pushback list is already flushed, so skip flush */
1590 dm_table_postsuspend_targets(map
);
1592 set_bit(DMF_SUSPENDED
, &md
->flags
);
1595 if (r
&& md
->suspended_bdev
) {
1596 bdput(md
->suspended_bdev
);
1597 md
->suspended_bdev
= NULL
;
1603 mutex_unlock(&md
->suspend_lock
);
1607 int dm_resume(struct mapped_device
*md
)
1610 struct dm_table
*map
= NULL
;
1612 mutex_lock(&md
->suspend_lock
);
1613 if (!dm_suspended(md
))
1616 map
= dm_get_table(md
);
1617 if (!map
|| !dm_table_get_size(map
))
1620 r
= dm_table_resume_targets(map
);
1628 if (md
->suspended_bdev
) {
1629 bdput(md
->suspended_bdev
);
1630 md
->suspended_bdev
= NULL
;
1633 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1635 dm_table_unplug_all(map
);
1637 dm_kobject_uevent(md
);
1643 mutex_unlock(&md
->suspend_lock
);
1648 /*-----------------------------------------------------------------
1649 * Event notification.
1650 *---------------------------------------------------------------*/
1651 void dm_kobject_uevent(struct mapped_device
*md
)
1653 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, KOBJ_CHANGE
);
1656 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
1658 return atomic_add_return(1, &md
->uevent_seq
);
1661 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1663 return atomic_read(&md
->event_nr
);
1666 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1668 return wait_event_interruptible(md
->eventq
,
1669 (event_nr
!= atomic_read(&md
->event_nr
)));
1672 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
1674 unsigned long flags
;
1676 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1677 list_add(elist
, &md
->uevent_list
);
1678 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1682 * The gendisk is only valid as long as you have a reference
1685 struct gendisk
*dm_disk(struct mapped_device
*md
)
1690 struct kobject
*dm_kobject(struct mapped_device
*md
)
1696 * struct mapped_device should not be exported outside of dm.c
1697 * so use this check to verify that kobj is part of md structure
1699 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
1701 struct mapped_device
*md
;
1703 md
= container_of(kobj
, struct mapped_device
, kobj
);
1704 if (&md
->kobj
!= kobj
)
1711 int dm_suspended(struct mapped_device
*md
)
1713 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1716 int dm_noflush_suspending(struct dm_target
*ti
)
1718 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1719 int r
= __noflush_suspending(md
);
1725 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1727 static struct block_device_operations dm_blk_dops
= {
1728 .open
= dm_blk_open
,
1729 .release
= dm_blk_close
,
1730 .ioctl
= dm_blk_ioctl
,
1731 .getgeo
= dm_blk_getgeo
,
1732 .owner
= THIS_MODULE
1735 EXPORT_SYMBOL(dm_get_mapinfo
);
1740 module_init(dm_init
);
1741 module_exit(dm_exit
);
1743 module_param(major
, uint
, 0);
1744 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1745 MODULE_DESCRIPTION(DM_NAME
" driver");
1746 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1747 MODULE_LICENSE("GPL");