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 <linux/smp_lock.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 union map_info
*dm_get_mapinfo(struct bio
*bio
)
57 if (bio
&& bio
->bi_private
)
58 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
62 #define MINOR_ALLOCED ((void *)-1)
65 * Bits for the md->flags field.
67 #define DMF_BLOCK_IO 0
68 #define DMF_SUSPENDED 1
71 #define DMF_DELETING 4
72 #define DMF_NOFLUSH_SUSPENDING 5
75 * Work processed by per-device workqueue.
82 struct work_struct work
;
83 struct mapped_device
*md
;
87 struct mapped_device
{
88 struct rw_semaphore io_lock
;
89 struct mutex suspend_lock
;
90 spinlock_t pushback_lock
;
97 struct request_queue
*queue
;
104 * A list of ios that arrived while we were suspended.
107 wait_queue_head_t wait
;
108 struct bio_list deferred
;
109 struct bio_list pushback
;
112 * Processing queue (flush/barriers)
114 struct workqueue_struct
*wq
;
117 * The current mapping.
119 struct dm_table
*map
;
122 * io objects are allocated from here.
133 wait_queue_head_t eventq
;
135 struct list_head uevent_list
;
136 spinlock_t uevent_lock
; /* Protect access to uevent_list */
139 * freeze/thaw support require holding onto a super block
141 struct super_block
*frozen_sb
;
142 struct block_device
*suspended_bdev
;
144 /* forced geometry settings */
145 struct hd_geometry geometry
;
149 static struct kmem_cache
*_io_cache
;
150 static struct kmem_cache
*_tio_cache
;
152 static int __init
local_init(void)
156 /* allocate a slab for the dm_ios */
157 _io_cache
= KMEM_CACHE(dm_io
, 0);
161 /* allocate a slab for the target ios */
162 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
164 kmem_cache_destroy(_io_cache
);
168 r
= dm_uevent_init();
170 kmem_cache_destroy(_tio_cache
);
171 kmem_cache_destroy(_io_cache
);
176 r
= register_blkdev(_major
, _name
);
178 kmem_cache_destroy(_tio_cache
);
179 kmem_cache_destroy(_io_cache
);
190 static void local_exit(void)
192 kmem_cache_destroy(_tio_cache
);
193 kmem_cache_destroy(_io_cache
);
194 unregister_blkdev(_major
, _name
);
199 DMINFO("cleaned up");
202 static int (*_inits
[])(void) __initdata
= {
211 static void (*_exits
[])(void) = {
220 static int __init
dm_init(void)
222 const int count
= ARRAY_SIZE(_inits
);
226 for (i
= 0; i
< count
; i
++) {
241 static void __exit
dm_exit(void)
243 int i
= ARRAY_SIZE(_exits
);
250 * Block device functions
252 static int dm_blk_open(struct inode
*inode
, struct file
*file
)
254 struct mapped_device
*md
;
256 spin_lock(&_minor_lock
);
258 md
= inode
->i_bdev
->bd_disk
->private_data
;
262 if (test_bit(DMF_FREEING
, &md
->flags
) ||
263 test_bit(DMF_DELETING
, &md
->flags
)) {
269 atomic_inc(&md
->open_count
);
272 spin_unlock(&_minor_lock
);
274 return md
? 0 : -ENXIO
;
277 static int dm_blk_close(struct inode
*inode
, struct file
*file
)
279 struct mapped_device
*md
;
281 md
= inode
->i_bdev
->bd_disk
->private_data
;
282 atomic_dec(&md
->open_count
);
287 int dm_open_count(struct mapped_device
*md
)
289 return atomic_read(&md
->open_count
);
293 * Guarantees nothing is using the device before it's deleted.
295 int dm_lock_for_deletion(struct mapped_device
*md
)
299 spin_lock(&_minor_lock
);
301 if (dm_open_count(md
))
304 set_bit(DMF_DELETING
, &md
->flags
);
306 spin_unlock(&_minor_lock
);
311 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
313 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
315 return dm_get_geometry(md
, geo
);
318 static int dm_blk_ioctl(struct inode
*inode
, struct file
*file
,
319 unsigned int cmd
, unsigned long arg
)
321 struct mapped_device
*md
;
322 struct dm_table
*map
;
323 struct dm_target
*tgt
;
326 /* We don't really need this lock, but we do need 'inode'. */
329 md
= inode
->i_bdev
->bd_disk
->private_data
;
331 map
= dm_get_table(md
);
333 if (!map
|| !dm_table_get_size(map
))
336 /* We only support devices that have a single target */
337 if (dm_table_get_num_targets(map
) != 1)
340 tgt
= dm_table_get_target(map
, 0);
342 if (dm_suspended(md
)) {
347 if (tgt
->type
->ioctl
)
348 r
= tgt
->type
->ioctl(tgt
, inode
, file
, cmd
, arg
);
357 static struct dm_io
*alloc_io(struct mapped_device
*md
)
359 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
362 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
364 mempool_free(io
, md
->io_pool
);
367 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
369 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
372 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
374 mempool_free(tio
, md
->tio_pool
);
377 static void start_io_acct(struct dm_io
*io
)
379 struct mapped_device
*md
= io
->md
;
382 io
->start_time
= jiffies
;
384 cpu
= part_stat_lock();
385 part_round_stats(cpu
, &dm_disk(md
)->part0
);
387 dm_disk(md
)->part0
.in_flight
= atomic_inc_return(&md
->pending
);
390 static int end_io_acct(struct dm_io
*io
)
392 struct mapped_device
*md
= io
->md
;
393 struct bio
*bio
= io
->bio
;
394 unsigned long duration
= jiffies
- io
->start_time
;
396 int rw
= bio_data_dir(bio
);
398 cpu
= part_stat_lock();
399 part_round_stats(cpu
, &dm_disk(md
)->part0
);
400 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
403 dm_disk(md
)->part0
.in_flight
= pending
=
404 atomic_dec_return(&md
->pending
);
410 * Add the bio to the list of deferred io.
412 static int queue_io(struct mapped_device
*md
, struct bio
*bio
)
414 down_write(&md
->io_lock
);
416 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
417 up_write(&md
->io_lock
);
421 bio_list_add(&md
->deferred
, bio
);
423 up_write(&md
->io_lock
);
424 return 0; /* deferred successfully */
428 * Everyone (including functions in this file), should use this
429 * function to access the md->map field, and make sure they call
430 * dm_table_put() when finished.
432 struct dm_table
*dm_get_table(struct mapped_device
*md
)
436 read_lock(&md
->map_lock
);
440 read_unlock(&md
->map_lock
);
446 * Get the geometry associated with a dm device
448 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
456 * Set the geometry of a device.
458 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
460 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
462 if (geo
->start
> sz
) {
463 DMWARN("Start sector is beyond the geometry limits.");
472 /*-----------------------------------------------------------------
474 * A more elegant soln is in the works that uses the queue
475 * merge fn, unfortunately there are a couple of changes to
476 * the block layer that I want to make for this. So in the
477 * interests of getting something for people to use I give
478 * you this clearly demarcated crap.
479 *---------------------------------------------------------------*/
481 static int __noflush_suspending(struct mapped_device
*md
)
483 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
487 * Decrements the number of outstanding ios that a bio has been
488 * cloned into, completing the original io if necc.
490 static void dec_pending(struct dm_io
*io
, int error
)
494 /* Push-back supersedes any I/O errors */
495 if (error
&& !(io
->error
> 0 && __noflush_suspending(io
->md
)))
498 if (atomic_dec_and_test(&io
->io_count
)) {
499 if (io
->error
== DM_ENDIO_REQUEUE
) {
501 * Target requested pushing back the I/O.
502 * This must be handled before the sleeper on
503 * suspend queue merges the pushback list.
505 spin_lock_irqsave(&io
->md
->pushback_lock
, flags
);
506 if (__noflush_suspending(io
->md
))
507 bio_list_add(&io
->md
->pushback
, io
->bio
);
509 /* noflush suspend was interrupted. */
511 spin_unlock_irqrestore(&io
->md
->pushback_lock
, flags
);
515 /* nudge anyone waiting on suspend queue */
516 wake_up(&io
->md
->wait
);
518 if (io
->error
!= DM_ENDIO_REQUEUE
) {
519 blk_add_trace_bio(io
->md
->queue
, io
->bio
,
522 bio_endio(io
->bio
, io
->error
);
529 static void clone_endio(struct bio
*bio
, int error
)
532 struct dm_target_io
*tio
= bio
->bi_private
;
533 struct mapped_device
*md
= tio
->io
->md
;
534 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
536 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
540 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
541 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
543 * error and requeue request are handled
547 else if (r
== DM_ENDIO_INCOMPLETE
)
548 /* The target will handle the io */
551 DMWARN("unimplemented target endio return value: %d", r
);
556 dec_pending(tio
->io
, error
);
559 * Store md for cleanup instead of tio which is about to get freed.
561 bio
->bi_private
= md
->bs
;
567 static sector_t
max_io_len(struct mapped_device
*md
,
568 sector_t sector
, struct dm_target
*ti
)
570 sector_t offset
= sector
- ti
->begin
;
571 sector_t len
= ti
->len
- offset
;
574 * Does the target need to split even further ?
578 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
587 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
588 struct dm_target_io
*tio
)
592 struct mapped_device
*md
;
597 BUG_ON(!clone
->bi_size
);
599 clone
->bi_end_io
= clone_endio
;
600 clone
->bi_private
= tio
;
603 * Map the clone. If r == 0 we don't need to do
604 * anything, the target has assumed ownership of
607 atomic_inc(&tio
->io
->io_count
);
608 sector
= clone
->bi_sector
;
609 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
610 if (r
== DM_MAPIO_REMAPPED
) {
611 /* the bio has been remapped so dispatch it */
613 blk_add_trace_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
614 tio
->io
->bio
->bi_bdev
->bd_dev
,
615 clone
->bi_sector
, sector
);
617 generic_make_request(clone
);
618 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
619 /* error the io and bail out, or requeue it if needed */
621 dec_pending(tio
->io
, r
);
623 * Store bio_set for cleanup.
625 clone
->bi_private
= md
->bs
;
629 DMWARN("unimplemented target map return value: %d", r
);
635 struct mapped_device
*md
;
636 struct dm_table
*map
;
640 sector_t sector_count
;
644 static void dm_bio_destructor(struct bio
*bio
)
646 struct bio_set
*bs
= bio
->bi_private
;
652 * Creates a little bio that is just does part of a bvec.
654 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
655 unsigned short idx
, unsigned int offset
,
656 unsigned int len
, struct bio_set
*bs
)
659 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
661 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
662 clone
->bi_destructor
= dm_bio_destructor
;
663 *clone
->bi_io_vec
= *bv
;
665 clone
->bi_sector
= sector
;
666 clone
->bi_bdev
= bio
->bi_bdev
;
667 clone
->bi_rw
= bio
->bi_rw
;
669 clone
->bi_size
= to_bytes(len
);
670 clone
->bi_io_vec
->bv_offset
= offset
;
671 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
672 clone
->bi_flags
|= 1 << BIO_CLONED
;
678 * Creates a bio that consists of range of complete bvecs.
680 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
681 unsigned short idx
, unsigned short bv_count
,
682 unsigned int len
, struct bio_set
*bs
)
686 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
687 __bio_clone(clone
, bio
);
688 clone
->bi_destructor
= dm_bio_destructor
;
689 clone
->bi_sector
= sector
;
691 clone
->bi_vcnt
= idx
+ bv_count
;
692 clone
->bi_size
= to_bytes(len
);
693 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
698 static int __clone_and_map(struct clone_info
*ci
)
700 struct bio
*clone
, *bio
= ci
->bio
;
701 struct dm_target
*ti
;
702 sector_t len
= 0, max
;
703 struct dm_target_io
*tio
;
705 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
706 if (!dm_target_is_valid(ti
))
709 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
712 * Allocate a target io object.
714 tio
= alloc_tio(ci
->md
);
717 memset(&tio
->info
, 0, sizeof(tio
->info
));
719 if (ci
->sector_count
<= max
) {
721 * Optimise for the simple case where we can do all of
722 * the remaining io with a single clone.
724 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
725 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
727 __map_bio(ti
, clone
, tio
);
728 ci
->sector_count
= 0;
730 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
732 * There are some bvecs that don't span targets.
733 * Do as many of these as possible.
736 sector_t remaining
= max
;
739 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
740 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
742 if (bv_len
> remaining
)
749 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
751 __map_bio(ti
, clone
, tio
);
754 ci
->sector_count
-= len
;
759 * Handle a bvec that must be split between two or more targets.
761 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
762 sector_t remaining
= to_sector(bv
->bv_len
);
763 unsigned int offset
= 0;
767 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
768 if (!dm_target_is_valid(ti
))
771 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
773 tio
= alloc_tio(ci
->md
);
776 memset(&tio
->info
, 0, sizeof(tio
->info
));
779 len
= min(remaining
, max
);
781 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
782 bv
->bv_offset
+ offset
, len
,
785 __map_bio(ti
, clone
, tio
);
788 ci
->sector_count
-= len
;
789 offset
+= to_bytes(len
);
790 } while (remaining
-= len
);
799 * Split the bio into several clones.
801 static int __split_bio(struct mapped_device
*md
, struct bio
*bio
)
803 struct clone_info ci
;
806 ci
.map
= dm_get_table(md
);
807 if (unlikely(!ci
.map
))
812 ci
.io
= alloc_io(md
);
814 atomic_set(&ci
.io
->io_count
, 1);
817 ci
.sector
= bio
->bi_sector
;
818 ci
.sector_count
= bio_sectors(bio
);
819 ci
.idx
= bio
->bi_idx
;
821 start_io_acct(ci
.io
);
822 while (ci
.sector_count
&& !error
)
823 error
= __clone_and_map(&ci
);
825 /* drop the extra reference count */
826 dec_pending(ci
.io
, error
);
827 dm_table_put(ci
.map
);
831 /*-----------------------------------------------------------------
833 *---------------------------------------------------------------*/
835 static int dm_merge_bvec(struct request_queue
*q
,
836 struct bvec_merge_data
*bvm
,
837 struct bio_vec
*biovec
)
839 struct mapped_device
*md
= q
->queuedata
;
840 struct dm_table
*map
= dm_get_table(md
);
841 struct dm_target
*ti
;
842 sector_t max_sectors
;
848 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
849 if (!dm_target_is_valid(ti
))
853 * Find maximum amount of I/O that won't need splitting
855 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
856 (sector_t
) BIO_MAX_SECTORS
);
857 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
862 * merge_bvec_fn() returns number of bytes
863 * it can accept at this offset
864 * max is precomputed maximal io size
866 if (max_size
&& ti
->type
->merge
)
867 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
874 * Always allow an entire first page
876 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
877 max_size
= biovec
->bv_len
;
883 * The request function that just remaps the bio built up by
886 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
889 int rw
= bio_data_dir(bio
);
890 struct mapped_device
*md
= q
->queuedata
;
894 * There is no use in forwarding any barrier request since we can't
895 * guarantee it is (or can be) handled by the targets correctly.
897 if (unlikely(bio_barrier(bio
))) {
898 bio_endio(bio
, -EOPNOTSUPP
);
902 down_read(&md
->io_lock
);
904 cpu
= part_stat_lock();
905 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
906 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
910 * If we're suspended we have to queue
913 while (test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
914 up_read(&md
->io_lock
);
916 if (bio_rw(bio
) != READA
)
917 r
= queue_io(md
, bio
);
923 * We're in a while loop, because someone could suspend
924 * before we get to the following read lock.
926 down_read(&md
->io_lock
);
929 r
= __split_bio(md
, bio
);
930 up_read(&md
->io_lock
);
939 static void dm_unplug_all(struct request_queue
*q
)
941 struct mapped_device
*md
= q
->queuedata
;
942 struct dm_table
*map
= dm_get_table(md
);
945 dm_table_unplug_all(map
);
950 static int dm_any_congested(void *congested_data
, int bdi_bits
)
953 struct mapped_device
*md
= (struct mapped_device
*) congested_data
;
954 struct dm_table
*map
= dm_get_table(md
);
956 if (!map
|| test_bit(DMF_BLOCK_IO
, &md
->flags
))
959 r
= dm_table_any_congested(map
, bdi_bits
);
965 /*-----------------------------------------------------------------
966 * An IDR is used to keep track of allocated minor numbers.
967 *---------------------------------------------------------------*/
968 static DEFINE_IDR(_minor_idr
);
970 static void free_minor(int minor
)
972 spin_lock(&_minor_lock
);
973 idr_remove(&_minor_idr
, minor
);
974 spin_unlock(&_minor_lock
);
978 * See if the device with a specific minor # is free.
980 static int specific_minor(int minor
)
984 if (minor
>= (1 << MINORBITS
))
987 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
991 spin_lock(&_minor_lock
);
993 if (idr_find(&_minor_idr
, minor
)) {
998 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1003 idr_remove(&_minor_idr
, m
);
1009 spin_unlock(&_minor_lock
);
1013 static int next_free_minor(int *minor
)
1017 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1021 spin_lock(&_minor_lock
);
1023 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1027 if (m
>= (1 << MINORBITS
)) {
1028 idr_remove(&_minor_idr
, m
);
1036 spin_unlock(&_minor_lock
);
1040 static struct block_device_operations dm_blk_dops
;
1043 * Allocate and initialise a blank device with a given minor.
1045 static struct mapped_device
*alloc_dev(int minor
)
1048 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1052 DMWARN("unable to allocate device, out of memory.");
1056 if (!try_module_get(THIS_MODULE
))
1057 goto bad_module_get
;
1059 /* get a minor number for the dev */
1060 if (minor
== DM_ANY_MINOR
)
1061 r
= next_free_minor(&minor
);
1063 r
= specific_minor(minor
);
1067 init_rwsem(&md
->io_lock
);
1068 mutex_init(&md
->suspend_lock
);
1069 spin_lock_init(&md
->pushback_lock
);
1070 rwlock_init(&md
->map_lock
);
1071 atomic_set(&md
->holders
, 1);
1072 atomic_set(&md
->open_count
, 0);
1073 atomic_set(&md
->event_nr
, 0);
1074 atomic_set(&md
->uevent_seq
, 0);
1075 INIT_LIST_HEAD(&md
->uevent_list
);
1076 spin_lock_init(&md
->uevent_lock
);
1078 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1082 md
->queue
->queuedata
= md
;
1083 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1084 md
->queue
->backing_dev_info
.congested_data
= md
;
1085 blk_queue_make_request(md
->queue
, dm_request
);
1086 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1087 md
->queue
->unplug_fn
= dm_unplug_all
;
1088 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1090 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1094 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1098 md
->bs
= bioset_create(16, 16);
1102 md
->disk
= alloc_disk(1);
1106 atomic_set(&md
->pending
, 0);
1107 init_waitqueue_head(&md
->wait
);
1108 init_waitqueue_head(&md
->eventq
);
1110 md
->disk
->major
= _major
;
1111 md
->disk
->first_minor
= minor
;
1112 md
->disk
->fops
= &dm_blk_dops
;
1113 md
->disk
->queue
= md
->queue
;
1114 md
->disk
->private_data
= md
;
1115 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1117 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1119 md
->wq
= create_singlethread_workqueue("kdmflush");
1123 /* Populate the mapping, nobody knows we exist yet */
1124 spin_lock(&_minor_lock
);
1125 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1126 spin_unlock(&_minor_lock
);
1128 BUG_ON(old_md
!= MINOR_ALLOCED
);
1135 bioset_free(md
->bs
);
1137 mempool_destroy(md
->tio_pool
);
1139 mempool_destroy(md
->io_pool
);
1141 blk_cleanup_queue(md
->queue
);
1145 module_put(THIS_MODULE
);
1151 static void unlock_fs(struct mapped_device
*md
);
1153 static void free_dev(struct mapped_device
*md
)
1155 int minor
= MINOR(disk_devt(md
->disk
));
1157 if (md
->suspended_bdev
) {
1159 bdput(md
->suspended_bdev
);
1161 destroy_workqueue(md
->wq
);
1162 mempool_destroy(md
->tio_pool
);
1163 mempool_destroy(md
->io_pool
);
1164 bioset_free(md
->bs
);
1165 del_gendisk(md
->disk
);
1168 spin_lock(&_minor_lock
);
1169 md
->disk
->private_data
= NULL
;
1170 spin_unlock(&_minor_lock
);
1173 blk_cleanup_queue(md
->queue
);
1174 module_put(THIS_MODULE
);
1179 * Bind a table to the device.
1181 static void event_callback(void *context
)
1183 unsigned long flags
;
1185 struct mapped_device
*md
= (struct mapped_device
*) context
;
1187 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1188 list_splice_init(&md
->uevent_list
, &uevents
);
1189 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1191 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1193 atomic_inc(&md
->event_nr
);
1194 wake_up(&md
->eventq
);
1197 static void __set_size(struct mapped_device
*md
, sector_t size
)
1199 set_capacity(md
->disk
, size
);
1201 mutex_lock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1202 i_size_write(md
->suspended_bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1203 mutex_unlock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1206 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1208 struct request_queue
*q
= md
->queue
;
1211 size
= dm_table_get_size(t
);
1214 * Wipe any geometry if the size of the table changed.
1216 if (size
!= get_capacity(md
->disk
))
1217 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1219 if (md
->suspended_bdev
)
1220 __set_size(md
, size
);
1225 dm_table_event_callback(t
, event_callback
, md
);
1227 write_lock(&md
->map_lock
);
1229 dm_table_set_restrictions(t
, q
);
1230 write_unlock(&md
->map_lock
);
1235 static void __unbind(struct mapped_device
*md
)
1237 struct dm_table
*map
= md
->map
;
1242 dm_table_event_callback(map
, NULL
, NULL
);
1243 write_lock(&md
->map_lock
);
1245 write_unlock(&md
->map_lock
);
1250 * Constructor for a new device.
1252 int dm_create(int minor
, struct mapped_device
**result
)
1254 struct mapped_device
*md
;
1256 md
= alloc_dev(minor
);
1264 static struct mapped_device
*dm_find_md(dev_t dev
)
1266 struct mapped_device
*md
;
1267 unsigned minor
= MINOR(dev
);
1269 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1272 spin_lock(&_minor_lock
);
1274 md
= idr_find(&_minor_idr
, minor
);
1275 if (md
&& (md
== MINOR_ALLOCED
||
1276 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1277 test_bit(DMF_FREEING
, &md
->flags
))) {
1283 spin_unlock(&_minor_lock
);
1288 struct mapped_device
*dm_get_md(dev_t dev
)
1290 struct mapped_device
*md
= dm_find_md(dev
);
1298 void *dm_get_mdptr(struct mapped_device
*md
)
1300 return md
->interface_ptr
;
1303 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1305 md
->interface_ptr
= ptr
;
1308 void dm_get(struct mapped_device
*md
)
1310 atomic_inc(&md
->holders
);
1313 const char *dm_device_name(struct mapped_device
*md
)
1317 EXPORT_SYMBOL_GPL(dm_device_name
);
1319 void dm_put(struct mapped_device
*md
)
1321 struct dm_table
*map
;
1323 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1325 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1326 map
= dm_get_table(md
);
1327 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
1328 MINOR(disk_devt(dm_disk(md
))));
1329 set_bit(DMF_FREEING
, &md
->flags
);
1330 spin_unlock(&_minor_lock
);
1331 if (!dm_suspended(md
)) {
1332 dm_table_presuspend_targets(map
);
1333 dm_table_postsuspend_targets(map
);
1340 EXPORT_SYMBOL_GPL(dm_put
);
1342 static int dm_wait_for_completion(struct mapped_device
*md
)
1347 set_current_state(TASK_INTERRUPTIBLE
);
1350 if (!atomic_read(&md
->pending
))
1353 if (signal_pending(current
)) {
1360 set_current_state(TASK_RUNNING
);
1366 * Process the deferred bios
1368 static void __flush_deferred_io(struct mapped_device
*md
)
1372 while ((c
= bio_list_pop(&md
->deferred
))) {
1373 if (__split_bio(md
, c
))
1377 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1380 static void __merge_pushback_list(struct mapped_device
*md
)
1382 unsigned long flags
;
1384 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1385 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1386 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1387 bio_list_init(&md
->pushback
);
1388 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1391 static void dm_wq_work(struct work_struct
*work
)
1393 struct dm_wq_req
*req
= container_of(work
, struct dm_wq_req
, work
);
1394 struct mapped_device
*md
= req
->md
;
1396 down_write(&md
->io_lock
);
1397 switch (req
->type
) {
1398 case DM_WQ_FLUSH_ALL
:
1399 __merge_pushback_list(md
);
1401 case DM_WQ_FLUSH_DEFERRED
:
1402 __flush_deferred_io(md
);
1405 DMERR("dm_wq_work: unrecognised work type %d", req
->type
);
1408 up_write(&md
->io_lock
);
1411 static void dm_wq_queue(struct mapped_device
*md
, int type
, void *context
,
1412 struct dm_wq_req
*req
)
1416 req
->context
= context
;
1417 INIT_WORK(&req
->work
, dm_wq_work
);
1418 queue_work(md
->wq
, &req
->work
);
1421 static void dm_queue_flush(struct mapped_device
*md
, int type
, void *context
)
1423 struct dm_wq_req req
;
1425 dm_wq_queue(md
, type
, context
, &req
);
1426 flush_workqueue(md
->wq
);
1430 * Swap in a new table (destroying old one).
1432 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1436 mutex_lock(&md
->suspend_lock
);
1438 /* device must be suspended */
1439 if (!dm_suspended(md
))
1442 /* without bdev, the device size cannot be changed */
1443 if (!md
->suspended_bdev
)
1444 if (get_capacity(md
->disk
) != dm_table_get_size(table
))
1448 r
= __bind(md
, table
);
1451 mutex_unlock(&md
->suspend_lock
);
1456 * Functions to lock and unlock any filesystem running on the
1459 static int lock_fs(struct mapped_device
*md
)
1463 WARN_ON(md
->frozen_sb
);
1465 md
->frozen_sb
= freeze_bdev(md
->suspended_bdev
);
1466 if (IS_ERR(md
->frozen_sb
)) {
1467 r
= PTR_ERR(md
->frozen_sb
);
1468 md
->frozen_sb
= NULL
;
1472 set_bit(DMF_FROZEN
, &md
->flags
);
1474 /* don't bdput right now, we don't want the bdev
1475 * to go away while it is locked.
1480 static void unlock_fs(struct mapped_device
*md
)
1482 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1485 thaw_bdev(md
->suspended_bdev
, md
->frozen_sb
);
1486 md
->frozen_sb
= NULL
;
1487 clear_bit(DMF_FROZEN
, &md
->flags
);
1491 * We need to be able to change a mapping table under a mounted
1492 * filesystem. For example we might want to move some data in
1493 * the background. Before the table can be swapped with
1494 * dm_bind_table, dm_suspend must be called to flush any in
1495 * flight bios and ensure that any further io gets deferred.
1497 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1499 struct dm_table
*map
= NULL
;
1500 DECLARE_WAITQUEUE(wait
, current
);
1502 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1503 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1505 mutex_lock(&md
->suspend_lock
);
1507 if (dm_suspended(md
)) {
1512 map
= dm_get_table(md
);
1515 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1516 * This flag is cleared before dm_suspend returns.
1519 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1521 /* This does not get reverted if there's an error later. */
1522 dm_table_presuspend_targets(map
);
1524 /* bdget() can stall if the pending I/Os are not flushed */
1526 md
->suspended_bdev
= bdget_disk(md
->disk
, 0);
1527 if (!md
->suspended_bdev
) {
1528 DMWARN("bdget failed in dm_suspend");
1534 * Flush I/O to the device. noflush supersedes do_lockfs,
1535 * because lock_fs() needs to flush I/Os.
1545 * First we set the BLOCK_IO flag so no more ios will be mapped.
1547 down_write(&md
->io_lock
);
1548 set_bit(DMF_BLOCK_IO
, &md
->flags
);
1550 add_wait_queue(&md
->wait
, &wait
);
1551 up_write(&md
->io_lock
);
1555 dm_table_unplug_all(map
);
1558 * Wait for the already-mapped ios to complete.
1560 r
= dm_wait_for_completion(md
);
1562 down_write(&md
->io_lock
);
1563 remove_wait_queue(&md
->wait
, &wait
);
1566 __merge_pushback_list(md
);
1567 up_write(&md
->io_lock
);
1569 /* were we interrupted ? */
1571 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1574 goto out
; /* pushback list is already flushed, so skip flush */
1577 dm_table_postsuspend_targets(map
);
1579 set_bit(DMF_SUSPENDED
, &md
->flags
);
1584 * Because there may be already I/Os in the pushback list,
1585 * flush them before return.
1587 dm_queue_flush(md
, DM_WQ_FLUSH_ALL
, NULL
);
1590 if (r
&& md
->suspended_bdev
) {
1591 bdput(md
->suspended_bdev
);
1592 md
->suspended_bdev
= NULL
;
1598 mutex_unlock(&md
->suspend_lock
);
1602 int dm_resume(struct mapped_device
*md
)
1605 struct dm_table
*map
= NULL
;
1607 mutex_lock(&md
->suspend_lock
);
1608 if (!dm_suspended(md
))
1611 map
= dm_get_table(md
);
1612 if (!map
|| !dm_table_get_size(map
))
1615 r
= dm_table_resume_targets(map
);
1619 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1623 if (md
->suspended_bdev
) {
1624 bdput(md
->suspended_bdev
);
1625 md
->suspended_bdev
= NULL
;
1628 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1630 dm_table_unplug_all(map
);
1632 dm_kobject_uevent(md
);
1638 mutex_unlock(&md
->suspend_lock
);
1643 /*-----------------------------------------------------------------
1644 * Event notification.
1645 *---------------------------------------------------------------*/
1646 void dm_kobject_uevent(struct mapped_device
*md
)
1648 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, KOBJ_CHANGE
);
1651 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
1653 return atomic_add_return(1, &md
->uevent_seq
);
1656 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1658 return atomic_read(&md
->event_nr
);
1661 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1663 return wait_event_interruptible(md
->eventq
,
1664 (event_nr
!= atomic_read(&md
->event_nr
)));
1667 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
1669 unsigned long flags
;
1671 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1672 list_add(elist
, &md
->uevent_list
);
1673 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1677 * The gendisk is only valid as long as you have a reference
1680 struct gendisk
*dm_disk(struct mapped_device
*md
)
1685 int dm_suspended(struct mapped_device
*md
)
1687 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1690 int dm_noflush_suspending(struct dm_target
*ti
)
1692 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1693 int r
= __noflush_suspending(md
);
1699 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1701 static struct block_device_operations dm_blk_dops
= {
1702 .open
= dm_blk_open
,
1703 .release
= dm_blk_close
,
1704 .ioctl
= dm_blk_ioctl
,
1705 .getgeo
= dm_blk_getgeo
,
1706 .owner
= THIS_MODULE
1709 EXPORT_SYMBOL(dm_get_mapinfo
);
1714 module_init(dm_init
);
1715 module_exit(dm_exit
);
1717 module_param(major
, uint
, 0);
1718 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1719 MODULE_DESCRIPTION(DM_NAME
" driver");
1720 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1721 MODULE_LICENSE("GPL");