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.
81 struct work_struct work
;
82 struct mapped_device
*md
;
86 struct mapped_device
{
87 struct rw_semaphore io_lock
;
88 struct mutex suspend_lock
;
89 spinlock_t pushback_lock
;
96 struct request_queue
*queue
;
103 * A list of ios that arrived while we were suspended.
106 wait_queue_head_t wait
;
107 struct bio_list deferred
;
108 struct bio_list pushback
;
111 * Processing queue (flush/barriers)
113 struct workqueue_struct
*wq
;
116 * The current mapping.
118 struct dm_table
*map
;
121 * io objects are allocated from here.
132 wait_queue_head_t eventq
;
134 struct list_head uevent_list
;
135 spinlock_t uevent_lock
; /* Protect access to uevent_list */
138 * freeze/thaw support require holding onto a super block
140 struct super_block
*frozen_sb
;
141 struct block_device
*suspended_bdev
;
143 /* forced geometry settings */
144 struct hd_geometry geometry
;
148 static struct kmem_cache
*_io_cache
;
149 static struct kmem_cache
*_tio_cache
;
151 static int __init
local_init(void)
155 /* allocate a slab for the dm_ios */
156 _io_cache
= KMEM_CACHE(dm_io
, 0);
160 /* allocate a slab for the target ios */
161 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
163 goto out_free_io_cache
;
165 r
= dm_uevent_init();
167 goto out_free_tio_cache
;
170 r
= register_blkdev(_major
, _name
);
172 goto out_uevent_exit
;
182 kmem_cache_destroy(_tio_cache
);
184 kmem_cache_destroy(_io_cache
);
189 static void local_exit(void)
191 kmem_cache_destroy(_tio_cache
);
192 kmem_cache_destroy(_io_cache
);
193 unregister_blkdev(_major
, _name
);
198 DMINFO("cleaned up");
201 static int (*_inits
[])(void) __initdata
= {
210 static void (*_exits
[])(void) = {
219 static int __init
dm_init(void)
221 const int count
= ARRAY_SIZE(_inits
);
225 for (i
= 0; i
< count
; i
++) {
240 static void __exit
dm_exit(void)
242 int i
= ARRAY_SIZE(_exits
);
249 * Block device functions
251 static int dm_blk_open(struct inode
*inode
, struct file
*file
)
253 struct mapped_device
*md
;
255 spin_lock(&_minor_lock
);
257 md
= inode
->i_bdev
->bd_disk
->private_data
;
261 if (test_bit(DMF_FREEING
, &md
->flags
) ||
262 test_bit(DMF_DELETING
, &md
->flags
)) {
268 atomic_inc(&md
->open_count
);
271 spin_unlock(&_minor_lock
);
273 return md
? 0 : -ENXIO
;
276 static int dm_blk_close(struct inode
*inode
, struct file
*file
)
278 struct mapped_device
*md
;
280 md
= inode
->i_bdev
->bd_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 inode
*inode
, struct file
*file
,
318 unsigned int cmd
, unsigned long arg
)
320 struct mapped_device
*md
;
321 struct dm_table
*map
;
322 struct dm_target
*tgt
;
325 /* We don't really need this lock, but we do need 'inode'. */
328 md
= inode
->i_bdev
->bd_disk
->private_data
;
330 map
= dm_get_table(md
);
332 if (!map
|| !dm_table_get_size(map
))
335 /* We only support devices that have a single target */
336 if (dm_table_get_num_targets(map
) != 1)
339 tgt
= dm_table_get_target(map
, 0);
341 if (dm_suspended(md
)) {
346 if (tgt
->type
->ioctl
)
347 r
= tgt
->type
->ioctl(tgt
, inode
, file
, cmd
, arg
);
356 static struct dm_io
*alloc_io(struct mapped_device
*md
)
358 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
361 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
363 mempool_free(io
, md
->io_pool
);
366 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
368 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
371 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
373 mempool_free(tio
, md
->tio_pool
);
376 static void start_io_acct(struct dm_io
*io
)
378 struct mapped_device
*md
= io
->md
;
381 io
->start_time
= jiffies
;
383 cpu
= part_stat_lock();
384 part_round_stats(cpu
, &dm_disk(md
)->part0
);
386 dm_disk(md
)->part0
.in_flight
= atomic_inc_return(&md
->pending
);
389 static int end_io_acct(struct dm_io
*io
)
391 struct mapped_device
*md
= io
->md
;
392 struct bio
*bio
= io
->bio
;
393 unsigned long duration
= jiffies
- io
->start_time
;
395 int rw
= bio_data_dir(bio
);
397 cpu
= part_stat_lock();
398 part_round_stats(cpu
, &dm_disk(md
)->part0
);
399 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
402 dm_disk(md
)->part0
.in_flight
= pending
=
403 atomic_dec_return(&md
->pending
);
409 * Add the bio to the list of deferred io.
411 static int queue_io(struct mapped_device
*md
, struct bio
*bio
)
413 down_write(&md
->io_lock
);
415 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
416 up_write(&md
->io_lock
);
420 bio_list_add(&md
->deferred
, bio
);
422 up_write(&md
->io_lock
);
423 return 0; /* deferred successfully */
427 * Everyone (including functions in this file), should use this
428 * function to access the md->map field, and make sure they call
429 * dm_table_put() when finished.
431 struct dm_table
*dm_get_table(struct mapped_device
*md
)
435 read_lock(&md
->map_lock
);
439 read_unlock(&md
->map_lock
);
445 * Get the geometry associated with a dm device
447 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
455 * Set the geometry of a device.
457 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
459 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
461 if (geo
->start
> sz
) {
462 DMWARN("Start sector is beyond the geometry limits.");
471 /*-----------------------------------------------------------------
473 * A more elegant soln is in the works that uses the queue
474 * merge fn, unfortunately there are a couple of changes to
475 * the block layer that I want to make for this. So in the
476 * interests of getting something for people to use I give
477 * you this clearly demarcated crap.
478 *---------------------------------------------------------------*/
480 static int __noflush_suspending(struct mapped_device
*md
)
482 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
486 * Decrements the number of outstanding ios that a bio has been
487 * cloned into, completing the original io if necc.
489 static void dec_pending(struct dm_io
*io
, int error
)
493 /* Push-back supersedes any I/O errors */
494 if (error
&& !(io
->error
> 0 && __noflush_suspending(io
->md
)))
497 if (atomic_dec_and_test(&io
->io_count
)) {
498 if (io
->error
== DM_ENDIO_REQUEUE
) {
500 * Target requested pushing back the I/O.
501 * This must be handled before the sleeper on
502 * suspend queue merges the pushback list.
504 spin_lock_irqsave(&io
->md
->pushback_lock
, flags
);
505 if (__noflush_suspending(io
->md
))
506 bio_list_add(&io
->md
->pushback
, io
->bio
);
508 /* noflush suspend was interrupted. */
510 spin_unlock_irqrestore(&io
->md
->pushback_lock
, flags
);
514 /* nudge anyone waiting on suspend queue */
515 wake_up(&io
->md
->wait
);
517 if (io
->error
!= DM_ENDIO_REQUEUE
) {
518 blk_add_trace_bio(io
->md
->queue
, io
->bio
,
521 bio_endio(io
->bio
, io
->error
);
528 static void clone_endio(struct bio
*bio
, int error
)
531 struct dm_target_io
*tio
= bio
->bi_private
;
532 struct mapped_device
*md
= tio
->io
->md
;
533 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
535 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
539 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
540 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
542 * error and requeue request are handled
546 else if (r
== DM_ENDIO_INCOMPLETE
)
547 /* The target will handle the io */
550 DMWARN("unimplemented target endio return value: %d", r
);
555 dec_pending(tio
->io
, error
);
558 * Store md for cleanup instead of tio which is about to get freed.
560 bio
->bi_private
= md
->bs
;
566 static sector_t
max_io_len(struct mapped_device
*md
,
567 sector_t sector
, struct dm_target
*ti
)
569 sector_t offset
= sector
- ti
->begin
;
570 sector_t len
= ti
->len
- offset
;
573 * Does the target need to split even further ?
577 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
586 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
587 struct dm_target_io
*tio
)
591 struct mapped_device
*md
;
596 BUG_ON(!clone
->bi_size
);
598 clone
->bi_end_io
= clone_endio
;
599 clone
->bi_private
= tio
;
602 * Map the clone. If r == 0 we don't need to do
603 * anything, the target has assumed ownership of
606 atomic_inc(&tio
->io
->io_count
);
607 sector
= clone
->bi_sector
;
608 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
609 if (r
== DM_MAPIO_REMAPPED
) {
610 /* the bio has been remapped so dispatch it */
612 blk_add_trace_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
613 tio
->io
->bio
->bi_bdev
->bd_dev
,
614 clone
->bi_sector
, sector
);
616 generic_make_request(clone
);
617 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
618 /* error the io and bail out, or requeue it if needed */
620 dec_pending(tio
->io
, r
);
622 * Store bio_set for cleanup.
624 clone
->bi_private
= md
->bs
;
628 DMWARN("unimplemented target map return value: %d", r
);
634 struct mapped_device
*md
;
635 struct dm_table
*map
;
639 sector_t sector_count
;
643 static void dm_bio_destructor(struct bio
*bio
)
645 struct bio_set
*bs
= bio
->bi_private
;
651 * Creates a little bio that is just does part of a bvec.
653 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
654 unsigned short idx
, unsigned int offset
,
655 unsigned int len
, struct bio_set
*bs
)
658 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
660 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
661 clone
->bi_destructor
= dm_bio_destructor
;
662 *clone
->bi_io_vec
= *bv
;
664 clone
->bi_sector
= sector
;
665 clone
->bi_bdev
= bio
->bi_bdev
;
666 clone
->bi_rw
= bio
->bi_rw
;
668 clone
->bi_size
= to_bytes(len
);
669 clone
->bi_io_vec
->bv_offset
= offset
;
670 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
671 clone
->bi_flags
|= 1 << BIO_CLONED
;
677 * Creates a bio that consists of range of complete bvecs.
679 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
680 unsigned short idx
, unsigned short bv_count
,
681 unsigned int len
, struct bio_set
*bs
)
685 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
686 __bio_clone(clone
, bio
);
687 clone
->bi_destructor
= dm_bio_destructor
;
688 clone
->bi_sector
= sector
;
690 clone
->bi_vcnt
= idx
+ bv_count
;
691 clone
->bi_size
= to_bytes(len
);
692 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
697 static int __clone_and_map(struct clone_info
*ci
)
699 struct bio
*clone
, *bio
= ci
->bio
;
700 struct dm_target
*ti
;
701 sector_t len
= 0, max
;
702 struct dm_target_io
*tio
;
704 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
705 if (!dm_target_is_valid(ti
))
708 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
711 * Allocate a target io object.
713 tio
= alloc_tio(ci
->md
);
716 memset(&tio
->info
, 0, sizeof(tio
->info
));
718 if (ci
->sector_count
<= max
) {
720 * Optimise for the simple case where we can do all of
721 * the remaining io with a single clone.
723 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
724 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
726 __map_bio(ti
, clone
, tio
);
727 ci
->sector_count
= 0;
729 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
731 * There are some bvecs that don't span targets.
732 * Do as many of these as possible.
735 sector_t remaining
= max
;
738 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
739 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
741 if (bv_len
> remaining
)
748 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
750 __map_bio(ti
, clone
, tio
);
753 ci
->sector_count
-= len
;
758 * Handle a bvec that must be split between two or more targets.
760 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
761 sector_t remaining
= to_sector(bv
->bv_len
);
762 unsigned int offset
= 0;
766 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
767 if (!dm_target_is_valid(ti
))
770 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
772 tio
= alloc_tio(ci
->md
);
775 memset(&tio
->info
, 0, sizeof(tio
->info
));
778 len
= min(remaining
, max
);
780 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
781 bv
->bv_offset
+ offset
, len
,
784 __map_bio(ti
, clone
, tio
);
787 ci
->sector_count
-= len
;
788 offset
+= to_bytes(len
);
789 } while (remaining
-= len
);
798 * Split the bio into several clones.
800 static int __split_bio(struct mapped_device
*md
, struct bio
*bio
)
802 struct clone_info ci
;
805 ci
.map
= dm_get_table(md
);
806 if (unlikely(!ci
.map
))
811 ci
.io
= alloc_io(md
);
813 atomic_set(&ci
.io
->io_count
, 1);
816 ci
.sector
= bio
->bi_sector
;
817 ci
.sector_count
= bio_sectors(bio
);
818 ci
.idx
= bio
->bi_idx
;
820 start_io_acct(ci
.io
);
821 while (ci
.sector_count
&& !error
)
822 error
= __clone_and_map(&ci
);
824 /* drop the extra reference count */
825 dec_pending(ci
.io
, error
);
826 dm_table_put(ci
.map
);
830 /*-----------------------------------------------------------------
832 *---------------------------------------------------------------*/
834 static int dm_merge_bvec(struct request_queue
*q
,
835 struct bvec_merge_data
*bvm
,
836 struct bio_vec
*biovec
)
838 struct mapped_device
*md
= q
->queuedata
;
839 struct dm_table
*map
= dm_get_table(md
);
840 struct dm_target
*ti
;
841 sector_t max_sectors
;
847 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
848 if (!dm_target_is_valid(ti
))
852 * Find maximum amount of I/O that won't need splitting
854 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
855 (sector_t
) BIO_MAX_SECTORS
);
856 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
861 * merge_bvec_fn() returns number of bytes
862 * it can accept at this offset
863 * max is precomputed maximal io size
865 if (max_size
&& ti
->type
->merge
)
866 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
873 * Always allow an entire first page
875 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
876 max_size
= biovec
->bv_len
;
882 * The request function that just remaps the bio built up by
885 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
888 int rw
= bio_data_dir(bio
);
889 struct mapped_device
*md
= q
->queuedata
;
893 * There is no use in forwarding any barrier request since we can't
894 * guarantee it is (or can be) handled by the targets correctly.
896 if (unlikely(bio_barrier(bio
))) {
897 bio_endio(bio
, -EOPNOTSUPP
);
901 down_read(&md
->io_lock
);
903 cpu
= part_stat_lock();
904 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
905 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
909 * If we're suspended we have to queue
912 while (test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
913 up_read(&md
->io_lock
);
915 if (bio_rw(bio
) != READA
)
916 r
= queue_io(md
, bio
);
922 * We're in a while loop, because someone could suspend
923 * before we get to the following read lock.
925 down_read(&md
->io_lock
);
928 r
= __split_bio(md
, bio
);
929 up_read(&md
->io_lock
);
938 static void dm_unplug_all(struct request_queue
*q
)
940 struct mapped_device
*md
= q
->queuedata
;
941 struct dm_table
*map
= dm_get_table(md
);
944 dm_table_unplug_all(map
);
949 static int dm_any_congested(void *congested_data
, int bdi_bits
)
952 struct mapped_device
*md
= (struct mapped_device
*) congested_data
;
953 struct dm_table
*map
= dm_get_table(md
);
955 if (!map
|| test_bit(DMF_BLOCK_IO
, &md
->flags
))
958 r
= dm_table_any_congested(map
, bdi_bits
);
964 /*-----------------------------------------------------------------
965 * An IDR is used to keep track of allocated minor numbers.
966 *---------------------------------------------------------------*/
967 static DEFINE_IDR(_minor_idr
);
969 static void free_minor(int minor
)
971 spin_lock(&_minor_lock
);
972 idr_remove(&_minor_idr
, minor
);
973 spin_unlock(&_minor_lock
);
977 * See if the device with a specific minor # is free.
979 static int specific_minor(int minor
)
983 if (minor
>= (1 << MINORBITS
))
986 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
990 spin_lock(&_minor_lock
);
992 if (idr_find(&_minor_idr
, minor
)) {
997 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1002 idr_remove(&_minor_idr
, m
);
1008 spin_unlock(&_minor_lock
);
1012 static int next_free_minor(int *minor
)
1016 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1020 spin_lock(&_minor_lock
);
1022 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1026 if (m
>= (1 << MINORBITS
)) {
1027 idr_remove(&_minor_idr
, m
);
1035 spin_unlock(&_minor_lock
);
1039 static struct block_device_operations dm_blk_dops
;
1042 * Allocate and initialise a blank device with a given minor.
1044 static struct mapped_device
*alloc_dev(int minor
)
1047 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1051 DMWARN("unable to allocate device, out of memory.");
1055 if (!try_module_get(THIS_MODULE
))
1056 goto bad_module_get
;
1058 /* get a minor number for the dev */
1059 if (minor
== DM_ANY_MINOR
)
1060 r
= next_free_minor(&minor
);
1062 r
= specific_minor(minor
);
1066 init_rwsem(&md
->io_lock
);
1067 mutex_init(&md
->suspend_lock
);
1068 spin_lock_init(&md
->pushback_lock
);
1069 rwlock_init(&md
->map_lock
);
1070 atomic_set(&md
->holders
, 1);
1071 atomic_set(&md
->open_count
, 0);
1072 atomic_set(&md
->event_nr
, 0);
1073 atomic_set(&md
->uevent_seq
, 0);
1074 INIT_LIST_HEAD(&md
->uevent_list
);
1075 spin_lock_init(&md
->uevent_lock
);
1077 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1081 md
->queue
->queuedata
= md
;
1082 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1083 md
->queue
->backing_dev_info
.congested_data
= md
;
1084 blk_queue_make_request(md
->queue
, dm_request
);
1085 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1086 md
->queue
->unplug_fn
= dm_unplug_all
;
1087 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1089 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1093 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1097 md
->bs
= bioset_create(16, 16);
1101 md
->disk
= alloc_disk(1);
1105 atomic_set(&md
->pending
, 0);
1106 init_waitqueue_head(&md
->wait
);
1107 init_waitqueue_head(&md
->eventq
);
1109 md
->disk
->major
= _major
;
1110 md
->disk
->first_minor
= minor
;
1111 md
->disk
->fops
= &dm_blk_dops
;
1112 md
->disk
->queue
= md
->queue
;
1113 md
->disk
->private_data
= md
;
1114 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1116 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1118 md
->wq
= create_singlethread_workqueue("kdmflush");
1122 /* Populate the mapping, nobody knows we exist yet */
1123 spin_lock(&_minor_lock
);
1124 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1125 spin_unlock(&_minor_lock
);
1127 BUG_ON(old_md
!= MINOR_ALLOCED
);
1134 bioset_free(md
->bs
);
1136 mempool_destroy(md
->tio_pool
);
1138 mempool_destroy(md
->io_pool
);
1140 blk_cleanup_queue(md
->queue
);
1144 module_put(THIS_MODULE
);
1150 static void unlock_fs(struct mapped_device
*md
);
1152 static void free_dev(struct mapped_device
*md
)
1154 int minor
= MINOR(disk_devt(md
->disk
));
1156 if (md
->suspended_bdev
) {
1158 bdput(md
->suspended_bdev
);
1160 destroy_workqueue(md
->wq
);
1161 mempool_destroy(md
->tio_pool
);
1162 mempool_destroy(md
->io_pool
);
1163 bioset_free(md
->bs
);
1164 del_gendisk(md
->disk
);
1167 spin_lock(&_minor_lock
);
1168 md
->disk
->private_data
= NULL
;
1169 spin_unlock(&_minor_lock
);
1172 blk_cleanup_queue(md
->queue
);
1173 module_put(THIS_MODULE
);
1178 * Bind a table to the device.
1180 static void event_callback(void *context
)
1182 unsigned long flags
;
1184 struct mapped_device
*md
= (struct mapped_device
*) context
;
1186 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1187 list_splice_init(&md
->uevent_list
, &uevents
);
1188 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1190 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1192 atomic_inc(&md
->event_nr
);
1193 wake_up(&md
->eventq
);
1196 static void __set_size(struct mapped_device
*md
, sector_t size
)
1198 set_capacity(md
->disk
, size
);
1200 mutex_lock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1201 i_size_write(md
->suspended_bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1202 mutex_unlock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1205 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1207 struct request_queue
*q
= md
->queue
;
1210 size
= dm_table_get_size(t
);
1213 * Wipe any geometry if the size of the table changed.
1215 if (size
!= get_capacity(md
->disk
))
1216 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1218 if (md
->suspended_bdev
)
1219 __set_size(md
, size
);
1224 dm_table_event_callback(t
, event_callback
, md
);
1226 write_lock(&md
->map_lock
);
1228 dm_table_set_restrictions(t
, q
);
1229 write_unlock(&md
->map_lock
);
1234 static void __unbind(struct mapped_device
*md
)
1236 struct dm_table
*map
= md
->map
;
1241 dm_table_event_callback(map
, NULL
, NULL
);
1242 write_lock(&md
->map_lock
);
1244 write_unlock(&md
->map_lock
);
1249 * Constructor for a new device.
1251 int dm_create(int minor
, struct mapped_device
**result
)
1253 struct mapped_device
*md
;
1255 md
= alloc_dev(minor
);
1263 static struct mapped_device
*dm_find_md(dev_t dev
)
1265 struct mapped_device
*md
;
1266 unsigned minor
= MINOR(dev
);
1268 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1271 spin_lock(&_minor_lock
);
1273 md
= idr_find(&_minor_idr
, minor
);
1274 if (md
&& (md
== MINOR_ALLOCED
||
1275 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1276 test_bit(DMF_FREEING
, &md
->flags
))) {
1282 spin_unlock(&_minor_lock
);
1287 struct mapped_device
*dm_get_md(dev_t dev
)
1289 struct mapped_device
*md
= dm_find_md(dev
);
1297 void *dm_get_mdptr(struct mapped_device
*md
)
1299 return md
->interface_ptr
;
1302 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1304 md
->interface_ptr
= ptr
;
1307 void dm_get(struct mapped_device
*md
)
1309 atomic_inc(&md
->holders
);
1312 const char *dm_device_name(struct mapped_device
*md
)
1316 EXPORT_SYMBOL_GPL(dm_device_name
);
1318 void dm_put(struct mapped_device
*md
)
1320 struct dm_table
*map
;
1322 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1324 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1325 map
= dm_get_table(md
);
1326 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
1327 MINOR(disk_devt(dm_disk(md
))));
1328 set_bit(DMF_FREEING
, &md
->flags
);
1329 spin_unlock(&_minor_lock
);
1330 if (!dm_suspended(md
)) {
1331 dm_table_presuspend_targets(map
);
1332 dm_table_postsuspend_targets(map
);
1339 EXPORT_SYMBOL_GPL(dm_put
);
1341 static int dm_wait_for_completion(struct mapped_device
*md
)
1346 set_current_state(TASK_INTERRUPTIBLE
);
1349 if (!atomic_read(&md
->pending
))
1352 if (signal_pending(current
)) {
1359 set_current_state(TASK_RUNNING
);
1365 * Process the deferred bios
1367 static void __flush_deferred_io(struct mapped_device
*md
)
1371 while ((c
= bio_list_pop(&md
->deferred
))) {
1372 if (__split_bio(md
, c
))
1376 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1379 static void __merge_pushback_list(struct mapped_device
*md
)
1381 unsigned long flags
;
1383 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1384 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1385 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1386 bio_list_init(&md
->pushback
);
1387 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1390 static void dm_wq_work(struct work_struct
*work
)
1392 struct dm_wq_req
*req
= container_of(work
, struct dm_wq_req
, work
);
1393 struct mapped_device
*md
= req
->md
;
1395 down_write(&md
->io_lock
);
1396 switch (req
->type
) {
1397 case DM_WQ_FLUSH_DEFERRED
:
1398 __flush_deferred_io(md
);
1401 DMERR("dm_wq_work: unrecognised work type %d", req
->type
);
1404 up_write(&md
->io_lock
);
1407 static void dm_wq_queue(struct mapped_device
*md
, int type
, void *context
,
1408 struct dm_wq_req
*req
)
1412 req
->context
= context
;
1413 INIT_WORK(&req
->work
, dm_wq_work
);
1414 queue_work(md
->wq
, &req
->work
);
1417 static void dm_queue_flush(struct mapped_device
*md
, int type
, void *context
)
1419 struct dm_wq_req req
;
1421 dm_wq_queue(md
, type
, context
, &req
);
1422 flush_workqueue(md
->wq
);
1426 * Swap in a new table (destroying old one).
1428 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1432 mutex_lock(&md
->suspend_lock
);
1434 /* device must be suspended */
1435 if (!dm_suspended(md
))
1438 /* without bdev, the device size cannot be changed */
1439 if (!md
->suspended_bdev
)
1440 if (get_capacity(md
->disk
) != dm_table_get_size(table
))
1444 r
= __bind(md
, table
);
1447 mutex_unlock(&md
->suspend_lock
);
1452 * Functions to lock and unlock any filesystem running on the
1455 static int lock_fs(struct mapped_device
*md
)
1459 WARN_ON(md
->frozen_sb
);
1461 md
->frozen_sb
= freeze_bdev(md
->suspended_bdev
);
1462 if (IS_ERR(md
->frozen_sb
)) {
1463 r
= PTR_ERR(md
->frozen_sb
);
1464 md
->frozen_sb
= NULL
;
1468 set_bit(DMF_FROZEN
, &md
->flags
);
1470 /* don't bdput right now, we don't want the bdev
1471 * to go away while it is locked.
1476 static void unlock_fs(struct mapped_device
*md
)
1478 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1481 thaw_bdev(md
->suspended_bdev
, md
->frozen_sb
);
1482 md
->frozen_sb
= NULL
;
1483 clear_bit(DMF_FROZEN
, &md
->flags
);
1487 * We need to be able to change a mapping table under a mounted
1488 * filesystem. For example we might want to move some data in
1489 * the background. Before the table can be swapped with
1490 * dm_bind_table, dm_suspend must be called to flush any in
1491 * flight bios and ensure that any further io gets deferred.
1493 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1495 struct dm_table
*map
= NULL
;
1496 DECLARE_WAITQUEUE(wait
, current
);
1498 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1499 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1501 mutex_lock(&md
->suspend_lock
);
1503 if (dm_suspended(md
)) {
1508 map
= dm_get_table(md
);
1511 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1512 * This flag is cleared before dm_suspend returns.
1515 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1517 /* This does not get reverted if there's an error later. */
1518 dm_table_presuspend_targets(map
);
1520 /* bdget() can stall if the pending I/Os are not flushed */
1522 md
->suspended_bdev
= bdget_disk(md
->disk
, 0);
1523 if (!md
->suspended_bdev
) {
1524 DMWARN("bdget failed in dm_suspend");
1530 * Flush I/O to the device. noflush supersedes do_lockfs,
1531 * because lock_fs() needs to flush I/Os.
1541 * First we set the BLOCK_IO flag so no more ios will be mapped.
1543 down_write(&md
->io_lock
);
1544 set_bit(DMF_BLOCK_IO
, &md
->flags
);
1546 add_wait_queue(&md
->wait
, &wait
);
1547 up_write(&md
->io_lock
);
1551 dm_table_unplug_all(map
);
1554 * Wait for the already-mapped ios to complete.
1556 r
= dm_wait_for_completion(md
);
1558 down_write(&md
->io_lock
);
1559 remove_wait_queue(&md
->wait
, &wait
);
1562 __merge_pushback_list(md
);
1563 up_write(&md
->io_lock
);
1565 /* were we interrupted ? */
1567 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1570 goto out
; /* pushback list is already flushed, so skip flush */
1573 dm_table_postsuspend_targets(map
);
1575 set_bit(DMF_SUSPENDED
, &md
->flags
);
1578 if (r
&& md
->suspended_bdev
) {
1579 bdput(md
->suspended_bdev
);
1580 md
->suspended_bdev
= NULL
;
1586 mutex_unlock(&md
->suspend_lock
);
1590 int dm_resume(struct mapped_device
*md
)
1593 struct dm_table
*map
= NULL
;
1595 mutex_lock(&md
->suspend_lock
);
1596 if (!dm_suspended(md
))
1599 map
= dm_get_table(md
);
1600 if (!map
|| !dm_table_get_size(map
))
1603 r
= dm_table_resume_targets(map
);
1607 dm_queue_flush(md
, DM_WQ_FLUSH_DEFERRED
, NULL
);
1611 if (md
->suspended_bdev
) {
1612 bdput(md
->suspended_bdev
);
1613 md
->suspended_bdev
= NULL
;
1616 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1618 dm_table_unplug_all(map
);
1620 dm_kobject_uevent(md
);
1626 mutex_unlock(&md
->suspend_lock
);
1631 /*-----------------------------------------------------------------
1632 * Event notification.
1633 *---------------------------------------------------------------*/
1634 void dm_kobject_uevent(struct mapped_device
*md
)
1636 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, KOBJ_CHANGE
);
1639 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
1641 return atomic_add_return(1, &md
->uevent_seq
);
1644 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1646 return atomic_read(&md
->event_nr
);
1649 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1651 return wait_event_interruptible(md
->eventq
,
1652 (event_nr
!= atomic_read(&md
->event_nr
)));
1655 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
1657 unsigned long flags
;
1659 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1660 list_add(elist
, &md
->uevent_list
);
1661 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1665 * The gendisk is only valid as long as you have a reference
1668 struct gendisk
*dm_disk(struct mapped_device
*md
)
1673 int dm_suspended(struct mapped_device
*md
)
1675 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1678 int dm_noflush_suspending(struct dm_target
*ti
)
1680 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1681 int r
= __noflush_suspending(md
);
1687 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1689 static struct block_device_operations dm_blk_dops
= {
1690 .open
= dm_blk_open
,
1691 .release
= dm_blk_close
,
1692 .ioctl
= dm_blk_ioctl
,
1693 .getgeo
= dm_blk_getgeo
,
1694 .owner
= THIS_MODULE
1697 EXPORT_SYMBOL(dm_get_mapinfo
);
1702 module_init(dm_init
);
1703 module_exit(dm_exit
);
1705 module_param(major
, uint
, 0);
1706 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1707 MODULE_DESCRIPTION(DM_NAME
" driver");
1708 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1709 MODULE_LICENSE("GPL");