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
74 struct mapped_device
{
75 struct rw_semaphore io_lock
;
76 struct semaphore suspend_lock
;
77 spinlock_t pushback_lock
;
84 struct request_queue
*queue
;
91 * A list of ios that arrived while we were suspended.
94 wait_queue_head_t wait
;
95 struct bio_list deferred
;
96 struct bio_list pushback
;
99 * The current mapping.
101 struct dm_table
*map
;
104 * io objects are allocated from here.
115 wait_queue_head_t eventq
;
118 * freeze/thaw support require holding onto a super block
120 struct super_block
*frozen_sb
;
121 struct block_device
*suspended_bdev
;
123 /* forced geometry settings */
124 struct hd_geometry geometry
;
128 static struct kmem_cache
*_io_cache
;
129 static struct kmem_cache
*_tio_cache
;
131 static int __init
local_init(void)
135 /* allocate a slab for the dm_ios */
136 _io_cache
= KMEM_CACHE(dm_io
, 0);
140 /* allocate a slab for the target ios */
141 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
143 kmem_cache_destroy(_io_cache
);
147 r
= dm_uevent_init();
149 kmem_cache_destroy(_tio_cache
);
150 kmem_cache_destroy(_io_cache
);
155 r
= register_blkdev(_major
, _name
);
157 kmem_cache_destroy(_tio_cache
);
158 kmem_cache_destroy(_io_cache
);
169 static void local_exit(void)
171 kmem_cache_destroy(_tio_cache
);
172 kmem_cache_destroy(_io_cache
);
173 unregister_blkdev(_major
, _name
);
178 DMINFO("cleaned up");
181 int (*_inits
[])(void) __initdata
= {
189 void (*_exits
[])(void) = {
197 static int __init
dm_init(void)
199 const int count
= ARRAY_SIZE(_inits
);
203 for (i
= 0; i
< count
; i
++) {
218 static void __exit
dm_exit(void)
220 int i
= ARRAY_SIZE(_exits
);
227 * Block device functions
229 static int dm_blk_open(struct inode
*inode
, struct file
*file
)
231 struct mapped_device
*md
;
233 spin_lock(&_minor_lock
);
235 md
= inode
->i_bdev
->bd_disk
->private_data
;
239 if (test_bit(DMF_FREEING
, &md
->flags
) ||
240 test_bit(DMF_DELETING
, &md
->flags
)) {
246 atomic_inc(&md
->open_count
);
249 spin_unlock(&_minor_lock
);
251 return md
? 0 : -ENXIO
;
254 static int dm_blk_close(struct inode
*inode
, struct file
*file
)
256 struct mapped_device
*md
;
258 md
= inode
->i_bdev
->bd_disk
->private_data
;
259 atomic_dec(&md
->open_count
);
264 int dm_open_count(struct mapped_device
*md
)
266 return atomic_read(&md
->open_count
);
270 * Guarantees nothing is using the device before it's deleted.
272 int dm_lock_for_deletion(struct mapped_device
*md
)
276 spin_lock(&_minor_lock
);
278 if (dm_open_count(md
))
281 set_bit(DMF_DELETING
, &md
->flags
);
283 spin_unlock(&_minor_lock
);
288 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
290 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
292 return dm_get_geometry(md
, geo
);
295 static int dm_blk_ioctl(struct inode
*inode
, struct file
*file
,
296 unsigned int cmd
, unsigned long arg
)
298 struct mapped_device
*md
;
299 struct dm_table
*map
;
300 struct dm_target
*tgt
;
303 /* We don't really need this lock, but we do need 'inode'. */
306 md
= inode
->i_bdev
->bd_disk
->private_data
;
308 map
= dm_get_table(md
);
310 if (!map
|| !dm_table_get_size(map
))
313 /* We only support devices that have a single target */
314 if (dm_table_get_num_targets(map
) != 1)
317 tgt
= dm_table_get_target(map
, 0);
319 if (dm_suspended(md
)) {
324 if (tgt
->type
->ioctl
)
325 r
= tgt
->type
->ioctl(tgt
, inode
, file
, cmd
, arg
);
334 static struct dm_io
*alloc_io(struct mapped_device
*md
)
336 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
339 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
341 mempool_free(io
, md
->io_pool
);
344 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
346 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
349 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
351 mempool_free(tio
, md
->tio_pool
);
354 static void start_io_acct(struct dm_io
*io
)
356 struct mapped_device
*md
= io
->md
;
358 io
->start_time
= jiffies
;
361 disk_round_stats(dm_disk(md
));
363 dm_disk(md
)->in_flight
= atomic_inc_return(&md
->pending
);
366 static int end_io_acct(struct dm_io
*io
)
368 struct mapped_device
*md
= io
->md
;
369 struct bio
*bio
= io
->bio
;
370 unsigned long duration
= jiffies
- io
->start_time
;
372 int rw
= bio_data_dir(bio
);
375 disk_round_stats(dm_disk(md
));
377 dm_disk(md
)->in_flight
= pending
= atomic_dec_return(&md
->pending
);
379 disk_stat_add(dm_disk(md
), ticks
[rw
], duration
);
385 * Add the bio to the list of deferred io.
387 static int queue_io(struct mapped_device
*md
, struct bio
*bio
)
389 down_write(&md
->io_lock
);
391 if (!test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
392 up_write(&md
->io_lock
);
396 bio_list_add(&md
->deferred
, bio
);
398 up_write(&md
->io_lock
);
399 return 0; /* deferred successfully */
403 * Everyone (including functions in this file), should use this
404 * function to access the md->map field, and make sure they call
405 * dm_table_put() when finished.
407 struct dm_table
*dm_get_table(struct mapped_device
*md
)
411 read_lock(&md
->map_lock
);
415 read_unlock(&md
->map_lock
);
421 * Get the geometry associated with a dm device
423 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
431 * Set the geometry of a device.
433 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
435 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
437 if (geo
->start
> sz
) {
438 DMWARN("Start sector is beyond the geometry limits.");
447 /*-----------------------------------------------------------------
449 * A more elegant soln is in the works that uses the queue
450 * merge fn, unfortunately there are a couple of changes to
451 * the block layer that I want to make for this. So in the
452 * interests of getting something for people to use I give
453 * you this clearly demarcated crap.
454 *---------------------------------------------------------------*/
456 static int __noflush_suspending(struct mapped_device
*md
)
458 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
462 * Decrements the number of outstanding ios that a bio has been
463 * cloned into, completing the original io if necc.
465 static void dec_pending(struct dm_io
*io
, int error
)
469 /* Push-back supersedes any I/O errors */
470 if (error
&& !(io
->error
> 0 && __noflush_suspending(io
->md
)))
473 if (atomic_dec_and_test(&io
->io_count
)) {
474 if (io
->error
== DM_ENDIO_REQUEUE
) {
476 * Target requested pushing back the I/O.
477 * This must be handled before the sleeper on
478 * suspend queue merges the pushback list.
480 spin_lock_irqsave(&io
->md
->pushback_lock
, flags
);
481 if (__noflush_suspending(io
->md
))
482 bio_list_add(&io
->md
->pushback
, io
->bio
);
484 /* noflush suspend was interrupted. */
486 spin_unlock_irqrestore(&io
->md
->pushback_lock
, flags
);
490 /* nudge anyone waiting on suspend queue */
491 wake_up(&io
->md
->wait
);
493 if (io
->error
!= DM_ENDIO_REQUEUE
) {
494 blk_add_trace_bio(io
->md
->queue
, io
->bio
,
497 bio_endio(io
->bio
, io
->error
);
504 static void clone_endio(struct bio
*bio
, int error
)
507 struct dm_target_io
*tio
= bio
->bi_private
;
508 struct mapped_device
*md
= tio
->io
->md
;
509 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
511 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
515 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
516 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
518 * error and requeue request are handled
522 else if (r
== DM_ENDIO_INCOMPLETE
)
523 /* The target will handle the io */
526 DMWARN("unimplemented target endio return value: %d", r
);
531 dec_pending(tio
->io
, error
);
534 * Store md for cleanup instead of tio which is about to get freed.
536 bio
->bi_private
= md
->bs
;
542 static sector_t
max_io_len(struct mapped_device
*md
,
543 sector_t sector
, struct dm_target
*ti
)
545 sector_t offset
= sector
- ti
->begin
;
546 sector_t len
= ti
->len
- offset
;
549 * Does the target need to split even further ?
553 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
562 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
563 struct dm_target_io
*tio
)
567 struct mapped_device
*md
;
572 BUG_ON(!clone
->bi_size
);
574 clone
->bi_end_io
= clone_endio
;
575 clone
->bi_private
= tio
;
578 * Map the clone. If r == 0 we don't need to do
579 * anything, the target has assumed ownership of
582 atomic_inc(&tio
->io
->io_count
);
583 sector
= clone
->bi_sector
;
584 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
585 if (r
== DM_MAPIO_REMAPPED
) {
586 /* the bio has been remapped so dispatch it */
588 blk_add_trace_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
589 tio
->io
->bio
->bi_bdev
->bd_dev
,
590 clone
->bi_sector
, sector
);
592 generic_make_request(clone
);
593 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
594 /* error the io and bail out, or requeue it if needed */
596 dec_pending(tio
->io
, r
);
598 * Store bio_set for cleanup.
600 clone
->bi_private
= md
->bs
;
604 DMWARN("unimplemented target map return value: %d", r
);
610 struct mapped_device
*md
;
611 struct dm_table
*map
;
615 sector_t sector_count
;
619 static void dm_bio_destructor(struct bio
*bio
)
621 struct bio_set
*bs
= bio
->bi_private
;
627 * Creates a little bio that is just does part of a bvec.
629 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
630 unsigned short idx
, unsigned int offset
,
631 unsigned int len
, struct bio_set
*bs
)
634 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
636 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
637 clone
->bi_destructor
= dm_bio_destructor
;
638 *clone
->bi_io_vec
= *bv
;
640 clone
->bi_sector
= sector
;
641 clone
->bi_bdev
= bio
->bi_bdev
;
642 clone
->bi_rw
= bio
->bi_rw
;
644 clone
->bi_size
= to_bytes(len
);
645 clone
->bi_io_vec
->bv_offset
= offset
;
646 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
652 * Creates a bio that consists of range of complete bvecs.
654 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
655 unsigned short idx
, unsigned short bv_count
,
656 unsigned int len
, struct bio_set
*bs
)
660 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
661 __bio_clone(clone
, bio
);
662 clone
->bi_destructor
= dm_bio_destructor
;
663 clone
->bi_sector
= sector
;
665 clone
->bi_vcnt
= idx
+ bv_count
;
666 clone
->bi_size
= to_bytes(len
);
667 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
672 static void __clone_and_map(struct clone_info
*ci
)
674 struct bio
*clone
, *bio
= ci
->bio
;
675 struct dm_target
*ti
= dm_table_find_target(ci
->map
, ci
->sector
);
676 sector_t len
= 0, max
= max_io_len(ci
->md
, ci
->sector
, ti
);
677 struct dm_target_io
*tio
;
680 * Allocate a target io object.
682 tio
= alloc_tio(ci
->md
);
685 memset(&tio
->info
, 0, sizeof(tio
->info
));
687 if (ci
->sector_count
<= max
) {
689 * Optimise for the simple case where we can do all of
690 * the remaining io with a single clone.
692 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
693 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
695 __map_bio(ti
, clone
, tio
);
696 ci
->sector_count
= 0;
698 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
700 * There are some bvecs that don't span targets.
701 * Do as many of these as possible.
704 sector_t remaining
= max
;
707 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
708 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
710 if (bv_len
> remaining
)
717 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
719 __map_bio(ti
, clone
, tio
);
722 ci
->sector_count
-= len
;
727 * Handle a bvec that must be split between two or more targets.
729 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
730 sector_t remaining
= to_sector(bv
->bv_len
);
731 unsigned int offset
= 0;
735 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
736 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
738 tio
= alloc_tio(ci
->md
);
741 memset(&tio
->info
, 0, sizeof(tio
->info
));
744 len
= min(remaining
, max
);
746 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
747 bv
->bv_offset
+ offset
, len
,
750 __map_bio(ti
, clone
, tio
);
753 ci
->sector_count
-= len
;
754 offset
+= to_bytes(len
);
755 } while (remaining
-= len
);
762 * Split the bio into several clones.
764 static int __split_bio(struct mapped_device
*md
, struct bio
*bio
)
766 struct clone_info ci
;
768 ci
.map
= dm_get_table(md
);
769 if (unlikely(!ci
.map
))
774 ci
.io
= alloc_io(md
);
776 atomic_set(&ci
.io
->io_count
, 1);
779 ci
.sector
= bio
->bi_sector
;
780 ci
.sector_count
= bio_sectors(bio
);
781 ci
.idx
= bio
->bi_idx
;
783 start_io_acct(ci
.io
);
784 while (ci
.sector_count
)
785 __clone_and_map(&ci
);
787 /* drop the extra reference count */
788 dec_pending(ci
.io
, 0);
789 dm_table_put(ci
.map
);
793 /*-----------------------------------------------------------------
795 *---------------------------------------------------------------*/
798 * The request function that just remaps the bio built up by
801 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
804 int rw
= bio_data_dir(bio
);
805 struct mapped_device
*md
= q
->queuedata
;
808 * There is no use in forwarding any barrier request since we can't
809 * guarantee it is (or can be) handled by the targets correctly.
811 if (unlikely(bio_barrier(bio
))) {
812 bio_endio(bio
, -EOPNOTSUPP
);
816 down_read(&md
->io_lock
);
818 disk_stat_inc(dm_disk(md
), ios
[rw
]);
819 disk_stat_add(dm_disk(md
), sectors
[rw
], bio_sectors(bio
));
822 * If we're suspended we have to queue
825 while (test_bit(DMF_BLOCK_IO
, &md
->flags
)) {
826 up_read(&md
->io_lock
);
828 if (bio_rw(bio
) != READA
)
829 r
= queue_io(md
, bio
);
835 * We're in a while loop, because someone could suspend
836 * before we get to the following read lock.
838 down_read(&md
->io_lock
);
841 r
= __split_bio(md
, bio
);
842 up_read(&md
->io_lock
);
851 static void dm_unplug_all(struct request_queue
*q
)
853 struct mapped_device
*md
= q
->queuedata
;
854 struct dm_table
*map
= dm_get_table(md
);
857 dm_table_unplug_all(map
);
862 static int dm_any_congested(void *congested_data
, int bdi_bits
)
865 struct mapped_device
*md
= (struct mapped_device
*) congested_data
;
866 struct dm_table
*map
= dm_get_table(md
);
868 if (!map
|| test_bit(DMF_BLOCK_IO
, &md
->flags
))
871 r
= dm_table_any_congested(map
, bdi_bits
);
877 /*-----------------------------------------------------------------
878 * An IDR is used to keep track of allocated minor numbers.
879 *---------------------------------------------------------------*/
880 static DEFINE_IDR(_minor_idr
);
882 static void free_minor(int minor
)
884 spin_lock(&_minor_lock
);
885 idr_remove(&_minor_idr
, minor
);
886 spin_unlock(&_minor_lock
);
890 * See if the device with a specific minor # is free.
892 static int specific_minor(struct mapped_device
*md
, int minor
)
896 if (minor
>= (1 << MINORBITS
))
899 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
903 spin_lock(&_minor_lock
);
905 if (idr_find(&_minor_idr
, minor
)) {
910 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
915 idr_remove(&_minor_idr
, m
);
921 spin_unlock(&_minor_lock
);
925 static int next_free_minor(struct mapped_device
*md
, int *minor
)
929 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
933 spin_lock(&_minor_lock
);
935 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
940 if (m
>= (1 << MINORBITS
)) {
941 idr_remove(&_minor_idr
, m
);
949 spin_unlock(&_minor_lock
);
953 static struct block_device_operations dm_blk_dops
;
956 * Allocate and initialise a blank device with a given minor.
958 static struct mapped_device
*alloc_dev(int minor
)
961 struct mapped_device
*md
= kmalloc(sizeof(*md
), GFP_KERNEL
);
965 DMWARN("unable to allocate device, out of memory.");
969 if (!try_module_get(THIS_MODULE
))
972 /* get a minor number for the dev */
973 if (minor
== DM_ANY_MINOR
)
974 r
= next_free_minor(md
, &minor
);
976 r
= specific_minor(md
, minor
);
980 memset(md
, 0, sizeof(*md
));
981 init_rwsem(&md
->io_lock
);
982 init_MUTEX(&md
->suspend_lock
);
983 spin_lock_init(&md
->pushback_lock
);
984 rwlock_init(&md
->map_lock
);
985 atomic_set(&md
->holders
, 1);
986 atomic_set(&md
->open_count
, 0);
987 atomic_set(&md
->event_nr
, 0);
989 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
991 goto bad1_free_minor
;
993 md
->queue
->queuedata
= md
;
994 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
995 md
->queue
->backing_dev_info
.congested_data
= md
;
996 blk_queue_make_request(md
->queue
, dm_request
);
997 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
998 md
->queue
->unplug_fn
= dm_unplug_all
;
1000 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1004 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1008 md
->bs
= bioset_create(16, 16);
1012 md
->disk
= alloc_disk(1);
1016 atomic_set(&md
->pending
, 0);
1017 init_waitqueue_head(&md
->wait
);
1018 init_waitqueue_head(&md
->eventq
);
1020 md
->disk
->major
= _major
;
1021 md
->disk
->first_minor
= minor
;
1022 md
->disk
->fops
= &dm_blk_dops
;
1023 md
->disk
->queue
= md
->queue
;
1024 md
->disk
->private_data
= md
;
1025 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1027 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1029 /* Populate the mapping, nobody knows we exist yet */
1030 spin_lock(&_minor_lock
);
1031 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1032 spin_unlock(&_minor_lock
);
1034 BUG_ON(old_md
!= MINOR_ALLOCED
);
1039 bioset_free(md
->bs
);
1041 mempool_destroy(md
->tio_pool
);
1043 mempool_destroy(md
->io_pool
);
1045 blk_cleanup_queue(md
->queue
);
1049 module_put(THIS_MODULE
);
1055 static void unlock_fs(struct mapped_device
*md
);
1057 static void free_dev(struct mapped_device
*md
)
1059 int minor
= md
->disk
->first_minor
;
1061 if (md
->suspended_bdev
) {
1063 bdput(md
->suspended_bdev
);
1065 mempool_destroy(md
->tio_pool
);
1066 mempool_destroy(md
->io_pool
);
1067 bioset_free(md
->bs
);
1068 del_gendisk(md
->disk
);
1071 spin_lock(&_minor_lock
);
1072 md
->disk
->private_data
= NULL
;
1073 spin_unlock(&_minor_lock
);
1076 blk_cleanup_queue(md
->queue
);
1077 module_put(THIS_MODULE
);
1082 * Bind a table to the device.
1084 static void event_callback(void *context
)
1086 struct mapped_device
*md
= (struct mapped_device
*) context
;
1088 atomic_inc(&md
->event_nr
);
1089 wake_up(&md
->eventq
);
1092 static void __set_size(struct mapped_device
*md
, sector_t size
)
1094 set_capacity(md
->disk
, size
);
1096 mutex_lock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1097 i_size_write(md
->suspended_bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1098 mutex_unlock(&md
->suspended_bdev
->bd_inode
->i_mutex
);
1101 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1103 struct request_queue
*q
= md
->queue
;
1106 size
= dm_table_get_size(t
);
1109 * Wipe any geometry if the size of the table changed.
1111 if (size
!= get_capacity(md
->disk
))
1112 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1114 if (md
->suspended_bdev
)
1115 __set_size(md
, size
);
1120 dm_table_event_callback(t
, event_callback
, md
);
1122 write_lock(&md
->map_lock
);
1124 dm_table_set_restrictions(t
, q
);
1125 write_unlock(&md
->map_lock
);
1130 static void __unbind(struct mapped_device
*md
)
1132 struct dm_table
*map
= md
->map
;
1137 dm_table_event_callback(map
, NULL
, NULL
);
1138 write_lock(&md
->map_lock
);
1140 write_unlock(&md
->map_lock
);
1145 * Constructor for a new device.
1147 int dm_create(int minor
, struct mapped_device
**result
)
1149 struct mapped_device
*md
;
1151 md
= alloc_dev(minor
);
1159 static struct mapped_device
*dm_find_md(dev_t dev
)
1161 struct mapped_device
*md
;
1162 unsigned minor
= MINOR(dev
);
1164 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1167 spin_lock(&_minor_lock
);
1169 md
= idr_find(&_minor_idr
, minor
);
1170 if (md
&& (md
== MINOR_ALLOCED
||
1171 (dm_disk(md
)->first_minor
!= minor
) ||
1172 test_bit(DMF_FREEING
, &md
->flags
))) {
1178 spin_unlock(&_minor_lock
);
1183 struct mapped_device
*dm_get_md(dev_t dev
)
1185 struct mapped_device
*md
= dm_find_md(dev
);
1193 void *dm_get_mdptr(struct mapped_device
*md
)
1195 return md
->interface_ptr
;
1198 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1200 md
->interface_ptr
= ptr
;
1203 void dm_get(struct mapped_device
*md
)
1205 atomic_inc(&md
->holders
);
1208 const char *dm_device_name(struct mapped_device
*md
)
1212 EXPORT_SYMBOL_GPL(dm_device_name
);
1214 void dm_put(struct mapped_device
*md
)
1216 struct dm_table
*map
;
1218 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1220 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1221 map
= dm_get_table(md
);
1222 idr_replace(&_minor_idr
, MINOR_ALLOCED
, dm_disk(md
)->first_minor
);
1223 set_bit(DMF_FREEING
, &md
->flags
);
1224 spin_unlock(&_minor_lock
);
1225 if (!dm_suspended(md
)) {
1226 dm_table_presuspend_targets(map
);
1227 dm_table_postsuspend_targets(map
);
1234 EXPORT_SYMBOL_GPL(dm_put
);
1237 * Process the deferred bios
1239 static void __flush_deferred_io(struct mapped_device
*md
, struct bio
*c
)
1246 if (__split_bio(md
, c
))
1253 * Swap in a new table (destroying old one).
1255 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1259 down(&md
->suspend_lock
);
1261 /* device must be suspended */
1262 if (!dm_suspended(md
))
1265 /* without bdev, the device size cannot be changed */
1266 if (!md
->suspended_bdev
)
1267 if (get_capacity(md
->disk
) != dm_table_get_size(table
))
1271 r
= __bind(md
, table
);
1274 up(&md
->suspend_lock
);
1279 * Functions to lock and unlock any filesystem running on the
1282 static int lock_fs(struct mapped_device
*md
)
1286 WARN_ON(md
->frozen_sb
);
1288 md
->frozen_sb
= freeze_bdev(md
->suspended_bdev
);
1289 if (IS_ERR(md
->frozen_sb
)) {
1290 r
= PTR_ERR(md
->frozen_sb
);
1291 md
->frozen_sb
= NULL
;
1295 set_bit(DMF_FROZEN
, &md
->flags
);
1297 /* don't bdput right now, we don't want the bdev
1298 * to go away while it is locked.
1303 static void unlock_fs(struct mapped_device
*md
)
1305 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1308 thaw_bdev(md
->suspended_bdev
, md
->frozen_sb
);
1309 md
->frozen_sb
= NULL
;
1310 clear_bit(DMF_FROZEN
, &md
->flags
);
1314 * We need to be able to change a mapping table under a mounted
1315 * filesystem. For example we might want to move some data in
1316 * the background. Before the table can be swapped with
1317 * dm_bind_table, dm_suspend must be called to flush any in
1318 * flight bios and ensure that any further io gets deferred.
1320 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1322 struct dm_table
*map
= NULL
;
1323 unsigned long flags
;
1324 DECLARE_WAITQUEUE(wait
, current
);
1327 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1328 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1330 down(&md
->suspend_lock
);
1332 if (dm_suspended(md
))
1335 map
= dm_get_table(md
);
1338 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1339 * This flag is cleared before dm_suspend returns.
1342 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1344 /* This does not get reverted if there's an error later. */
1345 dm_table_presuspend_targets(map
);
1347 /* bdget() can stall if the pending I/Os are not flushed */
1349 md
->suspended_bdev
= bdget_disk(md
->disk
, 0);
1350 if (!md
->suspended_bdev
) {
1351 DMWARN("bdget failed in dm_suspend");
1358 * Flush I/O to the device.
1359 * noflush supersedes do_lockfs, because lock_fs() needs to flush I/Os.
1361 if (do_lockfs
&& !noflush
) {
1368 * First we set the BLOCK_IO flag so no more ios will be mapped.
1370 down_write(&md
->io_lock
);
1371 set_bit(DMF_BLOCK_IO
, &md
->flags
);
1373 add_wait_queue(&md
->wait
, &wait
);
1374 up_write(&md
->io_lock
);
1378 dm_table_unplug_all(map
);
1381 * Then we wait for the already mapped ios to
1385 set_current_state(TASK_INTERRUPTIBLE
);
1387 if (!atomic_read(&md
->pending
) || signal_pending(current
))
1392 set_current_state(TASK_RUNNING
);
1394 down_write(&md
->io_lock
);
1395 remove_wait_queue(&md
->wait
, &wait
);
1398 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1399 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1400 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1401 bio_list_init(&md
->pushback
);
1402 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1405 /* were we interrupted ? */
1407 if (atomic_read(&md
->pending
)) {
1408 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1409 def
= bio_list_get(&md
->deferred
);
1410 __flush_deferred_io(md
, def
);
1411 up_write(&md
->io_lock
);
1413 goto out
; /* pushback list is already flushed, so skip flush */
1415 up_write(&md
->io_lock
);
1417 dm_table_postsuspend_targets(map
);
1419 set_bit(DMF_SUSPENDED
, &md
->flags
);
1426 * Because there may be already I/Os in the pushback list,
1427 * flush them before return.
1429 down_write(&md
->io_lock
);
1431 spin_lock_irqsave(&md
->pushback_lock
, flags
);
1432 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1433 bio_list_merge_head(&md
->deferred
, &md
->pushback
);
1434 bio_list_init(&md
->pushback
);
1435 spin_unlock_irqrestore(&md
->pushback_lock
, flags
);
1437 def
= bio_list_get(&md
->deferred
);
1438 __flush_deferred_io(md
, def
);
1439 up_write(&md
->io_lock
);
1443 if (r
&& md
->suspended_bdev
) {
1444 bdput(md
->suspended_bdev
);
1445 md
->suspended_bdev
= NULL
;
1451 up(&md
->suspend_lock
);
1455 int dm_resume(struct mapped_device
*md
)
1459 struct dm_table
*map
= NULL
;
1461 down(&md
->suspend_lock
);
1462 if (!dm_suspended(md
))
1465 map
= dm_get_table(md
);
1466 if (!map
|| !dm_table_get_size(map
))
1469 r
= dm_table_resume_targets(map
);
1473 down_write(&md
->io_lock
);
1474 clear_bit(DMF_BLOCK_IO
, &md
->flags
);
1476 def
= bio_list_get(&md
->deferred
);
1477 __flush_deferred_io(md
, def
);
1478 up_write(&md
->io_lock
);
1482 if (md
->suspended_bdev
) {
1483 bdput(md
->suspended_bdev
);
1484 md
->suspended_bdev
= NULL
;
1487 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1489 dm_table_unplug_all(map
);
1491 kobject_uevent(&md
->disk
->kobj
, KOBJ_CHANGE
);
1497 up(&md
->suspend_lock
);
1502 /*-----------------------------------------------------------------
1503 * Event notification.
1504 *---------------------------------------------------------------*/
1505 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1507 return atomic_read(&md
->event_nr
);
1510 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1512 return wait_event_interruptible(md
->eventq
,
1513 (event_nr
!= atomic_read(&md
->event_nr
)));
1517 * The gendisk is only valid as long as you have a reference
1520 struct gendisk
*dm_disk(struct mapped_device
*md
)
1525 int dm_suspended(struct mapped_device
*md
)
1527 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1530 int dm_noflush_suspending(struct dm_target
*ti
)
1532 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1533 int r
= __noflush_suspending(md
);
1539 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1541 static struct block_device_operations dm_blk_dops
= {
1542 .open
= dm_blk_open
,
1543 .release
= dm_blk_close
,
1544 .ioctl
= dm_blk_ioctl
,
1545 .getgeo
= dm_blk_getgeo
,
1546 .owner
= THIS_MODULE
1549 EXPORT_SYMBOL(dm_get_mapinfo
);
1554 module_init(dm_init
);
1555 module_exit(dm_exit
);
1557 module_param(major
, uint
, 0);
1558 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1559 MODULE_DESCRIPTION(DM_NAME
" driver");
1560 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1561 MODULE_LICENSE("GPL");