2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
27 static const char *_name
= DM_NAME
;
29 static unsigned int major
= 0;
30 static unsigned int _major
= 0;
32 static DEFINE_SPINLOCK(_minor_lock
);
35 * One of these is allocated per bio.
38 struct mapped_device
*md
;
42 unsigned long start_time
;
47 * One of these is allocated per target within a bio. Hopefully
48 * this will be simplified out one day.
57 * For request-based dm.
58 * One of these is allocated per request.
60 struct dm_rq_target_io
{
61 struct mapped_device
*md
;
63 struct request
*orig
, clone
;
69 * For request-based dm.
70 * One of these is allocated per bio.
72 struct dm_rq_clone_bio_info
{
77 union map_info
*dm_get_mapinfo(struct bio
*bio
)
79 if (bio
&& bio
->bi_private
)
80 return &((struct dm_target_io
*)bio
->bi_private
)->info
;
84 #define MINOR_ALLOCED ((void *)-1)
87 * Bits for the md->flags field.
89 #define DMF_BLOCK_IO_FOR_SUSPEND 0
90 #define DMF_SUSPENDED 1
93 #define DMF_DELETING 4
94 #define DMF_NOFLUSH_SUSPENDING 5
95 #define DMF_QUEUE_IO_TO_THREAD 6
98 * Work processed by per-device workqueue.
100 struct mapped_device
{
101 struct rw_semaphore io_lock
;
102 struct mutex suspend_lock
;
109 struct request_queue
*queue
;
110 struct gendisk
*disk
;
116 * A list of ios that arrived while we were suspended.
119 wait_queue_head_t wait
;
120 struct work_struct work
;
121 struct bio_list deferred
;
122 spinlock_t deferred_lock
;
125 * An error from the barrier request currently being processed.
130 * Processing queue (flush/barriers)
132 struct workqueue_struct
*wq
;
135 * The current mapping.
137 struct dm_table
*map
;
140 * io objects are allocated from here.
151 wait_queue_head_t eventq
;
153 struct list_head uevent_list
;
154 spinlock_t uevent_lock
; /* Protect access to uevent_list */
157 * freeze/thaw support require holding onto a super block
159 struct super_block
*frozen_sb
;
160 struct block_device
*bdev
;
162 /* forced geometry settings */
163 struct hd_geometry geometry
;
170 static struct kmem_cache
*_io_cache
;
171 static struct kmem_cache
*_tio_cache
;
172 static struct kmem_cache
*_rq_tio_cache
;
173 static struct kmem_cache
*_rq_bio_info_cache
;
175 static int __init
local_init(void)
179 /* allocate a slab for the dm_ios */
180 _io_cache
= KMEM_CACHE(dm_io
, 0);
184 /* allocate a slab for the target ios */
185 _tio_cache
= KMEM_CACHE(dm_target_io
, 0);
187 goto out_free_io_cache
;
189 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
191 goto out_free_tio_cache
;
193 _rq_bio_info_cache
= KMEM_CACHE(dm_rq_clone_bio_info
, 0);
194 if (!_rq_bio_info_cache
)
195 goto out_free_rq_tio_cache
;
197 r
= dm_uevent_init();
199 goto out_free_rq_bio_info_cache
;
202 r
= register_blkdev(_major
, _name
);
204 goto out_uevent_exit
;
213 out_free_rq_bio_info_cache
:
214 kmem_cache_destroy(_rq_bio_info_cache
);
215 out_free_rq_tio_cache
:
216 kmem_cache_destroy(_rq_tio_cache
);
218 kmem_cache_destroy(_tio_cache
);
220 kmem_cache_destroy(_io_cache
);
225 static void local_exit(void)
227 kmem_cache_destroy(_rq_bio_info_cache
);
228 kmem_cache_destroy(_rq_tio_cache
);
229 kmem_cache_destroy(_tio_cache
);
230 kmem_cache_destroy(_io_cache
);
231 unregister_blkdev(_major
, _name
);
236 DMINFO("cleaned up");
239 static int (*_inits
[])(void) __initdata
= {
248 static void (*_exits
[])(void) = {
257 static int __init
dm_init(void)
259 const int count
= ARRAY_SIZE(_inits
);
263 for (i
= 0; i
< count
; i
++) {
278 static void __exit
dm_exit(void)
280 int i
= ARRAY_SIZE(_exits
);
287 * Block device functions
289 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
291 struct mapped_device
*md
;
293 spin_lock(&_minor_lock
);
295 md
= bdev
->bd_disk
->private_data
;
299 if (test_bit(DMF_FREEING
, &md
->flags
) ||
300 test_bit(DMF_DELETING
, &md
->flags
)) {
306 atomic_inc(&md
->open_count
);
309 spin_unlock(&_minor_lock
);
311 return md
? 0 : -ENXIO
;
314 static int dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
316 struct mapped_device
*md
= disk
->private_data
;
317 atomic_dec(&md
->open_count
);
322 int dm_open_count(struct mapped_device
*md
)
324 return atomic_read(&md
->open_count
);
328 * Guarantees nothing is using the device before it's deleted.
330 int dm_lock_for_deletion(struct mapped_device
*md
)
334 spin_lock(&_minor_lock
);
336 if (dm_open_count(md
))
339 set_bit(DMF_DELETING
, &md
->flags
);
341 spin_unlock(&_minor_lock
);
346 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
348 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
350 return dm_get_geometry(md
, geo
);
353 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
354 unsigned int cmd
, unsigned long arg
)
356 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
357 struct dm_table
*map
= dm_get_table(md
);
358 struct dm_target
*tgt
;
361 if (!map
|| !dm_table_get_size(map
))
364 /* We only support devices that have a single target */
365 if (dm_table_get_num_targets(map
) != 1)
368 tgt
= dm_table_get_target(map
, 0);
370 if (dm_suspended(md
)) {
375 if (tgt
->type
->ioctl
)
376 r
= tgt
->type
->ioctl(tgt
, cmd
, arg
);
384 static struct dm_io
*alloc_io(struct mapped_device
*md
)
386 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
389 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
391 mempool_free(io
, md
->io_pool
);
394 static struct dm_target_io
*alloc_tio(struct mapped_device
*md
)
396 return mempool_alloc(md
->tio_pool
, GFP_NOIO
);
399 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
401 mempool_free(tio
, md
->tio_pool
);
404 static void start_io_acct(struct dm_io
*io
)
406 struct mapped_device
*md
= io
->md
;
409 io
->start_time
= jiffies
;
411 cpu
= part_stat_lock();
412 part_round_stats(cpu
, &dm_disk(md
)->part0
);
414 dm_disk(md
)->part0
.in_flight
= atomic_inc_return(&md
->pending
);
417 static void end_io_acct(struct dm_io
*io
)
419 struct mapped_device
*md
= io
->md
;
420 struct bio
*bio
= io
->bio
;
421 unsigned long duration
= jiffies
- io
->start_time
;
423 int rw
= bio_data_dir(bio
);
425 cpu
= part_stat_lock();
426 part_round_stats(cpu
, &dm_disk(md
)->part0
);
427 part_stat_add(cpu
, &dm_disk(md
)->part0
, ticks
[rw
], duration
);
431 * After this is decremented the bio must not be touched if it is
434 dm_disk(md
)->part0
.in_flight
= pending
=
435 atomic_dec_return(&md
->pending
);
437 /* nudge anyone waiting on suspend queue */
443 * Add the bio to the list of deferred io.
445 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
447 down_write(&md
->io_lock
);
449 spin_lock_irq(&md
->deferred_lock
);
450 bio_list_add(&md
->deferred
, bio
);
451 spin_unlock_irq(&md
->deferred_lock
);
453 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
))
454 queue_work(md
->wq
, &md
->work
);
456 up_write(&md
->io_lock
);
460 * Everyone (including functions in this file), should use this
461 * function to access the md->map field, and make sure they call
462 * dm_table_put() when finished.
464 struct dm_table
*dm_get_table(struct mapped_device
*md
)
468 read_lock(&md
->map_lock
);
472 read_unlock(&md
->map_lock
);
478 * Get the geometry associated with a dm device
480 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
488 * Set the geometry of a device.
490 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
492 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
494 if (geo
->start
> sz
) {
495 DMWARN("Start sector is beyond the geometry limits.");
504 /*-----------------------------------------------------------------
506 * A more elegant soln is in the works that uses the queue
507 * merge fn, unfortunately there are a couple of changes to
508 * the block layer that I want to make for this. So in the
509 * interests of getting something for people to use I give
510 * you this clearly demarcated crap.
511 *---------------------------------------------------------------*/
513 static int __noflush_suspending(struct mapped_device
*md
)
515 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
519 * Decrements the number of outstanding ios that a bio has been
520 * cloned into, completing the original io if necc.
522 static void dec_pending(struct dm_io
*io
, int error
)
527 struct mapped_device
*md
= io
->md
;
529 /* Push-back supersedes any I/O errors */
530 if (error
&& !(io
->error
> 0 && __noflush_suspending(md
)))
533 if (atomic_dec_and_test(&io
->io_count
)) {
534 if (io
->error
== DM_ENDIO_REQUEUE
) {
536 * Target requested pushing back the I/O.
538 spin_lock_irqsave(&md
->deferred_lock
, flags
);
539 if (__noflush_suspending(md
))
540 bio_list_add_head(&md
->deferred
, io
->bio
);
542 /* noflush suspend was interrupted. */
544 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
547 io_error
= io
->error
;
550 if (bio_barrier(bio
)) {
552 * There can be just one barrier request so we use
553 * a per-device variable for error reporting.
554 * Note that you can't touch the bio after end_io_acct
556 md
->barrier_error
= io_error
;
561 if (io_error
!= DM_ENDIO_REQUEUE
) {
562 trace_block_bio_complete(md
->queue
, bio
);
564 bio_endio(bio
, io_error
);
572 static void clone_endio(struct bio
*bio
, int error
)
575 struct dm_target_io
*tio
= bio
->bi_private
;
576 struct dm_io
*io
= tio
->io
;
577 struct mapped_device
*md
= tio
->io
->md
;
578 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
580 if (!bio_flagged(bio
, BIO_UPTODATE
) && !error
)
584 r
= endio(tio
->ti
, bio
, error
, &tio
->info
);
585 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
587 * error and requeue request are handled
591 else if (r
== DM_ENDIO_INCOMPLETE
)
592 /* The target will handle the io */
595 DMWARN("unimplemented target endio return value: %d", r
);
601 * Store md for cleanup instead of tio which is about to get freed.
603 bio
->bi_private
= md
->bs
;
607 dec_pending(io
, error
);
610 static sector_t
max_io_len(struct mapped_device
*md
,
611 sector_t sector
, struct dm_target
*ti
)
613 sector_t offset
= sector
- ti
->begin
;
614 sector_t len
= ti
->len
- offset
;
617 * Does the target need to split even further ?
621 boundary
= ((offset
+ ti
->split_io
) & ~(ti
->split_io
- 1))
630 static void __map_bio(struct dm_target
*ti
, struct bio
*clone
,
631 struct dm_target_io
*tio
)
635 struct mapped_device
*md
;
640 BUG_ON(!clone
->bi_size
);
642 clone
->bi_end_io
= clone_endio
;
643 clone
->bi_private
= tio
;
646 * Map the clone. If r == 0 we don't need to do
647 * anything, the target has assumed ownership of
650 atomic_inc(&tio
->io
->io_count
);
651 sector
= clone
->bi_sector
;
652 r
= ti
->type
->map(ti
, clone
, &tio
->info
);
653 if (r
== DM_MAPIO_REMAPPED
) {
654 /* the bio has been remapped so dispatch it */
656 trace_block_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
657 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
659 generic_make_request(clone
);
660 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
661 /* error the io and bail out, or requeue it if needed */
663 dec_pending(tio
->io
, r
);
665 * Store bio_set for cleanup.
667 clone
->bi_private
= md
->bs
;
671 DMWARN("unimplemented target map return value: %d", r
);
677 struct mapped_device
*md
;
678 struct dm_table
*map
;
682 sector_t sector_count
;
686 static void dm_bio_destructor(struct bio
*bio
)
688 struct bio_set
*bs
= bio
->bi_private
;
694 * Creates a little bio that is just does part of a bvec.
696 static struct bio
*split_bvec(struct bio
*bio
, sector_t sector
,
697 unsigned short idx
, unsigned int offset
,
698 unsigned int len
, struct bio_set
*bs
)
701 struct bio_vec
*bv
= bio
->bi_io_vec
+ idx
;
703 clone
= bio_alloc_bioset(GFP_NOIO
, 1, bs
);
704 clone
->bi_destructor
= dm_bio_destructor
;
705 *clone
->bi_io_vec
= *bv
;
707 clone
->bi_sector
= sector
;
708 clone
->bi_bdev
= bio
->bi_bdev
;
709 clone
->bi_rw
= bio
->bi_rw
& ~(1 << BIO_RW_BARRIER
);
711 clone
->bi_size
= to_bytes(len
);
712 clone
->bi_io_vec
->bv_offset
= offset
;
713 clone
->bi_io_vec
->bv_len
= clone
->bi_size
;
714 clone
->bi_flags
|= 1 << BIO_CLONED
;
716 if (bio_integrity(bio
)) {
717 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
718 bio_integrity_trim(clone
,
719 bio_sector_offset(bio
, idx
, offset
), len
);
726 * Creates a bio that consists of range of complete bvecs.
728 static struct bio
*clone_bio(struct bio
*bio
, sector_t sector
,
729 unsigned short idx
, unsigned short bv_count
,
730 unsigned int len
, struct bio_set
*bs
)
734 clone
= bio_alloc_bioset(GFP_NOIO
, bio
->bi_max_vecs
, bs
);
735 __bio_clone(clone
, bio
);
736 clone
->bi_rw
&= ~(1 << BIO_RW_BARRIER
);
737 clone
->bi_destructor
= dm_bio_destructor
;
738 clone
->bi_sector
= sector
;
740 clone
->bi_vcnt
= idx
+ bv_count
;
741 clone
->bi_size
= to_bytes(len
);
742 clone
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
744 if (bio_integrity(bio
)) {
745 bio_integrity_clone(clone
, bio
, GFP_NOIO
);
747 if (idx
!= bio
->bi_idx
|| clone
->bi_size
< bio
->bi_size
)
748 bio_integrity_trim(clone
,
749 bio_sector_offset(bio
, idx
, 0), len
);
755 static int __clone_and_map(struct clone_info
*ci
)
757 struct bio
*clone
, *bio
= ci
->bio
;
758 struct dm_target
*ti
;
759 sector_t len
= 0, max
;
760 struct dm_target_io
*tio
;
762 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
763 if (!dm_target_is_valid(ti
))
766 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
769 * Allocate a target io object.
771 tio
= alloc_tio(ci
->md
);
774 memset(&tio
->info
, 0, sizeof(tio
->info
));
776 if (ci
->sector_count
<= max
) {
778 * Optimise for the simple case where we can do all of
779 * the remaining io with a single clone.
781 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
,
782 bio
->bi_vcnt
- ci
->idx
, ci
->sector_count
,
784 __map_bio(ti
, clone
, tio
);
785 ci
->sector_count
= 0;
787 } else if (to_sector(bio
->bi_io_vec
[ci
->idx
].bv_len
) <= max
) {
789 * There are some bvecs that don't span targets.
790 * Do as many of these as possible.
793 sector_t remaining
= max
;
796 for (i
= ci
->idx
; remaining
&& (i
< bio
->bi_vcnt
); i
++) {
797 bv_len
= to_sector(bio
->bi_io_vec
[i
].bv_len
);
799 if (bv_len
> remaining
)
806 clone
= clone_bio(bio
, ci
->sector
, ci
->idx
, i
- ci
->idx
, len
,
808 __map_bio(ti
, clone
, tio
);
811 ci
->sector_count
-= len
;
816 * Handle a bvec that must be split between two or more targets.
818 struct bio_vec
*bv
= bio
->bi_io_vec
+ ci
->idx
;
819 sector_t remaining
= to_sector(bv
->bv_len
);
820 unsigned int offset
= 0;
824 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
825 if (!dm_target_is_valid(ti
))
828 max
= max_io_len(ci
->md
, ci
->sector
, ti
);
830 tio
= alloc_tio(ci
->md
);
833 memset(&tio
->info
, 0, sizeof(tio
->info
));
836 len
= min(remaining
, max
);
838 clone
= split_bvec(bio
, ci
->sector
, ci
->idx
,
839 bv
->bv_offset
+ offset
, len
,
842 __map_bio(ti
, clone
, tio
);
845 ci
->sector_count
-= len
;
846 offset
+= to_bytes(len
);
847 } while (remaining
-= len
);
856 * Split the bio into several clones and submit it to targets.
858 static void __split_and_process_bio(struct mapped_device
*md
, struct bio
*bio
)
860 struct clone_info ci
;
863 ci
.map
= dm_get_table(md
);
864 if (unlikely(!ci
.map
)) {
865 if (!bio_barrier(bio
))
868 md
->barrier_error
= -EIO
;
874 ci
.io
= alloc_io(md
);
876 atomic_set(&ci
.io
->io_count
, 1);
879 ci
.sector
= bio
->bi_sector
;
880 ci
.sector_count
= bio_sectors(bio
);
881 ci
.idx
= bio
->bi_idx
;
883 start_io_acct(ci
.io
);
884 while (ci
.sector_count
&& !error
)
885 error
= __clone_and_map(&ci
);
887 /* drop the extra reference count */
888 dec_pending(ci
.io
, error
);
889 dm_table_put(ci
.map
);
891 /*-----------------------------------------------------------------
893 *---------------------------------------------------------------*/
895 static int dm_merge_bvec(struct request_queue
*q
,
896 struct bvec_merge_data
*bvm
,
897 struct bio_vec
*biovec
)
899 struct mapped_device
*md
= q
->queuedata
;
900 struct dm_table
*map
= dm_get_table(md
);
901 struct dm_target
*ti
;
902 sector_t max_sectors
;
908 ti
= dm_table_find_target(map
, bvm
->bi_sector
);
909 if (!dm_target_is_valid(ti
))
913 * Find maximum amount of I/O that won't need splitting
915 max_sectors
= min(max_io_len(md
, bvm
->bi_sector
, ti
),
916 (sector_t
) BIO_MAX_SECTORS
);
917 max_size
= (max_sectors
<< SECTOR_SHIFT
) - bvm
->bi_size
;
922 * merge_bvec_fn() returns number of bytes
923 * it can accept at this offset
924 * max is precomputed maximal io size
926 if (max_size
&& ti
->type
->merge
)
927 max_size
= ti
->type
->merge(ti
, bvm
, biovec
, max_size
);
929 * If the target doesn't support merge method and some of the devices
930 * provided their merge_bvec method (we know this by looking at
931 * queue_max_hw_sectors), then we can't allow bios with multiple vector
932 * entries. So always set max_size to 0, and the code below allows
935 else if (queue_max_hw_sectors(q
) <= PAGE_SIZE
>> 9)
944 * Always allow an entire first page
946 if (max_size
<= biovec
->bv_len
&& !(bvm
->bi_size
>> SECTOR_SHIFT
))
947 max_size
= biovec
->bv_len
;
953 * The request function that just remaps the bio built up by
956 static int dm_request(struct request_queue
*q
, struct bio
*bio
)
958 int rw
= bio_data_dir(bio
);
959 struct mapped_device
*md
= q
->queuedata
;
962 down_read(&md
->io_lock
);
964 cpu
= part_stat_lock();
965 part_stat_inc(cpu
, &dm_disk(md
)->part0
, ios
[rw
]);
966 part_stat_add(cpu
, &dm_disk(md
)->part0
, sectors
[rw
], bio_sectors(bio
));
970 * If we're suspended or the thread is processing barriers
971 * we have to queue this io for later.
973 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
)) ||
974 unlikely(bio_barrier(bio
))) {
975 up_read(&md
->io_lock
);
977 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) &&
978 bio_rw(bio
) == READA
) {
988 __split_and_process_bio(md
, bio
);
989 up_read(&md
->io_lock
);
993 static void dm_unplug_all(struct request_queue
*q
)
995 struct mapped_device
*md
= q
->queuedata
;
996 struct dm_table
*map
= dm_get_table(md
);
999 dm_table_unplug_all(map
);
1004 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1007 struct mapped_device
*md
= congested_data
;
1008 struct dm_table
*map
;
1010 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1011 map
= dm_get_table(md
);
1013 r
= dm_table_any_congested(map
, bdi_bits
);
1021 /*-----------------------------------------------------------------
1022 * An IDR is used to keep track of allocated minor numbers.
1023 *---------------------------------------------------------------*/
1024 static DEFINE_IDR(_minor_idr
);
1026 static void free_minor(int minor
)
1028 spin_lock(&_minor_lock
);
1029 idr_remove(&_minor_idr
, minor
);
1030 spin_unlock(&_minor_lock
);
1034 * See if the device with a specific minor # is free.
1036 static int specific_minor(int minor
)
1040 if (minor
>= (1 << MINORBITS
))
1043 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1047 spin_lock(&_minor_lock
);
1049 if (idr_find(&_minor_idr
, minor
)) {
1054 r
= idr_get_new_above(&_minor_idr
, MINOR_ALLOCED
, minor
, &m
);
1059 idr_remove(&_minor_idr
, m
);
1065 spin_unlock(&_minor_lock
);
1069 static int next_free_minor(int *minor
)
1073 r
= idr_pre_get(&_minor_idr
, GFP_KERNEL
);
1077 spin_lock(&_minor_lock
);
1079 r
= idr_get_new(&_minor_idr
, MINOR_ALLOCED
, &m
);
1083 if (m
>= (1 << MINORBITS
)) {
1084 idr_remove(&_minor_idr
, m
);
1092 spin_unlock(&_minor_lock
);
1096 static struct block_device_operations dm_blk_dops
;
1098 static void dm_wq_work(struct work_struct
*work
);
1101 * Allocate and initialise a blank device with a given minor.
1103 static struct mapped_device
*alloc_dev(int minor
)
1106 struct mapped_device
*md
= kzalloc(sizeof(*md
), GFP_KERNEL
);
1110 DMWARN("unable to allocate device, out of memory.");
1114 if (!try_module_get(THIS_MODULE
))
1115 goto bad_module_get
;
1117 /* get a minor number for the dev */
1118 if (minor
== DM_ANY_MINOR
)
1119 r
= next_free_minor(&minor
);
1121 r
= specific_minor(minor
);
1125 init_rwsem(&md
->io_lock
);
1126 mutex_init(&md
->suspend_lock
);
1127 spin_lock_init(&md
->deferred_lock
);
1128 rwlock_init(&md
->map_lock
);
1129 atomic_set(&md
->holders
, 1);
1130 atomic_set(&md
->open_count
, 0);
1131 atomic_set(&md
->event_nr
, 0);
1132 atomic_set(&md
->uevent_seq
, 0);
1133 INIT_LIST_HEAD(&md
->uevent_list
);
1134 spin_lock_init(&md
->uevent_lock
);
1136 md
->queue
= blk_alloc_queue(GFP_KERNEL
);
1140 md
->queue
->queuedata
= md
;
1141 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
1142 md
->queue
->backing_dev_info
.congested_data
= md
;
1143 blk_queue_make_request(md
->queue
, dm_request
);
1144 blk_queue_ordered(md
->queue
, QUEUE_ORDERED_DRAIN
, NULL
);
1145 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
1146 md
->queue
->unplug_fn
= dm_unplug_all
;
1147 blk_queue_merge_bvec(md
->queue
, dm_merge_bvec
);
1149 md
->io_pool
= mempool_create_slab_pool(MIN_IOS
, _io_cache
);
1153 md
->tio_pool
= mempool_create_slab_pool(MIN_IOS
, _tio_cache
);
1157 md
->bs
= bioset_create(16, 0);
1161 md
->disk
= alloc_disk(1);
1165 atomic_set(&md
->pending
, 0);
1166 init_waitqueue_head(&md
->wait
);
1167 INIT_WORK(&md
->work
, dm_wq_work
);
1168 init_waitqueue_head(&md
->eventq
);
1170 md
->disk
->major
= _major
;
1171 md
->disk
->first_minor
= minor
;
1172 md
->disk
->fops
= &dm_blk_dops
;
1173 md
->disk
->queue
= md
->queue
;
1174 md
->disk
->private_data
= md
;
1175 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1177 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1179 md
->wq
= create_singlethread_workqueue("kdmflush");
1183 md
->bdev
= bdget_disk(md
->disk
, 0);
1187 /* Populate the mapping, nobody knows we exist yet */
1188 spin_lock(&_minor_lock
);
1189 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1190 spin_unlock(&_minor_lock
);
1192 BUG_ON(old_md
!= MINOR_ALLOCED
);
1197 destroy_workqueue(md
->wq
);
1201 bioset_free(md
->bs
);
1203 mempool_destroy(md
->tio_pool
);
1205 mempool_destroy(md
->io_pool
);
1207 blk_cleanup_queue(md
->queue
);
1211 module_put(THIS_MODULE
);
1217 static void unlock_fs(struct mapped_device
*md
);
1219 static void free_dev(struct mapped_device
*md
)
1221 int minor
= MINOR(disk_devt(md
->disk
));
1225 destroy_workqueue(md
->wq
);
1226 mempool_destroy(md
->tio_pool
);
1227 mempool_destroy(md
->io_pool
);
1228 bioset_free(md
->bs
);
1229 blk_integrity_unregister(md
->disk
);
1230 del_gendisk(md
->disk
);
1233 spin_lock(&_minor_lock
);
1234 md
->disk
->private_data
= NULL
;
1235 spin_unlock(&_minor_lock
);
1238 blk_cleanup_queue(md
->queue
);
1239 module_put(THIS_MODULE
);
1244 * Bind a table to the device.
1246 static void event_callback(void *context
)
1248 unsigned long flags
;
1250 struct mapped_device
*md
= (struct mapped_device
*) context
;
1252 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1253 list_splice_init(&md
->uevent_list
, &uevents
);
1254 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1256 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1258 atomic_inc(&md
->event_nr
);
1259 wake_up(&md
->eventq
);
1262 static void __set_size(struct mapped_device
*md
, sector_t size
)
1264 set_capacity(md
->disk
, size
);
1266 mutex_lock(&md
->bdev
->bd_inode
->i_mutex
);
1267 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1268 mutex_unlock(&md
->bdev
->bd_inode
->i_mutex
);
1271 static int __bind(struct mapped_device
*md
, struct dm_table
*t
)
1273 struct request_queue
*q
= md
->queue
;
1276 size
= dm_table_get_size(t
);
1279 * Wipe any geometry if the size of the table changed.
1281 if (size
!= get_capacity(md
->disk
))
1282 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1284 __set_size(md
, size
);
1287 dm_table_destroy(t
);
1291 dm_table_event_callback(t
, event_callback
, md
);
1293 write_lock(&md
->map_lock
);
1295 dm_table_set_restrictions(t
, q
);
1296 write_unlock(&md
->map_lock
);
1301 static void __unbind(struct mapped_device
*md
)
1303 struct dm_table
*map
= md
->map
;
1308 dm_table_event_callback(map
, NULL
, NULL
);
1309 write_lock(&md
->map_lock
);
1311 write_unlock(&md
->map_lock
);
1312 dm_table_destroy(map
);
1316 * Constructor for a new device.
1318 int dm_create(int minor
, struct mapped_device
**result
)
1320 struct mapped_device
*md
;
1322 md
= alloc_dev(minor
);
1332 static struct mapped_device
*dm_find_md(dev_t dev
)
1334 struct mapped_device
*md
;
1335 unsigned minor
= MINOR(dev
);
1337 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
1340 spin_lock(&_minor_lock
);
1342 md
= idr_find(&_minor_idr
, minor
);
1343 if (md
&& (md
== MINOR_ALLOCED
||
1344 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
1345 test_bit(DMF_FREEING
, &md
->flags
))) {
1351 spin_unlock(&_minor_lock
);
1356 struct mapped_device
*dm_get_md(dev_t dev
)
1358 struct mapped_device
*md
= dm_find_md(dev
);
1366 void *dm_get_mdptr(struct mapped_device
*md
)
1368 return md
->interface_ptr
;
1371 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
1373 md
->interface_ptr
= ptr
;
1376 void dm_get(struct mapped_device
*md
)
1378 atomic_inc(&md
->holders
);
1381 const char *dm_device_name(struct mapped_device
*md
)
1385 EXPORT_SYMBOL_GPL(dm_device_name
);
1387 void dm_put(struct mapped_device
*md
)
1389 struct dm_table
*map
;
1391 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
1393 if (atomic_dec_and_lock(&md
->holders
, &_minor_lock
)) {
1394 map
= dm_get_table(md
);
1395 idr_replace(&_minor_idr
, MINOR_ALLOCED
,
1396 MINOR(disk_devt(dm_disk(md
))));
1397 set_bit(DMF_FREEING
, &md
->flags
);
1398 spin_unlock(&_minor_lock
);
1399 if (!dm_suspended(md
)) {
1400 dm_table_presuspend_targets(map
);
1401 dm_table_postsuspend_targets(map
);
1409 EXPORT_SYMBOL_GPL(dm_put
);
1411 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
1414 DECLARE_WAITQUEUE(wait
, current
);
1416 dm_unplug_all(md
->queue
);
1418 add_wait_queue(&md
->wait
, &wait
);
1421 set_current_state(interruptible
);
1424 if (!atomic_read(&md
->pending
))
1427 if (interruptible
== TASK_INTERRUPTIBLE
&&
1428 signal_pending(current
)) {
1435 set_current_state(TASK_RUNNING
);
1437 remove_wait_queue(&md
->wait
, &wait
);
1442 static void dm_flush(struct mapped_device
*md
)
1444 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
1447 static void process_barrier(struct mapped_device
*md
, struct bio
*bio
)
1451 if (bio_empty_barrier(bio
)) {
1456 __split_and_process_bio(md
, bio
);
1459 if (md
->barrier_error
!= DM_ENDIO_REQUEUE
)
1460 bio_endio(bio
, md
->barrier_error
);
1464 * Process the deferred bios
1466 static void dm_wq_work(struct work_struct
*work
)
1468 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
1472 down_write(&md
->io_lock
);
1474 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1475 spin_lock_irq(&md
->deferred_lock
);
1476 c
= bio_list_pop(&md
->deferred
);
1477 spin_unlock_irq(&md
->deferred_lock
);
1480 clear_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
);
1484 up_write(&md
->io_lock
);
1487 process_barrier(md
, c
);
1489 __split_and_process_bio(md
, c
);
1491 down_write(&md
->io_lock
);
1494 up_write(&md
->io_lock
);
1497 static void dm_queue_flush(struct mapped_device
*md
)
1499 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
1500 smp_mb__after_clear_bit();
1501 queue_work(md
->wq
, &md
->work
);
1505 * Swap in a new table (destroying old one).
1507 int dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
1511 mutex_lock(&md
->suspend_lock
);
1513 /* device must be suspended */
1514 if (!dm_suspended(md
))
1518 r
= __bind(md
, table
);
1521 mutex_unlock(&md
->suspend_lock
);
1526 * Functions to lock and unlock any filesystem running on the
1529 static int lock_fs(struct mapped_device
*md
)
1533 WARN_ON(md
->frozen_sb
);
1535 md
->frozen_sb
= freeze_bdev(md
->bdev
);
1536 if (IS_ERR(md
->frozen_sb
)) {
1537 r
= PTR_ERR(md
->frozen_sb
);
1538 md
->frozen_sb
= NULL
;
1542 set_bit(DMF_FROZEN
, &md
->flags
);
1547 static void unlock_fs(struct mapped_device
*md
)
1549 if (!test_bit(DMF_FROZEN
, &md
->flags
))
1552 thaw_bdev(md
->bdev
, md
->frozen_sb
);
1553 md
->frozen_sb
= NULL
;
1554 clear_bit(DMF_FROZEN
, &md
->flags
);
1558 * We need to be able to change a mapping table under a mounted
1559 * filesystem. For example we might want to move some data in
1560 * the background. Before the table can be swapped with
1561 * dm_bind_table, dm_suspend must be called to flush any in
1562 * flight bios and ensure that any further io gets deferred.
1564 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
1566 struct dm_table
*map
= NULL
;
1568 int do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
? 1 : 0;
1569 int noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
? 1 : 0;
1571 mutex_lock(&md
->suspend_lock
);
1573 if (dm_suspended(md
)) {
1578 map
= dm_get_table(md
);
1581 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
1582 * This flag is cleared before dm_suspend returns.
1585 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1587 /* This does not get reverted if there's an error later. */
1588 dm_table_presuspend_targets(map
);
1591 * Flush I/O to the device. noflush supersedes do_lockfs,
1592 * because lock_fs() needs to flush I/Os.
1594 if (!noflush
&& do_lockfs
) {
1601 * Here we must make sure that no processes are submitting requests
1602 * to target drivers i.e. no one may be executing
1603 * __split_and_process_bio. This is called from dm_request and
1606 * To get all processes out of __split_and_process_bio in dm_request,
1607 * we take the write lock. To prevent any process from reentering
1608 * __split_and_process_bio from dm_request, we set
1609 * DMF_QUEUE_IO_TO_THREAD.
1611 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
1612 * and call flush_workqueue(md->wq). flush_workqueue will wait until
1613 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
1614 * further calls to __split_and_process_bio from dm_wq_work.
1616 down_write(&md
->io_lock
);
1617 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
1618 set_bit(DMF_QUEUE_IO_TO_THREAD
, &md
->flags
);
1619 up_write(&md
->io_lock
);
1621 flush_workqueue(md
->wq
);
1624 * At this point no more requests are entering target request routines.
1625 * We call dm_wait_for_completion to wait for all existing requests
1628 r
= dm_wait_for_completion(md
, TASK_INTERRUPTIBLE
);
1630 down_write(&md
->io_lock
);
1632 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
1633 up_write(&md
->io_lock
);
1635 /* were we interrupted ? */
1640 goto out
; /* pushback list is already flushed, so skip flush */
1644 * If dm_wait_for_completion returned 0, the device is completely
1645 * quiescent now. There is no request-processing activity. All new
1646 * requests are being added to md->deferred list.
1649 dm_table_postsuspend_targets(map
);
1651 set_bit(DMF_SUSPENDED
, &md
->flags
);
1657 mutex_unlock(&md
->suspend_lock
);
1661 int dm_resume(struct mapped_device
*md
)
1664 struct dm_table
*map
= NULL
;
1666 mutex_lock(&md
->suspend_lock
);
1667 if (!dm_suspended(md
))
1670 map
= dm_get_table(md
);
1671 if (!map
|| !dm_table_get_size(map
))
1674 r
= dm_table_resume_targets(map
);
1682 clear_bit(DMF_SUSPENDED
, &md
->flags
);
1684 dm_table_unplug_all(map
);
1686 dm_kobject_uevent(md
);
1692 mutex_unlock(&md
->suspend_lock
);
1697 /*-----------------------------------------------------------------
1698 * Event notification.
1699 *---------------------------------------------------------------*/
1700 void dm_kobject_uevent(struct mapped_device
*md
)
1702 kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, KOBJ_CHANGE
);
1705 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
1707 return atomic_add_return(1, &md
->uevent_seq
);
1710 uint32_t dm_get_event_nr(struct mapped_device
*md
)
1712 return atomic_read(&md
->event_nr
);
1715 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
1717 return wait_event_interruptible(md
->eventq
,
1718 (event_nr
!= atomic_read(&md
->event_nr
)));
1721 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
1723 unsigned long flags
;
1725 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1726 list_add(elist
, &md
->uevent_list
);
1727 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1731 * The gendisk is only valid as long as you have a reference
1734 struct gendisk
*dm_disk(struct mapped_device
*md
)
1739 struct kobject
*dm_kobject(struct mapped_device
*md
)
1745 * struct mapped_device should not be exported outside of dm.c
1746 * so use this check to verify that kobj is part of md structure
1748 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
1750 struct mapped_device
*md
;
1752 md
= container_of(kobj
, struct mapped_device
, kobj
);
1753 if (&md
->kobj
!= kobj
)
1756 if (test_bit(DMF_FREEING
, &md
->flags
) ||
1757 test_bit(DMF_DELETING
, &md
->flags
))
1764 int dm_suspended(struct mapped_device
*md
)
1766 return test_bit(DMF_SUSPENDED
, &md
->flags
);
1769 int dm_noflush_suspending(struct dm_target
*ti
)
1771 struct mapped_device
*md
= dm_table_get_md(ti
->table
);
1772 int r
= __noflush_suspending(md
);
1778 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
1780 static struct block_device_operations dm_blk_dops
= {
1781 .open
= dm_blk_open
,
1782 .release
= dm_blk_close
,
1783 .ioctl
= dm_blk_ioctl
,
1784 .getgeo
= dm_blk_getgeo
,
1785 .owner
= THIS_MODULE
1788 EXPORT_SYMBOL(dm_get_mapinfo
);
1793 module_init(dm_init
);
1794 module_exit(dm_exit
);
1796 module_param(major
, uint
, 0);
1797 MODULE_PARM_DESC(major
, "The major number of the device mapper");
1798 MODULE_DESCRIPTION(DM_NAME
" driver");
1799 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
1800 MODULE_LICENSE("GPL");