4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
6 * This file is released under the GPL.
9 #include <linux/blkdev.h>
10 #include <linux/device-mapper.h>
11 #include <linux/delay.h>
13 #include <linux/init.h>
14 #include <linux/kdev_t.h>
15 #include <linux/list.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/slab.h>
19 #include <linux/vmalloc.h>
20 #include <linux/log2.h>
21 #include <linux/dm-kcopyd.h>
22 #include <linux/workqueue.h>
24 #include "dm-exception-store.h"
26 #define DM_MSG_PREFIX "snapshots"
29 * The percentage increment we will wake up users at
31 #define WAKE_UP_PERCENT 5
34 * kcopyd priority of snapshot operations
36 #define SNAPSHOT_COPY_PRIORITY 2
39 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
41 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
44 * The size of the mempool used to track chunks in use.
48 #define DM_TRACKED_CHUNK_HASH_SIZE 16
49 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
50 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
52 struct exception_table
{
55 struct list_head
*table
;
59 struct rw_semaphore lock
;
61 struct dm_dev
*origin
;
63 /* List of snapshots per Origin */
64 struct list_head list
;
66 /* You can't use a snapshot if this is 0 (e.g. if full) */
69 /* Origin writes don't trigger exceptions until this is set */
72 mempool_t
*pending_pool
;
74 atomic_t pending_exceptions_count
;
76 struct exception_table pending
;
77 struct exception_table complete
;
80 * pe_lock protects all pending_exception operations and access
81 * as well as the snapshot_bios list.
85 /* The on disk metadata handler */
86 struct dm_exception_store
*store
;
88 struct dm_kcopyd_client
*kcopyd_client
;
90 /* Queue of snapshot writes for ksnapd to flush */
91 struct bio_list queued_bios
;
92 struct work_struct queued_bios_work
;
94 /* Chunks with outstanding reads */
95 mempool_t
*tracked_chunk_pool
;
96 spinlock_t tracked_chunk_lock
;
97 struct hlist_head tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH_SIZE
];
100 static struct workqueue_struct
*ksnapd
;
101 static void flush_queued_bios(struct work_struct
*work
);
103 static sector_t
chunk_to_sector(struct dm_exception_store
*store
,
106 return chunk
<< store
->chunk_shift
;
109 static int bdev_equal(struct block_device
*lhs
, struct block_device
*rhs
)
112 * There is only ever one instance of a particular block
113 * device so we can compare pointers safely.
118 struct dm_snap_pending_exception
{
119 struct dm_snap_exception e
;
122 * Origin buffers waiting for this to complete are held
125 struct bio_list origin_bios
;
126 struct bio_list snapshot_bios
;
129 * Short-term queue of pending exceptions prior to submission.
131 struct list_head list
;
134 * The primary pending_exception is the one that holds
135 * the ref_count and the list of origin_bios for a
136 * group of pending_exceptions. It is always last to get freed.
137 * These fields get set up when writing to the origin.
139 struct dm_snap_pending_exception
*primary_pe
;
142 * Number of pending_exceptions processing this chunk.
143 * When this drops to zero we must complete the origin bios.
144 * If incrementing or decrementing this, hold pe->snap->lock for
145 * the sibling concerned and not pe->primary_pe->snap->lock unless
150 /* Pointer back to snapshot context */
151 struct dm_snapshot
*snap
;
154 * 1 indicates the exception has already been sent to
161 * Hash table mapping origin volumes to lists of snapshots and
162 * a lock to protect it
164 static struct kmem_cache
*exception_cache
;
165 static struct kmem_cache
*pending_cache
;
167 struct dm_snap_tracked_chunk
{
168 struct hlist_node node
;
172 static struct kmem_cache
*tracked_chunk_cache
;
174 static struct dm_snap_tracked_chunk
*track_chunk(struct dm_snapshot
*s
,
177 struct dm_snap_tracked_chunk
*c
= mempool_alloc(s
->tracked_chunk_pool
,
183 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
184 hlist_add_head(&c
->node
,
185 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)]);
186 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
191 static void stop_tracking_chunk(struct dm_snapshot
*s
,
192 struct dm_snap_tracked_chunk
*c
)
196 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
198 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
200 mempool_free(c
, s
->tracked_chunk_pool
);
203 static int __chunk_is_tracked(struct dm_snapshot
*s
, chunk_t chunk
)
205 struct dm_snap_tracked_chunk
*c
;
206 struct hlist_node
*hn
;
209 spin_lock_irq(&s
->tracked_chunk_lock
);
211 hlist_for_each_entry(c
, hn
,
212 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)], node
) {
213 if (c
->chunk
== chunk
) {
219 spin_unlock_irq(&s
->tracked_chunk_lock
);
225 * One of these per registered origin, held in the snapshot_origins hash
228 /* The origin device */
229 struct block_device
*bdev
;
231 struct list_head hash_list
;
233 /* List of snapshots for this origin */
234 struct list_head snapshots
;
238 * Size of the hash table for origin volumes. If we make this
239 * the size of the minors list then it should be nearly perfect
241 #define ORIGIN_HASH_SIZE 256
242 #define ORIGIN_MASK 0xFF
243 static struct list_head
*_origins
;
244 static struct rw_semaphore _origins_lock
;
246 static int init_origin_hash(void)
250 _origins
= kmalloc(ORIGIN_HASH_SIZE
* sizeof(struct list_head
),
253 DMERR("unable to allocate memory");
257 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
258 INIT_LIST_HEAD(_origins
+ i
);
259 init_rwsem(&_origins_lock
);
264 static void exit_origin_hash(void)
269 static unsigned origin_hash(struct block_device
*bdev
)
271 return bdev
->bd_dev
& ORIGIN_MASK
;
274 static struct origin
*__lookup_origin(struct block_device
*origin
)
276 struct list_head
*ol
;
279 ol
= &_origins
[origin_hash(origin
)];
280 list_for_each_entry (o
, ol
, hash_list
)
281 if (bdev_equal(o
->bdev
, origin
))
287 static void __insert_origin(struct origin
*o
)
289 struct list_head
*sl
= &_origins
[origin_hash(o
->bdev
)];
290 list_add_tail(&o
->hash_list
, sl
);
294 * Make a note of the snapshot and its origin so we can look it
295 * up when the origin has a write on it.
297 static int register_snapshot(struct dm_snapshot
*snap
)
299 struct dm_snapshot
*l
;
300 struct origin
*o
, *new_o
;
301 struct block_device
*bdev
= snap
->origin
->bdev
;
303 new_o
= kmalloc(sizeof(*new_o
), GFP_KERNEL
);
307 down_write(&_origins_lock
);
308 o
= __lookup_origin(bdev
);
316 /* Initialise the struct */
317 INIT_LIST_HEAD(&o
->snapshots
);
323 /* Sort the list according to chunk size, largest-first smallest-last */
324 list_for_each_entry(l
, &o
->snapshots
, list
)
325 if (l
->store
->chunk_size
< snap
->store
->chunk_size
)
327 list_add_tail(&snap
->list
, &l
->list
);
329 up_write(&_origins_lock
);
333 static void unregister_snapshot(struct dm_snapshot
*s
)
337 down_write(&_origins_lock
);
338 o
= __lookup_origin(s
->origin
->bdev
);
341 if (list_empty(&o
->snapshots
)) {
342 list_del(&o
->hash_list
);
346 up_write(&_origins_lock
);
350 * Implementation of the exception hash tables.
351 * The lowest hash_shift bits of the chunk number are ignored, allowing
352 * some consecutive chunks to be grouped together.
354 static int init_exception_table(struct exception_table
*et
, uint32_t size
,
359 et
->hash_shift
= hash_shift
;
360 et
->hash_mask
= size
- 1;
361 et
->table
= dm_vcalloc(size
, sizeof(struct list_head
));
365 for (i
= 0; i
< size
; i
++)
366 INIT_LIST_HEAD(et
->table
+ i
);
371 static void exit_exception_table(struct exception_table
*et
, struct kmem_cache
*mem
)
373 struct list_head
*slot
;
374 struct dm_snap_exception
*ex
, *next
;
377 size
= et
->hash_mask
+ 1;
378 for (i
= 0; i
< size
; i
++) {
379 slot
= et
->table
+ i
;
381 list_for_each_entry_safe (ex
, next
, slot
, hash_list
)
382 kmem_cache_free(mem
, ex
);
388 static uint32_t exception_hash(struct exception_table
*et
, chunk_t chunk
)
390 return (chunk
>> et
->hash_shift
) & et
->hash_mask
;
393 static void insert_exception(struct exception_table
*eh
,
394 struct dm_snap_exception
*e
)
396 struct list_head
*l
= &eh
->table
[exception_hash(eh
, e
->old_chunk
)];
397 list_add(&e
->hash_list
, l
);
400 static void remove_exception(struct dm_snap_exception
*e
)
402 list_del(&e
->hash_list
);
406 * Return the exception data for a sector, or NULL if not
409 static struct dm_snap_exception
*lookup_exception(struct exception_table
*et
,
412 struct list_head
*slot
;
413 struct dm_snap_exception
*e
;
415 slot
= &et
->table
[exception_hash(et
, chunk
)];
416 list_for_each_entry (e
, slot
, hash_list
)
417 if (chunk
>= e
->old_chunk
&&
418 chunk
<= e
->old_chunk
+ dm_consecutive_chunk_count(e
))
424 static struct dm_snap_exception
*alloc_exception(void)
426 struct dm_snap_exception
*e
;
428 e
= kmem_cache_alloc(exception_cache
, GFP_NOIO
);
430 e
= kmem_cache_alloc(exception_cache
, GFP_ATOMIC
);
435 static void free_exception(struct dm_snap_exception
*e
)
437 kmem_cache_free(exception_cache
, e
);
440 static struct dm_snap_pending_exception
*alloc_pending_exception(struct dm_snapshot
*s
)
442 struct dm_snap_pending_exception
*pe
= mempool_alloc(s
->pending_pool
,
445 atomic_inc(&s
->pending_exceptions_count
);
451 static void free_pending_exception(struct dm_snap_pending_exception
*pe
)
453 struct dm_snapshot
*s
= pe
->snap
;
455 mempool_free(pe
, s
->pending_pool
);
456 smp_mb__before_atomic_dec();
457 atomic_dec(&s
->pending_exceptions_count
);
460 static void insert_completed_exception(struct dm_snapshot
*s
,
461 struct dm_snap_exception
*new_e
)
463 struct exception_table
*eh
= &s
->complete
;
465 struct dm_snap_exception
*e
= NULL
;
467 l
= &eh
->table
[exception_hash(eh
, new_e
->old_chunk
)];
469 /* Add immediately if this table doesn't support consecutive chunks */
473 /* List is ordered by old_chunk */
474 list_for_each_entry_reverse(e
, l
, hash_list
) {
475 /* Insert after an existing chunk? */
476 if (new_e
->old_chunk
== (e
->old_chunk
+
477 dm_consecutive_chunk_count(e
) + 1) &&
478 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) +
479 dm_consecutive_chunk_count(e
) + 1)) {
480 dm_consecutive_chunk_count_inc(e
);
481 free_exception(new_e
);
485 /* Insert before an existing chunk? */
486 if (new_e
->old_chunk
== (e
->old_chunk
- 1) &&
487 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) - 1)) {
488 dm_consecutive_chunk_count_inc(e
);
491 free_exception(new_e
);
495 if (new_e
->old_chunk
> e
->old_chunk
)
500 list_add(&new_e
->hash_list
, e
? &e
->hash_list
: l
);
504 * Callback used by the exception stores to load exceptions when
507 static int dm_add_exception(void *context
, chunk_t old
, chunk_t
new)
509 struct dm_snapshot
*s
= context
;
510 struct dm_snap_exception
*e
;
512 e
= alloc_exception();
518 /* Consecutive_count is implicitly initialised to zero */
521 insert_completed_exception(s
, e
);
529 static int calc_max_buckets(void)
531 /* use a fixed size of 2MB */
532 unsigned long mem
= 2 * 1024 * 1024;
533 mem
/= sizeof(struct list_head
);
539 * Allocate room for a suitable hash table.
541 static int init_hash_tables(struct dm_snapshot
*s
)
543 sector_t hash_size
, cow_dev_size
, origin_dev_size
, max_buckets
;
546 * Calculate based on the size of the original volume or
549 cow_dev_size
= get_dev_size(s
->store
->cow
->bdev
);
550 origin_dev_size
= get_dev_size(s
->origin
->bdev
);
551 max_buckets
= calc_max_buckets();
553 hash_size
= min(origin_dev_size
, cow_dev_size
) >> s
->store
->chunk_shift
;
554 hash_size
= min(hash_size
, max_buckets
);
556 hash_size
= rounddown_pow_of_two(hash_size
);
557 if (init_exception_table(&s
->complete
, hash_size
,
558 DM_CHUNK_CONSECUTIVE_BITS
))
562 * Allocate hash table for in-flight exceptions
563 * Make this smaller than the real hash table
569 if (init_exception_table(&s
->pending
, hash_size
, 0)) {
570 exit_exception_table(&s
->complete
, exception_cache
);
578 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
580 static int snapshot_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
582 struct dm_snapshot
*s
;
586 struct dm_exception_store
*store
;
590 ti
->error
= "requires exactly 4 arguments";
595 origin_path
= argv
[0];
599 r
= dm_exception_store_create(ti
, argc
, argv
, &args_used
, &store
);
601 ti
->error
= "Couldn't create exception store";
609 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
611 ti
->error
= "Cannot allocate snapshot context private "
617 r
= dm_get_device(ti
, origin_path
, 0, ti
->len
, FMODE_READ
, &s
->origin
);
619 ti
->error
= "Cannot get origin device";
626 atomic_set(&s
->pending_exceptions_count
, 0);
627 init_rwsem(&s
->lock
);
628 spin_lock_init(&s
->pe_lock
);
630 /* Allocate hash table for COW data */
631 if (init_hash_tables(s
)) {
632 ti
->error
= "Unable to allocate hash table space";
634 goto bad_hash_tables
;
637 r
= dm_kcopyd_client_create(SNAPSHOT_PAGES
, &s
->kcopyd_client
);
639 ti
->error
= "Could not create kcopyd client";
643 s
->pending_pool
= mempool_create_slab_pool(MIN_IOS
, pending_cache
);
644 if (!s
->pending_pool
) {
645 ti
->error
= "Could not allocate mempool for pending exceptions";
646 goto bad_pending_pool
;
649 s
->tracked_chunk_pool
= mempool_create_slab_pool(MIN_IOS
,
650 tracked_chunk_cache
);
651 if (!s
->tracked_chunk_pool
) {
652 ti
->error
= "Could not allocate tracked_chunk mempool for "
654 goto bad_tracked_chunk_pool
;
657 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
658 INIT_HLIST_HEAD(&s
->tracked_chunk_hash
[i
]);
660 spin_lock_init(&s
->tracked_chunk_lock
);
662 /* Metadata must only be loaded into one table at once */
663 r
= s
->store
->type
->read_metadata(s
->store
, dm_add_exception
,
666 ti
->error
= "Failed to read snapshot metadata";
667 goto bad_load_and_register
;
670 DMWARN("Snapshot is marked invalid.");
673 bio_list_init(&s
->queued_bios
);
674 INIT_WORK(&s
->queued_bios_work
, flush_queued_bios
);
676 /* Add snapshot to the list of snapshots for this origin */
677 /* Exceptions aren't triggered till snapshot_resume() is called */
678 if (register_snapshot(s
)) {
680 ti
->error
= "Cannot register snapshot origin";
681 goto bad_load_and_register
;
685 ti
->split_io
= s
->store
->chunk_size
;
686 ti
->num_flush_requests
= 1;
690 bad_load_and_register
:
691 mempool_destroy(s
->tracked_chunk_pool
);
693 bad_tracked_chunk_pool
:
694 mempool_destroy(s
->pending_pool
);
697 dm_kcopyd_client_destroy(s
->kcopyd_client
);
700 exit_exception_table(&s
->pending
, pending_cache
);
701 exit_exception_table(&s
->complete
, exception_cache
);
704 dm_put_device(ti
, s
->origin
);
710 dm_exception_store_destroy(store
);
716 static void __free_exceptions(struct dm_snapshot
*s
)
718 dm_kcopyd_client_destroy(s
->kcopyd_client
);
719 s
->kcopyd_client
= NULL
;
721 exit_exception_table(&s
->pending
, pending_cache
);
722 exit_exception_table(&s
->complete
, exception_cache
);
725 static void snapshot_dtr(struct dm_target
*ti
)
727 #ifdef CONFIG_DM_DEBUG
730 struct dm_snapshot
*s
= ti
->private;
732 flush_workqueue(ksnapd
);
734 /* Prevent further origin writes from using this snapshot. */
735 /* After this returns there can be no new kcopyd jobs. */
736 unregister_snapshot(s
);
738 while (atomic_read(&s
->pending_exceptions_count
))
741 * Ensure instructions in mempool_destroy aren't reordered
742 * before atomic_read.
746 #ifdef CONFIG_DM_DEBUG
747 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
748 BUG_ON(!hlist_empty(&s
->tracked_chunk_hash
[i
]));
751 mempool_destroy(s
->tracked_chunk_pool
);
753 __free_exceptions(s
);
755 mempool_destroy(s
->pending_pool
);
757 dm_put_device(ti
, s
->origin
);
759 dm_exception_store_destroy(s
->store
);
765 * Flush a list of buffers.
767 static void flush_bios(struct bio
*bio
)
774 generic_make_request(bio
);
779 static void flush_queued_bios(struct work_struct
*work
)
781 struct dm_snapshot
*s
=
782 container_of(work
, struct dm_snapshot
, queued_bios_work
);
783 struct bio
*queued_bios
;
786 spin_lock_irqsave(&s
->pe_lock
, flags
);
787 queued_bios
= bio_list_get(&s
->queued_bios
);
788 spin_unlock_irqrestore(&s
->pe_lock
, flags
);
790 flush_bios(queued_bios
);
794 * Error a list of buffers.
796 static void error_bios(struct bio
*bio
)
808 static void __invalidate_snapshot(struct dm_snapshot
*s
, int err
)
814 DMERR("Invalidating snapshot: Error reading/writing.");
815 else if (err
== -ENOMEM
)
816 DMERR("Invalidating snapshot: Unable to allocate exception.");
818 if (s
->store
->type
->drop_snapshot
)
819 s
->store
->type
->drop_snapshot(s
->store
);
823 dm_table_event(s
->store
->ti
->table
);
826 static void get_pending_exception(struct dm_snap_pending_exception
*pe
)
828 atomic_inc(&pe
->ref_count
);
831 static struct bio
*put_pending_exception(struct dm_snap_pending_exception
*pe
)
833 struct dm_snap_pending_exception
*primary_pe
;
834 struct bio
*origin_bios
= NULL
;
836 primary_pe
= pe
->primary_pe
;
839 * If this pe is involved in a write to the origin and
840 * it is the last sibling to complete then release
841 * the bios for the original write to the origin.
844 atomic_dec_and_test(&primary_pe
->ref_count
)) {
845 origin_bios
= bio_list_get(&primary_pe
->origin_bios
);
846 free_pending_exception(primary_pe
);
850 * Free the pe if it's not linked to an origin write or if
851 * it's not itself a primary pe.
853 if (!primary_pe
|| primary_pe
!= pe
)
854 free_pending_exception(pe
);
859 static void pending_complete(struct dm_snap_pending_exception
*pe
, int success
)
861 struct dm_snap_exception
*e
;
862 struct dm_snapshot
*s
= pe
->snap
;
863 struct bio
*origin_bios
= NULL
;
864 struct bio
*snapshot_bios
= NULL
;
868 /* Read/write error - snapshot is unusable */
869 down_write(&s
->lock
);
870 __invalidate_snapshot(s
, -EIO
);
875 e
= alloc_exception();
877 down_write(&s
->lock
);
878 __invalidate_snapshot(s
, -ENOMEM
);
884 down_write(&s
->lock
);
892 * Check for conflicting reads. This is extremely improbable,
893 * so msleep(1) is sufficient and there is no need for a wait queue.
895 while (__chunk_is_tracked(s
, pe
->e
.old_chunk
))
899 * Add a proper exception, and remove the
900 * in-flight exception from the list.
902 insert_completed_exception(s
, e
);
905 remove_exception(&pe
->e
);
906 snapshot_bios
= bio_list_get(&pe
->snapshot_bios
);
907 origin_bios
= put_pending_exception(pe
);
911 /* Submit any pending write bios */
913 error_bios(snapshot_bios
);
915 flush_bios(snapshot_bios
);
917 flush_bios(origin_bios
);
920 static void commit_callback(void *context
, int success
)
922 struct dm_snap_pending_exception
*pe
= context
;
924 pending_complete(pe
, success
);
928 * Called when the copy I/O has finished. kcopyd actually runs
929 * this code so don't block.
931 static void copy_callback(int read_err
, unsigned long write_err
, void *context
)
933 struct dm_snap_pending_exception
*pe
= context
;
934 struct dm_snapshot
*s
= pe
->snap
;
936 if (read_err
|| write_err
)
937 pending_complete(pe
, 0);
940 /* Update the metadata if we are persistent */
941 s
->store
->type
->commit_exception(s
->store
, &pe
->e
,
942 commit_callback
, pe
);
946 * Dispatches the copy operation to kcopyd.
948 static void start_copy(struct dm_snap_pending_exception
*pe
)
950 struct dm_snapshot
*s
= pe
->snap
;
951 struct dm_io_region src
, dest
;
952 struct block_device
*bdev
= s
->origin
->bdev
;
955 dev_size
= get_dev_size(bdev
);
958 src
.sector
= chunk_to_sector(s
->store
, pe
->e
.old_chunk
);
959 src
.count
= min(s
->store
->chunk_size
, dev_size
- src
.sector
);
961 dest
.bdev
= s
->store
->cow
->bdev
;
962 dest
.sector
= chunk_to_sector(s
->store
, pe
->e
.new_chunk
);
963 dest
.count
= src
.count
;
965 /* Hand over to kcopyd */
966 dm_kcopyd_copy(s
->kcopyd_client
,
967 &src
, 1, &dest
, 0, copy_callback
, pe
);
970 static struct dm_snap_pending_exception
*
971 __lookup_pending_exception(struct dm_snapshot
*s
, chunk_t chunk
)
973 struct dm_snap_exception
*e
= lookup_exception(&s
->pending
, chunk
);
978 return container_of(e
, struct dm_snap_pending_exception
, e
);
982 * Looks to see if this snapshot already has a pending exception
983 * for this chunk, otherwise it allocates a new one and inserts
984 * it into the pending table.
986 * NOTE: a write lock must be held on snap->lock before calling
989 static struct dm_snap_pending_exception
*
990 __find_pending_exception(struct dm_snapshot
*s
,
991 struct dm_snap_pending_exception
*pe
, chunk_t chunk
)
993 struct dm_snap_pending_exception
*pe2
;
995 pe2
= __lookup_pending_exception(s
, chunk
);
997 free_pending_exception(pe
);
1001 pe
->e
.old_chunk
= chunk
;
1002 bio_list_init(&pe
->origin_bios
);
1003 bio_list_init(&pe
->snapshot_bios
);
1004 pe
->primary_pe
= NULL
;
1005 atomic_set(&pe
->ref_count
, 0);
1008 if (s
->store
->type
->prepare_exception(s
->store
, &pe
->e
)) {
1009 free_pending_exception(pe
);
1013 get_pending_exception(pe
);
1014 insert_exception(&s
->pending
, &pe
->e
);
1019 static void remap_exception(struct dm_snapshot
*s
, struct dm_snap_exception
*e
,
1020 struct bio
*bio
, chunk_t chunk
)
1022 bio
->bi_bdev
= s
->store
->cow
->bdev
;
1023 bio
->bi_sector
= chunk_to_sector(s
->store
,
1024 dm_chunk_number(e
->new_chunk
) +
1025 (chunk
- e
->old_chunk
)) +
1027 s
->store
->chunk_mask
);
1030 static int snapshot_map(struct dm_target
*ti
, struct bio
*bio
,
1031 union map_info
*map_context
)
1033 struct dm_snap_exception
*e
;
1034 struct dm_snapshot
*s
= ti
->private;
1035 int r
= DM_MAPIO_REMAPPED
;
1037 struct dm_snap_pending_exception
*pe
= NULL
;
1039 if (unlikely(bio_empty_barrier(bio
))) {
1040 bio
->bi_bdev
= s
->store
->cow
->bdev
;
1041 return DM_MAPIO_REMAPPED
;
1044 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1046 /* Full snapshots are not usable */
1047 /* To get here the table must be live so s->active is always set. */
1051 /* FIXME: should only take write lock if we need
1052 * to copy an exception */
1053 down_write(&s
->lock
);
1060 /* If the block is already remapped - use that, else remap it */
1061 e
= lookup_exception(&s
->complete
, chunk
);
1063 remap_exception(s
, e
, bio
, chunk
);
1068 * Write to snapshot - higher level takes care of RW/RO
1069 * flags so we should only get this if we are
1072 if (bio_rw(bio
) == WRITE
) {
1073 pe
= __lookup_pending_exception(s
, chunk
);
1076 pe
= alloc_pending_exception(s
);
1077 down_write(&s
->lock
);
1080 free_pending_exception(pe
);
1085 e
= lookup_exception(&s
->complete
, chunk
);
1087 free_pending_exception(pe
);
1088 remap_exception(s
, e
, bio
, chunk
);
1092 pe
= __find_pending_exception(s
, pe
, chunk
);
1094 __invalidate_snapshot(s
, -ENOMEM
);
1100 remap_exception(s
, &pe
->e
, bio
, chunk
);
1101 bio_list_add(&pe
->snapshot_bios
, bio
);
1103 r
= DM_MAPIO_SUBMITTED
;
1106 /* this is protected by snap->lock */
1113 bio
->bi_bdev
= s
->origin
->bdev
;
1114 map_context
->ptr
= track_chunk(s
, chunk
);
1123 static int snapshot_end_io(struct dm_target
*ti
, struct bio
*bio
,
1124 int error
, union map_info
*map_context
)
1126 struct dm_snapshot
*s
= ti
->private;
1127 struct dm_snap_tracked_chunk
*c
= map_context
->ptr
;
1130 stop_tracking_chunk(s
, c
);
1135 static void snapshot_resume(struct dm_target
*ti
)
1137 struct dm_snapshot
*s
= ti
->private;
1139 down_write(&s
->lock
);
1144 static int snapshot_status(struct dm_target
*ti
, status_type_t type
,
1145 char *result
, unsigned int maxlen
)
1148 struct dm_snapshot
*snap
= ti
->private;
1151 case STATUSTYPE_INFO
:
1155 if (snap
->store
->type
->fraction_full
) {
1156 sector_t numerator
, denominator
;
1157 snap
->store
->type
->fraction_full(snap
->store
,
1161 (unsigned long long)numerator
,
1162 (unsigned long long)denominator
);
1169 case STATUSTYPE_TABLE
:
1171 * kdevname returns a static pointer so we need
1172 * to make private copies if the output is to
1175 DMEMIT("%s", snap
->origin
->name
);
1176 snap
->store
->type
->status(snap
->store
, type
, result
+ sz
,
1184 static int snapshot_iterate_devices(struct dm_target
*ti
,
1185 iterate_devices_callout_fn fn
, void *data
)
1187 struct dm_snapshot
*snap
= ti
->private;
1189 return fn(ti
, snap
->origin
, 0, ti
->len
, data
);
1193 /*-----------------------------------------------------------------
1195 *---------------------------------------------------------------*/
1196 static int __origin_write(struct list_head
*snapshots
, struct bio
*bio
)
1198 int r
= DM_MAPIO_REMAPPED
, first
= 0;
1199 struct dm_snapshot
*snap
;
1200 struct dm_snap_exception
*e
;
1201 struct dm_snap_pending_exception
*pe
, *next_pe
, *primary_pe
= NULL
;
1203 LIST_HEAD(pe_queue
);
1205 /* Do all the snapshots on this origin */
1206 list_for_each_entry (snap
, snapshots
, list
) {
1208 down_write(&snap
->lock
);
1210 /* Only deal with valid and active snapshots */
1211 if (!snap
->valid
|| !snap
->active
)
1214 /* Nothing to do if writing beyond end of snapshot */
1215 if (bio
->bi_sector
>= dm_table_get_size(snap
->store
->ti
->table
))
1219 * Remember, different snapshots can have
1220 * different chunk sizes.
1222 chunk
= sector_to_chunk(snap
->store
, bio
->bi_sector
);
1225 * Check exception table to see if block
1226 * is already remapped in this snapshot
1227 * and trigger an exception if not.
1229 * ref_count is initialised to 1 so pending_complete()
1230 * won't destroy the primary_pe while we're inside this loop.
1232 e
= lookup_exception(&snap
->complete
, chunk
);
1236 pe
= __lookup_pending_exception(snap
, chunk
);
1238 up_write(&snap
->lock
);
1239 pe
= alloc_pending_exception(snap
);
1240 down_write(&snap
->lock
);
1243 free_pending_exception(pe
);
1247 e
= lookup_exception(&snap
->complete
, chunk
);
1249 free_pending_exception(pe
);
1253 pe
= __find_pending_exception(snap
, pe
, chunk
);
1255 __invalidate_snapshot(snap
, -ENOMEM
);
1262 * Either every pe here has same
1263 * primary_pe or none has one yet.
1266 primary_pe
= pe
->primary_pe
;
1272 bio_list_add(&primary_pe
->origin_bios
, bio
);
1274 r
= DM_MAPIO_SUBMITTED
;
1277 if (!pe
->primary_pe
) {
1278 pe
->primary_pe
= primary_pe
;
1279 get_pending_exception(primary_pe
);
1284 list_add_tail(&pe
->list
, &pe_queue
);
1288 up_write(&snap
->lock
);
1295 * If this is the first time we're processing this chunk and
1296 * ref_count is now 1 it means all the pending exceptions
1297 * got completed while we were in the loop above, so it falls to
1298 * us here to remove the primary_pe and submit any origin_bios.
1301 if (first
&& atomic_dec_and_test(&primary_pe
->ref_count
)) {
1302 flush_bios(bio_list_get(&primary_pe
->origin_bios
));
1303 free_pending_exception(primary_pe
);
1304 /* If we got here, pe_queue is necessarily empty. */
1309 * Now that we have a complete pe list we can start the copying.
1311 list_for_each_entry_safe(pe
, next_pe
, &pe_queue
, list
)
1318 * Called on a write from the origin driver.
1320 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
)
1323 int r
= DM_MAPIO_REMAPPED
;
1325 down_read(&_origins_lock
);
1326 o
= __lookup_origin(origin
->bdev
);
1328 r
= __origin_write(&o
->snapshots
, bio
);
1329 up_read(&_origins_lock
);
1335 * Origin: maps a linear range of a device, with hooks for snapshotting.
1339 * Construct an origin mapping: <dev_path>
1340 * The context for an origin is merely a 'struct dm_dev *'
1341 * pointing to the real device.
1343 static int origin_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
1349 ti
->error
= "origin: incorrect number of arguments";
1353 r
= dm_get_device(ti
, argv
[0], 0, ti
->len
,
1354 dm_table_get_mode(ti
->table
), &dev
);
1356 ti
->error
= "Cannot get target device";
1361 ti
->num_flush_requests
= 1;
1366 static void origin_dtr(struct dm_target
*ti
)
1368 struct dm_dev
*dev
= ti
->private;
1369 dm_put_device(ti
, dev
);
1372 static int origin_map(struct dm_target
*ti
, struct bio
*bio
,
1373 union map_info
*map_context
)
1375 struct dm_dev
*dev
= ti
->private;
1376 bio
->bi_bdev
= dev
->bdev
;
1378 if (unlikely(bio_empty_barrier(bio
)))
1379 return DM_MAPIO_REMAPPED
;
1381 /* Only tell snapshots if this is a write */
1382 return (bio_rw(bio
) == WRITE
) ? do_origin(dev
, bio
) : DM_MAPIO_REMAPPED
;
1385 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1388 * Set the target "split_io" field to the minimum of all the snapshots'
1391 static void origin_resume(struct dm_target
*ti
)
1393 struct dm_dev
*dev
= ti
->private;
1394 struct dm_snapshot
*snap
;
1396 chunk_t chunk_size
= 0;
1398 down_read(&_origins_lock
);
1399 o
= __lookup_origin(dev
->bdev
);
1401 list_for_each_entry (snap
, &o
->snapshots
, list
)
1402 chunk_size
= min_not_zero(chunk_size
,
1403 snap
->store
->chunk_size
);
1404 up_read(&_origins_lock
);
1406 ti
->split_io
= chunk_size
;
1409 static int origin_status(struct dm_target
*ti
, status_type_t type
, char *result
,
1410 unsigned int maxlen
)
1412 struct dm_dev
*dev
= ti
->private;
1415 case STATUSTYPE_INFO
:
1419 case STATUSTYPE_TABLE
:
1420 snprintf(result
, maxlen
, "%s", dev
->name
);
1427 static int origin_iterate_devices(struct dm_target
*ti
,
1428 iterate_devices_callout_fn fn
, void *data
)
1430 struct dm_dev
*dev
= ti
->private;
1432 return fn(ti
, dev
, 0, ti
->len
, data
);
1435 static struct target_type origin_target
= {
1436 .name
= "snapshot-origin",
1437 .version
= {1, 7, 0},
1438 .module
= THIS_MODULE
,
1442 .resume
= origin_resume
,
1443 .status
= origin_status
,
1444 .iterate_devices
= origin_iterate_devices
,
1447 static struct target_type snapshot_target
= {
1449 .version
= {1, 7, 0},
1450 .module
= THIS_MODULE
,
1451 .ctr
= snapshot_ctr
,
1452 .dtr
= snapshot_dtr
,
1453 .map
= snapshot_map
,
1454 .end_io
= snapshot_end_io
,
1455 .resume
= snapshot_resume
,
1456 .status
= snapshot_status
,
1457 .iterate_devices
= snapshot_iterate_devices
,
1460 static int __init
dm_snapshot_init(void)
1464 r
= dm_exception_store_init();
1466 DMERR("Failed to initialize exception stores");
1470 r
= dm_register_target(&snapshot_target
);
1472 DMERR("snapshot target register failed %d", r
);
1473 goto bad_register_snapshot_target
;
1476 r
= dm_register_target(&origin_target
);
1478 DMERR("Origin target register failed %d", r
);
1482 r
= init_origin_hash();
1484 DMERR("init_origin_hash failed.");
1488 exception_cache
= KMEM_CACHE(dm_snap_exception
, 0);
1489 if (!exception_cache
) {
1490 DMERR("Couldn't create exception cache.");
1495 pending_cache
= KMEM_CACHE(dm_snap_pending_exception
, 0);
1496 if (!pending_cache
) {
1497 DMERR("Couldn't create pending cache.");
1502 tracked_chunk_cache
= KMEM_CACHE(dm_snap_tracked_chunk
, 0);
1503 if (!tracked_chunk_cache
) {
1504 DMERR("Couldn't create cache to track chunks in use.");
1509 ksnapd
= create_singlethread_workqueue("ksnapd");
1511 DMERR("Failed to create ksnapd workqueue.");
1513 goto bad_pending_pool
;
1519 kmem_cache_destroy(tracked_chunk_cache
);
1521 kmem_cache_destroy(pending_cache
);
1523 kmem_cache_destroy(exception_cache
);
1527 dm_unregister_target(&origin_target
);
1529 dm_unregister_target(&snapshot_target
);
1531 bad_register_snapshot_target
:
1532 dm_exception_store_exit();
1536 static void __exit
dm_snapshot_exit(void)
1538 destroy_workqueue(ksnapd
);
1540 dm_unregister_target(&snapshot_target
);
1541 dm_unregister_target(&origin_target
);
1544 kmem_cache_destroy(pending_cache
);
1545 kmem_cache_destroy(exception_cache
);
1546 kmem_cache_destroy(tracked_chunk_cache
);
1548 dm_exception_store_exit();
1552 module_init(dm_snapshot_init
);
1553 module_exit(dm_snapshot_exit
);
1555 MODULE_DESCRIPTION(DM_NAME
" snapshot target");
1556 MODULE_AUTHOR("Joe Thornber");
1557 MODULE_LICENSE("GPL");