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"
28 static const char dm_snapshot_merge_target_name
[] = "snapshot-merge";
30 #define dm_target_is_snapshot_merge(ti) \
31 ((ti)->type->name == dm_snapshot_merge_target_name)
34 * The percentage increment we will wake up users at
36 #define WAKE_UP_PERCENT 5
39 * kcopyd priority of snapshot operations
41 #define SNAPSHOT_COPY_PRIORITY 2
44 * Reserve 1MB for each snapshot initially (with minimum of 1 page).
46 #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
49 * The size of the mempool used to track chunks in use.
53 #define DM_TRACKED_CHUNK_HASH_SIZE 16
54 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
55 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
57 struct dm_exception_table
{
60 struct list_head
*table
;
64 struct rw_semaphore lock
;
66 struct dm_dev
*origin
;
71 /* List of snapshots per Origin */
72 struct list_head list
;
74 /* You can't use a snapshot if this is 0 (e.g. if full) */
77 /* Origin writes don't trigger exceptions until this is set */
80 /* Whether or not owning mapped_device is suspended */
83 mempool_t
*pending_pool
;
85 atomic_t pending_exceptions_count
;
87 struct dm_exception_table pending
;
88 struct dm_exception_table complete
;
91 * pe_lock protects all pending_exception operations and access
92 * as well as the snapshot_bios list.
96 /* The on disk metadata handler */
97 struct dm_exception_store
*store
;
99 struct dm_kcopyd_client
*kcopyd_client
;
101 /* Queue of snapshot writes for ksnapd to flush */
102 struct bio_list queued_bios
;
103 struct work_struct queued_bios_work
;
105 /* Chunks with outstanding reads */
106 mempool_t
*tracked_chunk_pool
;
107 spinlock_t tracked_chunk_lock
;
108 struct hlist_head tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH_SIZE
];
110 /* Wait for events based on state_bits */
111 unsigned long state_bits
;
113 /* Range of chunks currently being merged. */
114 chunk_t first_merging_chunk
;
115 int num_merging_chunks
;
118 * Incoming bios that overlap with chunks being merged must wait
119 * for them to be committed.
121 struct bio_list bios_queued_during_merge
;
126 * RUNNING_MERGE - Merge operation is in progress.
127 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
128 * cleared afterwards.
130 #define RUNNING_MERGE 0
131 #define SHUTDOWN_MERGE 1
133 struct dm_dev
*dm_snap_cow(struct dm_snapshot
*s
)
137 EXPORT_SYMBOL(dm_snap_cow
);
139 static struct workqueue_struct
*ksnapd
;
140 static void flush_queued_bios(struct work_struct
*work
);
142 static sector_t
chunk_to_sector(struct dm_exception_store
*store
,
145 return chunk
<< store
->chunk_shift
;
148 static int bdev_equal(struct block_device
*lhs
, struct block_device
*rhs
)
151 * There is only ever one instance of a particular block
152 * device so we can compare pointers safely.
157 struct dm_snap_pending_exception
{
158 struct dm_exception e
;
161 * Origin buffers waiting for this to complete are held
164 struct bio_list origin_bios
;
165 struct bio_list snapshot_bios
;
167 /* Pointer back to snapshot context */
168 struct dm_snapshot
*snap
;
171 * 1 indicates the exception has already been sent to
178 * Hash table mapping origin volumes to lists of snapshots and
179 * a lock to protect it
181 static struct kmem_cache
*exception_cache
;
182 static struct kmem_cache
*pending_cache
;
184 struct dm_snap_tracked_chunk
{
185 struct hlist_node node
;
189 static struct kmem_cache
*tracked_chunk_cache
;
191 static struct dm_snap_tracked_chunk
*track_chunk(struct dm_snapshot
*s
,
194 struct dm_snap_tracked_chunk
*c
= mempool_alloc(s
->tracked_chunk_pool
,
200 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
201 hlist_add_head(&c
->node
,
202 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)]);
203 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
208 static void stop_tracking_chunk(struct dm_snapshot
*s
,
209 struct dm_snap_tracked_chunk
*c
)
213 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
215 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
217 mempool_free(c
, s
->tracked_chunk_pool
);
220 static int __chunk_is_tracked(struct dm_snapshot
*s
, chunk_t chunk
)
222 struct dm_snap_tracked_chunk
*c
;
223 struct hlist_node
*hn
;
226 spin_lock_irq(&s
->tracked_chunk_lock
);
228 hlist_for_each_entry(c
, hn
,
229 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)], node
) {
230 if (c
->chunk
== chunk
) {
236 spin_unlock_irq(&s
->tracked_chunk_lock
);
242 * This conflicting I/O is extremely improbable in the caller,
243 * so msleep(1) is sufficient and there is no need for a wait queue.
245 static void __check_for_conflicting_io(struct dm_snapshot
*s
, chunk_t chunk
)
247 while (__chunk_is_tracked(s
, chunk
))
252 * One of these per registered origin, held in the snapshot_origins hash
255 /* The origin device */
256 struct block_device
*bdev
;
258 struct list_head hash_list
;
260 /* List of snapshots for this origin */
261 struct list_head snapshots
;
265 * Size of the hash table for origin volumes. If we make this
266 * the size of the minors list then it should be nearly perfect
268 #define ORIGIN_HASH_SIZE 256
269 #define ORIGIN_MASK 0xFF
270 static struct list_head
*_origins
;
271 static struct rw_semaphore _origins_lock
;
273 static int init_origin_hash(void)
277 _origins
= kmalloc(ORIGIN_HASH_SIZE
* sizeof(struct list_head
),
280 DMERR("unable to allocate memory");
284 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
285 INIT_LIST_HEAD(_origins
+ i
);
286 init_rwsem(&_origins_lock
);
291 static void exit_origin_hash(void)
296 static unsigned origin_hash(struct block_device
*bdev
)
298 return bdev
->bd_dev
& ORIGIN_MASK
;
301 static struct origin
*__lookup_origin(struct block_device
*origin
)
303 struct list_head
*ol
;
306 ol
= &_origins
[origin_hash(origin
)];
307 list_for_each_entry (o
, ol
, hash_list
)
308 if (bdev_equal(o
->bdev
, origin
))
314 static void __insert_origin(struct origin
*o
)
316 struct list_head
*sl
= &_origins
[origin_hash(o
->bdev
)];
317 list_add_tail(&o
->hash_list
, sl
);
321 * _origins_lock must be held when calling this function.
322 * Returns number of snapshots registered using the supplied cow device, plus:
323 * snap_src - a snapshot suitable for use as a source of exception handover
324 * snap_dest - a snapshot capable of receiving exception handover.
325 * snap_merge - an existing snapshot-merge target linked to the same origin.
326 * There can be at most one snapshot-merge target. The parameter is optional.
328 * Possible return values and states of snap_src and snap_dest.
329 * 0: NULL, NULL - first new snapshot
330 * 1: snap_src, NULL - normal snapshot
331 * 2: snap_src, snap_dest - waiting for handover
332 * 2: snap_src, NULL - handed over, waiting for old to be deleted
333 * 1: NULL, snap_dest - source got destroyed without handover
335 static int __find_snapshots_sharing_cow(struct dm_snapshot
*snap
,
336 struct dm_snapshot
**snap_src
,
337 struct dm_snapshot
**snap_dest
,
338 struct dm_snapshot
**snap_merge
)
340 struct dm_snapshot
*s
;
345 o
= __lookup_origin(snap
->origin
->bdev
);
349 list_for_each_entry(s
, &o
->snapshots
, list
) {
350 if (dm_target_is_snapshot_merge(s
->ti
) && snap_merge
)
352 if (!bdev_equal(s
->cow
->bdev
, snap
->cow
->bdev
))
362 } else if (snap_dest
)
373 * On success, returns 1 if this snapshot is a handover destination,
374 * otherwise returns 0.
376 static int __validate_exception_handover(struct dm_snapshot
*snap
)
378 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
379 struct dm_snapshot
*snap_merge
= NULL
;
381 /* Does snapshot need exceptions handed over to it? */
382 if ((__find_snapshots_sharing_cow(snap
, &snap_src
, &snap_dest
,
383 &snap_merge
) == 2) ||
385 snap
->ti
->error
= "Snapshot cow pairing for exception "
386 "table handover failed";
391 * If no snap_src was found, snap cannot become a handover
398 * Non-snapshot-merge handover?
400 if (!dm_target_is_snapshot_merge(snap
->ti
))
404 * Do not allow more than one merging snapshot.
407 snap
->ti
->error
= "A snapshot is already merging.";
411 if (!snap_src
->store
->type
->prepare_merge
||
412 !snap_src
->store
->type
->commit_merge
) {
413 snap
->ti
->error
= "Snapshot exception store does not "
414 "support snapshot-merge.";
421 static void __insert_snapshot(struct origin
*o
, struct dm_snapshot
*s
)
423 struct dm_snapshot
*l
;
425 /* Sort the list according to chunk size, largest-first smallest-last */
426 list_for_each_entry(l
, &o
->snapshots
, list
)
427 if (l
->store
->chunk_size
< s
->store
->chunk_size
)
429 list_add_tail(&s
->list
, &l
->list
);
433 * Make a note of the snapshot and its origin so we can look it
434 * up when the origin has a write on it.
436 * Also validate snapshot exception store handovers.
437 * On success, returns 1 if this registration is a handover destination,
438 * otherwise returns 0.
440 static int register_snapshot(struct dm_snapshot
*snap
)
442 struct origin
*o
, *new_o
= NULL
;
443 struct block_device
*bdev
= snap
->origin
->bdev
;
446 new_o
= kmalloc(sizeof(*new_o
), GFP_KERNEL
);
450 down_write(&_origins_lock
);
452 r
= __validate_exception_handover(snap
);
458 o
= __lookup_origin(bdev
);
465 /* Initialise the struct */
466 INIT_LIST_HEAD(&o
->snapshots
);
472 __insert_snapshot(o
, snap
);
475 up_write(&_origins_lock
);
481 * Move snapshot to correct place in list according to chunk size.
483 static void reregister_snapshot(struct dm_snapshot
*s
)
485 struct block_device
*bdev
= s
->origin
->bdev
;
487 down_write(&_origins_lock
);
490 __insert_snapshot(__lookup_origin(bdev
), s
);
492 up_write(&_origins_lock
);
495 static void unregister_snapshot(struct dm_snapshot
*s
)
499 down_write(&_origins_lock
);
500 o
= __lookup_origin(s
->origin
->bdev
);
503 if (o
&& list_empty(&o
->snapshots
)) {
504 list_del(&o
->hash_list
);
508 up_write(&_origins_lock
);
512 * Implementation of the exception hash tables.
513 * The lowest hash_shift bits of the chunk number are ignored, allowing
514 * some consecutive chunks to be grouped together.
516 static int dm_exception_table_init(struct dm_exception_table
*et
,
517 uint32_t size
, unsigned hash_shift
)
521 et
->hash_shift
= hash_shift
;
522 et
->hash_mask
= size
- 1;
523 et
->table
= dm_vcalloc(size
, sizeof(struct list_head
));
527 for (i
= 0; i
< size
; i
++)
528 INIT_LIST_HEAD(et
->table
+ i
);
533 static void dm_exception_table_exit(struct dm_exception_table
*et
,
534 struct kmem_cache
*mem
)
536 struct list_head
*slot
;
537 struct dm_exception
*ex
, *next
;
540 size
= et
->hash_mask
+ 1;
541 for (i
= 0; i
< size
; i
++) {
542 slot
= et
->table
+ i
;
544 list_for_each_entry_safe (ex
, next
, slot
, hash_list
)
545 kmem_cache_free(mem
, ex
);
551 static uint32_t exception_hash(struct dm_exception_table
*et
, chunk_t chunk
)
553 return (chunk
>> et
->hash_shift
) & et
->hash_mask
;
556 static void dm_remove_exception(struct dm_exception
*e
)
558 list_del(&e
->hash_list
);
562 * Return the exception data for a sector, or NULL if not
565 static struct dm_exception
*dm_lookup_exception(struct dm_exception_table
*et
,
568 struct list_head
*slot
;
569 struct dm_exception
*e
;
571 slot
= &et
->table
[exception_hash(et
, chunk
)];
572 list_for_each_entry (e
, slot
, hash_list
)
573 if (chunk
>= e
->old_chunk
&&
574 chunk
<= e
->old_chunk
+ dm_consecutive_chunk_count(e
))
580 static struct dm_exception
*alloc_completed_exception(void)
582 struct dm_exception
*e
;
584 e
= kmem_cache_alloc(exception_cache
, GFP_NOIO
);
586 e
= kmem_cache_alloc(exception_cache
, GFP_ATOMIC
);
591 static void free_completed_exception(struct dm_exception
*e
)
593 kmem_cache_free(exception_cache
, e
);
596 static struct dm_snap_pending_exception
*alloc_pending_exception(struct dm_snapshot
*s
)
598 struct dm_snap_pending_exception
*pe
= mempool_alloc(s
->pending_pool
,
601 atomic_inc(&s
->pending_exceptions_count
);
607 static void free_pending_exception(struct dm_snap_pending_exception
*pe
)
609 struct dm_snapshot
*s
= pe
->snap
;
611 mempool_free(pe
, s
->pending_pool
);
612 smp_mb__before_atomic_dec();
613 atomic_dec(&s
->pending_exceptions_count
);
616 static void dm_insert_exception(struct dm_exception_table
*eh
,
617 struct dm_exception
*new_e
)
620 struct dm_exception
*e
= NULL
;
622 l
= &eh
->table
[exception_hash(eh
, new_e
->old_chunk
)];
624 /* Add immediately if this table doesn't support consecutive chunks */
628 /* List is ordered by old_chunk */
629 list_for_each_entry_reverse(e
, l
, hash_list
) {
630 /* Insert after an existing chunk? */
631 if (new_e
->old_chunk
== (e
->old_chunk
+
632 dm_consecutive_chunk_count(e
) + 1) &&
633 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) +
634 dm_consecutive_chunk_count(e
) + 1)) {
635 dm_consecutive_chunk_count_inc(e
);
636 free_completed_exception(new_e
);
640 /* Insert before an existing chunk? */
641 if (new_e
->old_chunk
== (e
->old_chunk
- 1) &&
642 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) - 1)) {
643 dm_consecutive_chunk_count_inc(e
);
646 free_completed_exception(new_e
);
650 if (new_e
->old_chunk
> e
->old_chunk
)
655 list_add(&new_e
->hash_list
, e
? &e
->hash_list
: l
);
659 * Callback used by the exception stores to load exceptions when
662 static int dm_add_exception(void *context
, chunk_t old
, chunk_t
new)
664 struct dm_snapshot
*s
= context
;
665 struct dm_exception
*e
;
667 e
= alloc_completed_exception();
673 /* Consecutive_count is implicitly initialised to zero */
676 dm_insert_exception(&s
->complete
, e
);
681 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
684 * Return a minimum chunk size of all snapshots that have the specified origin.
685 * Return zero if the origin has no snapshots.
687 static sector_t
__minimum_chunk_size(struct origin
*o
)
689 struct dm_snapshot
*snap
;
690 unsigned chunk_size
= 0;
693 list_for_each_entry(snap
, &o
->snapshots
, list
)
694 chunk_size
= min_not_zero(chunk_size
,
695 snap
->store
->chunk_size
);
703 static int calc_max_buckets(void)
705 /* use a fixed size of 2MB */
706 unsigned long mem
= 2 * 1024 * 1024;
707 mem
/= sizeof(struct list_head
);
713 * Allocate room for a suitable hash table.
715 static int init_hash_tables(struct dm_snapshot
*s
)
717 sector_t hash_size
, cow_dev_size
, origin_dev_size
, max_buckets
;
720 * Calculate based on the size of the original volume or
723 cow_dev_size
= get_dev_size(s
->cow
->bdev
);
724 origin_dev_size
= get_dev_size(s
->origin
->bdev
);
725 max_buckets
= calc_max_buckets();
727 hash_size
= min(origin_dev_size
, cow_dev_size
) >> s
->store
->chunk_shift
;
728 hash_size
= min(hash_size
, max_buckets
);
732 hash_size
= rounddown_pow_of_two(hash_size
);
733 if (dm_exception_table_init(&s
->complete
, hash_size
,
734 DM_CHUNK_CONSECUTIVE_BITS
))
738 * Allocate hash table for in-flight exceptions
739 * Make this smaller than the real hash table
745 if (dm_exception_table_init(&s
->pending
, hash_size
, 0)) {
746 dm_exception_table_exit(&s
->complete
, exception_cache
);
753 static void merge_shutdown(struct dm_snapshot
*s
)
755 clear_bit_unlock(RUNNING_MERGE
, &s
->state_bits
);
756 smp_mb__after_clear_bit();
757 wake_up_bit(&s
->state_bits
, RUNNING_MERGE
);
760 static struct bio
*__release_queued_bios_after_merge(struct dm_snapshot
*s
)
762 s
->first_merging_chunk
= 0;
763 s
->num_merging_chunks
= 0;
765 return bio_list_get(&s
->bios_queued_during_merge
);
769 * Remove one chunk from the index of completed exceptions.
771 static int __remove_single_exception_chunk(struct dm_snapshot
*s
,
774 struct dm_exception
*e
;
776 e
= dm_lookup_exception(&s
->complete
, old_chunk
);
778 DMERR("Corruption detected: exception for block %llu is "
779 "on disk but not in memory",
780 (unsigned long long)old_chunk
);
785 * If this is the only chunk using this exception, remove exception.
787 if (!dm_consecutive_chunk_count(e
)) {
788 dm_remove_exception(e
);
789 free_completed_exception(e
);
794 * The chunk may be either at the beginning or the end of a
795 * group of consecutive chunks - never in the middle. We are
796 * removing chunks in the opposite order to that in which they
797 * were added, so this should always be true.
798 * Decrement the consecutive chunk counter and adjust the
799 * starting point if necessary.
801 if (old_chunk
== e
->old_chunk
) {
804 } else if (old_chunk
!= e
->old_chunk
+
805 dm_consecutive_chunk_count(e
)) {
806 DMERR("Attempt to merge block %llu from the "
807 "middle of a chunk range [%llu - %llu]",
808 (unsigned long long)old_chunk
,
809 (unsigned long long)e
->old_chunk
,
811 e
->old_chunk
+ dm_consecutive_chunk_count(e
));
815 dm_consecutive_chunk_count_dec(e
);
820 static void flush_bios(struct bio
*bio
);
822 static int remove_single_exception_chunk(struct dm_snapshot
*s
)
824 struct bio
*b
= NULL
;
826 chunk_t old_chunk
= s
->first_merging_chunk
+ s
->num_merging_chunks
- 1;
828 down_write(&s
->lock
);
831 * Process chunks (and associated exceptions) in reverse order
832 * so that dm_consecutive_chunk_count_dec() accounting works.
835 r
= __remove_single_exception_chunk(s
, old_chunk
);
838 } while (old_chunk
-- > s
->first_merging_chunk
);
840 b
= __release_queued_bios_after_merge(s
);
850 static void merge_callback(int read_err
, unsigned long write_err
,
853 static void snapshot_merge_next_chunks(struct dm_snapshot
*s
)
856 chunk_t old_chunk
, new_chunk
;
857 struct dm_io_region src
, dest
;
859 BUG_ON(!test_bit(RUNNING_MERGE
, &s
->state_bits
));
860 if (unlikely(test_bit(SHUTDOWN_MERGE
, &s
->state_bits
)))
864 * valid flag never changes during merge, so no lock required.
867 DMERR("Snapshot is invalid: can't merge");
871 r
= s
->store
->type
->prepare_merge(s
->store
, &old_chunk
, &new_chunk
);
874 DMERR("Read error in exception store: "
875 "shutting down merge");
879 /* TODO: use larger I/O size once we verify that kcopyd handles it */
881 dest
.bdev
= s
->origin
->bdev
;
882 dest
.sector
= chunk_to_sector(s
->store
, old_chunk
);
883 dest
.count
= min((sector_t
)s
->store
->chunk_size
,
884 get_dev_size(dest
.bdev
) - dest
.sector
);
886 src
.bdev
= s
->cow
->bdev
;
887 src
.sector
= chunk_to_sector(s
->store
, new_chunk
);
888 src
.count
= dest
.count
;
890 down_write(&s
->lock
);
891 s
->first_merging_chunk
= old_chunk
;
892 s
->num_merging_chunks
= 1;
895 /* !!! FIXME: wait until writes to this chunk drain */
897 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, merge_callback
, s
);
904 static void error_bios(struct bio
*bio
);
906 static void merge_callback(int read_err
, unsigned long write_err
, void *context
)
908 struct dm_snapshot
*s
= context
;
909 struct bio
*b
= NULL
;
911 if (read_err
|| write_err
) {
913 DMERR("Read error: shutting down merge.");
915 DMERR("Write error: shutting down merge.");
919 if (s
->store
->type
->commit_merge(s
->store
,
920 s
->num_merging_chunks
) < 0) {
921 DMERR("Write error in exception store: shutting down merge");
925 if (remove_single_exception_chunk(s
) < 0)
928 snapshot_merge_next_chunks(s
);
933 down_write(&s
->lock
);
934 b
= __release_queued_bios_after_merge(s
);
941 static void start_merge(struct dm_snapshot
*s
)
943 if (!test_and_set_bit(RUNNING_MERGE
, &s
->state_bits
))
944 snapshot_merge_next_chunks(s
);
947 static int wait_schedule(void *ptr
)
955 * Stop the merging process and wait until it finishes.
957 static void stop_merge(struct dm_snapshot
*s
)
959 set_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
960 wait_on_bit(&s
->state_bits
, RUNNING_MERGE
, wait_schedule
,
961 TASK_UNINTERRUPTIBLE
);
962 clear_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
966 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
968 static int snapshot_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
970 struct dm_snapshot
*s
;
973 char *origin_path
, *cow_path
;
974 unsigned args_used
, num_flush_requests
= 1;
975 fmode_t origin_mode
= FMODE_READ
;
978 ti
->error
= "requires exactly 4 arguments";
983 if (dm_target_is_snapshot_merge(ti
)) {
984 num_flush_requests
= 2;
985 origin_mode
= FMODE_WRITE
;
988 origin_path
= argv
[0];
992 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
994 ti
->error
= "Cannot allocate snapshot context private "
1004 r
= dm_get_device(ti
, cow_path
, 0, 0,
1005 FMODE_READ
| FMODE_WRITE
, &s
->cow
);
1007 ti
->error
= "Cannot get COW device";
1011 r
= dm_exception_store_create(ti
, argc
, argv
, s
, &args_used
, &s
->store
);
1013 ti
->error
= "Couldn't create exception store";
1021 r
= dm_get_device(ti
, origin_path
, 0, ti
->len
, origin_mode
, &s
->origin
);
1023 ti
->error
= "Cannot get origin device";
1031 atomic_set(&s
->pending_exceptions_count
, 0);
1032 init_rwsem(&s
->lock
);
1033 INIT_LIST_HEAD(&s
->list
);
1034 spin_lock_init(&s
->pe_lock
);
1036 s
->first_merging_chunk
= 0;
1037 s
->num_merging_chunks
= 0;
1038 bio_list_init(&s
->bios_queued_during_merge
);
1040 /* Allocate hash table for COW data */
1041 if (init_hash_tables(s
)) {
1042 ti
->error
= "Unable to allocate hash table space";
1044 goto bad_hash_tables
;
1047 r
= dm_kcopyd_client_create(SNAPSHOT_PAGES
, &s
->kcopyd_client
);
1049 ti
->error
= "Could not create kcopyd client";
1053 s
->pending_pool
= mempool_create_slab_pool(MIN_IOS
, pending_cache
);
1054 if (!s
->pending_pool
) {
1055 ti
->error
= "Could not allocate mempool for pending exceptions";
1056 goto bad_pending_pool
;
1059 s
->tracked_chunk_pool
= mempool_create_slab_pool(MIN_IOS
,
1060 tracked_chunk_cache
);
1061 if (!s
->tracked_chunk_pool
) {
1062 ti
->error
= "Could not allocate tracked_chunk mempool for "
1064 goto bad_tracked_chunk_pool
;
1067 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1068 INIT_HLIST_HEAD(&s
->tracked_chunk_hash
[i
]);
1070 spin_lock_init(&s
->tracked_chunk_lock
);
1072 bio_list_init(&s
->queued_bios
);
1073 INIT_WORK(&s
->queued_bios_work
, flush_queued_bios
);
1076 ti
->num_flush_requests
= num_flush_requests
;
1078 /* Add snapshot to the list of snapshots for this origin */
1079 /* Exceptions aren't triggered till snapshot_resume() is called */
1080 r
= register_snapshot(s
);
1082 ti
->error
= "Snapshot origin struct allocation failed";
1083 goto bad_load_and_register
;
1085 /* invalid handover, register_snapshot has set ti->error */
1086 goto bad_load_and_register
;
1090 * Metadata must only be loaded into one table at once, so skip this
1091 * if metadata will be handed over during resume.
1092 * Chunk size will be set during the handover - set it to zero to
1093 * ensure it's ignored.
1096 s
->store
->chunk_size
= 0;
1100 r
= s
->store
->type
->read_metadata(s
->store
, dm_add_exception
,
1103 ti
->error
= "Failed to read snapshot metadata";
1104 goto bad_read_metadata
;
1107 DMWARN("Snapshot is marked invalid.");
1110 if (!s
->store
->chunk_size
) {
1111 ti
->error
= "Chunk size not set";
1112 goto bad_read_metadata
;
1114 ti
->split_io
= s
->store
->chunk_size
;
1119 unregister_snapshot(s
);
1121 bad_load_and_register
:
1122 mempool_destroy(s
->tracked_chunk_pool
);
1124 bad_tracked_chunk_pool
:
1125 mempool_destroy(s
->pending_pool
);
1128 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1131 dm_exception_table_exit(&s
->pending
, pending_cache
);
1132 dm_exception_table_exit(&s
->complete
, exception_cache
);
1135 dm_put_device(ti
, s
->origin
);
1138 dm_exception_store_destroy(s
->store
);
1141 dm_put_device(ti
, s
->cow
);
1150 static void __free_exceptions(struct dm_snapshot
*s
)
1152 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1153 s
->kcopyd_client
= NULL
;
1155 dm_exception_table_exit(&s
->pending
, pending_cache
);
1156 dm_exception_table_exit(&s
->complete
, exception_cache
);
1159 static void __handover_exceptions(struct dm_snapshot
*snap_src
,
1160 struct dm_snapshot
*snap_dest
)
1163 struct dm_exception_table table_swap
;
1164 struct dm_exception_store
*store_swap
;
1168 * Swap all snapshot context information between the two instances.
1170 u
.table_swap
= snap_dest
->complete
;
1171 snap_dest
->complete
= snap_src
->complete
;
1172 snap_src
->complete
= u
.table_swap
;
1174 u
.store_swap
= snap_dest
->store
;
1175 snap_dest
->store
= snap_src
->store
;
1176 snap_src
->store
= u
.store_swap
;
1178 snap_dest
->store
->snap
= snap_dest
;
1179 snap_src
->store
->snap
= snap_src
;
1181 snap_dest
->ti
->split_io
= snap_dest
->store
->chunk_size
;
1182 snap_dest
->valid
= snap_src
->valid
;
1185 * Set source invalid to ensure it receives no further I/O.
1187 snap_src
->valid
= 0;
1190 static void snapshot_dtr(struct dm_target
*ti
)
1192 #ifdef CONFIG_DM_DEBUG
1195 struct dm_snapshot
*s
= ti
->private;
1196 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1198 flush_workqueue(ksnapd
);
1200 down_read(&_origins_lock
);
1201 /* Check whether exception handover must be cancelled */
1202 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1203 if (snap_src
&& snap_dest
&& (s
== snap_src
)) {
1204 down_write(&snap_dest
->lock
);
1205 snap_dest
->valid
= 0;
1206 up_write(&snap_dest
->lock
);
1207 DMERR("Cancelling snapshot handover.");
1209 up_read(&_origins_lock
);
1211 if (dm_target_is_snapshot_merge(ti
))
1214 /* Prevent further origin writes from using this snapshot. */
1215 /* After this returns there can be no new kcopyd jobs. */
1216 unregister_snapshot(s
);
1218 while (atomic_read(&s
->pending_exceptions_count
))
1221 * Ensure instructions in mempool_destroy aren't reordered
1222 * before atomic_read.
1226 #ifdef CONFIG_DM_DEBUG
1227 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1228 BUG_ON(!hlist_empty(&s
->tracked_chunk_hash
[i
]));
1231 mempool_destroy(s
->tracked_chunk_pool
);
1233 __free_exceptions(s
);
1235 mempool_destroy(s
->pending_pool
);
1237 dm_put_device(ti
, s
->origin
);
1239 dm_exception_store_destroy(s
->store
);
1241 dm_put_device(ti
, s
->cow
);
1247 * Flush a list of buffers.
1249 static void flush_bios(struct bio
*bio
)
1255 bio
->bi_next
= NULL
;
1256 generic_make_request(bio
);
1261 static void flush_queued_bios(struct work_struct
*work
)
1263 struct dm_snapshot
*s
=
1264 container_of(work
, struct dm_snapshot
, queued_bios_work
);
1265 struct bio
*queued_bios
;
1266 unsigned long flags
;
1268 spin_lock_irqsave(&s
->pe_lock
, flags
);
1269 queued_bios
= bio_list_get(&s
->queued_bios
);
1270 spin_unlock_irqrestore(&s
->pe_lock
, flags
);
1272 flush_bios(queued_bios
);
1275 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
);
1278 * Flush a list of buffers.
1280 static void retry_origin_bios(struct dm_snapshot
*s
, struct bio
*bio
)
1287 bio
->bi_next
= NULL
;
1288 r
= do_origin(s
->origin
, bio
);
1289 if (r
== DM_MAPIO_REMAPPED
)
1290 generic_make_request(bio
);
1296 * Error a list of buffers.
1298 static void error_bios(struct bio
*bio
)
1304 bio
->bi_next
= NULL
;
1310 static void __invalidate_snapshot(struct dm_snapshot
*s
, int err
)
1316 DMERR("Invalidating snapshot: Error reading/writing.");
1317 else if (err
== -ENOMEM
)
1318 DMERR("Invalidating snapshot: Unable to allocate exception.");
1320 if (s
->store
->type
->drop_snapshot
)
1321 s
->store
->type
->drop_snapshot(s
->store
);
1325 dm_table_event(s
->ti
->table
);
1328 static void pending_complete(struct dm_snap_pending_exception
*pe
, int success
)
1330 struct dm_exception
*e
;
1331 struct dm_snapshot
*s
= pe
->snap
;
1332 struct bio
*origin_bios
= NULL
;
1333 struct bio
*snapshot_bios
= NULL
;
1337 /* Read/write error - snapshot is unusable */
1338 down_write(&s
->lock
);
1339 __invalidate_snapshot(s
, -EIO
);
1344 e
= alloc_completed_exception();
1346 down_write(&s
->lock
);
1347 __invalidate_snapshot(s
, -ENOMEM
);
1353 down_write(&s
->lock
);
1355 free_completed_exception(e
);
1360 /* Check for conflicting reads */
1361 __check_for_conflicting_io(s
, pe
->e
.old_chunk
);
1364 * Add a proper exception, and remove the
1365 * in-flight exception from the list.
1367 dm_insert_exception(&s
->complete
, e
);
1370 dm_remove_exception(&pe
->e
);
1371 snapshot_bios
= bio_list_get(&pe
->snapshot_bios
);
1372 origin_bios
= bio_list_get(&pe
->origin_bios
);
1373 free_pending_exception(pe
);
1377 /* Submit any pending write bios */
1379 error_bios(snapshot_bios
);
1381 flush_bios(snapshot_bios
);
1383 retry_origin_bios(s
, origin_bios
);
1386 static void commit_callback(void *context
, int success
)
1388 struct dm_snap_pending_exception
*pe
= context
;
1390 pending_complete(pe
, success
);
1394 * Called when the copy I/O has finished. kcopyd actually runs
1395 * this code so don't block.
1397 static void copy_callback(int read_err
, unsigned long write_err
, void *context
)
1399 struct dm_snap_pending_exception
*pe
= context
;
1400 struct dm_snapshot
*s
= pe
->snap
;
1402 if (read_err
|| write_err
)
1403 pending_complete(pe
, 0);
1406 /* Update the metadata if we are persistent */
1407 s
->store
->type
->commit_exception(s
->store
, &pe
->e
,
1408 commit_callback
, pe
);
1412 * Dispatches the copy operation to kcopyd.
1414 static void start_copy(struct dm_snap_pending_exception
*pe
)
1416 struct dm_snapshot
*s
= pe
->snap
;
1417 struct dm_io_region src
, dest
;
1418 struct block_device
*bdev
= s
->origin
->bdev
;
1421 dev_size
= get_dev_size(bdev
);
1424 src
.sector
= chunk_to_sector(s
->store
, pe
->e
.old_chunk
);
1425 src
.count
= min((sector_t
)s
->store
->chunk_size
, dev_size
- src
.sector
);
1427 dest
.bdev
= s
->cow
->bdev
;
1428 dest
.sector
= chunk_to_sector(s
->store
, pe
->e
.new_chunk
);
1429 dest
.count
= src
.count
;
1431 /* Hand over to kcopyd */
1432 dm_kcopyd_copy(s
->kcopyd_client
,
1433 &src
, 1, &dest
, 0, copy_callback
, pe
);
1436 static struct dm_snap_pending_exception
*
1437 __lookup_pending_exception(struct dm_snapshot
*s
, chunk_t chunk
)
1439 struct dm_exception
*e
= dm_lookup_exception(&s
->pending
, chunk
);
1444 return container_of(e
, struct dm_snap_pending_exception
, e
);
1448 * Looks to see if this snapshot already has a pending exception
1449 * for this chunk, otherwise it allocates a new one and inserts
1450 * it into the pending table.
1452 * NOTE: a write lock must be held on snap->lock before calling
1455 static struct dm_snap_pending_exception
*
1456 __find_pending_exception(struct dm_snapshot
*s
,
1457 struct dm_snap_pending_exception
*pe
, chunk_t chunk
)
1459 struct dm_snap_pending_exception
*pe2
;
1461 pe2
= __lookup_pending_exception(s
, chunk
);
1463 free_pending_exception(pe
);
1467 pe
->e
.old_chunk
= chunk
;
1468 bio_list_init(&pe
->origin_bios
);
1469 bio_list_init(&pe
->snapshot_bios
);
1472 if (s
->store
->type
->prepare_exception(s
->store
, &pe
->e
)) {
1473 free_pending_exception(pe
);
1477 dm_insert_exception(&s
->pending
, &pe
->e
);
1482 static void remap_exception(struct dm_snapshot
*s
, struct dm_exception
*e
,
1483 struct bio
*bio
, chunk_t chunk
)
1485 bio
->bi_bdev
= s
->cow
->bdev
;
1486 bio
->bi_sector
= chunk_to_sector(s
->store
,
1487 dm_chunk_number(e
->new_chunk
) +
1488 (chunk
- e
->old_chunk
)) +
1490 s
->store
->chunk_mask
);
1493 static int snapshot_map(struct dm_target
*ti
, struct bio
*bio
,
1494 union map_info
*map_context
)
1496 struct dm_exception
*e
;
1497 struct dm_snapshot
*s
= ti
->private;
1498 int r
= DM_MAPIO_REMAPPED
;
1500 struct dm_snap_pending_exception
*pe
= NULL
;
1502 if (unlikely(bio_empty_barrier(bio
))) {
1503 bio
->bi_bdev
= s
->cow
->bdev
;
1504 return DM_MAPIO_REMAPPED
;
1507 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1509 /* Full snapshots are not usable */
1510 /* To get here the table must be live so s->active is always set. */
1514 /* FIXME: should only take write lock if we need
1515 * to copy an exception */
1516 down_write(&s
->lock
);
1523 /* If the block is already remapped - use that, else remap it */
1524 e
= dm_lookup_exception(&s
->complete
, chunk
);
1526 remap_exception(s
, e
, bio
, chunk
);
1531 * Write to snapshot - higher level takes care of RW/RO
1532 * flags so we should only get this if we are
1535 if (bio_rw(bio
) == WRITE
) {
1536 pe
= __lookup_pending_exception(s
, chunk
);
1539 pe
= alloc_pending_exception(s
);
1540 down_write(&s
->lock
);
1543 free_pending_exception(pe
);
1548 e
= dm_lookup_exception(&s
->complete
, chunk
);
1550 free_pending_exception(pe
);
1551 remap_exception(s
, e
, bio
, chunk
);
1555 pe
= __find_pending_exception(s
, pe
, chunk
);
1557 __invalidate_snapshot(s
, -ENOMEM
);
1563 remap_exception(s
, &pe
->e
, bio
, chunk
);
1564 bio_list_add(&pe
->snapshot_bios
, bio
);
1566 r
= DM_MAPIO_SUBMITTED
;
1569 /* this is protected by snap->lock */
1576 bio
->bi_bdev
= s
->origin
->bdev
;
1577 map_context
->ptr
= track_chunk(s
, chunk
);
1587 * A snapshot-merge target behaves like a combination of a snapshot
1588 * target and a snapshot-origin target. It only generates new
1589 * exceptions in other snapshots and not in the one that is being
1592 * For each chunk, if there is an existing exception, it is used to
1593 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1594 * which in turn might generate exceptions in other snapshots.
1595 * If merging is currently taking place on the chunk in question, the
1596 * I/O is deferred by adding it to s->bios_queued_during_merge.
1598 static int snapshot_merge_map(struct dm_target
*ti
, struct bio
*bio
,
1599 union map_info
*map_context
)
1601 struct dm_exception
*e
;
1602 struct dm_snapshot
*s
= ti
->private;
1603 int r
= DM_MAPIO_REMAPPED
;
1606 if (unlikely(bio_empty_barrier(bio
))) {
1607 if (!map_context
->flush_request
)
1608 bio
->bi_bdev
= s
->origin
->bdev
;
1610 bio
->bi_bdev
= s
->cow
->bdev
;
1611 map_context
->ptr
= NULL
;
1612 return DM_MAPIO_REMAPPED
;
1615 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1617 down_write(&s
->lock
);
1619 /* Full snapshots are not usable */
1625 /* If the block is already remapped - use that */
1626 e
= dm_lookup_exception(&s
->complete
, chunk
);
1628 /* Queue writes overlapping with chunks being merged */
1629 if (bio_rw(bio
) == WRITE
&&
1630 chunk
>= s
->first_merging_chunk
&&
1631 chunk
< (s
->first_merging_chunk
+
1632 s
->num_merging_chunks
)) {
1633 bio
->bi_bdev
= s
->origin
->bdev
;
1634 bio_list_add(&s
->bios_queued_during_merge
, bio
);
1635 r
= DM_MAPIO_SUBMITTED
;
1638 remap_exception(s
, e
, bio
, chunk
);
1642 bio
->bi_bdev
= s
->origin
->bdev
;
1644 if (bio_rw(bio
) == WRITE
) {
1646 return do_origin(s
->origin
, bio
);
1655 static int snapshot_end_io(struct dm_target
*ti
, struct bio
*bio
,
1656 int error
, union map_info
*map_context
)
1658 struct dm_snapshot
*s
= ti
->private;
1659 struct dm_snap_tracked_chunk
*c
= map_context
->ptr
;
1662 stop_tracking_chunk(s
, c
);
1667 static void snapshot_merge_presuspend(struct dm_target
*ti
)
1669 struct dm_snapshot
*s
= ti
->private;
1674 static void snapshot_postsuspend(struct dm_target
*ti
)
1676 struct dm_snapshot
*s
= ti
->private;
1678 down_write(&s
->lock
);
1683 static int snapshot_preresume(struct dm_target
*ti
)
1686 struct dm_snapshot
*s
= ti
->private;
1687 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1689 down_read(&_origins_lock
);
1690 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1691 if (snap_src
&& snap_dest
) {
1692 down_read(&snap_src
->lock
);
1693 if (s
== snap_src
) {
1694 DMERR("Unable to resume snapshot source until "
1695 "handover completes.");
1697 } else if (!snap_src
->suspended
) {
1698 DMERR("Unable to perform snapshot handover until "
1699 "source is suspended.");
1702 up_read(&snap_src
->lock
);
1704 up_read(&_origins_lock
);
1709 static void snapshot_resume(struct dm_target
*ti
)
1711 struct dm_snapshot
*s
= ti
->private;
1712 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1714 down_read(&_origins_lock
);
1715 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1716 if (snap_src
&& snap_dest
) {
1717 down_write(&snap_src
->lock
);
1718 down_write_nested(&snap_dest
->lock
, SINGLE_DEPTH_NESTING
);
1719 __handover_exceptions(snap_src
, snap_dest
);
1720 up_write(&snap_dest
->lock
);
1721 up_write(&snap_src
->lock
);
1723 up_read(&_origins_lock
);
1725 /* Now we have correct chunk size, reregister */
1726 reregister_snapshot(s
);
1728 down_write(&s
->lock
);
1734 static sector_t
get_origin_minimum_chunksize(struct block_device
*bdev
)
1736 sector_t min_chunksize
;
1738 down_read(&_origins_lock
);
1739 min_chunksize
= __minimum_chunk_size(__lookup_origin(bdev
));
1740 up_read(&_origins_lock
);
1742 return min_chunksize
;
1745 static void snapshot_merge_resume(struct dm_target
*ti
)
1747 struct dm_snapshot
*s
= ti
->private;
1750 * Handover exceptions from existing snapshot.
1752 snapshot_resume(ti
);
1755 * snapshot-merge acts as an origin, so set ti->split_io
1757 ti
->split_io
= get_origin_minimum_chunksize(s
->origin
->bdev
);
1762 static int snapshot_status(struct dm_target
*ti
, status_type_t type
,
1763 char *result
, unsigned int maxlen
)
1766 struct dm_snapshot
*snap
= ti
->private;
1769 case STATUSTYPE_INFO
:
1771 down_write(&snap
->lock
);
1776 if (snap
->store
->type
->usage
) {
1777 sector_t total_sectors
, sectors_allocated
,
1779 snap
->store
->type
->usage(snap
->store
,
1783 DMEMIT("%llu/%llu %llu",
1784 (unsigned long long)sectors_allocated
,
1785 (unsigned long long)total_sectors
,
1786 (unsigned long long)metadata_sectors
);
1792 up_write(&snap
->lock
);
1796 case STATUSTYPE_TABLE
:
1798 * kdevname returns a static pointer so we need
1799 * to make private copies if the output is to
1802 DMEMIT("%s %s", snap
->origin
->name
, snap
->cow
->name
);
1803 snap
->store
->type
->status(snap
->store
, type
, result
+ sz
,
1811 static int snapshot_iterate_devices(struct dm_target
*ti
,
1812 iterate_devices_callout_fn fn
, void *data
)
1814 struct dm_snapshot
*snap
= ti
->private;
1816 return fn(ti
, snap
->origin
, 0, ti
->len
, data
);
1820 /*-----------------------------------------------------------------
1822 *---------------------------------------------------------------*/
1825 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1826 * supplied bio was ignored. The caller may submit it immediately.
1827 * (No remapping actually occurs as the origin is always a direct linear
1830 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1831 * and any supplied bio is added to a list to be submitted once all
1832 * the necessary exceptions exist.
1834 static int __origin_write(struct list_head
*snapshots
, sector_t sector
,
1837 int r
= DM_MAPIO_REMAPPED
;
1838 struct dm_snapshot
*snap
;
1839 struct dm_exception
*e
;
1840 struct dm_snap_pending_exception
*pe
;
1841 struct dm_snap_pending_exception
*pe_to_start_now
= NULL
;
1842 struct dm_snap_pending_exception
*pe_to_start_last
= NULL
;
1845 /* Do all the snapshots on this origin */
1846 list_for_each_entry (snap
, snapshots
, list
) {
1848 * Don't make new exceptions in a merging snapshot
1849 * because it has effectively been deleted
1851 if (dm_target_is_snapshot_merge(snap
->ti
))
1854 down_write(&snap
->lock
);
1856 /* Only deal with valid and active snapshots */
1857 if (!snap
->valid
|| !snap
->active
)
1860 /* Nothing to do if writing beyond end of snapshot */
1861 if (sector
>= dm_table_get_size(snap
->ti
->table
))
1865 * Remember, different snapshots can have
1866 * different chunk sizes.
1868 chunk
= sector_to_chunk(snap
->store
, sector
);
1871 * Check exception table to see if block
1872 * is already remapped in this snapshot
1873 * and trigger an exception if not.
1875 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1879 pe
= __lookup_pending_exception(snap
, chunk
);
1881 up_write(&snap
->lock
);
1882 pe
= alloc_pending_exception(snap
);
1883 down_write(&snap
->lock
);
1886 free_pending_exception(pe
);
1890 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1892 free_pending_exception(pe
);
1896 pe
= __find_pending_exception(snap
, pe
, chunk
);
1898 __invalidate_snapshot(snap
, -ENOMEM
);
1903 r
= DM_MAPIO_SUBMITTED
;
1906 * If an origin bio was supplied, queue it to wait for the
1907 * completion of this exception, and start this one last,
1908 * at the end of the function.
1911 bio_list_add(&pe
->origin_bios
, bio
);
1916 pe_to_start_last
= pe
;
1922 pe_to_start_now
= pe
;
1926 up_write(&snap
->lock
);
1928 if (pe_to_start_now
) {
1929 start_copy(pe_to_start_now
);
1930 pe_to_start_now
= NULL
;
1935 * Submit the exception against which the bio is queued last,
1936 * to give the other exceptions a head start.
1938 if (pe_to_start_last
)
1939 start_copy(pe_to_start_last
);
1945 * Called on a write from the origin driver.
1947 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
)
1950 int r
= DM_MAPIO_REMAPPED
;
1952 down_read(&_origins_lock
);
1953 o
= __lookup_origin(origin
->bdev
);
1955 r
= __origin_write(&o
->snapshots
, bio
->bi_sector
, bio
);
1956 up_read(&_origins_lock
);
1962 * Origin: maps a linear range of a device, with hooks for snapshotting.
1966 * Construct an origin mapping: <dev_path>
1967 * The context for an origin is merely a 'struct dm_dev *'
1968 * pointing to the real device.
1970 static int origin_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
1976 ti
->error
= "origin: incorrect number of arguments";
1980 r
= dm_get_device(ti
, argv
[0], 0, ti
->len
,
1981 dm_table_get_mode(ti
->table
), &dev
);
1983 ti
->error
= "Cannot get target device";
1988 ti
->num_flush_requests
= 1;
1993 static void origin_dtr(struct dm_target
*ti
)
1995 struct dm_dev
*dev
= ti
->private;
1996 dm_put_device(ti
, dev
);
1999 static int origin_map(struct dm_target
*ti
, struct bio
*bio
,
2000 union map_info
*map_context
)
2002 struct dm_dev
*dev
= ti
->private;
2003 bio
->bi_bdev
= dev
->bdev
;
2005 if (unlikely(bio_empty_barrier(bio
)))
2006 return DM_MAPIO_REMAPPED
;
2008 /* Only tell snapshots if this is a write */
2009 return (bio_rw(bio
) == WRITE
) ? do_origin(dev
, bio
) : DM_MAPIO_REMAPPED
;
2013 * Set the target "split_io" field to the minimum of all the snapshots'
2016 static void origin_resume(struct dm_target
*ti
)
2018 struct dm_dev
*dev
= ti
->private;
2020 ti
->split_io
= get_origin_minimum_chunksize(dev
->bdev
);
2023 static int origin_status(struct dm_target
*ti
, status_type_t type
, char *result
,
2024 unsigned int maxlen
)
2026 struct dm_dev
*dev
= ti
->private;
2029 case STATUSTYPE_INFO
:
2033 case STATUSTYPE_TABLE
:
2034 snprintf(result
, maxlen
, "%s", dev
->name
);
2041 static int origin_iterate_devices(struct dm_target
*ti
,
2042 iterate_devices_callout_fn fn
, void *data
)
2044 struct dm_dev
*dev
= ti
->private;
2046 return fn(ti
, dev
, 0, ti
->len
, data
);
2049 static struct target_type origin_target
= {
2050 .name
= "snapshot-origin",
2051 .version
= {1, 7, 0},
2052 .module
= THIS_MODULE
,
2056 .resume
= origin_resume
,
2057 .status
= origin_status
,
2058 .iterate_devices
= origin_iterate_devices
,
2061 static struct target_type snapshot_target
= {
2063 .version
= {1, 9, 0},
2064 .module
= THIS_MODULE
,
2065 .ctr
= snapshot_ctr
,
2066 .dtr
= snapshot_dtr
,
2067 .map
= snapshot_map
,
2068 .end_io
= snapshot_end_io
,
2069 .postsuspend
= snapshot_postsuspend
,
2070 .preresume
= snapshot_preresume
,
2071 .resume
= snapshot_resume
,
2072 .status
= snapshot_status
,
2073 .iterate_devices
= snapshot_iterate_devices
,
2076 static struct target_type merge_target
= {
2077 .name
= dm_snapshot_merge_target_name
,
2078 .version
= {1, 0, 0},
2079 .module
= THIS_MODULE
,
2080 .ctr
= snapshot_ctr
,
2081 .dtr
= snapshot_dtr
,
2082 .map
= snapshot_merge_map
,
2083 .end_io
= snapshot_end_io
,
2084 .presuspend
= snapshot_merge_presuspend
,
2085 .postsuspend
= snapshot_postsuspend
,
2086 .preresume
= snapshot_preresume
,
2087 .resume
= snapshot_merge_resume
,
2088 .status
= snapshot_status
,
2089 .iterate_devices
= snapshot_iterate_devices
,
2092 static int __init
dm_snapshot_init(void)
2096 r
= dm_exception_store_init();
2098 DMERR("Failed to initialize exception stores");
2102 r
= dm_register_target(&snapshot_target
);
2104 DMERR("snapshot target register failed %d", r
);
2105 goto bad_register_snapshot_target
;
2108 r
= dm_register_target(&origin_target
);
2110 DMERR("Origin target register failed %d", r
);
2111 goto bad_register_origin_target
;
2114 r
= dm_register_target(&merge_target
);
2116 DMERR("Merge target register failed %d", r
);
2117 goto bad_register_merge_target
;
2120 r
= init_origin_hash();
2122 DMERR("init_origin_hash failed.");
2123 goto bad_origin_hash
;
2126 exception_cache
= KMEM_CACHE(dm_exception
, 0);
2127 if (!exception_cache
) {
2128 DMERR("Couldn't create exception cache.");
2130 goto bad_exception_cache
;
2133 pending_cache
= KMEM_CACHE(dm_snap_pending_exception
, 0);
2134 if (!pending_cache
) {
2135 DMERR("Couldn't create pending cache.");
2137 goto bad_pending_cache
;
2140 tracked_chunk_cache
= KMEM_CACHE(dm_snap_tracked_chunk
, 0);
2141 if (!tracked_chunk_cache
) {
2142 DMERR("Couldn't create cache to track chunks in use.");
2144 goto bad_tracked_chunk_cache
;
2147 ksnapd
= create_singlethread_workqueue("ksnapd");
2149 DMERR("Failed to create ksnapd workqueue.");
2151 goto bad_pending_pool
;
2157 kmem_cache_destroy(tracked_chunk_cache
);
2158 bad_tracked_chunk_cache
:
2159 kmem_cache_destroy(pending_cache
);
2161 kmem_cache_destroy(exception_cache
);
2162 bad_exception_cache
:
2165 dm_unregister_target(&merge_target
);
2166 bad_register_merge_target
:
2167 dm_unregister_target(&origin_target
);
2168 bad_register_origin_target
:
2169 dm_unregister_target(&snapshot_target
);
2170 bad_register_snapshot_target
:
2171 dm_exception_store_exit();
2176 static void __exit
dm_snapshot_exit(void)
2178 destroy_workqueue(ksnapd
);
2180 dm_unregister_target(&snapshot_target
);
2181 dm_unregister_target(&origin_target
);
2182 dm_unregister_target(&merge_target
);
2185 kmem_cache_destroy(pending_cache
);
2186 kmem_cache_destroy(exception_cache
);
2187 kmem_cache_destroy(tracked_chunk_cache
);
2189 dm_exception_store_exit();
2193 module_init(dm_snapshot_init
);
2194 module_exit(dm_snapshot_exit
);
2196 MODULE_DESCRIPTION(DM_NAME
" snapshot target");
2197 MODULE_AUTHOR("Joe Thornber");
2198 MODULE_LICENSE("GPL");