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>
23 #include "dm-exception-store.h"
25 #define DM_MSG_PREFIX "snapshots"
27 static const char dm_snapshot_merge_target_name
[] = "snapshot-merge";
29 #define dm_target_is_snapshot_merge(ti) \
30 ((ti)->type->name == dm_snapshot_merge_target_name)
33 * The size of the mempool used to track chunks in use.
37 #define DM_TRACKED_CHUNK_HASH_SIZE 16
38 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
39 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
41 struct dm_exception_table
{
44 struct list_head
*table
;
48 struct rw_semaphore lock
;
50 struct dm_dev
*origin
;
55 /* List of snapshots per Origin */
56 struct list_head list
;
59 * You can't use a snapshot if this is 0 (e.g. if full).
60 * A snapshot-merge target never clears this.
64 /* Origin writes don't trigger exceptions until this is set */
67 atomic_t pending_exceptions_count
;
69 mempool_t
*pending_pool
;
71 struct dm_exception_table pending
;
72 struct dm_exception_table complete
;
75 * pe_lock protects all pending_exception operations and access
76 * as well as the snapshot_bios list.
80 /* Chunks with outstanding reads */
81 spinlock_t tracked_chunk_lock
;
82 struct hlist_head tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH_SIZE
];
84 /* The on disk metadata handler */
85 struct dm_exception_store
*store
;
87 struct dm_kcopyd_client
*kcopyd_client
;
89 /* Wait for events based on state_bits */
90 unsigned long state_bits
;
92 /* Range of chunks currently being merged. */
93 chunk_t first_merging_chunk
;
94 int num_merging_chunks
;
97 * The merge operation failed if this flag is set.
98 * Failure modes are handled as follows:
99 * - I/O error reading the header
100 * => don't load the target; abort.
101 * - Header does not have "valid" flag set
102 * => use the origin; forget about the snapshot.
103 * - I/O error when reading exceptions
104 * => don't load the target; abort.
105 * (We can't use the intermediate origin state.)
106 * - I/O error while merging
107 * => stop merging; set merge_failed; process I/O normally.
112 * Incoming bios that overlap with chunks being merged must wait
113 * for them to be committed.
115 struct bio_list bios_queued_during_merge
;
120 * RUNNING_MERGE - Merge operation is in progress.
121 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
122 * cleared afterwards.
124 #define RUNNING_MERGE 0
125 #define SHUTDOWN_MERGE 1
127 struct dm_dev
*dm_snap_origin(struct dm_snapshot
*s
)
131 EXPORT_SYMBOL(dm_snap_origin
);
133 struct dm_dev
*dm_snap_cow(struct dm_snapshot
*s
)
137 EXPORT_SYMBOL(dm_snap_cow
);
139 static sector_t
chunk_to_sector(struct dm_exception_store
*store
,
142 return chunk
<< store
->chunk_shift
;
145 static int bdev_equal(struct block_device
*lhs
, struct block_device
*rhs
)
148 * There is only ever one instance of a particular block
149 * device so we can compare pointers safely.
154 struct dm_snap_pending_exception
{
155 struct dm_exception e
;
158 * Origin buffers waiting for this to complete are held
161 struct bio_list origin_bios
;
162 struct bio_list snapshot_bios
;
164 /* Pointer back to snapshot context */
165 struct dm_snapshot
*snap
;
168 * 1 indicates the exception has already been sent to
174 * For writing a complete chunk, bypassing the copy.
176 struct bio
*full_bio
;
177 bio_end_io_t
*full_bio_end_io
;
178 void *full_bio_private
;
182 * Hash table mapping origin volumes to lists of snapshots and
183 * a lock to protect it
185 static struct kmem_cache
*exception_cache
;
186 static struct kmem_cache
*pending_cache
;
188 struct dm_snap_tracked_chunk
{
189 struct hlist_node node
;
193 static void init_tracked_chunk(struct bio
*bio
)
195 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
196 INIT_HLIST_NODE(&c
->node
);
199 static bool is_bio_tracked(struct bio
*bio
)
201 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
202 return !hlist_unhashed(&c
->node
);
205 static void track_chunk(struct dm_snapshot
*s
, struct bio
*bio
, chunk_t chunk
)
207 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
211 spin_lock_irq(&s
->tracked_chunk_lock
);
212 hlist_add_head(&c
->node
,
213 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)]);
214 spin_unlock_irq(&s
->tracked_chunk_lock
);
217 static void stop_tracking_chunk(struct dm_snapshot
*s
, struct bio
*bio
)
219 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
222 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
224 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
227 static int __chunk_is_tracked(struct dm_snapshot
*s
, chunk_t chunk
)
229 struct dm_snap_tracked_chunk
*c
;
230 struct hlist_node
*hn
;
233 spin_lock_irq(&s
->tracked_chunk_lock
);
235 hlist_for_each_entry(c
, hn
,
236 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)], node
) {
237 if (c
->chunk
== chunk
) {
243 spin_unlock_irq(&s
->tracked_chunk_lock
);
249 * This conflicting I/O is extremely improbable in the caller,
250 * so msleep(1) is sufficient and there is no need for a wait queue.
252 static void __check_for_conflicting_io(struct dm_snapshot
*s
, chunk_t chunk
)
254 while (__chunk_is_tracked(s
, chunk
))
259 * One of these per registered origin, held in the snapshot_origins hash
262 /* The origin device */
263 struct block_device
*bdev
;
265 struct list_head hash_list
;
267 /* List of snapshots for this origin */
268 struct list_head snapshots
;
272 * Size of the hash table for origin volumes. If we make this
273 * the size of the minors list then it should be nearly perfect
275 #define ORIGIN_HASH_SIZE 256
276 #define ORIGIN_MASK 0xFF
277 static struct list_head
*_origins
;
278 static struct rw_semaphore _origins_lock
;
280 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done
);
281 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock
);
282 static uint64_t _pending_exceptions_done_count
;
284 static int init_origin_hash(void)
288 _origins
= kmalloc(ORIGIN_HASH_SIZE
* sizeof(struct list_head
),
291 DMERR("unable to allocate memory");
295 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
296 INIT_LIST_HEAD(_origins
+ i
);
297 init_rwsem(&_origins_lock
);
302 static void exit_origin_hash(void)
307 static unsigned origin_hash(struct block_device
*bdev
)
309 return bdev
->bd_dev
& ORIGIN_MASK
;
312 static struct origin
*__lookup_origin(struct block_device
*origin
)
314 struct list_head
*ol
;
317 ol
= &_origins
[origin_hash(origin
)];
318 list_for_each_entry (o
, ol
, hash_list
)
319 if (bdev_equal(o
->bdev
, origin
))
325 static void __insert_origin(struct origin
*o
)
327 struct list_head
*sl
= &_origins
[origin_hash(o
->bdev
)];
328 list_add_tail(&o
->hash_list
, sl
);
332 * _origins_lock must be held when calling this function.
333 * Returns number of snapshots registered using the supplied cow device, plus:
334 * snap_src - a snapshot suitable for use as a source of exception handover
335 * snap_dest - a snapshot capable of receiving exception handover.
336 * snap_merge - an existing snapshot-merge target linked to the same origin.
337 * There can be at most one snapshot-merge target. The parameter is optional.
339 * Possible return values and states of snap_src and snap_dest.
340 * 0: NULL, NULL - first new snapshot
341 * 1: snap_src, NULL - normal snapshot
342 * 2: snap_src, snap_dest - waiting for handover
343 * 2: snap_src, NULL - handed over, waiting for old to be deleted
344 * 1: NULL, snap_dest - source got destroyed without handover
346 static int __find_snapshots_sharing_cow(struct dm_snapshot
*snap
,
347 struct dm_snapshot
**snap_src
,
348 struct dm_snapshot
**snap_dest
,
349 struct dm_snapshot
**snap_merge
)
351 struct dm_snapshot
*s
;
356 o
= __lookup_origin(snap
->origin
->bdev
);
360 list_for_each_entry(s
, &o
->snapshots
, list
) {
361 if (dm_target_is_snapshot_merge(s
->ti
) && snap_merge
)
363 if (!bdev_equal(s
->cow
->bdev
, snap
->cow
->bdev
))
373 } else if (snap_dest
)
384 * On success, returns 1 if this snapshot is a handover destination,
385 * otherwise returns 0.
387 static int __validate_exception_handover(struct dm_snapshot
*snap
)
389 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
390 struct dm_snapshot
*snap_merge
= NULL
;
392 /* Does snapshot need exceptions handed over to it? */
393 if ((__find_snapshots_sharing_cow(snap
, &snap_src
, &snap_dest
,
394 &snap_merge
) == 2) ||
396 snap
->ti
->error
= "Snapshot cow pairing for exception "
397 "table handover failed";
402 * If no snap_src was found, snap cannot become a handover
409 * Non-snapshot-merge handover?
411 if (!dm_target_is_snapshot_merge(snap
->ti
))
415 * Do not allow more than one merging snapshot.
418 snap
->ti
->error
= "A snapshot is already merging.";
422 if (!snap_src
->store
->type
->prepare_merge
||
423 !snap_src
->store
->type
->commit_merge
) {
424 snap
->ti
->error
= "Snapshot exception store does not "
425 "support snapshot-merge.";
432 static void __insert_snapshot(struct origin
*o
, struct dm_snapshot
*s
)
434 struct dm_snapshot
*l
;
436 /* Sort the list according to chunk size, largest-first smallest-last */
437 list_for_each_entry(l
, &o
->snapshots
, list
)
438 if (l
->store
->chunk_size
< s
->store
->chunk_size
)
440 list_add_tail(&s
->list
, &l
->list
);
444 * Make a note of the snapshot and its origin so we can look it
445 * up when the origin has a write on it.
447 * Also validate snapshot exception store handovers.
448 * On success, returns 1 if this registration is a handover destination,
449 * otherwise returns 0.
451 static int register_snapshot(struct dm_snapshot
*snap
)
453 struct origin
*o
, *new_o
= NULL
;
454 struct block_device
*bdev
= snap
->origin
->bdev
;
457 new_o
= kmalloc(sizeof(*new_o
), GFP_KERNEL
);
461 down_write(&_origins_lock
);
463 r
= __validate_exception_handover(snap
);
469 o
= __lookup_origin(bdev
);
476 /* Initialise the struct */
477 INIT_LIST_HEAD(&o
->snapshots
);
483 __insert_snapshot(o
, snap
);
486 up_write(&_origins_lock
);
492 * Move snapshot to correct place in list according to chunk size.
494 static void reregister_snapshot(struct dm_snapshot
*s
)
496 struct block_device
*bdev
= s
->origin
->bdev
;
498 down_write(&_origins_lock
);
501 __insert_snapshot(__lookup_origin(bdev
), s
);
503 up_write(&_origins_lock
);
506 static void unregister_snapshot(struct dm_snapshot
*s
)
510 down_write(&_origins_lock
);
511 o
= __lookup_origin(s
->origin
->bdev
);
514 if (o
&& list_empty(&o
->snapshots
)) {
515 list_del(&o
->hash_list
);
519 up_write(&_origins_lock
);
523 * Implementation of the exception hash tables.
524 * The lowest hash_shift bits of the chunk number are ignored, allowing
525 * some consecutive chunks to be grouped together.
527 static int dm_exception_table_init(struct dm_exception_table
*et
,
528 uint32_t size
, unsigned hash_shift
)
532 et
->hash_shift
= hash_shift
;
533 et
->hash_mask
= size
- 1;
534 et
->table
= dm_vcalloc(size
, sizeof(struct list_head
));
538 for (i
= 0; i
< size
; i
++)
539 INIT_LIST_HEAD(et
->table
+ i
);
544 static void dm_exception_table_exit(struct dm_exception_table
*et
,
545 struct kmem_cache
*mem
)
547 struct list_head
*slot
;
548 struct dm_exception
*ex
, *next
;
551 size
= et
->hash_mask
+ 1;
552 for (i
= 0; i
< size
; i
++) {
553 slot
= et
->table
+ i
;
555 list_for_each_entry_safe (ex
, next
, slot
, hash_list
)
556 kmem_cache_free(mem
, ex
);
562 static uint32_t exception_hash(struct dm_exception_table
*et
, chunk_t chunk
)
564 return (chunk
>> et
->hash_shift
) & et
->hash_mask
;
567 static void dm_remove_exception(struct dm_exception
*e
)
569 list_del(&e
->hash_list
);
573 * Return the exception data for a sector, or NULL if not
576 static struct dm_exception
*dm_lookup_exception(struct dm_exception_table
*et
,
579 struct list_head
*slot
;
580 struct dm_exception
*e
;
582 slot
= &et
->table
[exception_hash(et
, chunk
)];
583 list_for_each_entry (e
, slot
, hash_list
)
584 if (chunk
>= e
->old_chunk
&&
585 chunk
<= e
->old_chunk
+ dm_consecutive_chunk_count(e
))
591 static struct dm_exception
*alloc_completed_exception(void)
593 struct dm_exception
*e
;
595 e
= kmem_cache_alloc(exception_cache
, GFP_NOIO
);
597 e
= kmem_cache_alloc(exception_cache
, GFP_ATOMIC
);
602 static void free_completed_exception(struct dm_exception
*e
)
604 kmem_cache_free(exception_cache
, e
);
607 static struct dm_snap_pending_exception
*alloc_pending_exception(struct dm_snapshot
*s
)
609 struct dm_snap_pending_exception
*pe
= mempool_alloc(s
->pending_pool
,
612 atomic_inc(&s
->pending_exceptions_count
);
618 static void free_pending_exception(struct dm_snap_pending_exception
*pe
)
620 struct dm_snapshot
*s
= pe
->snap
;
622 mempool_free(pe
, s
->pending_pool
);
623 smp_mb__before_atomic_dec();
624 atomic_dec(&s
->pending_exceptions_count
);
627 static void dm_insert_exception(struct dm_exception_table
*eh
,
628 struct dm_exception
*new_e
)
631 struct dm_exception
*e
= NULL
;
633 l
= &eh
->table
[exception_hash(eh
, new_e
->old_chunk
)];
635 /* Add immediately if this table doesn't support consecutive chunks */
639 /* List is ordered by old_chunk */
640 list_for_each_entry_reverse(e
, l
, hash_list
) {
641 /* Insert after an existing chunk? */
642 if (new_e
->old_chunk
== (e
->old_chunk
+
643 dm_consecutive_chunk_count(e
) + 1) &&
644 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) +
645 dm_consecutive_chunk_count(e
) + 1)) {
646 dm_consecutive_chunk_count_inc(e
);
647 free_completed_exception(new_e
);
651 /* Insert before an existing chunk? */
652 if (new_e
->old_chunk
== (e
->old_chunk
- 1) &&
653 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) - 1)) {
654 dm_consecutive_chunk_count_inc(e
);
657 free_completed_exception(new_e
);
661 if (new_e
->old_chunk
> e
->old_chunk
)
666 list_add(&new_e
->hash_list
, e
? &e
->hash_list
: l
);
670 * Callback used by the exception stores to load exceptions when
673 static int dm_add_exception(void *context
, chunk_t old
, chunk_t
new)
675 struct dm_snapshot
*s
= context
;
676 struct dm_exception
*e
;
678 e
= alloc_completed_exception();
684 /* Consecutive_count is implicitly initialised to zero */
687 dm_insert_exception(&s
->complete
, e
);
693 * Return a minimum chunk size of all snapshots that have the specified origin.
694 * Return zero if the origin has no snapshots.
696 static uint32_t __minimum_chunk_size(struct origin
*o
)
698 struct dm_snapshot
*snap
;
699 unsigned chunk_size
= 0;
702 list_for_each_entry(snap
, &o
->snapshots
, list
)
703 chunk_size
= min_not_zero(chunk_size
,
704 snap
->store
->chunk_size
);
706 return (uint32_t) chunk_size
;
712 static int calc_max_buckets(void)
714 /* use a fixed size of 2MB */
715 unsigned long mem
= 2 * 1024 * 1024;
716 mem
/= sizeof(struct list_head
);
722 * Allocate room for a suitable hash table.
724 static int init_hash_tables(struct dm_snapshot
*s
)
726 sector_t hash_size
, cow_dev_size
, origin_dev_size
, max_buckets
;
729 * Calculate based on the size of the original volume or
732 cow_dev_size
= get_dev_size(s
->cow
->bdev
);
733 origin_dev_size
= get_dev_size(s
->origin
->bdev
);
734 max_buckets
= calc_max_buckets();
736 hash_size
= min(origin_dev_size
, cow_dev_size
) >> s
->store
->chunk_shift
;
737 hash_size
= min(hash_size
, max_buckets
);
741 hash_size
= rounddown_pow_of_two(hash_size
);
742 if (dm_exception_table_init(&s
->complete
, hash_size
,
743 DM_CHUNK_CONSECUTIVE_BITS
))
747 * Allocate hash table for in-flight exceptions
748 * Make this smaller than the real hash table
754 if (dm_exception_table_init(&s
->pending
, hash_size
, 0)) {
755 dm_exception_table_exit(&s
->complete
, exception_cache
);
762 static void merge_shutdown(struct dm_snapshot
*s
)
764 clear_bit_unlock(RUNNING_MERGE
, &s
->state_bits
);
765 smp_mb__after_clear_bit();
766 wake_up_bit(&s
->state_bits
, RUNNING_MERGE
);
769 static struct bio
*__release_queued_bios_after_merge(struct dm_snapshot
*s
)
771 s
->first_merging_chunk
= 0;
772 s
->num_merging_chunks
= 0;
774 return bio_list_get(&s
->bios_queued_during_merge
);
778 * Remove one chunk from the index of completed exceptions.
780 static int __remove_single_exception_chunk(struct dm_snapshot
*s
,
783 struct dm_exception
*e
;
785 e
= dm_lookup_exception(&s
->complete
, old_chunk
);
787 DMERR("Corruption detected: exception for block %llu is "
788 "on disk but not in memory",
789 (unsigned long long)old_chunk
);
794 * If this is the only chunk using this exception, remove exception.
796 if (!dm_consecutive_chunk_count(e
)) {
797 dm_remove_exception(e
);
798 free_completed_exception(e
);
803 * The chunk may be either at the beginning or the end of a
804 * group of consecutive chunks - never in the middle. We are
805 * removing chunks in the opposite order to that in which they
806 * were added, so this should always be true.
807 * Decrement the consecutive chunk counter and adjust the
808 * starting point if necessary.
810 if (old_chunk
== e
->old_chunk
) {
813 } else if (old_chunk
!= e
->old_chunk
+
814 dm_consecutive_chunk_count(e
)) {
815 DMERR("Attempt to merge block %llu from the "
816 "middle of a chunk range [%llu - %llu]",
817 (unsigned long long)old_chunk
,
818 (unsigned long long)e
->old_chunk
,
820 e
->old_chunk
+ dm_consecutive_chunk_count(e
));
824 dm_consecutive_chunk_count_dec(e
);
829 static void flush_bios(struct bio
*bio
);
831 static int remove_single_exception_chunk(struct dm_snapshot
*s
)
833 struct bio
*b
= NULL
;
835 chunk_t old_chunk
= s
->first_merging_chunk
+ s
->num_merging_chunks
- 1;
837 down_write(&s
->lock
);
840 * Process chunks (and associated exceptions) in reverse order
841 * so that dm_consecutive_chunk_count_dec() accounting works.
844 r
= __remove_single_exception_chunk(s
, old_chunk
);
847 } while (old_chunk
-- > s
->first_merging_chunk
);
849 b
= __release_queued_bios_after_merge(s
);
859 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
860 sector_t sector
, unsigned chunk_size
);
862 static void merge_callback(int read_err
, unsigned long write_err
,
865 static uint64_t read_pending_exceptions_done_count(void)
867 uint64_t pending_exceptions_done
;
869 spin_lock(&_pending_exceptions_done_spinlock
);
870 pending_exceptions_done
= _pending_exceptions_done_count
;
871 spin_unlock(&_pending_exceptions_done_spinlock
);
873 return pending_exceptions_done
;
876 static void increment_pending_exceptions_done_count(void)
878 spin_lock(&_pending_exceptions_done_spinlock
);
879 _pending_exceptions_done_count
++;
880 spin_unlock(&_pending_exceptions_done_spinlock
);
882 wake_up_all(&_pending_exceptions_done
);
885 static void snapshot_merge_next_chunks(struct dm_snapshot
*s
)
887 int i
, linear_chunks
;
888 chunk_t old_chunk
, new_chunk
;
889 struct dm_io_region src
, dest
;
891 uint64_t previous_count
;
893 BUG_ON(!test_bit(RUNNING_MERGE
, &s
->state_bits
));
894 if (unlikely(test_bit(SHUTDOWN_MERGE
, &s
->state_bits
)))
898 * valid flag never changes during merge, so no lock required.
901 DMERR("Snapshot is invalid: can't merge");
905 linear_chunks
= s
->store
->type
->prepare_merge(s
->store
, &old_chunk
,
907 if (linear_chunks
<= 0) {
908 if (linear_chunks
< 0) {
909 DMERR("Read error in exception store: "
910 "shutting down merge");
911 down_write(&s
->lock
);
918 /* Adjust old_chunk and new_chunk to reflect start of linear region */
919 old_chunk
= old_chunk
+ 1 - linear_chunks
;
920 new_chunk
= new_chunk
+ 1 - linear_chunks
;
923 * Use one (potentially large) I/O to copy all 'linear_chunks'
924 * from the exception store to the origin
926 io_size
= linear_chunks
* s
->store
->chunk_size
;
928 dest
.bdev
= s
->origin
->bdev
;
929 dest
.sector
= chunk_to_sector(s
->store
, old_chunk
);
930 dest
.count
= min(io_size
, get_dev_size(dest
.bdev
) - dest
.sector
);
932 src
.bdev
= s
->cow
->bdev
;
933 src
.sector
= chunk_to_sector(s
->store
, new_chunk
);
934 src
.count
= dest
.count
;
937 * Reallocate any exceptions needed in other snapshots then
938 * wait for the pending exceptions to complete.
939 * Each time any pending exception (globally on the system)
940 * completes we are woken and repeat the process to find out
941 * if we can proceed. While this may not seem a particularly
942 * efficient algorithm, it is not expected to have any
943 * significant impact on performance.
945 previous_count
= read_pending_exceptions_done_count();
946 while (origin_write_extent(s
, dest
.sector
, io_size
)) {
947 wait_event(_pending_exceptions_done
,
948 (read_pending_exceptions_done_count() !=
950 /* Retry after the wait, until all exceptions are done. */
951 previous_count
= read_pending_exceptions_done_count();
954 down_write(&s
->lock
);
955 s
->first_merging_chunk
= old_chunk
;
956 s
->num_merging_chunks
= linear_chunks
;
959 /* Wait until writes to all 'linear_chunks' drain */
960 for (i
= 0; i
< linear_chunks
; i
++)
961 __check_for_conflicting_io(s
, old_chunk
+ i
);
963 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, merge_callback
, s
);
970 static void error_bios(struct bio
*bio
);
972 static void merge_callback(int read_err
, unsigned long write_err
, void *context
)
974 struct dm_snapshot
*s
= context
;
975 struct bio
*b
= NULL
;
977 if (read_err
|| write_err
) {
979 DMERR("Read error: shutting down merge.");
981 DMERR("Write error: shutting down merge.");
985 if (s
->store
->type
->commit_merge(s
->store
,
986 s
->num_merging_chunks
) < 0) {
987 DMERR("Write error in exception store: shutting down merge");
991 if (remove_single_exception_chunk(s
) < 0)
994 snapshot_merge_next_chunks(s
);
999 down_write(&s
->lock
);
1000 s
->merge_failed
= 1;
1001 b
= __release_queued_bios_after_merge(s
);
1008 static void start_merge(struct dm_snapshot
*s
)
1010 if (!test_and_set_bit(RUNNING_MERGE
, &s
->state_bits
))
1011 snapshot_merge_next_chunks(s
);
1014 static int wait_schedule(void *ptr
)
1022 * Stop the merging process and wait until it finishes.
1024 static void stop_merge(struct dm_snapshot
*s
)
1026 set_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1027 wait_on_bit(&s
->state_bits
, RUNNING_MERGE
, wait_schedule
,
1028 TASK_UNINTERRUPTIBLE
);
1029 clear_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1033 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1035 static int snapshot_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
1037 struct dm_snapshot
*s
;
1040 char *origin_path
, *cow_path
;
1041 unsigned args_used
, num_flush_requests
= 1;
1042 fmode_t origin_mode
= FMODE_READ
;
1045 ti
->error
= "requires exactly 4 arguments";
1050 if (dm_target_is_snapshot_merge(ti
)) {
1051 num_flush_requests
= 2;
1052 origin_mode
= FMODE_WRITE
;
1055 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1057 ti
->error
= "Cannot allocate private snapshot structure";
1062 origin_path
= argv
[0];
1066 r
= dm_get_device(ti
, origin_path
, origin_mode
, &s
->origin
);
1068 ti
->error
= "Cannot get origin device";
1076 r
= dm_get_device(ti
, cow_path
, dm_table_get_mode(ti
->table
), &s
->cow
);
1078 ti
->error
= "Cannot get COW device";
1082 r
= dm_exception_store_create(ti
, argc
, argv
, s
, &args_used
, &s
->store
);
1084 ti
->error
= "Couldn't create exception store";
1095 atomic_set(&s
->pending_exceptions_count
, 0);
1096 init_rwsem(&s
->lock
);
1097 INIT_LIST_HEAD(&s
->list
);
1098 spin_lock_init(&s
->pe_lock
);
1100 s
->merge_failed
= 0;
1101 s
->first_merging_chunk
= 0;
1102 s
->num_merging_chunks
= 0;
1103 bio_list_init(&s
->bios_queued_during_merge
);
1105 /* Allocate hash table for COW data */
1106 if (init_hash_tables(s
)) {
1107 ti
->error
= "Unable to allocate hash table space";
1109 goto bad_hash_tables
;
1112 s
->kcopyd_client
= dm_kcopyd_client_create();
1113 if (IS_ERR(s
->kcopyd_client
)) {
1114 r
= PTR_ERR(s
->kcopyd_client
);
1115 ti
->error
= "Could not create kcopyd client";
1119 s
->pending_pool
= mempool_create_slab_pool(MIN_IOS
, pending_cache
);
1120 if (!s
->pending_pool
) {
1121 ti
->error
= "Could not allocate mempool for pending exceptions";
1122 goto bad_pending_pool
;
1125 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1126 INIT_HLIST_HEAD(&s
->tracked_chunk_hash
[i
]);
1128 spin_lock_init(&s
->tracked_chunk_lock
);
1131 ti
->num_flush_requests
= num_flush_requests
;
1132 ti
->per_bio_data_size
= sizeof(struct dm_snap_tracked_chunk
);
1134 /* Add snapshot to the list of snapshots for this origin */
1135 /* Exceptions aren't triggered till snapshot_resume() is called */
1136 r
= register_snapshot(s
);
1138 ti
->error
= "Snapshot origin struct allocation failed";
1139 goto bad_load_and_register
;
1141 /* invalid handover, register_snapshot has set ti->error */
1142 goto bad_load_and_register
;
1146 * Metadata must only be loaded into one table at once, so skip this
1147 * if metadata will be handed over during resume.
1148 * Chunk size will be set during the handover - set it to zero to
1149 * ensure it's ignored.
1152 s
->store
->chunk_size
= 0;
1156 r
= s
->store
->type
->read_metadata(s
->store
, dm_add_exception
,
1159 ti
->error
= "Failed to read snapshot metadata";
1160 goto bad_read_metadata
;
1163 DMWARN("Snapshot is marked invalid.");
1166 if (!s
->store
->chunk_size
) {
1167 ti
->error
= "Chunk size not set";
1168 goto bad_read_metadata
;
1171 r
= dm_set_target_max_io_len(ti
, s
->store
->chunk_size
);
1173 goto bad_read_metadata
;
1178 unregister_snapshot(s
);
1180 bad_load_and_register
:
1181 mempool_destroy(s
->pending_pool
);
1184 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1187 dm_exception_table_exit(&s
->pending
, pending_cache
);
1188 dm_exception_table_exit(&s
->complete
, exception_cache
);
1191 dm_exception_store_destroy(s
->store
);
1194 dm_put_device(ti
, s
->cow
);
1197 dm_put_device(ti
, s
->origin
);
1206 static void __free_exceptions(struct dm_snapshot
*s
)
1208 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1209 s
->kcopyd_client
= NULL
;
1211 dm_exception_table_exit(&s
->pending
, pending_cache
);
1212 dm_exception_table_exit(&s
->complete
, exception_cache
);
1215 static void __handover_exceptions(struct dm_snapshot
*snap_src
,
1216 struct dm_snapshot
*snap_dest
)
1219 struct dm_exception_table table_swap
;
1220 struct dm_exception_store
*store_swap
;
1224 * Swap all snapshot context information between the two instances.
1226 u
.table_swap
= snap_dest
->complete
;
1227 snap_dest
->complete
= snap_src
->complete
;
1228 snap_src
->complete
= u
.table_swap
;
1230 u
.store_swap
= snap_dest
->store
;
1231 snap_dest
->store
= snap_src
->store
;
1232 snap_src
->store
= u
.store_swap
;
1234 snap_dest
->store
->snap
= snap_dest
;
1235 snap_src
->store
->snap
= snap_src
;
1237 snap_dest
->ti
->max_io_len
= snap_dest
->store
->chunk_size
;
1238 snap_dest
->valid
= snap_src
->valid
;
1241 * Set source invalid to ensure it receives no further I/O.
1243 snap_src
->valid
= 0;
1246 static void snapshot_dtr(struct dm_target
*ti
)
1248 #ifdef CONFIG_DM_DEBUG
1251 struct dm_snapshot
*s
= ti
->private;
1252 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1254 down_read(&_origins_lock
);
1255 /* Check whether exception handover must be cancelled */
1256 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1257 if (snap_src
&& snap_dest
&& (s
== snap_src
)) {
1258 down_write(&snap_dest
->lock
);
1259 snap_dest
->valid
= 0;
1260 up_write(&snap_dest
->lock
);
1261 DMERR("Cancelling snapshot handover.");
1263 up_read(&_origins_lock
);
1265 if (dm_target_is_snapshot_merge(ti
))
1268 /* Prevent further origin writes from using this snapshot. */
1269 /* After this returns there can be no new kcopyd jobs. */
1270 unregister_snapshot(s
);
1272 while (atomic_read(&s
->pending_exceptions_count
))
1275 * Ensure instructions in mempool_destroy aren't reordered
1276 * before atomic_read.
1280 #ifdef CONFIG_DM_DEBUG
1281 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1282 BUG_ON(!hlist_empty(&s
->tracked_chunk_hash
[i
]));
1285 __free_exceptions(s
);
1287 mempool_destroy(s
->pending_pool
);
1289 dm_exception_store_destroy(s
->store
);
1291 dm_put_device(ti
, s
->cow
);
1293 dm_put_device(ti
, s
->origin
);
1299 * Flush a list of buffers.
1301 static void flush_bios(struct bio
*bio
)
1307 bio
->bi_next
= NULL
;
1308 generic_make_request(bio
);
1313 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
);
1316 * Flush a list of buffers.
1318 static void retry_origin_bios(struct dm_snapshot
*s
, struct bio
*bio
)
1325 bio
->bi_next
= NULL
;
1326 r
= do_origin(s
->origin
, bio
);
1327 if (r
== DM_MAPIO_REMAPPED
)
1328 generic_make_request(bio
);
1334 * Error a list of buffers.
1336 static void error_bios(struct bio
*bio
)
1342 bio
->bi_next
= NULL
;
1348 static void __invalidate_snapshot(struct dm_snapshot
*s
, int err
)
1354 DMERR("Invalidating snapshot: Error reading/writing.");
1355 else if (err
== -ENOMEM
)
1356 DMERR("Invalidating snapshot: Unable to allocate exception.");
1358 if (s
->store
->type
->drop_snapshot
)
1359 s
->store
->type
->drop_snapshot(s
->store
);
1363 dm_table_event(s
->ti
->table
);
1366 static void pending_complete(struct dm_snap_pending_exception
*pe
, int success
)
1368 struct dm_exception
*e
;
1369 struct dm_snapshot
*s
= pe
->snap
;
1370 struct bio
*origin_bios
= NULL
;
1371 struct bio
*snapshot_bios
= NULL
;
1372 struct bio
*full_bio
= NULL
;
1376 /* Read/write error - snapshot is unusable */
1377 down_write(&s
->lock
);
1378 __invalidate_snapshot(s
, -EIO
);
1383 e
= alloc_completed_exception();
1385 down_write(&s
->lock
);
1386 __invalidate_snapshot(s
, -ENOMEM
);
1392 down_write(&s
->lock
);
1394 free_completed_exception(e
);
1399 /* Check for conflicting reads */
1400 __check_for_conflicting_io(s
, pe
->e
.old_chunk
);
1403 * Add a proper exception, and remove the
1404 * in-flight exception from the list.
1406 dm_insert_exception(&s
->complete
, e
);
1409 dm_remove_exception(&pe
->e
);
1410 snapshot_bios
= bio_list_get(&pe
->snapshot_bios
);
1411 origin_bios
= bio_list_get(&pe
->origin_bios
);
1412 full_bio
= pe
->full_bio
;
1414 full_bio
->bi_end_io
= pe
->full_bio_end_io
;
1415 full_bio
->bi_private
= pe
->full_bio_private
;
1417 free_pending_exception(pe
);
1419 increment_pending_exceptions_done_count();
1423 /* Submit any pending write bios */
1426 bio_io_error(full_bio
);
1427 error_bios(snapshot_bios
);
1430 bio_endio(full_bio
, 0);
1431 flush_bios(snapshot_bios
);
1434 retry_origin_bios(s
, origin_bios
);
1437 static void commit_callback(void *context
, int success
)
1439 struct dm_snap_pending_exception
*pe
= context
;
1441 pending_complete(pe
, success
);
1445 * Called when the copy I/O has finished. kcopyd actually runs
1446 * this code so don't block.
1448 static void copy_callback(int read_err
, unsigned long write_err
, void *context
)
1450 struct dm_snap_pending_exception
*pe
= context
;
1451 struct dm_snapshot
*s
= pe
->snap
;
1453 if (read_err
|| write_err
)
1454 pending_complete(pe
, 0);
1457 /* Update the metadata if we are persistent */
1458 s
->store
->type
->commit_exception(s
->store
, &pe
->e
,
1459 commit_callback
, pe
);
1463 * Dispatches the copy operation to kcopyd.
1465 static void start_copy(struct dm_snap_pending_exception
*pe
)
1467 struct dm_snapshot
*s
= pe
->snap
;
1468 struct dm_io_region src
, dest
;
1469 struct block_device
*bdev
= s
->origin
->bdev
;
1472 dev_size
= get_dev_size(bdev
);
1475 src
.sector
= chunk_to_sector(s
->store
, pe
->e
.old_chunk
);
1476 src
.count
= min((sector_t
)s
->store
->chunk_size
, dev_size
- src
.sector
);
1478 dest
.bdev
= s
->cow
->bdev
;
1479 dest
.sector
= chunk_to_sector(s
->store
, pe
->e
.new_chunk
);
1480 dest
.count
= src
.count
;
1482 /* Hand over to kcopyd */
1483 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, copy_callback
, pe
);
1486 static void full_bio_end_io(struct bio
*bio
, int error
)
1488 void *callback_data
= bio
->bi_private
;
1490 dm_kcopyd_do_callback(callback_data
, 0, error
? 1 : 0);
1493 static void start_full_bio(struct dm_snap_pending_exception
*pe
,
1496 struct dm_snapshot
*s
= pe
->snap
;
1497 void *callback_data
;
1500 pe
->full_bio_end_io
= bio
->bi_end_io
;
1501 pe
->full_bio_private
= bio
->bi_private
;
1503 callback_data
= dm_kcopyd_prepare_callback(s
->kcopyd_client
,
1506 bio
->bi_end_io
= full_bio_end_io
;
1507 bio
->bi_private
= callback_data
;
1509 generic_make_request(bio
);
1512 static struct dm_snap_pending_exception
*
1513 __lookup_pending_exception(struct dm_snapshot
*s
, chunk_t chunk
)
1515 struct dm_exception
*e
= dm_lookup_exception(&s
->pending
, chunk
);
1520 return container_of(e
, struct dm_snap_pending_exception
, e
);
1524 * Looks to see if this snapshot already has a pending exception
1525 * for this chunk, otherwise it allocates a new one and inserts
1526 * it into the pending table.
1528 * NOTE: a write lock must be held on snap->lock before calling
1531 static struct dm_snap_pending_exception
*
1532 __find_pending_exception(struct dm_snapshot
*s
,
1533 struct dm_snap_pending_exception
*pe
, chunk_t chunk
)
1535 struct dm_snap_pending_exception
*pe2
;
1537 pe2
= __lookup_pending_exception(s
, chunk
);
1539 free_pending_exception(pe
);
1543 pe
->e
.old_chunk
= chunk
;
1544 bio_list_init(&pe
->origin_bios
);
1545 bio_list_init(&pe
->snapshot_bios
);
1547 pe
->full_bio
= NULL
;
1549 if (s
->store
->type
->prepare_exception(s
->store
, &pe
->e
)) {
1550 free_pending_exception(pe
);
1554 dm_insert_exception(&s
->pending
, &pe
->e
);
1559 static void remap_exception(struct dm_snapshot
*s
, struct dm_exception
*e
,
1560 struct bio
*bio
, chunk_t chunk
)
1562 bio
->bi_bdev
= s
->cow
->bdev
;
1563 bio
->bi_sector
= chunk_to_sector(s
->store
,
1564 dm_chunk_number(e
->new_chunk
) +
1565 (chunk
- e
->old_chunk
)) +
1567 s
->store
->chunk_mask
);
1570 static int snapshot_map(struct dm_target
*ti
, struct bio
*bio
)
1572 struct dm_exception
*e
;
1573 struct dm_snapshot
*s
= ti
->private;
1574 int r
= DM_MAPIO_REMAPPED
;
1576 struct dm_snap_pending_exception
*pe
= NULL
;
1578 init_tracked_chunk(bio
);
1580 if (bio
->bi_rw
& REQ_FLUSH
) {
1581 bio
->bi_bdev
= s
->cow
->bdev
;
1582 return DM_MAPIO_REMAPPED
;
1585 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1587 /* Full snapshots are not usable */
1588 /* To get here the table must be live so s->active is always set. */
1592 /* FIXME: should only take write lock if we need
1593 * to copy an exception */
1594 down_write(&s
->lock
);
1601 /* If the block is already remapped - use that, else remap it */
1602 e
= dm_lookup_exception(&s
->complete
, chunk
);
1604 remap_exception(s
, e
, bio
, chunk
);
1609 * Write to snapshot - higher level takes care of RW/RO
1610 * flags so we should only get this if we are
1613 if (bio_rw(bio
) == WRITE
) {
1614 pe
= __lookup_pending_exception(s
, chunk
);
1617 pe
= alloc_pending_exception(s
);
1618 down_write(&s
->lock
);
1621 free_pending_exception(pe
);
1626 e
= dm_lookup_exception(&s
->complete
, chunk
);
1628 free_pending_exception(pe
);
1629 remap_exception(s
, e
, bio
, chunk
);
1633 pe
= __find_pending_exception(s
, pe
, chunk
);
1635 __invalidate_snapshot(s
, -ENOMEM
);
1641 remap_exception(s
, &pe
->e
, bio
, chunk
);
1643 r
= DM_MAPIO_SUBMITTED
;
1646 bio
->bi_size
== (s
->store
->chunk_size
<< SECTOR_SHIFT
)) {
1649 start_full_bio(pe
, bio
);
1653 bio_list_add(&pe
->snapshot_bios
, bio
);
1656 /* this is protected by snap->lock */
1663 bio
->bi_bdev
= s
->origin
->bdev
;
1664 track_chunk(s
, bio
, chunk
);
1674 * A snapshot-merge target behaves like a combination of a snapshot
1675 * target and a snapshot-origin target. It only generates new
1676 * exceptions in other snapshots and not in the one that is being
1679 * For each chunk, if there is an existing exception, it is used to
1680 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1681 * which in turn might generate exceptions in other snapshots.
1682 * If merging is currently taking place on the chunk in question, the
1683 * I/O is deferred by adding it to s->bios_queued_during_merge.
1685 static int snapshot_merge_map(struct dm_target
*ti
, struct bio
*bio
)
1687 struct dm_exception
*e
;
1688 struct dm_snapshot
*s
= ti
->private;
1689 int r
= DM_MAPIO_REMAPPED
;
1692 init_tracked_chunk(bio
);
1694 if (bio
->bi_rw
& REQ_FLUSH
) {
1695 if (!dm_bio_get_target_request_nr(bio
))
1696 bio
->bi_bdev
= s
->origin
->bdev
;
1698 bio
->bi_bdev
= s
->cow
->bdev
;
1699 return DM_MAPIO_REMAPPED
;
1702 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1704 down_write(&s
->lock
);
1706 /* Full merging snapshots are redirected to the origin */
1708 goto redirect_to_origin
;
1710 /* If the block is already remapped - use that */
1711 e
= dm_lookup_exception(&s
->complete
, chunk
);
1713 /* Queue writes overlapping with chunks being merged */
1714 if (bio_rw(bio
) == WRITE
&&
1715 chunk
>= s
->first_merging_chunk
&&
1716 chunk
< (s
->first_merging_chunk
+
1717 s
->num_merging_chunks
)) {
1718 bio
->bi_bdev
= s
->origin
->bdev
;
1719 bio_list_add(&s
->bios_queued_during_merge
, bio
);
1720 r
= DM_MAPIO_SUBMITTED
;
1724 remap_exception(s
, e
, bio
, chunk
);
1726 if (bio_rw(bio
) == WRITE
)
1727 track_chunk(s
, bio
, chunk
);
1732 bio
->bi_bdev
= s
->origin
->bdev
;
1734 if (bio_rw(bio
) == WRITE
) {
1736 return do_origin(s
->origin
, bio
);
1745 static int snapshot_end_io(struct dm_target
*ti
, struct bio
*bio
, int error
)
1747 struct dm_snapshot
*s
= ti
->private;
1749 if (is_bio_tracked(bio
))
1750 stop_tracking_chunk(s
, bio
);
1755 static void snapshot_merge_presuspend(struct dm_target
*ti
)
1757 struct dm_snapshot
*s
= ti
->private;
1762 static int snapshot_preresume(struct dm_target
*ti
)
1765 struct dm_snapshot
*s
= ti
->private;
1766 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1768 down_read(&_origins_lock
);
1769 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1770 if (snap_src
&& snap_dest
) {
1771 down_read(&snap_src
->lock
);
1772 if (s
== snap_src
) {
1773 DMERR("Unable to resume snapshot source until "
1774 "handover completes.");
1776 } else if (!dm_suspended(snap_src
->ti
)) {
1777 DMERR("Unable to perform snapshot handover until "
1778 "source is suspended.");
1781 up_read(&snap_src
->lock
);
1783 up_read(&_origins_lock
);
1788 static void snapshot_resume(struct dm_target
*ti
)
1790 struct dm_snapshot
*s
= ti
->private;
1791 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1793 down_read(&_origins_lock
);
1794 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1795 if (snap_src
&& snap_dest
) {
1796 down_write(&snap_src
->lock
);
1797 down_write_nested(&snap_dest
->lock
, SINGLE_DEPTH_NESTING
);
1798 __handover_exceptions(snap_src
, snap_dest
);
1799 up_write(&snap_dest
->lock
);
1800 up_write(&snap_src
->lock
);
1802 up_read(&_origins_lock
);
1804 /* Now we have correct chunk size, reregister */
1805 reregister_snapshot(s
);
1807 down_write(&s
->lock
);
1812 static uint32_t get_origin_minimum_chunksize(struct block_device
*bdev
)
1814 uint32_t min_chunksize
;
1816 down_read(&_origins_lock
);
1817 min_chunksize
= __minimum_chunk_size(__lookup_origin(bdev
));
1818 up_read(&_origins_lock
);
1820 return min_chunksize
;
1823 static void snapshot_merge_resume(struct dm_target
*ti
)
1825 struct dm_snapshot
*s
= ti
->private;
1828 * Handover exceptions from existing snapshot.
1830 snapshot_resume(ti
);
1833 * snapshot-merge acts as an origin, so set ti->max_io_len
1835 ti
->max_io_len
= get_origin_minimum_chunksize(s
->origin
->bdev
);
1840 static int snapshot_status(struct dm_target
*ti
, status_type_t type
,
1841 unsigned status_flags
, char *result
, unsigned maxlen
)
1844 struct dm_snapshot
*snap
= ti
->private;
1847 case STATUSTYPE_INFO
:
1849 down_write(&snap
->lock
);
1853 else if (snap
->merge_failed
)
1854 DMEMIT("Merge failed");
1856 if (snap
->store
->type
->usage
) {
1857 sector_t total_sectors
, sectors_allocated
,
1859 snap
->store
->type
->usage(snap
->store
,
1863 DMEMIT("%llu/%llu %llu",
1864 (unsigned long long)sectors_allocated
,
1865 (unsigned long long)total_sectors
,
1866 (unsigned long long)metadata_sectors
);
1872 up_write(&snap
->lock
);
1876 case STATUSTYPE_TABLE
:
1878 * kdevname returns a static pointer so we need
1879 * to make private copies if the output is to
1882 DMEMIT("%s %s", snap
->origin
->name
, snap
->cow
->name
);
1883 snap
->store
->type
->status(snap
->store
, type
, result
+ sz
,
1891 static int snapshot_iterate_devices(struct dm_target
*ti
,
1892 iterate_devices_callout_fn fn
, void *data
)
1894 struct dm_snapshot
*snap
= ti
->private;
1897 r
= fn(ti
, snap
->origin
, 0, ti
->len
, data
);
1900 r
= fn(ti
, snap
->cow
, 0, get_dev_size(snap
->cow
->bdev
), data
);
1906 /*-----------------------------------------------------------------
1908 *---------------------------------------------------------------*/
1911 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1912 * supplied bio was ignored. The caller may submit it immediately.
1913 * (No remapping actually occurs as the origin is always a direct linear
1916 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1917 * and any supplied bio is added to a list to be submitted once all
1918 * the necessary exceptions exist.
1920 static int __origin_write(struct list_head
*snapshots
, sector_t sector
,
1923 int r
= DM_MAPIO_REMAPPED
;
1924 struct dm_snapshot
*snap
;
1925 struct dm_exception
*e
;
1926 struct dm_snap_pending_exception
*pe
;
1927 struct dm_snap_pending_exception
*pe_to_start_now
= NULL
;
1928 struct dm_snap_pending_exception
*pe_to_start_last
= NULL
;
1931 /* Do all the snapshots on this origin */
1932 list_for_each_entry (snap
, snapshots
, list
) {
1934 * Don't make new exceptions in a merging snapshot
1935 * because it has effectively been deleted
1937 if (dm_target_is_snapshot_merge(snap
->ti
))
1940 down_write(&snap
->lock
);
1942 /* Only deal with valid and active snapshots */
1943 if (!snap
->valid
|| !snap
->active
)
1946 /* Nothing to do if writing beyond end of snapshot */
1947 if (sector
>= dm_table_get_size(snap
->ti
->table
))
1951 * Remember, different snapshots can have
1952 * different chunk sizes.
1954 chunk
= sector_to_chunk(snap
->store
, sector
);
1957 * Check exception table to see if block
1958 * is already remapped in this snapshot
1959 * and trigger an exception if not.
1961 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1965 pe
= __lookup_pending_exception(snap
, chunk
);
1967 up_write(&snap
->lock
);
1968 pe
= alloc_pending_exception(snap
);
1969 down_write(&snap
->lock
);
1972 free_pending_exception(pe
);
1976 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1978 free_pending_exception(pe
);
1982 pe
= __find_pending_exception(snap
, pe
, chunk
);
1984 __invalidate_snapshot(snap
, -ENOMEM
);
1989 r
= DM_MAPIO_SUBMITTED
;
1992 * If an origin bio was supplied, queue it to wait for the
1993 * completion of this exception, and start this one last,
1994 * at the end of the function.
1997 bio_list_add(&pe
->origin_bios
, bio
);
2002 pe_to_start_last
= pe
;
2008 pe_to_start_now
= pe
;
2012 up_write(&snap
->lock
);
2014 if (pe_to_start_now
) {
2015 start_copy(pe_to_start_now
);
2016 pe_to_start_now
= NULL
;
2021 * Submit the exception against which the bio is queued last,
2022 * to give the other exceptions a head start.
2024 if (pe_to_start_last
)
2025 start_copy(pe_to_start_last
);
2031 * Called on a write from the origin driver.
2033 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
)
2036 int r
= DM_MAPIO_REMAPPED
;
2038 down_read(&_origins_lock
);
2039 o
= __lookup_origin(origin
->bdev
);
2041 r
= __origin_write(&o
->snapshots
, bio
->bi_sector
, bio
);
2042 up_read(&_origins_lock
);
2048 * Trigger exceptions in all non-merging snapshots.
2050 * The chunk size of the merging snapshot may be larger than the chunk
2051 * size of some other snapshot so we may need to reallocate multiple
2052 * chunks in other snapshots.
2054 * We scan all the overlapping exceptions in the other snapshots.
2055 * Returns 1 if anything was reallocated and must be waited for,
2056 * otherwise returns 0.
2058 * size must be a multiple of merging_snap's chunk_size.
2060 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
2061 sector_t sector
, unsigned size
)
2068 * The origin's __minimum_chunk_size() got stored in max_io_len
2069 * by snapshot_merge_resume().
2071 down_read(&_origins_lock
);
2072 o
= __lookup_origin(merging_snap
->origin
->bdev
);
2073 for (n
= 0; n
< size
; n
+= merging_snap
->ti
->max_io_len
)
2074 if (__origin_write(&o
->snapshots
, sector
+ n
, NULL
) ==
2077 up_read(&_origins_lock
);
2083 * Origin: maps a linear range of a device, with hooks for snapshotting.
2087 * Construct an origin mapping: <dev_path>
2088 * The context for an origin is merely a 'struct dm_dev *'
2089 * pointing to the real device.
2091 static int origin_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
2097 ti
->error
= "origin: incorrect number of arguments";
2101 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &dev
);
2103 ti
->error
= "Cannot get target device";
2108 ti
->num_flush_requests
= 1;
2113 static void origin_dtr(struct dm_target
*ti
)
2115 struct dm_dev
*dev
= ti
->private;
2116 dm_put_device(ti
, dev
);
2119 static int origin_map(struct dm_target
*ti
, struct bio
*bio
)
2121 struct dm_dev
*dev
= ti
->private;
2122 bio
->bi_bdev
= dev
->bdev
;
2124 if (bio
->bi_rw
& REQ_FLUSH
)
2125 return DM_MAPIO_REMAPPED
;
2127 /* Only tell snapshots if this is a write */
2128 return (bio_rw(bio
) == WRITE
) ? do_origin(dev
, bio
) : DM_MAPIO_REMAPPED
;
2132 * Set the target "max_io_len" field to the minimum of all the snapshots'
2135 static void origin_resume(struct dm_target
*ti
)
2137 struct dm_dev
*dev
= ti
->private;
2139 ti
->max_io_len
= get_origin_minimum_chunksize(dev
->bdev
);
2142 static int origin_status(struct dm_target
*ti
, status_type_t type
,
2143 unsigned status_flags
, char *result
, unsigned maxlen
)
2145 struct dm_dev
*dev
= ti
->private;
2148 case STATUSTYPE_INFO
:
2152 case STATUSTYPE_TABLE
:
2153 snprintf(result
, maxlen
, "%s", dev
->name
);
2160 static int origin_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2161 struct bio_vec
*biovec
, int max_size
)
2163 struct dm_dev
*dev
= ti
->private;
2164 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
2166 if (!q
->merge_bvec_fn
)
2169 bvm
->bi_bdev
= dev
->bdev
;
2171 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2174 static int origin_iterate_devices(struct dm_target
*ti
,
2175 iterate_devices_callout_fn fn
, void *data
)
2177 struct dm_dev
*dev
= ti
->private;
2179 return fn(ti
, dev
, 0, ti
->len
, data
);
2182 static struct target_type origin_target
= {
2183 .name
= "snapshot-origin",
2184 .version
= {1, 8, 0},
2185 .module
= THIS_MODULE
,
2189 .resume
= origin_resume
,
2190 .status
= origin_status
,
2191 .merge
= origin_merge
,
2192 .iterate_devices
= origin_iterate_devices
,
2195 static struct target_type snapshot_target
= {
2197 .version
= {1, 11, 0},
2198 .module
= THIS_MODULE
,
2199 .ctr
= snapshot_ctr
,
2200 .dtr
= snapshot_dtr
,
2201 .map
= snapshot_map
,
2202 .end_io
= snapshot_end_io
,
2203 .preresume
= snapshot_preresume
,
2204 .resume
= snapshot_resume
,
2205 .status
= snapshot_status
,
2206 .iterate_devices
= snapshot_iterate_devices
,
2209 static struct target_type merge_target
= {
2210 .name
= dm_snapshot_merge_target_name
,
2211 .version
= {1, 2, 0},
2212 .module
= THIS_MODULE
,
2213 .ctr
= snapshot_ctr
,
2214 .dtr
= snapshot_dtr
,
2215 .map
= snapshot_merge_map
,
2216 .end_io
= snapshot_end_io
,
2217 .presuspend
= snapshot_merge_presuspend
,
2218 .preresume
= snapshot_preresume
,
2219 .resume
= snapshot_merge_resume
,
2220 .status
= snapshot_status
,
2221 .iterate_devices
= snapshot_iterate_devices
,
2224 static int __init
dm_snapshot_init(void)
2228 r
= dm_exception_store_init();
2230 DMERR("Failed to initialize exception stores");
2234 r
= dm_register_target(&snapshot_target
);
2236 DMERR("snapshot target register failed %d", r
);
2237 goto bad_register_snapshot_target
;
2240 r
= dm_register_target(&origin_target
);
2242 DMERR("Origin target register failed %d", r
);
2243 goto bad_register_origin_target
;
2246 r
= dm_register_target(&merge_target
);
2248 DMERR("Merge target register failed %d", r
);
2249 goto bad_register_merge_target
;
2252 r
= init_origin_hash();
2254 DMERR("init_origin_hash failed.");
2255 goto bad_origin_hash
;
2258 exception_cache
= KMEM_CACHE(dm_exception
, 0);
2259 if (!exception_cache
) {
2260 DMERR("Couldn't create exception cache.");
2262 goto bad_exception_cache
;
2265 pending_cache
= KMEM_CACHE(dm_snap_pending_exception
, 0);
2266 if (!pending_cache
) {
2267 DMERR("Couldn't create pending cache.");
2269 goto bad_pending_cache
;
2275 kmem_cache_destroy(exception_cache
);
2276 bad_exception_cache
:
2279 dm_unregister_target(&merge_target
);
2280 bad_register_merge_target
:
2281 dm_unregister_target(&origin_target
);
2282 bad_register_origin_target
:
2283 dm_unregister_target(&snapshot_target
);
2284 bad_register_snapshot_target
:
2285 dm_exception_store_exit();
2290 static void __exit
dm_snapshot_exit(void)
2292 dm_unregister_target(&snapshot_target
);
2293 dm_unregister_target(&origin_target
);
2294 dm_unregister_target(&merge_target
);
2297 kmem_cache_destroy(pending_cache
);
2298 kmem_cache_destroy(exception_cache
);
2300 dm_exception_store_exit();
2304 module_init(dm_snapshot_init
);
2305 module_exit(dm_snapshot_exit
);
2307 MODULE_DESCRIPTION(DM_NAME
" snapshot target");
2308 MODULE_AUTHOR("Joe Thornber");
2309 MODULE_LICENSE("GPL");