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
;
75 * You can't use a snapshot if this is 0 (e.g. if full).
76 * A snapshot-merge target never clears this.
80 /* Origin writes don't trigger exceptions until this is set */
83 /* Whether or not owning mapped_device is suspended */
86 atomic_t pending_exceptions_count
;
88 mempool_t
*pending_pool
;
90 struct dm_exception_table pending
;
91 struct dm_exception_table complete
;
94 * pe_lock protects all pending_exception operations and access
95 * as well as the snapshot_bios list.
99 /* Chunks with outstanding reads */
100 spinlock_t tracked_chunk_lock
;
101 mempool_t
*tracked_chunk_pool
;
102 struct hlist_head tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH_SIZE
];
104 /* The on disk metadata handler */
105 struct dm_exception_store
*store
;
107 struct dm_kcopyd_client
*kcopyd_client
;
109 /* Queue of snapshot writes for ksnapd to flush */
110 struct bio_list queued_bios
;
111 struct work_struct queued_bios_work
;
113 /* Wait for events based on state_bits */
114 unsigned long state_bits
;
116 /* Range of chunks currently being merged. */
117 chunk_t first_merging_chunk
;
118 int num_merging_chunks
;
121 * The merge operation failed if this flag is set.
122 * Failure modes are handled as follows:
123 * - I/O error reading the header
124 * => don't load the target; abort.
125 * - Header does not have "valid" flag set
126 * => use the origin; forget about the snapshot.
127 * - I/O error when reading exceptions
128 * => don't load the target; abort.
129 * (We can't use the intermediate origin state.)
130 * - I/O error while merging
131 * => stop merging; set merge_failed; process I/O normally.
136 * Incoming bios that overlap with chunks being merged must wait
137 * for them to be committed.
139 struct bio_list bios_queued_during_merge
;
144 * RUNNING_MERGE - Merge operation is in progress.
145 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
146 * cleared afterwards.
148 #define RUNNING_MERGE 0
149 #define SHUTDOWN_MERGE 1
151 struct dm_dev
*dm_snap_origin(struct dm_snapshot
*s
)
155 EXPORT_SYMBOL(dm_snap_origin
);
157 struct dm_dev
*dm_snap_cow(struct dm_snapshot
*s
)
161 EXPORT_SYMBOL(dm_snap_cow
);
163 static struct workqueue_struct
*ksnapd
;
164 static void flush_queued_bios(struct work_struct
*work
);
166 static sector_t
chunk_to_sector(struct dm_exception_store
*store
,
169 return chunk
<< store
->chunk_shift
;
172 static int bdev_equal(struct block_device
*lhs
, struct block_device
*rhs
)
175 * There is only ever one instance of a particular block
176 * device so we can compare pointers safely.
181 struct dm_snap_pending_exception
{
182 struct dm_exception e
;
185 * Origin buffers waiting for this to complete are held
188 struct bio_list origin_bios
;
189 struct bio_list snapshot_bios
;
191 /* Pointer back to snapshot context */
192 struct dm_snapshot
*snap
;
195 * 1 indicates the exception has already been sent to
202 * Hash table mapping origin volumes to lists of snapshots and
203 * a lock to protect it
205 static struct kmem_cache
*exception_cache
;
206 static struct kmem_cache
*pending_cache
;
208 struct dm_snap_tracked_chunk
{
209 struct hlist_node node
;
213 static struct kmem_cache
*tracked_chunk_cache
;
215 static struct dm_snap_tracked_chunk
*track_chunk(struct dm_snapshot
*s
,
218 struct dm_snap_tracked_chunk
*c
= mempool_alloc(s
->tracked_chunk_pool
,
224 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
225 hlist_add_head(&c
->node
,
226 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)]);
227 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
232 static void stop_tracking_chunk(struct dm_snapshot
*s
,
233 struct dm_snap_tracked_chunk
*c
)
237 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
239 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
241 mempool_free(c
, s
->tracked_chunk_pool
);
244 static int __chunk_is_tracked(struct dm_snapshot
*s
, chunk_t chunk
)
246 struct dm_snap_tracked_chunk
*c
;
247 struct hlist_node
*hn
;
250 spin_lock_irq(&s
->tracked_chunk_lock
);
252 hlist_for_each_entry(c
, hn
,
253 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)], node
) {
254 if (c
->chunk
== chunk
) {
260 spin_unlock_irq(&s
->tracked_chunk_lock
);
266 * This conflicting I/O is extremely improbable in the caller,
267 * so msleep(1) is sufficient and there is no need for a wait queue.
269 static void __check_for_conflicting_io(struct dm_snapshot
*s
, chunk_t chunk
)
271 while (__chunk_is_tracked(s
, chunk
))
276 * One of these per registered origin, held in the snapshot_origins hash
279 /* The origin device */
280 struct block_device
*bdev
;
282 struct list_head hash_list
;
284 /* List of snapshots for this origin */
285 struct list_head snapshots
;
289 * Size of the hash table for origin volumes. If we make this
290 * the size of the minors list then it should be nearly perfect
292 #define ORIGIN_HASH_SIZE 256
293 #define ORIGIN_MASK 0xFF
294 static struct list_head
*_origins
;
295 static struct rw_semaphore _origins_lock
;
297 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done
);
298 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock
);
299 static uint64_t _pending_exceptions_done_count
;
301 static int init_origin_hash(void)
305 _origins
= kmalloc(ORIGIN_HASH_SIZE
* sizeof(struct list_head
),
308 DMERR("unable to allocate memory");
312 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
313 INIT_LIST_HEAD(_origins
+ i
);
314 init_rwsem(&_origins_lock
);
319 static void exit_origin_hash(void)
324 static unsigned origin_hash(struct block_device
*bdev
)
326 return bdev
->bd_dev
& ORIGIN_MASK
;
329 static struct origin
*__lookup_origin(struct block_device
*origin
)
331 struct list_head
*ol
;
334 ol
= &_origins
[origin_hash(origin
)];
335 list_for_each_entry (o
, ol
, hash_list
)
336 if (bdev_equal(o
->bdev
, origin
))
342 static void __insert_origin(struct origin
*o
)
344 struct list_head
*sl
= &_origins
[origin_hash(o
->bdev
)];
345 list_add_tail(&o
->hash_list
, sl
);
349 * _origins_lock must be held when calling this function.
350 * Returns number of snapshots registered using the supplied cow device, plus:
351 * snap_src - a snapshot suitable for use as a source of exception handover
352 * snap_dest - a snapshot capable of receiving exception handover.
353 * snap_merge - an existing snapshot-merge target linked to the same origin.
354 * There can be at most one snapshot-merge target. The parameter is optional.
356 * Possible return values and states of snap_src and snap_dest.
357 * 0: NULL, NULL - first new snapshot
358 * 1: snap_src, NULL - normal snapshot
359 * 2: snap_src, snap_dest - waiting for handover
360 * 2: snap_src, NULL - handed over, waiting for old to be deleted
361 * 1: NULL, snap_dest - source got destroyed without handover
363 static int __find_snapshots_sharing_cow(struct dm_snapshot
*snap
,
364 struct dm_snapshot
**snap_src
,
365 struct dm_snapshot
**snap_dest
,
366 struct dm_snapshot
**snap_merge
)
368 struct dm_snapshot
*s
;
373 o
= __lookup_origin(snap
->origin
->bdev
);
377 list_for_each_entry(s
, &o
->snapshots
, list
) {
378 if (dm_target_is_snapshot_merge(s
->ti
) && snap_merge
)
380 if (!bdev_equal(s
->cow
->bdev
, snap
->cow
->bdev
))
390 } else if (snap_dest
)
401 * On success, returns 1 if this snapshot is a handover destination,
402 * otherwise returns 0.
404 static int __validate_exception_handover(struct dm_snapshot
*snap
)
406 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
407 struct dm_snapshot
*snap_merge
= NULL
;
409 /* Does snapshot need exceptions handed over to it? */
410 if ((__find_snapshots_sharing_cow(snap
, &snap_src
, &snap_dest
,
411 &snap_merge
) == 2) ||
413 snap
->ti
->error
= "Snapshot cow pairing for exception "
414 "table handover failed";
419 * If no snap_src was found, snap cannot become a handover
426 * Non-snapshot-merge handover?
428 if (!dm_target_is_snapshot_merge(snap
->ti
))
432 * Do not allow more than one merging snapshot.
435 snap
->ti
->error
= "A snapshot is already merging.";
439 if (!snap_src
->store
->type
->prepare_merge
||
440 !snap_src
->store
->type
->commit_merge
) {
441 snap
->ti
->error
= "Snapshot exception store does not "
442 "support snapshot-merge.";
449 static void __insert_snapshot(struct origin
*o
, struct dm_snapshot
*s
)
451 struct dm_snapshot
*l
;
453 /* Sort the list according to chunk size, largest-first smallest-last */
454 list_for_each_entry(l
, &o
->snapshots
, list
)
455 if (l
->store
->chunk_size
< s
->store
->chunk_size
)
457 list_add_tail(&s
->list
, &l
->list
);
461 * Make a note of the snapshot and its origin so we can look it
462 * up when the origin has a write on it.
464 * Also validate snapshot exception store handovers.
465 * On success, returns 1 if this registration is a handover destination,
466 * otherwise returns 0.
468 static int register_snapshot(struct dm_snapshot
*snap
)
470 struct origin
*o
, *new_o
= NULL
;
471 struct block_device
*bdev
= snap
->origin
->bdev
;
474 new_o
= kmalloc(sizeof(*new_o
), GFP_KERNEL
);
478 down_write(&_origins_lock
);
480 r
= __validate_exception_handover(snap
);
486 o
= __lookup_origin(bdev
);
493 /* Initialise the struct */
494 INIT_LIST_HEAD(&o
->snapshots
);
500 __insert_snapshot(o
, snap
);
503 up_write(&_origins_lock
);
509 * Move snapshot to correct place in list according to chunk size.
511 static void reregister_snapshot(struct dm_snapshot
*s
)
513 struct block_device
*bdev
= s
->origin
->bdev
;
515 down_write(&_origins_lock
);
518 __insert_snapshot(__lookup_origin(bdev
), s
);
520 up_write(&_origins_lock
);
523 static void unregister_snapshot(struct dm_snapshot
*s
)
527 down_write(&_origins_lock
);
528 o
= __lookup_origin(s
->origin
->bdev
);
531 if (o
&& list_empty(&o
->snapshots
)) {
532 list_del(&o
->hash_list
);
536 up_write(&_origins_lock
);
540 * Implementation of the exception hash tables.
541 * The lowest hash_shift bits of the chunk number are ignored, allowing
542 * some consecutive chunks to be grouped together.
544 static int dm_exception_table_init(struct dm_exception_table
*et
,
545 uint32_t size
, unsigned hash_shift
)
549 et
->hash_shift
= hash_shift
;
550 et
->hash_mask
= size
- 1;
551 et
->table
= dm_vcalloc(size
, sizeof(struct list_head
));
555 for (i
= 0; i
< size
; i
++)
556 INIT_LIST_HEAD(et
->table
+ i
);
561 static void dm_exception_table_exit(struct dm_exception_table
*et
,
562 struct kmem_cache
*mem
)
564 struct list_head
*slot
;
565 struct dm_exception
*ex
, *next
;
568 size
= et
->hash_mask
+ 1;
569 for (i
= 0; i
< size
; i
++) {
570 slot
= et
->table
+ i
;
572 list_for_each_entry_safe (ex
, next
, slot
, hash_list
)
573 kmem_cache_free(mem
, ex
);
579 static uint32_t exception_hash(struct dm_exception_table
*et
, chunk_t chunk
)
581 return (chunk
>> et
->hash_shift
) & et
->hash_mask
;
584 static void dm_remove_exception(struct dm_exception
*e
)
586 list_del(&e
->hash_list
);
590 * Return the exception data for a sector, or NULL if not
593 static struct dm_exception
*dm_lookup_exception(struct dm_exception_table
*et
,
596 struct list_head
*slot
;
597 struct dm_exception
*e
;
599 slot
= &et
->table
[exception_hash(et
, chunk
)];
600 list_for_each_entry (e
, slot
, hash_list
)
601 if (chunk
>= e
->old_chunk
&&
602 chunk
<= e
->old_chunk
+ dm_consecutive_chunk_count(e
))
608 static struct dm_exception
*alloc_completed_exception(void)
610 struct dm_exception
*e
;
612 e
= kmem_cache_alloc(exception_cache
, GFP_NOIO
);
614 e
= kmem_cache_alloc(exception_cache
, GFP_ATOMIC
);
619 static void free_completed_exception(struct dm_exception
*e
)
621 kmem_cache_free(exception_cache
, e
);
624 static struct dm_snap_pending_exception
*alloc_pending_exception(struct dm_snapshot
*s
)
626 struct dm_snap_pending_exception
*pe
= mempool_alloc(s
->pending_pool
,
629 atomic_inc(&s
->pending_exceptions_count
);
635 static void free_pending_exception(struct dm_snap_pending_exception
*pe
)
637 struct dm_snapshot
*s
= pe
->snap
;
639 mempool_free(pe
, s
->pending_pool
);
640 smp_mb__before_atomic_dec();
641 atomic_dec(&s
->pending_exceptions_count
);
644 static void dm_insert_exception(struct dm_exception_table
*eh
,
645 struct dm_exception
*new_e
)
648 struct dm_exception
*e
= NULL
;
650 l
= &eh
->table
[exception_hash(eh
, new_e
->old_chunk
)];
652 /* Add immediately if this table doesn't support consecutive chunks */
656 /* List is ordered by old_chunk */
657 list_for_each_entry_reverse(e
, l
, hash_list
) {
658 /* Insert after an existing chunk? */
659 if (new_e
->old_chunk
== (e
->old_chunk
+
660 dm_consecutive_chunk_count(e
) + 1) &&
661 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) +
662 dm_consecutive_chunk_count(e
) + 1)) {
663 dm_consecutive_chunk_count_inc(e
);
664 free_completed_exception(new_e
);
668 /* Insert before an existing chunk? */
669 if (new_e
->old_chunk
== (e
->old_chunk
- 1) &&
670 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) - 1)) {
671 dm_consecutive_chunk_count_inc(e
);
674 free_completed_exception(new_e
);
678 if (new_e
->old_chunk
> e
->old_chunk
)
683 list_add(&new_e
->hash_list
, e
? &e
->hash_list
: l
);
687 * Callback used by the exception stores to load exceptions when
690 static int dm_add_exception(void *context
, chunk_t old
, chunk_t
new)
692 struct dm_snapshot
*s
= context
;
693 struct dm_exception
*e
;
695 e
= alloc_completed_exception();
701 /* Consecutive_count is implicitly initialised to zero */
704 dm_insert_exception(&s
->complete
, e
);
709 #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r)))
712 * Return a minimum chunk size of all snapshots that have the specified origin.
713 * Return zero if the origin has no snapshots.
715 static sector_t
__minimum_chunk_size(struct origin
*o
)
717 struct dm_snapshot
*snap
;
718 unsigned chunk_size
= 0;
721 list_for_each_entry(snap
, &o
->snapshots
, list
)
722 chunk_size
= min_not_zero(chunk_size
,
723 snap
->store
->chunk_size
);
731 static int calc_max_buckets(void)
733 /* use a fixed size of 2MB */
734 unsigned long mem
= 2 * 1024 * 1024;
735 mem
/= sizeof(struct list_head
);
741 * Allocate room for a suitable hash table.
743 static int init_hash_tables(struct dm_snapshot
*s
)
745 sector_t hash_size
, cow_dev_size
, origin_dev_size
, max_buckets
;
748 * Calculate based on the size of the original volume or
751 cow_dev_size
= get_dev_size(s
->cow
->bdev
);
752 origin_dev_size
= get_dev_size(s
->origin
->bdev
);
753 max_buckets
= calc_max_buckets();
755 hash_size
= min(origin_dev_size
, cow_dev_size
) >> s
->store
->chunk_shift
;
756 hash_size
= min(hash_size
, max_buckets
);
760 hash_size
= rounddown_pow_of_two(hash_size
);
761 if (dm_exception_table_init(&s
->complete
, hash_size
,
762 DM_CHUNK_CONSECUTIVE_BITS
))
766 * Allocate hash table for in-flight exceptions
767 * Make this smaller than the real hash table
773 if (dm_exception_table_init(&s
->pending
, hash_size
, 0)) {
774 dm_exception_table_exit(&s
->complete
, exception_cache
);
781 static void merge_shutdown(struct dm_snapshot
*s
)
783 clear_bit_unlock(RUNNING_MERGE
, &s
->state_bits
);
784 smp_mb__after_clear_bit();
785 wake_up_bit(&s
->state_bits
, RUNNING_MERGE
);
788 static struct bio
*__release_queued_bios_after_merge(struct dm_snapshot
*s
)
790 s
->first_merging_chunk
= 0;
791 s
->num_merging_chunks
= 0;
793 return bio_list_get(&s
->bios_queued_during_merge
);
797 * Remove one chunk from the index of completed exceptions.
799 static int __remove_single_exception_chunk(struct dm_snapshot
*s
,
802 struct dm_exception
*e
;
804 e
= dm_lookup_exception(&s
->complete
, old_chunk
);
806 DMERR("Corruption detected: exception for block %llu is "
807 "on disk but not in memory",
808 (unsigned long long)old_chunk
);
813 * If this is the only chunk using this exception, remove exception.
815 if (!dm_consecutive_chunk_count(e
)) {
816 dm_remove_exception(e
);
817 free_completed_exception(e
);
822 * The chunk may be either at the beginning or the end of a
823 * group of consecutive chunks - never in the middle. We are
824 * removing chunks in the opposite order to that in which they
825 * were added, so this should always be true.
826 * Decrement the consecutive chunk counter and adjust the
827 * starting point if necessary.
829 if (old_chunk
== e
->old_chunk
) {
832 } else if (old_chunk
!= e
->old_chunk
+
833 dm_consecutive_chunk_count(e
)) {
834 DMERR("Attempt to merge block %llu from the "
835 "middle of a chunk range [%llu - %llu]",
836 (unsigned long long)old_chunk
,
837 (unsigned long long)e
->old_chunk
,
839 e
->old_chunk
+ dm_consecutive_chunk_count(e
));
843 dm_consecutive_chunk_count_dec(e
);
848 static void flush_bios(struct bio
*bio
);
850 static int remove_single_exception_chunk(struct dm_snapshot
*s
)
852 struct bio
*b
= NULL
;
854 chunk_t old_chunk
= s
->first_merging_chunk
+ s
->num_merging_chunks
- 1;
856 down_write(&s
->lock
);
859 * Process chunks (and associated exceptions) in reverse order
860 * so that dm_consecutive_chunk_count_dec() accounting works.
863 r
= __remove_single_exception_chunk(s
, old_chunk
);
866 } while (old_chunk
-- > s
->first_merging_chunk
);
868 b
= __release_queued_bios_after_merge(s
);
878 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
879 sector_t sector
, unsigned chunk_size
);
881 static void merge_callback(int read_err
, unsigned long write_err
,
884 static uint64_t read_pending_exceptions_done_count(void)
886 uint64_t pending_exceptions_done
;
888 spin_lock(&_pending_exceptions_done_spinlock
);
889 pending_exceptions_done
= _pending_exceptions_done_count
;
890 spin_unlock(&_pending_exceptions_done_spinlock
);
892 return pending_exceptions_done
;
895 static void increment_pending_exceptions_done_count(void)
897 spin_lock(&_pending_exceptions_done_spinlock
);
898 _pending_exceptions_done_count
++;
899 spin_unlock(&_pending_exceptions_done_spinlock
);
901 wake_up_all(&_pending_exceptions_done
);
904 static void snapshot_merge_next_chunks(struct dm_snapshot
*s
)
906 int i
, linear_chunks
;
907 chunk_t old_chunk
, new_chunk
;
908 struct dm_io_region src
, dest
;
910 uint64_t previous_count
;
912 BUG_ON(!test_bit(RUNNING_MERGE
, &s
->state_bits
));
913 if (unlikely(test_bit(SHUTDOWN_MERGE
, &s
->state_bits
)))
917 * valid flag never changes during merge, so no lock required.
920 DMERR("Snapshot is invalid: can't merge");
924 linear_chunks
= s
->store
->type
->prepare_merge(s
->store
, &old_chunk
,
926 if (linear_chunks
<= 0) {
927 if (linear_chunks
< 0) {
928 DMERR("Read error in exception store: "
929 "shutting down merge");
930 down_write(&s
->lock
);
937 /* Adjust old_chunk and new_chunk to reflect start of linear region */
938 old_chunk
= old_chunk
+ 1 - linear_chunks
;
939 new_chunk
= new_chunk
+ 1 - linear_chunks
;
942 * Use one (potentially large) I/O to copy all 'linear_chunks'
943 * from the exception store to the origin
945 io_size
= linear_chunks
* s
->store
->chunk_size
;
947 dest
.bdev
= s
->origin
->bdev
;
948 dest
.sector
= chunk_to_sector(s
->store
, old_chunk
);
949 dest
.count
= min(io_size
, get_dev_size(dest
.bdev
) - dest
.sector
);
951 src
.bdev
= s
->cow
->bdev
;
952 src
.sector
= chunk_to_sector(s
->store
, new_chunk
);
953 src
.count
= dest
.count
;
956 * Reallocate any exceptions needed in other snapshots then
957 * wait for the pending exceptions to complete.
958 * Each time any pending exception (globally on the system)
959 * completes we are woken and repeat the process to find out
960 * if we can proceed. While this may not seem a particularly
961 * efficient algorithm, it is not expected to have any
962 * significant impact on performance.
964 previous_count
= read_pending_exceptions_done_count();
965 while (origin_write_extent(s
, dest
.sector
, io_size
)) {
966 wait_event(_pending_exceptions_done
,
967 (read_pending_exceptions_done_count() !=
969 /* Retry after the wait, until all exceptions are done. */
970 previous_count
= read_pending_exceptions_done_count();
973 down_write(&s
->lock
);
974 s
->first_merging_chunk
= old_chunk
;
975 s
->num_merging_chunks
= linear_chunks
;
978 /* Wait until writes to all 'linear_chunks' drain */
979 for (i
= 0; i
< linear_chunks
; i
++)
980 __check_for_conflicting_io(s
, old_chunk
+ i
);
982 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, merge_callback
, s
);
989 static void error_bios(struct bio
*bio
);
991 static void merge_callback(int read_err
, unsigned long write_err
, void *context
)
993 struct dm_snapshot
*s
= context
;
994 struct bio
*b
= NULL
;
996 if (read_err
|| write_err
) {
998 DMERR("Read error: shutting down merge.");
1000 DMERR("Write error: shutting down merge.");
1004 if (s
->store
->type
->commit_merge(s
->store
,
1005 s
->num_merging_chunks
) < 0) {
1006 DMERR("Write error in exception store: shutting down merge");
1010 if (remove_single_exception_chunk(s
) < 0)
1013 snapshot_merge_next_chunks(s
);
1018 down_write(&s
->lock
);
1019 s
->merge_failed
= 1;
1020 b
= __release_queued_bios_after_merge(s
);
1027 static void start_merge(struct dm_snapshot
*s
)
1029 if (!test_and_set_bit(RUNNING_MERGE
, &s
->state_bits
))
1030 snapshot_merge_next_chunks(s
);
1033 static int wait_schedule(void *ptr
)
1041 * Stop the merging process and wait until it finishes.
1043 static void stop_merge(struct dm_snapshot
*s
)
1045 set_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1046 wait_on_bit(&s
->state_bits
, RUNNING_MERGE
, wait_schedule
,
1047 TASK_UNINTERRUPTIBLE
);
1048 clear_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1052 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
1054 static int snapshot_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
1056 struct dm_snapshot
*s
;
1059 char *origin_path
, *cow_path
;
1060 unsigned args_used
, num_flush_requests
= 1;
1061 fmode_t origin_mode
= FMODE_READ
;
1064 ti
->error
= "requires exactly 4 arguments";
1069 if (dm_target_is_snapshot_merge(ti
)) {
1070 num_flush_requests
= 2;
1071 origin_mode
= FMODE_WRITE
;
1074 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
1076 ti
->error
= "Cannot allocate snapshot context private "
1082 origin_path
= argv
[0];
1086 r
= dm_get_device(ti
, origin_path
, origin_mode
, &s
->origin
);
1088 ti
->error
= "Cannot get origin device";
1096 r
= dm_get_device(ti
, cow_path
, FMODE_READ
| FMODE_WRITE
, &s
->cow
);
1098 ti
->error
= "Cannot get COW device";
1102 r
= dm_exception_store_create(ti
, argc
, argv
, s
, &args_used
, &s
->store
);
1104 ti
->error
= "Couldn't create exception store";
1116 atomic_set(&s
->pending_exceptions_count
, 0);
1117 init_rwsem(&s
->lock
);
1118 INIT_LIST_HEAD(&s
->list
);
1119 spin_lock_init(&s
->pe_lock
);
1121 s
->merge_failed
= 0;
1122 s
->first_merging_chunk
= 0;
1123 s
->num_merging_chunks
= 0;
1124 bio_list_init(&s
->bios_queued_during_merge
);
1126 /* Allocate hash table for COW data */
1127 if (init_hash_tables(s
)) {
1128 ti
->error
= "Unable to allocate hash table space";
1130 goto bad_hash_tables
;
1133 r
= dm_kcopyd_client_create(SNAPSHOT_PAGES
, &s
->kcopyd_client
);
1135 ti
->error
= "Could not create kcopyd client";
1139 s
->pending_pool
= mempool_create_slab_pool(MIN_IOS
, pending_cache
);
1140 if (!s
->pending_pool
) {
1141 ti
->error
= "Could not allocate mempool for pending exceptions";
1142 goto bad_pending_pool
;
1145 s
->tracked_chunk_pool
= mempool_create_slab_pool(MIN_IOS
,
1146 tracked_chunk_cache
);
1147 if (!s
->tracked_chunk_pool
) {
1148 ti
->error
= "Could not allocate tracked_chunk mempool for "
1150 goto bad_tracked_chunk_pool
;
1153 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1154 INIT_HLIST_HEAD(&s
->tracked_chunk_hash
[i
]);
1156 spin_lock_init(&s
->tracked_chunk_lock
);
1158 bio_list_init(&s
->queued_bios
);
1159 INIT_WORK(&s
->queued_bios_work
, flush_queued_bios
);
1162 ti
->num_flush_requests
= num_flush_requests
;
1164 /* Add snapshot to the list of snapshots for this origin */
1165 /* Exceptions aren't triggered till snapshot_resume() is called */
1166 r
= register_snapshot(s
);
1168 ti
->error
= "Snapshot origin struct allocation failed";
1169 goto bad_load_and_register
;
1171 /* invalid handover, register_snapshot has set ti->error */
1172 goto bad_load_and_register
;
1176 * Metadata must only be loaded into one table at once, so skip this
1177 * if metadata will be handed over during resume.
1178 * Chunk size will be set during the handover - set it to zero to
1179 * ensure it's ignored.
1182 s
->store
->chunk_size
= 0;
1186 r
= s
->store
->type
->read_metadata(s
->store
, dm_add_exception
,
1189 ti
->error
= "Failed to read snapshot metadata";
1190 goto bad_read_metadata
;
1193 DMWARN("Snapshot is marked invalid.");
1196 if (!s
->store
->chunk_size
) {
1197 ti
->error
= "Chunk size not set";
1198 goto bad_read_metadata
;
1200 ti
->split_io
= s
->store
->chunk_size
;
1205 unregister_snapshot(s
);
1207 bad_load_and_register
:
1208 mempool_destroy(s
->tracked_chunk_pool
);
1210 bad_tracked_chunk_pool
:
1211 mempool_destroy(s
->pending_pool
);
1214 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1217 dm_exception_table_exit(&s
->pending
, pending_cache
);
1218 dm_exception_table_exit(&s
->complete
, exception_cache
);
1221 dm_exception_store_destroy(s
->store
);
1224 dm_put_device(ti
, s
->cow
);
1227 dm_put_device(ti
, s
->origin
);
1236 static void __free_exceptions(struct dm_snapshot
*s
)
1238 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1239 s
->kcopyd_client
= NULL
;
1241 dm_exception_table_exit(&s
->pending
, pending_cache
);
1242 dm_exception_table_exit(&s
->complete
, exception_cache
);
1245 static void __handover_exceptions(struct dm_snapshot
*snap_src
,
1246 struct dm_snapshot
*snap_dest
)
1249 struct dm_exception_table table_swap
;
1250 struct dm_exception_store
*store_swap
;
1254 * Swap all snapshot context information between the two instances.
1256 u
.table_swap
= snap_dest
->complete
;
1257 snap_dest
->complete
= snap_src
->complete
;
1258 snap_src
->complete
= u
.table_swap
;
1260 u
.store_swap
= snap_dest
->store
;
1261 snap_dest
->store
= snap_src
->store
;
1262 snap_src
->store
= u
.store_swap
;
1264 snap_dest
->store
->snap
= snap_dest
;
1265 snap_src
->store
->snap
= snap_src
;
1267 snap_dest
->ti
->split_io
= snap_dest
->store
->chunk_size
;
1268 snap_dest
->valid
= snap_src
->valid
;
1271 * Set source invalid to ensure it receives no further I/O.
1273 snap_src
->valid
= 0;
1276 static void snapshot_dtr(struct dm_target
*ti
)
1278 #ifdef CONFIG_DM_DEBUG
1281 struct dm_snapshot
*s
= ti
->private;
1282 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1284 flush_workqueue(ksnapd
);
1286 down_read(&_origins_lock
);
1287 /* Check whether exception handover must be cancelled */
1288 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1289 if (snap_src
&& snap_dest
&& (s
== snap_src
)) {
1290 down_write(&snap_dest
->lock
);
1291 snap_dest
->valid
= 0;
1292 up_write(&snap_dest
->lock
);
1293 DMERR("Cancelling snapshot handover.");
1295 up_read(&_origins_lock
);
1297 if (dm_target_is_snapshot_merge(ti
))
1300 /* Prevent further origin writes from using this snapshot. */
1301 /* After this returns there can be no new kcopyd jobs. */
1302 unregister_snapshot(s
);
1304 while (atomic_read(&s
->pending_exceptions_count
))
1307 * Ensure instructions in mempool_destroy aren't reordered
1308 * before atomic_read.
1312 #ifdef CONFIG_DM_DEBUG
1313 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1314 BUG_ON(!hlist_empty(&s
->tracked_chunk_hash
[i
]));
1317 mempool_destroy(s
->tracked_chunk_pool
);
1319 __free_exceptions(s
);
1321 mempool_destroy(s
->pending_pool
);
1323 dm_exception_store_destroy(s
->store
);
1325 dm_put_device(ti
, s
->cow
);
1327 dm_put_device(ti
, s
->origin
);
1333 * Flush a list of buffers.
1335 static void flush_bios(struct bio
*bio
)
1341 bio
->bi_next
= NULL
;
1342 generic_make_request(bio
);
1347 static void flush_queued_bios(struct work_struct
*work
)
1349 struct dm_snapshot
*s
=
1350 container_of(work
, struct dm_snapshot
, queued_bios_work
);
1351 struct bio
*queued_bios
;
1352 unsigned long flags
;
1354 spin_lock_irqsave(&s
->pe_lock
, flags
);
1355 queued_bios
= bio_list_get(&s
->queued_bios
);
1356 spin_unlock_irqrestore(&s
->pe_lock
, flags
);
1358 flush_bios(queued_bios
);
1361 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
);
1364 * Flush a list of buffers.
1366 static void retry_origin_bios(struct dm_snapshot
*s
, struct bio
*bio
)
1373 bio
->bi_next
= NULL
;
1374 r
= do_origin(s
->origin
, bio
);
1375 if (r
== DM_MAPIO_REMAPPED
)
1376 generic_make_request(bio
);
1382 * Error a list of buffers.
1384 static void error_bios(struct bio
*bio
)
1390 bio
->bi_next
= NULL
;
1396 static void __invalidate_snapshot(struct dm_snapshot
*s
, int err
)
1402 DMERR("Invalidating snapshot: Error reading/writing.");
1403 else if (err
== -ENOMEM
)
1404 DMERR("Invalidating snapshot: Unable to allocate exception.");
1406 if (s
->store
->type
->drop_snapshot
)
1407 s
->store
->type
->drop_snapshot(s
->store
);
1411 dm_table_event(s
->ti
->table
);
1414 static void pending_complete(struct dm_snap_pending_exception
*pe
, int success
)
1416 struct dm_exception
*e
;
1417 struct dm_snapshot
*s
= pe
->snap
;
1418 struct bio
*origin_bios
= NULL
;
1419 struct bio
*snapshot_bios
= NULL
;
1423 /* Read/write error - snapshot is unusable */
1424 down_write(&s
->lock
);
1425 __invalidate_snapshot(s
, -EIO
);
1430 e
= alloc_completed_exception();
1432 down_write(&s
->lock
);
1433 __invalidate_snapshot(s
, -ENOMEM
);
1439 down_write(&s
->lock
);
1441 free_completed_exception(e
);
1446 /* Check for conflicting reads */
1447 __check_for_conflicting_io(s
, pe
->e
.old_chunk
);
1450 * Add a proper exception, and remove the
1451 * in-flight exception from the list.
1453 dm_insert_exception(&s
->complete
, e
);
1456 dm_remove_exception(&pe
->e
);
1457 snapshot_bios
= bio_list_get(&pe
->snapshot_bios
);
1458 origin_bios
= bio_list_get(&pe
->origin_bios
);
1459 free_pending_exception(pe
);
1461 increment_pending_exceptions_done_count();
1465 /* Submit any pending write bios */
1467 error_bios(snapshot_bios
);
1469 flush_bios(snapshot_bios
);
1471 retry_origin_bios(s
, origin_bios
);
1474 static void commit_callback(void *context
, int success
)
1476 struct dm_snap_pending_exception
*pe
= context
;
1478 pending_complete(pe
, success
);
1482 * Called when the copy I/O has finished. kcopyd actually runs
1483 * this code so don't block.
1485 static void copy_callback(int read_err
, unsigned long write_err
, void *context
)
1487 struct dm_snap_pending_exception
*pe
= context
;
1488 struct dm_snapshot
*s
= pe
->snap
;
1490 if (read_err
|| write_err
)
1491 pending_complete(pe
, 0);
1494 /* Update the metadata if we are persistent */
1495 s
->store
->type
->commit_exception(s
->store
, &pe
->e
,
1496 commit_callback
, pe
);
1500 * Dispatches the copy operation to kcopyd.
1502 static void start_copy(struct dm_snap_pending_exception
*pe
)
1504 struct dm_snapshot
*s
= pe
->snap
;
1505 struct dm_io_region src
, dest
;
1506 struct block_device
*bdev
= s
->origin
->bdev
;
1509 dev_size
= get_dev_size(bdev
);
1512 src
.sector
= chunk_to_sector(s
->store
, pe
->e
.old_chunk
);
1513 src
.count
= min((sector_t
)s
->store
->chunk_size
, dev_size
- src
.sector
);
1515 dest
.bdev
= s
->cow
->bdev
;
1516 dest
.sector
= chunk_to_sector(s
->store
, pe
->e
.new_chunk
);
1517 dest
.count
= src
.count
;
1519 /* Hand over to kcopyd */
1520 dm_kcopyd_copy(s
->kcopyd_client
,
1521 &src
, 1, &dest
, 0, copy_callback
, pe
);
1524 static struct dm_snap_pending_exception
*
1525 __lookup_pending_exception(struct dm_snapshot
*s
, chunk_t chunk
)
1527 struct dm_exception
*e
= dm_lookup_exception(&s
->pending
, chunk
);
1532 return container_of(e
, struct dm_snap_pending_exception
, e
);
1536 * Looks to see if this snapshot already has a pending exception
1537 * for this chunk, otherwise it allocates a new one and inserts
1538 * it into the pending table.
1540 * NOTE: a write lock must be held on snap->lock before calling
1543 static struct dm_snap_pending_exception
*
1544 __find_pending_exception(struct dm_snapshot
*s
,
1545 struct dm_snap_pending_exception
*pe
, chunk_t chunk
)
1547 struct dm_snap_pending_exception
*pe2
;
1549 pe2
= __lookup_pending_exception(s
, chunk
);
1551 free_pending_exception(pe
);
1555 pe
->e
.old_chunk
= chunk
;
1556 bio_list_init(&pe
->origin_bios
);
1557 bio_list_init(&pe
->snapshot_bios
);
1560 if (s
->store
->type
->prepare_exception(s
->store
, &pe
->e
)) {
1561 free_pending_exception(pe
);
1565 dm_insert_exception(&s
->pending
, &pe
->e
);
1570 static void remap_exception(struct dm_snapshot
*s
, struct dm_exception
*e
,
1571 struct bio
*bio
, chunk_t chunk
)
1573 bio
->bi_bdev
= s
->cow
->bdev
;
1574 bio
->bi_sector
= chunk_to_sector(s
->store
,
1575 dm_chunk_number(e
->new_chunk
) +
1576 (chunk
- e
->old_chunk
)) +
1578 s
->store
->chunk_mask
);
1581 static int snapshot_map(struct dm_target
*ti
, struct bio
*bio
,
1582 union map_info
*map_context
)
1584 struct dm_exception
*e
;
1585 struct dm_snapshot
*s
= ti
->private;
1586 int r
= DM_MAPIO_REMAPPED
;
1588 struct dm_snap_pending_exception
*pe
= NULL
;
1590 if (unlikely(bio_empty_barrier(bio
))) {
1591 bio
->bi_bdev
= s
->cow
->bdev
;
1592 return DM_MAPIO_REMAPPED
;
1595 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1597 /* Full snapshots are not usable */
1598 /* To get here the table must be live so s->active is always set. */
1602 /* FIXME: should only take write lock if we need
1603 * to copy an exception */
1604 down_write(&s
->lock
);
1611 /* If the block is already remapped - use that, else remap it */
1612 e
= dm_lookup_exception(&s
->complete
, chunk
);
1614 remap_exception(s
, e
, bio
, chunk
);
1619 * Write to snapshot - higher level takes care of RW/RO
1620 * flags so we should only get this if we are
1623 if (bio_rw(bio
) == WRITE
) {
1624 pe
= __lookup_pending_exception(s
, chunk
);
1627 pe
= alloc_pending_exception(s
);
1628 down_write(&s
->lock
);
1631 free_pending_exception(pe
);
1636 e
= dm_lookup_exception(&s
->complete
, chunk
);
1638 free_pending_exception(pe
);
1639 remap_exception(s
, e
, bio
, chunk
);
1643 pe
= __find_pending_exception(s
, pe
, chunk
);
1645 __invalidate_snapshot(s
, -ENOMEM
);
1651 remap_exception(s
, &pe
->e
, bio
, chunk
);
1652 bio_list_add(&pe
->snapshot_bios
, bio
);
1654 r
= DM_MAPIO_SUBMITTED
;
1657 /* this is protected by snap->lock */
1664 bio
->bi_bdev
= s
->origin
->bdev
;
1665 map_context
->ptr
= track_chunk(s
, chunk
);
1675 * A snapshot-merge target behaves like a combination of a snapshot
1676 * target and a snapshot-origin target. It only generates new
1677 * exceptions in other snapshots and not in the one that is being
1680 * For each chunk, if there is an existing exception, it is used to
1681 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1682 * which in turn might generate exceptions in other snapshots.
1683 * If merging is currently taking place on the chunk in question, the
1684 * I/O is deferred by adding it to s->bios_queued_during_merge.
1686 static int snapshot_merge_map(struct dm_target
*ti
, struct bio
*bio
,
1687 union map_info
*map_context
)
1689 struct dm_exception
*e
;
1690 struct dm_snapshot
*s
= ti
->private;
1691 int r
= DM_MAPIO_REMAPPED
;
1694 if (unlikely(bio_empty_barrier(bio
))) {
1695 if (!map_context
->target_request_nr
)
1696 bio
->bi_bdev
= s
->origin
->bdev
;
1698 bio
->bi_bdev
= s
->cow
->bdev
;
1699 map_context
->ptr
= NULL
;
1700 return DM_MAPIO_REMAPPED
;
1703 chunk
= sector_to_chunk(s
->store
, bio
->bi_sector
);
1705 down_write(&s
->lock
);
1707 /* Full merging snapshots are redirected to the origin */
1709 goto redirect_to_origin
;
1711 /* If the block is already remapped - use that */
1712 e
= dm_lookup_exception(&s
->complete
, chunk
);
1714 /* Queue writes overlapping with chunks being merged */
1715 if (bio_rw(bio
) == WRITE
&&
1716 chunk
>= s
->first_merging_chunk
&&
1717 chunk
< (s
->first_merging_chunk
+
1718 s
->num_merging_chunks
)) {
1719 bio
->bi_bdev
= s
->origin
->bdev
;
1720 bio_list_add(&s
->bios_queued_during_merge
, bio
);
1721 r
= DM_MAPIO_SUBMITTED
;
1725 remap_exception(s
, e
, bio
, chunk
);
1727 if (bio_rw(bio
) == WRITE
)
1728 map_context
->ptr
= track_chunk(s
, chunk
);
1733 bio
->bi_bdev
= s
->origin
->bdev
;
1735 if (bio_rw(bio
) == WRITE
) {
1737 return do_origin(s
->origin
, bio
);
1746 static int snapshot_end_io(struct dm_target
*ti
, struct bio
*bio
,
1747 int error
, union map_info
*map_context
)
1749 struct dm_snapshot
*s
= ti
->private;
1750 struct dm_snap_tracked_chunk
*c
= map_context
->ptr
;
1753 stop_tracking_chunk(s
, c
);
1758 static void snapshot_merge_presuspend(struct dm_target
*ti
)
1760 struct dm_snapshot
*s
= ti
->private;
1765 static void snapshot_postsuspend(struct dm_target
*ti
)
1767 struct dm_snapshot
*s
= ti
->private;
1769 down_write(&s
->lock
);
1774 static int snapshot_preresume(struct dm_target
*ti
)
1777 struct dm_snapshot
*s
= ti
->private;
1778 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1780 down_read(&_origins_lock
);
1781 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1782 if (snap_src
&& snap_dest
) {
1783 down_read(&snap_src
->lock
);
1784 if (s
== snap_src
) {
1785 DMERR("Unable to resume snapshot source until "
1786 "handover completes.");
1788 } else if (!snap_src
->suspended
) {
1789 DMERR("Unable to perform snapshot handover until "
1790 "source is suspended.");
1793 up_read(&snap_src
->lock
);
1795 up_read(&_origins_lock
);
1800 static void snapshot_resume(struct dm_target
*ti
)
1802 struct dm_snapshot
*s
= ti
->private;
1803 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1805 down_read(&_origins_lock
);
1806 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1807 if (snap_src
&& snap_dest
) {
1808 down_write(&snap_src
->lock
);
1809 down_write_nested(&snap_dest
->lock
, SINGLE_DEPTH_NESTING
);
1810 __handover_exceptions(snap_src
, snap_dest
);
1811 up_write(&snap_dest
->lock
);
1812 up_write(&snap_src
->lock
);
1814 up_read(&_origins_lock
);
1816 /* Now we have correct chunk size, reregister */
1817 reregister_snapshot(s
);
1819 down_write(&s
->lock
);
1825 static sector_t
get_origin_minimum_chunksize(struct block_device
*bdev
)
1827 sector_t min_chunksize
;
1829 down_read(&_origins_lock
);
1830 min_chunksize
= __minimum_chunk_size(__lookup_origin(bdev
));
1831 up_read(&_origins_lock
);
1833 return min_chunksize
;
1836 static void snapshot_merge_resume(struct dm_target
*ti
)
1838 struct dm_snapshot
*s
= ti
->private;
1841 * Handover exceptions from existing snapshot.
1843 snapshot_resume(ti
);
1846 * snapshot-merge acts as an origin, so set ti->split_io
1848 ti
->split_io
= get_origin_minimum_chunksize(s
->origin
->bdev
);
1853 static int snapshot_status(struct dm_target
*ti
, status_type_t type
,
1854 char *result
, unsigned int maxlen
)
1857 struct dm_snapshot
*snap
= ti
->private;
1860 case STATUSTYPE_INFO
:
1862 down_write(&snap
->lock
);
1866 else if (snap
->merge_failed
)
1867 DMEMIT("Merge failed");
1869 if (snap
->store
->type
->usage
) {
1870 sector_t total_sectors
, sectors_allocated
,
1872 snap
->store
->type
->usage(snap
->store
,
1876 DMEMIT("%llu/%llu %llu",
1877 (unsigned long long)sectors_allocated
,
1878 (unsigned long long)total_sectors
,
1879 (unsigned long long)metadata_sectors
);
1885 up_write(&snap
->lock
);
1889 case STATUSTYPE_TABLE
:
1891 * kdevname returns a static pointer so we need
1892 * to make private copies if the output is to
1895 DMEMIT("%s %s", snap
->origin
->name
, snap
->cow
->name
);
1896 snap
->store
->type
->status(snap
->store
, type
, result
+ sz
,
1904 static int snapshot_iterate_devices(struct dm_target
*ti
,
1905 iterate_devices_callout_fn fn
, void *data
)
1907 struct dm_snapshot
*snap
= ti
->private;
1910 r
= fn(ti
, snap
->origin
, 0, ti
->len
, data
);
1913 r
= fn(ti
, snap
->cow
, 0, get_dev_size(snap
->cow
->bdev
), data
);
1919 /*-----------------------------------------------------------------
1921 *---------------------------------------------------------------*/
1924 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
1925 * supplied bio was ignored. The caller may submit it immediately.
1926 * (No remapping actually occurs as the origin is always a direct linear
1929 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
1930 * and any supplied bio is added to a list to be submitted once all
1931 * the necessary exceptions exist.
1933 static int __origin_write(struct list_head
*snapshots
, sector_t sector
,
1936 int r
= DM_MAPIO_REMAPPED
;
1937 struct dm_snapshot
*snap
;
1938 struct dm_exception
*e
;
1939 struct dm_snap_pending_exception
*pe
;
1940 struct dm_snap_pending_exception
*pe_to_start_now
= NULL
;
1941 struct dm_snap_pending_exception
*pe_to_start_last
= NULL
;
1944 /* Do all the snapshots on this origin */
1945 list_for_each_entry (snap
, snapshots
, list
) {
1947 * Don't make new exceptions in a merging snapshot
1948 * because it has effectively been deleted
1950 if (dm_target_is_snapshot_merge(snap
->ti
))
1953 down_write(&snap
->lock
);
1955 /* Only deal with valid and active snapshots */
1956 if (!snap
->valid
|| !snap
->active
)
1959 /* Nothing to do if writing beyond end of snapshot */
1960 if (sector
>= dm_table_get_size(snap
->ti
->table
))
1964 * Remember, different snapshots can have
1965 * different chunk sizes.
1967 chunk
= sector_to_chunk(snap
->store
, sector
);
1970 * Check exception table to see if block
1971 * is already remapped in this snapshot
1972 * and trigger an exception if not.
1974 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1978 pe
= __lookup_pending_exception(snap
, chunk
);
1980 up_write(&snap
->lock
);
1981 pe
= alloc_pending_exception(snap
);
1982 down_write(&snap
->lock
);
1985 free_pending_exception(pe
);
1989 e
= dm_lookup_exception(&snap
->complete
, chunk
);
1991 free_pending_exception(pe
);
1995 pe
= __find_pending_exception(snap
, pe
, chunk
);
1997 __invalidate_snapshot(snap
, -ENOMEM
);
2002 r
= DM_MAPIO_SUBMITTED
;
2005 * If an origin bio was supplied, queue it to wait for the
2006 * completion of this exception, and start this one last,
2007 * at the end of the function.
2010 bio_list_add(&pe
->origin_bios
, bio
);
2015 pe_to_start_last
= pe
;
2021 pe_to_start_now
= pe
;
2025 up_write(&snap
->lock
);
2027 if (pe_to_start_now
) {
2028 start_copy(pe_to_start_now
);
2029 pe_to_start_now
= NULL
;
2034 * Submit the exception against which the bio is queued last,
2035 * to give the other exceptions a head start.
2037 if (pe_to_start_last
)
2038 start_copy(pe_to_start_last
);
2044 * Called on a write from the origin driver.
2046 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
)
2049 int r
= DM_MAPIO_REMAPPED
;
2051 down_read(&_origins_lock
);
2052 o
= __lookup_origin(origin
->bdev
);
2054 r
= __origin_write(&o
->snapshots
, bio
->bi_sector
, bio
);
2055 up_read(&_origins_lock
);
2061 * Trigger exceptions in all non-merging snapshots.
2063 * The chunk size of the merging snapshot may be larger than the chunk
2064 * size of some other snapshot so we may need to reallocate multiple
2065 * chunks in other snapshots.
2067 * We scan all the overlapping exceptions in the other snapshots.
2068 * Returns 1 if anything was reallocated and must be waited for,
2069 * otherwise returns 0.
2071 * size must be a multiple of merging_snap's chunk_size.
2073 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
2074 sector_t sector
, unsigned size
)
2081 * The origin's __minimum_chunk_size() got stored in split_io
2082 * by snapshot_merge_resume().
2084 down_read(&_origins_lock
);
2085 o
= __lookup_origin(merging_snap
->origin
->bdev
);
2086 for (n
= 0; n
< size
; n
+= merging_snap
->ti
->split_io
)
2087 if (__origin_write(&o
->snapshots
, sector
+ n
, NULL
) ==
2090 up_read(&_origins_lock
);
2096 * Origin: maps a linear range of a device, with hooks for snapshotting.
2100 * Construct an origin mapping: <dev_path>
2101 * The context for an origin is merely a 'struct dm_dev *'
2102 * pointing to the real device.
2104 static int origin_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
2110 ti
->error
= "origin: incorrect number of arguments";
2114 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &dev
);
2116 ti
->error
= "Cannot get target device";
2121 ti
->num_flush_requests
= 1;
2126 static void origin_dtr(struct dm_target
*ti
)
2128 struct dm_dev
*dev
= ti
->private;
2129 dm_put_device(ti
, dev
);
2132 static int origin_map(struct dm_target
*ti
, struct bio
*bio
,
2133 union map_info
*map_context
)
2135 struct dm_dev
*dev
= ti
->private;
2136 bio
->bi_bdev
= dev
->bdev
;
2138 if (unlikely(bio_empty_barrier(bio
)))
2139 return DM_MAPIO_REMAPPED
;
2141 /* Only tell snapshots if this is a write */
2142 return (bio_rw(bio
) == WRITE
) ? do_origin(dev
, bio
) : DM_MAPIO_REMAPPED
;
2146 * Set the target "split_io" field to the minimum of all the snapshots'
2149 static void origin_resume(struct dm_target
*ti
)
2151 struct dm_dev
*dev
= ti
->private;
2153 ti
->split_io
= get_origin_minimum_chunksize(dev
->bdev
);
2156 static int origin_status(struct dm_target
*ti
, status_type_t type
, char *result
,
2157 unsigned int maxlen
)
2159 struct dm_dev
*dev
= ti
->private;
2162 case STATUSTYPE_INFO
:
2166 case STATUSTYPE_TABLE
:
2167 snprintf(result
, maxlen
, "%s", dev
->name
);
2174 static int origin_merge(struct dm_target
*ti
, struct bvec_merge_data
*bvm
,
2175 struct bio_vec
*biovec
, int max_size
)
2177 struct dm_dev
*dev
= ti
->private;
2178 struct request_queue
*q
= bdev_get_queue(dev
->bdev
);
2180 if (!q
->merge_bvec_fn
)
2183 bvm
->bi_bdev
= dev
->bdev
;
2184 bvm
->bi_sector
= bvm
->bi_sector
;
2186 return min(max_size
, q
->merge_bvec_fn(q
, bvm
, biovec
));
2189 static int origin_iterate_devices(struct dm_target
*ti
,
2190 iterate_devices_callout_fn fn
, void *data
)
2192 struct dm_dev
*dev
= ti
->private;
2194 return fn(ti
, dev
, 0, ti
->len
, data
);
2197 static struct target_type origin_target
= {
2198 .name
= "snapshot-origin",
2199 .version
= {1, 7, 0},
2200 .module
= THIS_MODULE
,
2204 .resume
= origin_resume
,
2205 .status
= origin_status
,
2206 .merge
= origin_merge
,
2207 .iterate_devices
= origin_iterate_devices
,
2210 static struct target_type snapshot_target
= {
2212 .version
= {1, 9, 0},
2213 .module
= THIS_MODULE
,
2214 .ctr
= snapshot_ctr
,
2215 .dtr
= snapshot_dtr
,
2216 .map
= snapshot_map
,
2217 .end_io
= snapshot_end_io
,
2218 .postsuspend
= snapshot_postsuspend
,
2219 .preresume
= snapshot_preresume
,
2220 .resume
= snapshot_resume
,
2221 .status
= snapshot_status
,
2222 .iterate_devices
= snapshot_iterate_devices
,
2225 static struct target_type merge_target
= {
2226 .name
= dm_snapshot_merge_target_name
,
2227 .version
= {1, 0, 0},
2228 .module
= THIS_MODULE
,
2229 .ctr
= snapshot_ctr
,
2230 .dtr
= snapshot_dtr
,
2231 .map
= snapshot_merge_map
,
2232 .end_io
= snapshot_end_io
,
2233 .presuspend
= snapshot_merge_presuspend
,
2234 .postsuspend
= snapshot_postsuspend
,
2235 .preresume
= snapshot_preresume
,
2236 .resume
= snapshot_merge_resume
,
2237 .status
= snapshot_status
,
2238 .iterate_devices
= snapshot_iterate_devices
,
2241 static int __init
dm_snapshot_init(void)
2245 r
= dm_exception_store_init();
2247 DMERR("Failed to initialize exception stores");
2251 r
= dm_register_target(&snapshot_target
);
2253 DMERR("snapshot target register failed %d", r
);
2254 goto bad_register_snapshot_target
;
2257 r
= dm_register_target(&origin_target
);
2259 DMERR("Origin target register failed %d", r
);
2260 goto bad_register_origin_target
;
2263 r
= dm_register_target(&merge_target
);
2265 DMERR("Merge target register failed %d", r
);
2266 goto bad_register_merge_target
;
2269 r
= init_origin_hash();
2271 DMERR("init_origin_hash failed.");
2272 goto bad_origin_hash
;
2275 exception_cache
= KMEM_CACHE(dm_exception
, 0);
2276 if (!exception_cache
) {
2277 DMERR("Couldn't create exception cache.");
2279 goto bad_exception_cache
;
2282 pending_cache
= KMEM_CACHE(dm_snap_pending_exception
, 0);
2283 if (!pending_cache
) {
2284 DMERR("Couldn't create pending cache.");
2286 goto bad_pending_cache
;
2289 tracked_chunk_cache
= KMEM_CACHE(dm_snap_tracked_chunk
, 0);
2290 if (!tracked_chunk_cache
) {
2291 DMERR("Couldn't create cache to track chunks in use.");
2293 goto bad_tracked_chunk_cache
;
2296 ksnapd
= create_singlethread_workqueue("ksnapd");
2298 DMERR("Failed to create ksnapd workqueue.");
2300 goto bad_pending_pool
;
2306 kmem_cache_destroy(tracked_chunk_cache
);
2307 bad_tracked_chunk_cache
:
2308 kmem_cache_destroy(pending_cache
);
2310 kmem_cache_destroy(exception_cache
);
2311 bad_exception_cache
:
2314 dm_unregister_target(&merge_target
);
2315 bad_register_merge_target
:
2316 dm_unregister_target(&origin_target
);
2317 bad_register_origin_target
:
2318 dm_unregister_target(&snapshot_target
);
2319 bad_register_snapshot_target
:
2320 dm_exception_store_exit();
2325 static void __exit
dm_snapshot_exit(void)
2327 destroy_workqueue(ksnapd
);
2329 dm_unregister_target(&snapshot_target
);
2330 dm_unregister_target(&origin_target
);
2331 dm_unregister_target(&merge_target
);
2334 kmem_cache_destroy(pending_cache
);
2335 kmem_cache_destroy(exception_cache
);
2336 kmem_cache_destroy(tracked_chunk_cache
);
2338 dm_exception_store_exit();
2342 module_init(dm_snapshot_init
);
2343 module_exit(dm_snapshot_exit
);
2345 MODULE_DESCRIPTION(DM_NAME
" snapshot target");
2346 MODULE_AUTHOR("Joe Thornber");
2347 MODULE_LICENSE("GPL");