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/semaphore.h>
26 #include "dm-exception-store.h"
28 #define DM_MSG_PREFIX "snapshots"
30 static const char dm_snapshot_merge_target_name
[] = "snapshot-merge";
32 #define dm_target_is_snapshot_merge(ti) \
33 ((ti)->type->name == dm_snapshot_merge_target_name)
36 * The size of the mempool used to track chunks in use.
40 #define DM_TRACKED_CHUNK_HASH_SIZE 16
41 #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \
42 (DM_TRACKED_CHUNK_HASH_SIZE - 1))
44 struct dm_exception_table
{
47 struct list_head
*table
;
53 struct dm_dev
*origin
;
58 /* List of snapshots per Origin */
59 struct list_head list
;
62 * You can't use a snapshot if this is 0 (e.g. if full).
63 * A snapshot-merge target never clears this.
68 * The snapshot overflowed because of a write to the snapshot device.
69 * We don't have to invalidate the snapshot in this case, but we need
70 * to prevent further writes.
72 int snapshot_overflowed
;
74 /* Origin writes don't trigger exceptions until this is set */
77 atomic_t pending_exceptions_count
;
79 /* Protected by "lock" */
80 sector_t exception_start_sequence
;
82 /* Protected by kcopyd single-threaded callback */
83 sector_t exception_complete_sequence
;
86 * A list of pending exceptions that completed out of order.
87 * Protected by kcopyd single-threaded callback.
89 struct rb_root out_of_order_tree
;
91 mempool_t pending_pool
;
93 struct dm_exception_table pending
;
94 struct dm_exception_table complete
;
97 * pe_lock protects all pending_exception operations and access
98 * as well as the snapshot_bios list.
102 /* Chunks with outstanding reads */
103 spinlock_t tracked_chunk_lock
;
104 struct hlist_head tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH_SIZE
];
106 /* The on disk metadata handler */
107 struct dm_exception_store
*store
;
109 /* Maximum number of in-flight COW jobs. */
110 struct semaphore cow_count
;
112 struct dm_kcopyd_client
*kcopyd_client
;
114 /* Wait for events based on state_bits */
115 unsigned long state_bits
;
117 /* Range of chunks currently being merged. */
118 chunk_t first_merging_chunk
;
119 int num_merging_chunks
;
122 * The merge operation failed if this flag is set.
123 * Failure modes are handled as follows:
124 * - I/O error reading the header
125 * => don't load the target; abort.
126 * - Header does not have "valid" flag set
127 * => use the origin; forget about the snapshot.
128 * - I/O error when reading exceptions
129 * => don't load the target; abort.
130 * (We can't use the intermediate origin state.)
131 * - I/O error while merging
132 * => stop merging; set merge_failed; process I/O normally.
137 * Incoming bios that overlap with chunks being merged must wait
138 * for them to be committed.
140 struct bio_list bios_queued_during_merge
;
145 * RUNNING_MERGE - Merge operation is in progress.
146 * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped;
147 * cleared afterwards.
149 #define RUNNING_MERGE 0
150 #define SHUTDOWN_MERGE 1
153 * Maximum number of chunks being copied on write.
155 * The value was decided experimentally as a trade-off between memory
156 * consumption, stalling the kernel's workqueues and maintaining a high enough
159 #define DEFAULT_COW_THRESHOLD 2048
161 static int cow_threshold
= DEFAULT_COW_THRESHOLD
;
162 module_param_named(snapshot_cow_threshold
, cow_threshold
, int, 0644);
163 MODULE_PARM_DESC(snapshot_cow_threshold
, "Maximum number of chunks being copied on write");
165 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle
,
166 "A percentage of time allocated for copy on write");
168 struct dm_dev
*dm_snap_origin(struct dm_snapshot
*s
)
172 EXPORT_SYMBOL(dm_snap_origin
);
174 struct dm_dev
*dm_snap_cow(struct dm_snapshot
*s
)
178 EXPORT_SYMBOL(dm_snap_cow
);
180 static sector_t
chunk_to_sector(struct dm_exception_store
*store
,
183 return chunk
<< store
->chunk_shift
;
186 static int bdev_equal(struct block_device
*lhs
, struct block_device
*rhs
)
189 * There is only ever one instance of a particular block
190 * device so we can compare pointers safely.
195 struct dm_snap_pending_exception
{
196 struct dm_exception e
;
199 * Origin buffers waiting for this to complete are held
202 struct bio_list origin_bios
;
203 struct bio_list snapshot_bios
;
205 /* Pointer back to snapshot context */
206 struct dm_snapshot
*snap
;
209 * 1 indicates the exception has already been sent to
214 /* There was copying error. */
217 /* A sequence number, it is used for in-order completion. */
218 sector_t exception_sequence
;
220 struct rb_node out_of_order_node
;
223 * For writing a complete chunk, bypassing the copy.
225 struct bio
*full_bio
;
226 bio_end_io_t
*full_bio_end_io
;
230 * Hash table mapping origin volumes to lists of snapshots and
231 * a lock to protect it
233 static struct kmem_cache
*exception_cache
;
234 static struct kmem_cache
*pending_cache
;
236 struct dm_snap_tracked_chunk
{
237 struct hlist_node node
;
241 static void init_tracked_chunk(struct bio
*bio
)
243 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
244 INIT_HLIST_NODE(&c
->node
);
247 static bool is_bio_tracked(struct bio
*bio
)
249 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
250 return !hlist_unhashed(&c
->node
);
253 static void track_chunk(struct dm_snapshot
*s
, struct bio
*bio
, chunk_t chunk
)
255 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
259 spin_lock_irq(&s
->tracked_chunk_lock
);
260 hlist_add_head(&c
->node
,
261 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)]);
262 spin_unlock_irq(&s
->tracked_chunk_lock
);
265 static void stop_tracking_chunk(struct dm_snapshot
*s
, struct bio
*bio
)
267 struct dm_snap_tracked_chunk
*c
= dm_per_bio_data(bio
, sizeof(struct dm_snap_tracked_chunk
));
270 spin_lock_irqsave(&s
->tracked_chunk_lock
, flags
);
272 spin_unlock_irqrestore(&s
->tracked_chunk_lock
, flags
);
275 static int __chunk_is_tracked(struct dm_snapshot
*s
, chunk_t chunk
)
277 struct dm_snap_tracked_chunk
*c
;
280 spin_lock_irq(&s
->tracked_chunk_lock
);
282 hlist_for_each_entry(c
,
283 &s
->tracked_chunk_hash
[DM_TRACKED_CHUNK_HASH(chunk
)], node
) {
284 if (c
->chunk
== chunk
) {
290 spin_unlock_irq(&s
->tracked_chunk_lock
);
296 * This conflicting I/O is extremely improbable in the caller,
297 * so msleep(1) is sufficient and there is no need for a wait queue.
299 static void __check_for_conflicting_io(struct dm_snapshot
*s
, chunk_t chunk
)
301 while (__chunk_is_tracked(s
, chunk
))
306 * One of these per registered origin, held in the snapshot_origins hash
309 /* The origin device */
310 struct block_device
*bdev
;
312 struct list_head hash_list
;
314 /* List of snapshots for this origin */
315 struct list_head snapshots
;
319 * This structure is allocated for each origin target
323 struct dm_target
*ti
;
324 unsigned split_boundary
;
325 struct list_head hash_list
;
329 * Size of the hash table for origin volumes. If we make this
330 * the size of the minors list then it should be nearly perfect
332 #define ORIGIN_HASH_SIZE 256
333 #define ORIGIN_MASK 0xFF
334 static struct list_head
*_origins
;
335 static struct list_head
*_dm_origins
;
336 static struct rw_semaphore _origins_lock
;
338 static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done
);
339 static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock
);
340 static uint64_t _pending_exceptions_done_count
;
342 static int init_origin_hash(void)
346 _origins
= kmalloc_array(ORIGIN_HASH_SIZE
, sizeof(struct list_head
),
349 DMERR("unable to allocate memory for _origins");
352 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
353 INIT_LIST_HEAD(_origins
+ i
);
355 _dm_origins
= kmalloc_array(ORIGIN_HASH_SIZE
,
356 sizeof(struct list_head
),
359 DMERR("unable to allocate memory for _dm_origins");
363 for (i
= 0; i
< ORIGIN_HASH_SIZE
; i
++)
364 INIT_LIST_HEAD(_dm_origins
+ i
);
366 init_rwsem(&_origins_lock
);
371 static void exit_origin_hash(void)
377 static unsigned origin_hash(struct block_device
*bdev
)
379 return bdev
->bd_dev
& ORIGIN_MASK
;
382 static struct origin
*__lookup_origin(struct block_device
*origin
)
384 struct list_head
*ol
;
387 ol
= &_origins
[origin_hash(origin
)];
388 list_for_each_entry (o
, ol
, hash_list
)
389 if (bdev_equal(o
->bdev
, origin
))
395 static void __insert_origin(struct origin
*o
)
397 struct list_head
*sl
= &_origins
[origin_hash(o
->bdev
)];
398 list_add_tail(&o
->hash_list
, sl
);
401 static struct dm_origin
*__lookup_dm_origin(struct block_device
*origin
)
403 struct list_head
*ol
;
406 ol
= &_dm_origins
[origin_hash(origin
)];
407 list_for_each_entry (o
, ol
, hash_list
)
408 if (bdev_equal(o
->dev
->bdev
, origin
))
414 static void __insert_dm_origin(struct dm_origin
*o
)
416 struct list_head
*sl
= &_dm_origins
[origin_hash(o
->dev
->bdev
)];
417 list_add_tail(&o
->hash_list
, sl
);
420 static void __remove_dm_origin(struct dm_origin
*o
)
422 list_del(&o
->hash_list
);
426 * _origins_lock must be held when calling this function.
427 * Returns number of snapshots registered using the supplied cow device, plus:
428 * snap_src - a snapshot suitable for use as a source of exception handover
429 * snap_dest - a snapshot capable of receiving exception handover.
430 * snap_merge - an existing snapshot-merge target linked to the same origin.
431 * There can be at most one snapshot-merge target. The parameter is optional.
433 * Possible return values and states of snap_src and snap_dest.
434 * 0: NULL, NULL - first new snapshot
435 * 1: snap_src, NULL - normal snapshot
436 * 2: snap_src, snap_dest - waiting for handover
437 * 2: snap_src, NULL - handed over, waiting for old to be deleted
438 * 1: NULL, snap_dest - source got destroyed without handover
440 static int __find_snapshots_sharing_cow(struct dm_snapshot
*snap
,
441 struct dm_snapshot
**snap_src
,
442 struct dm_snapshot
**snap_dest
,
443 struct dm_snapshot
**snap_merge
)
445 struct dm_snapshot
*s
;
450 o
= __lookup_origin(snap
->origin
->bdev
);
454 list_for_each_entry(s
, &o
->snapshots
, list
) {
455 if (dm_target_is_snapshot_merge(s
->ti
) && snap_merge
)
457 if (!bdev_equal(s
->cow
->bdev
, snap
->cow
->bdev
))
460 mutex_lock(&s
->lock
);
462 mutex_unlock(&s
->lock
);
467 } else if (snap_dest
)
478 * On success, returns 1 if this snapshot is a handover destination,
479 * otherwise returns 0.
481 static int __validate_exception_handover(struct dm_snapshot
*snap
)
483 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
484 struct dm_snapshot
*snap_merge
= NULL
;
486 /* Does snapshot need exceptions handed over to it? */
487 if ((__find_snapshots_sharing_cow(snap
, &snap_src
, &snap_dest
,
488 &snap_merge
) == 2) ||
490 snap
->ti
->error
= "Snapshot cow pairing for exception "
491 "table handover failed";
496 * If no snap_src was found, snap cannot become a handover
503 * Non-snapshot-merge handover?
505 if (!dm_target_is_snapshot_merge(snap
->ti
))
509 * Do not allow more than one merging snapshot.
512 snap
->ti
->error
= "A snapshot is already merging.";
516 if (!snap_src
->store
->type
->prepare_merge
||
517 !snap_src
->store
->type
->commit_merge
) {
518 snap
->ti
->error
= "Snapshot exception store does not "
519 "support snapshot-merge.";
526 static void __insert_snapshot(struct origin
*o
, struct dm_snapshot
*s
)
528 struct dm_snapshot
*l
;
530 /* Sort the list according to chunk size, largest-first smallest-last */
531 list_for_each_entry(l
, &o
->snapshots
, list
)
532 if (l
->store
->chunk_size
< s
->store
->chunk_size
)
534 list_add_tail(&s
->list
, &l
->list
);
538 * Make a note of the snapshot and its origin so we can look it
539 * up when the origin has a write on it.
541 * Also validate snapshot exception store handovers.
542 * On success, returns 1 if this registration is a handover destination,
543 * otherwise returns 0.
545 static int register_snapshot(struct dm_snapshot
*snap
)
547 struct origin
*o
, *new_o
= NULL
;
548 struct block_device
*bdev
= snap
->origin
->bdev
;
551 new_o
= kmalloc(sizeof(*new_o
), GFP_KERNEL
);
555 down_write(&_origins_lock
);
557 r
= __validate_exception_handover(snap
);
563 o
= __lookup_origin(bdev
);
570 /* Initialise the struct */
571 INIT_LIST_HEAD(&o
->snapshots
);
577 __insert_snapshot(o
, snap
);
580 up_write(&_origins_lock
);
586 * Move snapshot to correct place in list according to chunk size.
588 static void reregister_snapshot(struct dm_snapshot
*s
)
590 struct block_device
*bdev
= s
->origin
->bdev
;
592 down_write(&_origins_lock
);
595 __insert_snapshot(__lookup_origin(bdev
), s
);
597 up_write(&_origins_lock
);
600 static void unregister_snapshot(struct dm_snapshot
*s
)
604 down_write(&_origins_lock
);
605 o
= __lookup_origin(s
->origin
->bdev
);
608 if (o
&& list_empty(&o
->snapshots
)) {
609 list_del(&o
->hash_list
);
613 up_write(&_origins_lock
);
617 * Implementation of the exception hash tables.
618 * The lowest hash_shift bits of the chunk number are ignored, allowing
619 * some consecutive chunks to be grouped together.
621 static int dm_exception_table_init(struct dm_exception_table
*et
,
622 uint32_t size
, unsigned hash_shift
)
626 et
->hash_shift
= hash_shift
;
627 et
->hash_mask
= size
- 1;
628 et
->table
= dm_vcalloc(size
, sizeof(struct list_head
));
632 for (i
= 0; i
< size
; i
++)
633 INIT_LIST_HEAD(et
->table
+ i
);
638 static void dm_exception_table_exit(struct dm_exception_table
*et
,
639 struct kmem_cache
*mem
)
641 struct list_head
*slot
;
642 struct dm_exception
*ex
, *next
;
645 size
= et
->hash_mask
+ 1;
646 for (i
= 0; i
< size
; i
++) {
647 slot
= et
->table
+ i
;
649 list_for_each_entry_safe (ex
, next
, slot
, hash_list
)
650 kmem_cache_free(mem
, ex
);
656 static uint32_t exception_hash(struct dm_exception_table
*et
, chunk_t chunk
)
658 return (chunk
>> et
->hash_shift
) & et
->hash_mask
;
661 static void dm_remove_exception(struct dm_exception
*e
)
663 list_del(&e
->hash_list
);
667 * Return the exception data for a sector, or NULL if not
670 static struct dm_exception
*dm_lookup_exception(struct dm_exception_table
*et
,
673 struct list_head
*slot
;
674 struct dm_exception
*e
;
676 slot
= &et
->table
[exception_hash(et
, chunk
)];
677 list_for_each_entry (e
, slot
, hash_list
)
678 if (chunk
>= e
->old_chunk
&&
679 chunk
<= e
->old_chunk
+ dm_consecutive_chunk_count(e
))
685 static struct dm_exception
*alloc_completed_exception(gfp_t gfp
)
687 struct dm_exception
*e
;
689 e
= kmem_cache_alloc(exception_cache
, gfp
);
690 if (!e
&& gfp
== GFP_NOIO
)
691 e
= kmem_cache_alloc(exception_cache
, GFP_ATOMIC
);
696 static void free_completed_exception(struct dm_exception
*e
)
698 kmem_cache_free(exception_cache
, e
);
701 static struct dm_snap_pending_exception
*alloc_pending_exception(struct dm_snapshot
*s
)
703 struct dm_snap_pending_exception
*pe
= mempool_alloc(&s
->pending_pool
,
706 atomic_inc(&s
->pending_exceptions_count
);
712 static void free_pending_exception(struct dm_snap_pending_exception
*pe
)
714 struct dm_snapshot
*s
= pe
->snap
;
716 mempool_free(pe
, &s
->pending_pool
);
717 smp_mb__before_atomic();
718 atomic_dec(&s
->pending_exceptions_count
);
721 static void dm_insert_exception(struct dm_exception_table
*eh
,
722 struct dm_exception
*new_e
)
725 struct dm_exception
*e
= NULL
;
727 l
= &eh
->table
[exception_hash(eh
, new_e
->old_chunk
)];
729 /* Add immediately if this table doesn't support consecutive chunks */
733 /* List is ordered by old_chunk */
734 list_for_each_entry_reverse(e
, l
, hash_list
) {
735 /* Insert after an existing chunk? */
736 if (new_e
->old_chunk
== (e
->old_chunk
+
737 dm_consecutive_chunk_count(e
) + 1) &&
738 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) +
739 dm_consecutive_chunk_count(e
) + 1)) {
740 dm_consecutive_chunk_count_inc(e
);
741 free_completed_exception(new_e
);
745 /* Insert before an existing chunk? */
746 if (new_e
->old_chunk
== (e
->old_chunk
- 1) &&
747 new_e
->new_chunk
== (dm_chunk_number(e
->new_chunk
) - 1)) {
748 dm_consecutive_chunk_count_inc(e
);
751 free_completed_exception(new_e
);
755 if (new_e
->old_chunk
> e
->old_chunk
)
760 list_add(&new_e
->hash_list
, e
? &e
->hash_list
: l
);
764 * Callback used by the exception stores to load exceptions when
767 static int dm_add_exception(void *context
, chunk_t old
, chunk_t
new)
769 struct dm_snapshot
*s
= context
;
770 struct dm_exception
*e
;
772 e
= alloc_completed_exception(GFP_KERNEL
);
778 /* Consecutive_count is implicitly initialised to zero */
781 dm_insert_exception(&s
->complete
, e
);
787 * Return a minimum chunk size of all snapshots that have the specified origin.
788 * Return zero if the origin has no snapshots.
790 static uint32_t __minimum_chunk_size(struct origin
*o
)
792 struct dm_snapshot
*snap
;
793 unsigned chunk_size
= 0;
796 list_for_each_entry(snap
, &o
->snapshots
, list
)
797 chunk_size
= min_not_zero(chunk_size
,
798 snap
->store
->chunk_size
);
800 return (uint32_t) chunk_size
;
806 static int calc_max_buckets(void)
808 /* use a fixed size of 2MB */
809 unsigned long mem
= 2 * 1024 * 1024;
810 mem
/= sizeof(struct list_head
);
816 * Allocate room for a suitable hash table.
818 static int init_hash_tables(struct dm_snapshot
*s
)
820 sector_t hash_size
, cow_dev_size
, max_buckets
;
823 * Calculate based on the size of the original volume or
826 cow_dev_size
= get_dev_size(s
->cow
->bdev
);
827 max_buckets
= calc_max_buckets();
829 hash_size
= cow_dev_size
>> s
->store
->chunk_shift
;
830 hash_size
= min(hash_size
, max_buckets
);
834 hash_size
= rounddown_pow_of_two(hash_size
);
835 if (dm_exception_table_init(&s
->complete
, hash_size
,
836 DM_CHUNK_CONSECUTIVE_BITS
))
840 * Allocate hash table for in-flight exceptions
841 * Make this smaller than the real hash table
847 if (dm_exception_table_init(&s
->pending
, hash_size
, 0)) {
848 dm_exception_table_exit(&s
->complete
, exception_cache
);
855 static void merge_shutdown(struct dm_snapshot
*s
)
857 clear_bit_unlock(RUNNING_MERGE
, &s
->state_bits
);
858 smp_mb__after_atomic();
859 wake_up_bit(&s
->state_bits
, RUNNING_MERGE
);
862 static struct bio
*__release_queued_bios_after_merge(struct dm_snapshot
*s
)
864 s
->first_merging_chunk
= 0;
865 s
->num_merging_chunks
= 0;
867 return bio_list_get(&s
->bios_queued_during_merge
);
871 * Remove one chunk from the index of completed exceptions.
873 static int __remove_single_exception_chunk(struct dm_snapshot
*s
,
876 struct dm_exception
*e
;
878 e
= dm_lookup_exception(&s
->complete
, old_chunk
);
880 DMERR("Corruption detected: exception for block %llu is "
881 "on disk but not in memory",
882 (unsigned long long)old_chunk
);
887 * If this is the only chunk using this exception, remove exception.
889 if (!dm_consecutive_chunk_count(e
)) {
890 dm_remove_exception(e
);
891 free_completed_exception(e
);
896 * The chunk may be either at the beginning or the end of a
897 * group of consecutive chunks - never in the middle. We are
898 * removing chunks in the opposite order to that in which they
899 * were added, so this should always be true.
900 * Decrement the consecutive chunk counter and adjust the
901 * starting point if necessary.
903 if (old_chunk
== e
->old_chunk
) {
906 } else if (old_chunk
!= e
->old_chunk
+
907 dm_consecutive_chunk_count(e
)) {
908 DMERR("Attempt to merge block %llu from the "
909 "middle of a chunk range [%llu - %llu]",
910 (unsigned long long)old_chunk
,
911 (unsigned long long)e
->old_chunk
,
913 e
->old_chunk
+ dm_consecutive_chunk_count(e
));
917 dm_consecutive_chunk_count_dec(e
);
922 static void flush_bios(struct bio
*bio
);
924 static int remove_single_exception_chunk(struct dm_snapshot
*s
)
926 struct bio
*b
= NULL
;
928 chunk_t old_chunk
= s
->first_merging_chunk
+ s
->num_merging_chunks
- 1;
930 mutex_lock(&s
->lock
);
933 * Process chunks (and associated exceptions) in reverse order
934 * so that dm_consecutive_chunk_count_dec() accounting works.
937 r
= __remove_single_exception_chunk(s
, old_chunk
);
940 } while (old_chunk
-- > s
->first_merging_chunk
);
942 b
= __release_queued_bios_after_merge(s
);
945 mutex_unlock(&s
->lock
);
952 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
953 sector_t sector
, unsigned chunk_size
);
955 static void merge_callback(int read_err
, unsigned long write_err
,
958 static uint64_t read_pending_exceptions_done_count(void)
960 uint64_t pending_exceptions_done
;
962 spin_lock(&_pending_exceptions_done_spinlock
);
963 pending_exceptions_done
= _pending_exceptions_done_count
;
964 spin_unlock(&_pending_exceptions_done_spinlock
);
966 return pending_exceptions_done
;
969 static void increment_pending_exceptions_done_count(void)
971 spin_lock(&_pending_exceptions_done_spinlock
);
972 _pending_exceptions_done_count
++;
973 spin_unlock(&_pending_exceptions_done_spinlock
);
975 wake_up_all(&_pending_exceptions_done
);
978 static void snapshot_merge_next_chunks(struct dm_snapshot
*s
)
980 int i
, linear_chunks
;
981 chunk_t old_chunk
, new_chunk
;
982 struct dm_io_region src
, dest
;
984 uint64_t previous_count
;
986 BUG_ON(!test_bit(RUNNING_MERGE
, &s
->state_bits
));
987 if (unlikely(test_bit(SHUTDOWN_MERGE
, &s
->state_bits
)))
991 * valid flag never changes during merge, so no lock required.
994 DMERR("Snapshot is invalid: can't merge");
998 linear_chunks
= s
->store
->type
->prepare_merge(s
->store
, &old_chunk
,
1000 if (linear_chunks
<= 0) {
1001 if (linear_chunks
< 0) {
1002 DMERR("Read error in exception store: "
1003 "shutting down merge");
1004 mutex_lock(&s
->lock
);
1005 s
->merge_failed
= 1;
1006 mutex_unlock(&s
->lock
);
1011 /* Adjust old_chunk and new_chunk to reflect start of linear region */
1012 old_chunk
= old_chunk
+ 1 - linear_chunks
;
1013 new_chunk
= new_chunk
+ 1 - linear_chunks
;
1016 * Use one (potentially large) I/O to copy all 'linear_chunks'
1017 * from the exception store to the origin
1019 io_size
= linear_chunks
* s
->store
->chunk_size
;
1021 dest
.bdev
= s
->origin
->bdev
;
1022 dest
.sector
= chunk_to_sector(s
->store
, old_chunk
);
1023 dest
.count
= min(io_size
, get_dev_size(dest
.bdev
) - dest
.sector
);
1025 src
.bdev
= s
->cow
->bdev
;
1026 src
.sector
= chunk_to_sector(s
->store
, new_chunk
);
1027 src
.count
= dest
.count
;
1030 * Reallocate any exceptions needed in other snapshots then
1031 * wait for the pending exceptions to complete.
1032 * Each time any pending exception (globally on the system)
1033 * completes we are woken and repeat the process to find out
1034 * if we can proceed. While this may not seem a particularly
1035 * efficient algorithm, it is not expected to have any
1036 * significant impact on performance.
1038 previous_count
= read_pending_exceptions_done_count();
1039 while (origin_write_extent(s
, dest
.sector
, io_size
)) {
1040 wait_event(_pending_exceptions_done
,
1041 (read_pending_exceptions_done_count() !=
1043 /* Retry after the wait, until all exceptions are done. */
1044 previous_count
= read_pending_exceptions_done_count();
1047 mutex_lock(&s
->lock
);
1048 s
->first_merging_chunk
= old_chunk
;
1049 s
->num_merging_chunks
= linear_chunks
;
1050 mutex_unlock(&s
->lock
);
1052 /* Wait until writes to all 'linear_chunks' drain */
1053 for (i
= 0; i
< linear_chunks
; i
++)
1054 __check_for_conflicting_io(s
, old_chunk
+ i
);
1056 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, merge_callback
, s
);
1063 static void error_bios(struct bio
*bio
);
1065 static void merge_callback(int read_err
, unsigned long write_err
, void *context
)
1067 struct dm_snapshot
*s
= context
;
1068 struct bio
*b
= NULL
;
1070 if (read_err
|| write_err
) {
1072 DMERR("Read error: shutting down merge.");
1074 DMERR("Write error: shutting down merge.");
1078 if (s
->store
->type
->commit_merge(s
->store
,
1079 s
->num_merging_chunks
) < 0) {
1080 DMERR("Write error in exception store: shutting down merge");
1084 if (remove_single_exception_chunk(s
) < 0)
1087 snapshot_merge_next_chunks(s
);
1092 mutex_lock(&s
->lock
);
1093 s
->merge_failed
= 1;
1094 b
= __release_queued_bios_after_merge(s
);
1095 mutex_unlock(&s
->lock
);
1101 static void start_merge(struct dm_snapshot
*s
)
1103 if (!test_and_set_bit(RUNNING_MERGE
, &s
->state_bits
))
1104 snapshot_merge_next_chunks(s
);
1108 * Stop the merging process and wait until it finishes.
1110 static void stop_merge(struct dm_snapshot
*s
)
1112 set_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1113 wait_on_bit(&s
->state_bits
, RUNNING_MERGE
, TASK_UNINTERRUPTIBLE
);
1114 clear_bit(SHUTDOWN_MERGE
, &s
->state_bits
);
1118 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p|po|n> <chunk-size>
1120 static int snapshot_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
1122 struct dm_snapshot
*s
;
1125 char *origin_path
, *cow_path
;
1126 dev_t origin_dev
, cow_dev
;
1127 unsigned args_used
, num_flush_bios
= 1;
1128 fmode_t origin_mode
= FMODE_READ
;
1131 ti
->error
= "requires exactly 4 arguments";
1136 if (dm_target_is_snapshot_merge(ti
)) {
1138 origin_mode
= FMODE_WRITE
;
1141 s
= kzalloc(sizeof(*s
), GFP_KERNEL
);
1143 ti
->error
= "Cannot allocate private snapshot structure";
1148 origin_path
= argv
[0];
1152 r
= dm_get_device(ti
, origin_path
, origin_mode
, &s
->origin
);
1154 ti
->error
= "Cannot get origin device";
1157 origin_dev
= s
->origin
->bdev
->bd_dev
;
1163 cow_dev
= dm_get_dev_t(cow_path
);
1164 if (cow_dev
&& cow_dev
== origin_dev
) {
1165 ti
->error
= "COW device cannot be the same as origin device";
1170 r
= dm_get_device(ti
, cow_path
, dm_table_get_mode(ti
->table
), &s
->cow
);
1172 ti
->error
= "Cannot get COW device";
1176 r
= dm_exception_store_create(ti
, argc
, argv
, s
, &args_used
, &s
->store
);
1178 ti
->error
= "Couldn't create exception store";
1188 s
->snapshot_overflowed
= 0;
1190 atomic_set(&s
->pending_exceptions_count
, 0);
1191 s
->exception_start_sequence
= 0;
1192 s
->exception_complete_sequence
= 0;
1193 s
->out_of_order_tree
= RB_ROOT
;
1194 mutex_init(&s
->lock
);
1195 INIT_LIST_HEAD(&s
->list
);
1196 spin_lock_init(&s
->pe_lock
);
1198 s
->merge_failed
= 0;
1199 s
->first_merging_chunk
= 0;
1200 s
->num_merging_chunks
= 0;
1201 bio_list_init(&s
->bios_queued_during_merge
);
1203 /* Allocate hash table for COW data */
1204 if (init_hash_tables(s
)) {
1205 ti
->error
= "Unable to allocate hash table space";
1207 goto bad_hash_tables
;
1210 sema_init(&s
->cow_count
, (cow_threshold
> 0) ? cow_threshold
: INT_MAX
);
1212 s
->kcopyd_client
= dm_kcopyd_client_create(&dm_kcopyd_throttle
);
1213 if (IS_ERR(s
->kcopyd_client
)) {
1214 r
= PTR_ERR(s
->kcopyd_client
);
1215 ti
->error
= "Could not create kcopyd client";
1219 r
= mempool_init_slab_pool(&s
->pending_pool
, MIN_IOS
, pending_cache
);
1221 ti
->error
= "Could not allocate mempool for pending exceptions";
1222 goto bad_pending_pool
;
1225 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1226 INIT_HLIST_HEAD(&s
->tracked_chunk_hash
[i
]);
1228 spin_lock_init(&s
->tracked_chunk_lock
);
1231 ti
->num_flush_bios
= num_flush_bios
;
1232 ti
->per_io_data_size
= sizeof(struct dm_snap_tracked_chunk
);
1234 /* Add snapshot to the list of snapshots for this origin */
1235 /* Exceptions aren't triggered till snapshot_resume() is called */
1236 r
= register_snapshot(s
);
1238 ti
->error
= "Snapshot origin struct allocation failed";
1239 goto bad_load_and_register
;
1241 /* invalid handover, register_snapshot has set ti->error */
1242 goto bad_load_and_register
;
1246 * Metadata must only be loaded into one table at once, so skip this
1247 * if metadata will be handed over during resume.
1248 * Chunk size will be set during the handover - set it to zero to
1249 * ensure it's ignored.
1252 s
->store
->chunk_size
= 0;
1256 r
= s
->store
->type
->read_metadata(s
->store
, dm_add_exception
,
1259 ti
->error
= "Failed to read snapshot metadata";
1260 goto bad_read_metadata
;
1263 DMWARN("Snapshot is marked invalid.");
1266 if (!s
->store
->chunk_size
) {
1267 ti
->error
= "Chunk size not set";
1268 goto bad_read_metadata
;
1271 r
= dm_set_target_max_io_len(ti
, s
->store
->chunk_size
);
1273 goto bad_read_metadata
;
1278 unregister_snapshot(s
);
1280 bad_load_and_register
:
1281 mempool_exit(&s
->pending_pool
);
1284 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1287 dm_exception_table_exit(&s
->pending
, pending_cache
);
1288 dm_exception_table_exit(&s
->complete
, exception_cache
);
1291 dm_exception_store_destroy(s
->store
);
1294 dm_put_device(ti
, s
->cow
);
1297 dm_put_device(ti
, s
->origin
);
1306 static void __free_exceptions(struct dm_snapshot
*s
)
1308 dm_kcopyd_client_destroy(s
->kcopyd_client
);
1309 s
->kcopyd_client
= NULL
;
1311 dm_exception_table_exit(&s
->pending
, pending_cache
);
1312 dm_exception_table_exit(&s
->complete
, exception_cache
);
1315 static void __handover_exceptions(struct dm_snapshot
*snap_src
,
1316 struct dm_snapshot
*snap_dest
)
1319 struct dm_exception_table table_swap
;
1320 struct dm_exception_store
*store_swap
;
1324 * Swap all snapshot context information between the two instances.
1326 u
.table_swap
= snap_dest
->complete
;
1327 snap_dest
->complete
= snap_src
->complete
;
1328 snap_src
->complete
= u
.table_swap
;
1330 u
.store_swap
= snap_dest
->store
;
1331 snap_dest
->store
= snap_src
->store
;
1332 snap_dest
->store
->userspace_supports_overflow
= u
.store_swap
->userspace_supports_overflow
;
1333 snap_src
->store
= u
.store_swap
;
1335 snap_dest
->store
->snap
= snap_dest
;
1336 snap_src
->store
->snap
= snap_src
;
1338 snap_dest
->ti
->max_io_len
= snap_dest
->store
->chunk_size
;
1339 snap_dest
->valid
= snap_src
->valid
;
1340 snap_dest
->snapshot_overflowed
= snap_src
->snapshot_overflowed
;
1343 * Set source invalid to ensure it receives no further I/O.
1345 snap_src
->valid
= 0;
1348 static void snapshot_dtr(struct dm_target
*ti
)
1350 #ifdef CONFIG_DM_DEBUG
1353 struct dm_snapshot
*s
= ti
->private;
1354 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1356 down_read(&_origins_lock
);
1357 /* Check whether exception handover must be cancelled */
1358 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1359 if (snap_src
&& snap_dest
&& (s
== snap_src
)) {
1360 mutex_lock(&snap_dest
->lock
);
1361 snap_dest
->valid
= 0;
1362 mutex_unlock(&snap_dest
->lock
);
1363 DMERR("Cancelling snapshot handover.");
1365 up_read(&_origins_lock
);
1367 if (dm_target_is_snapshot_merge(ti
))
1370 /* Prevent further origin writes from using this snapshot. */
1371 /* After this returns there can be no new kcopyd jobs. */
1372 unregister_snapshot(s
);
1374 while (atomic_read(&s
->pending_exceptions_count
))
1377 * Ensure instructions in mempool_exit aren't reordered
1378 * before atomic_read.
1382 #ifdef CONFIG_DM_DEBUG
1383 for (i
= 0; i
< DM_TRACKED_CHUNK_HASH_SIZE
; i
++)
1384 BUG_ON(!hlist_empty(&s
->tracked_chunk_hash
[i
]));
1387 __free_exceptions(s
);
1389 mempool_exit(&s
->pending_pool
);
1391 dm_exception_store_destroy(s
->store
);
1393 mutex_destroy(&s
->lock
);
1395 dm_put_device(ti
, s
->cow
);
1397 dm_put_device(ti
, s
->origin
);
1403 * Flush a list of buffers.
1405 static void flush_bios(struct bio
*bio
)
1411 bio
->bi_next
= NULL
;
1412 generic_make_request(bio
);
1417 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
);
1420 * Flush a list of buffers.
1422 static void retry_origin_bios(struct dm_snapshot
*s
, struct bio
*bio
)
1429 bio
->bi_next
= NULL
;
1430 r
= do_origin(s
->origin
, bio
);
1431 if (r
== DM_MAPIO_REMAPPED
)
1432 generic_make_request(bio
);
1438 * Error a list of buffers.
1440 static void error_bios(struct bio
*bio
)
1446 bio
->bi_next
= NULL
;
1452 static void __invalidate_snapshot(struct dm_snapshot
*s
, int err
)
1458 DMERR("Invalidating snapshot: Error reading/writing.");
1459 else if (err
== -ENOMEM
)
1460 DMERR("Invalidating snapshot: Unable to allocate exception.");
1462 if (s
->store
->type
->drop_snapshot
)
1463 s
->store
->type
->drop_snapshot(s
->store
);
1467 dm_table_event(s
->ti
->table
);
1470 static void pending_complete(void *context
, int success
)
1472 struct dm_snap_pending_exception
*pe
= context
;
1473 struct dm_exception
*e
;
1474 struct dm_snapshot
*s
= pe
->snap
;
1475 struct bio
*origin_bios
= NULL
;
1476 struct bio
*snapshot_bios
= NULL
;
1477 struct bio
*full_bio
= NULL
;
1481 /* Read/write error - snapshot is unusable */
1482 mutex_lock(&s
->lock
);
1483 __invalidate_snapshot(s
, -EIO
);
1488 e
= alloc_completed_exception(GFP_NOIO
);
1490 mutex_lock(&s
->lock
);
1491 __invalidate_snapshot(s
, -ENOMEM
);
1497 mutex_lock(&s
->lock
);
1499 free_completed_exception(e
);
1504 /* Check for conflicting reads */
1505 __check_for_conflicting_io(s
, pe
->e
.old_chunk
);
1508 * Add a proper exception, and remove the
1509 * in-flight exception from the list.
1511 dm_insert_exception(&s
->complete
, e
);
1514 dm_remove_exception(&pe
->e
);
1515 snapshot_bios
= bio_list_get(&pe
->snapshot_bios
);
1516 origin_bios
= bio_list_get(&pe
->origin_bios
);
1517 full_bio
= pe
->full_bio
;
1519 full_bio
->bi_end_io
= pe
->full_bio_end_io
;
1520 increment_pending_exceptions_done_count();
1522 mutex_unlock(&s
->lock
);
1524 /* Submit any pending write bios */
1527 bio_io_error(full_bio
);
1528 error_bios(snapshot_bios
);
1531 bio_endio(full_bio
);
1532 flush_bios(snapshot_bios
);
1535 retry_origin_bios(s
, origin_bios
);
1537 free_pending_exception(pe
);
1540 static void complete_exception(struct dm_snap_pending_exception
*pe
)
1542 struct dm_snapshot
*s
= pe
->snap
;
1544 /* Update the metadata if we are persistent */
1545 s
->store
->type
->commit_exception(s
->store
, &pe
->e
, !pe
->copy_error
,
1546 pending_complete
, pe
);
1550 * Called when the copy I/O has finished. kcopyd actually runs
1551 * this code so don't block.
1553 static void copy_callback(int read_err
, unsigned long write_err
, void *context
)
1555 struct dm_snap_pending_exception
*pe
= context
;
1556 struct dm_snapshot
*s
= pe
->snap
;
1558 pe
->copy_error
= read_err
|| write_err
;
1560 if (pe
->exception_sequence
== s
->exception_complete_sequence
) {
1561 struct rb_node
*next
;
1563 s
->exception_complete_sequence
++;
1564 complete_exception(pe
);
1566 next
= rb_first(&s
->out_of_order_tree
);
1568 pe
= rb_entry(next
, struct dm_snap_pending_exception
,
1570 if (pe
->exception_sequence
!= s
->exception_complete_sequence
)
1572 next
= rb_next(next
);
1573 s
->exception_complete_sequence
++;
1574 rb_erase(&pe
->out_of_order_node
, &s
->out_of_order_tree
);
1575 complete_exception(pe
);
1579 struct rb_node
*parent
= NULL
;
1580 struct rb_node
**p
= &s
->out_of_order_tree
.rb_node
;
1581 struct dm_snap_pending_exception
*pe2
;
1584 pe2
= rb_entry(*p
, struct dm_snap_pending_exception
, out_of_order_node
);
1587 BUG_ON(pe
->exception_sequence
== pe2
->exception_sequence
);
1588 if (pe
->exception_sequence
< pe2
->exception_sequence
)
1589 p
= &((*p
)->rb_left
);
1591 p
= &((*p
)->rb_right
);
1594 rb_link_node(&pe
->out_of_order_node
, parent
, p
);
1595 rb_insert_color(&pe
->out_of_order_node
, &s
->out_of_order_tree
);
1601 * Dispatches the copy operation to kcopyd.
1603 static void start_copy(struct dm_snap_pending_exception
*pe
)
1605 struct dm_snapshot
*s
= pe
->snap
;
1606 struct dm_io_region src
, dest
;
1607 struct block_device
*bdev
= s
->origin
->bdev
;
1610 dev_size
= get_dev_size(bdev
);
1613 src
.sector
= chunk_to_sector(s
->store
, pe
->e
.old_chunk
);
1614 src
.count
= min((sector_t
)s
->store
->chunk_size
, dev_size
- src
.sector
);
1616 dest
.bdev
= s
->cow
->bdev
;
1617 dest
.sector
= chunk_to_sector(s
->store
, pe
->e
.new_chunk
);
1618 dest
.count
= src
.count
;
1620 /* Hand over to kcopyd */
1621 down(&s
->cow_count
);
1622 dm_kcopyd_copy(s
->kcopyd_client
, &src
, 1, &dest
, 0, copy_callback
, pe
);
1625 static void full_bio_end_io(struct bio
*bio
)
1627 void *callback_data
= bio
->bi_private
;
1629 dm_kcopyd_do_callback(callback_data
, 0, bio
->bi_status
? 1 : 0);
1632 static void start_full_bio(struct dm_snap_pending_exception
*pe
,
1635 struct dm_snapshot
*s
= pe
->snap
;
1636 void *callback_data
;
1639 pe
->full_bio_end_io
= bio
->bi_end_io
;
1641 down(&s
->cow_count
);
1642 callback_data
= dm_kcopyd_prepare_callback(s
->kcopyd_client
,
1645 bio
->bi_end_io
= full_bio_end_io
;
1646 bio
->bi_private
= callback_data
;
1648 generic_make_request(bio
);
1651 static struct dm_snap_pending_exception
*
1652 __lookup_pending_exception(struct dm_snapshot
*s
, chunk_t chunk
)
1654 struct dm_exception
*e
= dm_lookup_exception(&s
->pending
, chunk
);
1659 return container_of(e
, struct dm_snap_pending_exception
, e
);
1663 * Looks to see if this snapshot already has a pending exception
1664 * for this chunk, otherwise it allocates a new one and inserts
1665 * it into the pending table.
1667 * NOTE: a write lock must be held on snap->lock before calling
1670 static struct dm_snap_pending_exception
*
1671 __find_pending_exception(struct dm_snapshot
*s
,
1672 struct dm_snap_pending_exception
*pe
, chunk_t chunk
)
1674 struct dm_snap_pending_exception
*pe2
;
1676 pe2
= __lookup_pending_exception(s
, chunk
);
1678 free_pending_exception(pe
);
1682 pe
->e
.old_chunk
= chunk
;
1683 bio_list_init(&pe
->origin_bios
);
1684 bio_list_init(&pe
->snapshot_bios
);
1686 pe
->full_bio
= NULL
;
1688 if (s
->store
->type
->prepare_exception(s
->store
, &pe
->e
)) {
1689 free_pending_exception(pe
);
1693 pe
->exception_sequence
= s
->exception_start_sequence
++;
1695 dm_insert_exception(&s
->pending
, &pe
->e
);
1700 static void remap_exception(struct dm_snapshot
*s
, struct dm_exception
*e
,
1701 struct bio
*bio
, chunk_t chunk
)
1703 bio_set_dev(bio
, s
->cow
->bdev
);
1704 bio
->bi_iter
.bi_sector
=
1705 chunk_to_sector(s
->store
, dm_chunk_number(e
->new_chunk
) +
1706 (chunk
- e
->old_chunk
)) +
1707 (bio
->bi_iter
.bi_sector
& s
->store
->chunk_mask
);
1710 static int snapshot_map(struct dm_target
*ti
, struct bio
*bio
)
1712 struct dm_exception
*e
;
1713 struct dm_snapshot
*s
= ti
->private;
1714 int r
= DM_MAPIO_REMAPPED
;
1716 struct dm_snap_pending_exception
*pe
= NULL
;
1718 init_tracked_chunk(bio
);
1720 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1721 bio_set_dev(bio
, s
->cow
->bdev
);
1722 return DM_MAPIO_REMAPPED
;
1725 chunk
= sector_to_chunk(s
->store
, bio
->bi_iter
.bi_sector
);
1727 /* Full snapshots are not usable */
1728 /* To get here the table must be live so s->active is always set. */
1730 return DM_MAPIO_KILL
;
1732 mutex_lock(&s
->lock
);
1734 if (!s
->valid
|| (unlikely(s
->snapshot_overflowed
) &&
1735 bio_data_dir(bio
) == WRITE
)) {
1740 /* If the block is already remapped - use that, else remap it */
1741 e
= dm_lookup_exception(&s
->complete
, chunk
);
1743 remap_exception(s
, e
, bio
, chunk
);
1748 * Write to snapshot - higher level takes care of RW/RO
1749 * flags so we should only get this if we are
1752 if (bio_data_dir(bio
) == WRITE
) {
1753 pe
= __lookup_pending_exception(s
, chunk
);
1755 mutex_unlock(&s
->lock
);
1756 pe
= alloc_pending_exception(s
);
1757 mutex_lock(&s
->lock
);
1759 if (!s
->valid
|| s
->snapshot_overflowed
) {
1760 free_pending_exception(pe
);
1765 e
= dm_lookup_exception(&s
->complete
, chunk
);
1767 free_pending_exception(pe
);
1768 remap_exception(s
, e
, bio
, chunk
);
1772 pe
= __find_pending_exception(s
, pe
, chunk
);
1774 if (s
->store
->userspace_supports_overflow
) {
1775 s
->snapshot_overflowed
= 1;
1776 DMERR("Snapshot overflowed: Unable to allocate exception.");
1778 __invalidate_snapshot(s
, -ENOMEM
);
1784 remap_exception(s
, &pe
->e
, bio
, chunk
);
1786 r
= DM_MAPIO_SUBMITTED
;
1789 bio
->bi_iter
.bi_size
==
1790 (s
->store
->chunk_size
<< SECTOR_SHIFT
)) {
1792 mutex_unlock(&s
->lock
);
1793 start_full_bio(pe
, bio
);
1797 bio_list_add(&pe
->snapshot_bios
, bio
);
1800 /* this is protected by snap->lock */
1802 mutex_unlock(&s
->lock
);
1807 bio_set_dev(bio
, s
->origin
->bdev
);
1808 track_chunk(s
, bio
, chunk
);
1812 mutex_unlock(&s
->lock
);
1818 * A snapshot-merge target behaves like a combination of a snapshot
1819 * target and a snapshot-origin target. It only generates new
1820 * exceptions in other snapshots and not in the one that is being
1823 * For each chunk, if there is an existing exception, it is used to
1824 * redirect I/O to the cow device. Otherwise I/O is sent to the origin,
1825 * which in turn might generate exceptions in other snapshots.
1826 * If merging is currently taking place on the chunk in question, the
1827 * I/O is deferred by adding it to s->bios_queued_during_merge.
1829 static int snapshot_merge_map(struct dm_target
*ti
, struct bio
*bio
)
1831 struct dm_exception
*e
;
1832 struct dm_snapshot
*s
= ti
->private;
1833 int r
= DM_MAPIO_REMAPPED
;
1836 init_tracked_chunk(bio
);
1838 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1839 if (!dm_bio_get_target_bio_nr(bio
))
1840 bio_set_dev(bio
, s
->origin
->bdev
);
1842 bio_set_dev(bio
, s
->cow
->bdev
);
1843 return DM_MAPIO_REMAPPED
;
1846 chunk
= sector_to_chunk(s
->store
, bio
->bi_iter
.bi_sector
);
1848 mutex_lock(&s
->lock
);
1850 /* Full merging snapshots are redirected to the origin */
1852 goto redirect_to_origin
;
1854 /* If the block is already remapped - use that */
1855 e
= dm_lookup_exception(&s
->complete
, chunk
);
1857 /* Queue writes overlapping with chunks being merged */
1858 if (bio_data_dir(bio
) == WRITE
&&
1859 chunk
>= s
->first_merging_chunk
&&
1860 chunk
< (s
->first_merging_chunk
+
1861 s
->num_merging_chunks
)) {
1862 bio_set_dev(bio
, s
->origin
->bdev
);
1863 bio_list_add(&s
->bios_queued_during_merge
, bio
);
1864 r
= DM_MAPIO_SUBMITTED
;
1868 remap_exception(s
, e
, bio
, chunk
);
1870 if (bio_data_dir(bio
) == WRITE
)
1871 track_chunk(s
, bio
, chunk
);
1876 bio_set_dev(bio
, s
->origin
->bdev
);
1878 if (bio_data_dir(bio
) == WRITE
) {
1879 mutex_unlock(&s
->lock
);
1880 return do_origin(s
->origin
, bio
);
1884 mutex_unlock(&s
->lock
);
1889 static int snapshot_end_io(struct dm_target
*ti
, struct bio
*bio
,
1890 blk_status_t
*error
)
1892 struct dm_snapshot
*s
= ti
->private;
1894 if (is_bio_tracked(bio
))
1895 stop_tracking_chunk(s
, bio
);
1897 return DM_ENDIO_DONE
;
1900 static void snapshot_merge_presuspend(struct dm_target
*ti
)
1902 struct dm_snapshot
*s
= ti
->private;
1907 static int snapshot_preresume(struct dm_target
*ti
)
1910 struct dm_snapshot
*s
= ti
->private;
1911 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
;
1913 down_read(&_origins_lock
);
1914 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1915 if (snap_src
&& snap_dest
) {
1916 mutex_lock(&snap_src
->lock
);
1917 if (s
== snap_src
) {
1918 DMERR("Unable to resume snapshot source until "
1919 "handover completes.");
1921 } else if (!dm_suspended(snap_src
->ti
)) {
1922 DMERR("Unable to perform snapshot handover until "
1923 "source is suspended.");
1926 mutex_unlock(&snap_src
->lock
);
1928 up_read(&_origins_lock
);
1933 static void snapshot_resume(struct dm_target
*ti
)
1935 struct dm_snapshot
*s
= ti
->private;
1936 struct dm_snapshot
*snap_src
= NULL
, *snap_dest
= NULL
, *snap_merging
= NULL
;
1937 struct dm_origin
*o
;
1938 struct mapped_device
*origin_md
= NULL
;
1939 bool must_restart_merging
= false;
1941 down_read(&_origins_lock
);
1943 o
= __lookup_dm_origin(s
->origin
->bdev
);
1945 origin_md
= dm_table_get_md(o
->ti
->table
);
1947 (void) __find_snapshots_sharing_cow(s
, NULL
, NULL
, &snap_merging
);
1949 origin_md
= dm_table_get_md(snap_merging
->ti
->table
);
1951 if (origin_md
== dm_table_get_md(ti
->table
))
1954 if (dm_hold(origin_md
))
1958 up_read(&_origins_lock
);
1961 dm_internal_suspend_fast(origin_md
);
1962 if (snap_merging
&& test_bit(RUNNING_MERGE
, &snap_merging
->state_bits
)) {
1963 must_restart_merging
= true;
1964 stop_merge(snap_merging
);
1968 down_read(&_origins_lock
);
1970 (void) __find_snapshots_sharing_cow(s
, &snap_src
, &snap_dest
, NULL
);
1971 if (snap_src
&& snap_dest
) {
1972 mutex_lock(&snap_src
->lock
);
1973 mutex_lock_nested(&snap_dest
->lock
, SINGLE_DEPTH_NESTING
);
1974 __handover_exceptions(snap_src
, snap_dest
);
1975 mutex_unlock(&snap_dest
->lock
);
1976 mutex_unlock(&snap_src
->lock
);
1979 up_read(&_origins_lock
);
1982 if (must_restart_merging
)
1983 start_merge(snap_merging
);
1984 dm_internal_resume_fast(origin_md
);
1988 /* Now we have correct chunk size, reregister */
1989 reregister_snapshot(s
);
1991 mutex_lock(&s
->lock
);
1993 mutex_unlock(&s
->lock
);
1996 static uint32_t get_origin_minimum_chunksize(struct block_device
*bdev
)
1998 uint32_t min_chunksize
;
2000 down_read(&_origins_lock
);
2001 min_chunksize
= __minimum_chunk_size(__lookup_origin(bdev
));
2002 up_read(&_origins_lock
);
2004 return min_chunksize
;
2007 static void snapshot_merge_resume(struct dm_target
*ti
)
2009 struct dm_snapshot
*s
= ti
->private;
2012 * Handover exceptions from existing snapshot.
2014 snapshot_resume(ti
);
2017 * snapshot-merge acts as an origin, so set ti->max_io_len
2019 ti
->max_io_len
= get_origin_minimum_chunksize(s
->origin
->bdev
);
2024 static void snapshot_status(struct dm_target
*ti
, status_type_t type
,
2025 unsigned status_flags
, char *result
, unsigned maxlen
)
2028 struct dm_snapshot
*snap
= ti
->private;
2031 case STATUSTYPE_INFO
:
2033 mutex_lock(&snap
->lock
);
2037 else if (snap
->merge_failed
)
2038 DMEMIT("Merge failed");
2039 else if (snap
->snapshot_overflowed
)
2042 if (snap
->store
->type
->usage
) {
2043 sector_t total_sectors
, sectors_allocated
,
2045 snap
->store
->type
->usage(snap
->store
,
2049 DMEMIT("%llu/%llu %llu",
2050 (unsigned long long)sectors_allocated
,
2051 (unsigned long long)total_sectors
,
2052 (unsigned long long)metadata_sectors
);
2058 mutex_unlock(&snap
->lock
);
2062 case STATUSTYPE_TABLE
:
2064 * kdevname returns a static pointer so we need
2065 * to make private copies if the output is to
2068 DMEMIT("%s %s", snap
->origin
->name
, snap
->cow
->name
);
2069 snap
->store
->type
->status(snap
->store
, type
, result
+ sz
,
2075 static int snapshot_iterate_devices(struct dm_target
*ti
,
2076 iterate_devices_callout_fn fn
, void *data
)
2078 struct dm_snapshot
*snap
= ti
->private;
2081 r
= fn(ti
, snap
->origin
, 0, ti
->len
, data
);
2084 r
= fn(ti
, snap
->cow
, 0, get_dev_size(snap
->cow
->bdev
), data
);
2090 /*-----------------------------------------------------------------
2092 *---------------------------------------------------------------*/
2095 * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any
2096 * supplied bio was ignored. The caller may submit it immediately.
2097 * (No remapping actually occurs as the origin is always a direct linear
2100 * If further exceptions are required, DM_MAPIO_SUBMITTED is returned
2101 * and any supplied bio is added to a list to be submitted once all
2102 * the necessary exceptions exist.
2104 static int __origin_write(struct list_head
*snapshots
, sector_t sector
,
2107 int r
= DM_MAPIO_REMAPPED
;
2108 struct dm_snapshot
*snap
;
2109 struct dm_exception
*e
;
2110 struct dm_snap_pending_exception
*pe
;
2111 struct dm_snap_pending_exception
*pe_to_start_now
= NULL
;
2112 struct dm_snap_pending_exception
*pe_to_start_last
= NULL
;
2115 /* Do all the snapshots on this origin */
2116 list_for_each_entry (snap
, snapshots
, list
) {
2118 * Don't make new exceptions in a merging snapshot
2119 * because it has effectively been deleted
2121 if (dm_target_is_snapshot_merge(snap
->ti
))
2124 mutex_lock(&snap
->lock
);
2126 /* Only deal with valid and active snapshots */
2127 if (!snap
->valid
|| !snap
->active
)
2130 /* Nothing to do if writing beyond end of snapshot */
2131 if (sector
>= dm_table_get_size(snap
->ti
->table
))
2135 * Remember, different snapshots can have
2136 * different chunk sizes.
2138 chunk
= sector_to_chunk(snap
->store
, sector
);
2141 * Check exception table to see if block
2142 * is already remapped in this snapshot
2143 * and trigger an exception if not.
2145 e
= dm_lookup_exception(&snap
->complete
, chunk
);
2149 pe
= __lookup_pending_exception(snap
, chunk
);
2151 mutex_unlock(&snap
->lock
);
2152 pe
= alloc_pending_exception(snap
);
2153 mutex_lock(&snap
->lock
);
2156 free_pending_exception(pe
);
2160 e
= dm_lookup_exception(&snap
->complete
, chunk
);
2162 free_pending_exception(pe
);
2166 pe
= __find_pending_exception(snap
, pe
, chunk
);
2168 __invalidate_snapshot(snap
, -ENOMEM
);
2173 r
= DM_MAPIO_SUBMITTED
;
2176 * If an origin bio was supplied, queue it to wait for the
2177 * completion of this exception, and start this one last,
2178 * at the end of the function.
2181 bio_list_add(&pe
->origin_bios
, bio
);
2186 pe_to_start_last
= pe
;
2192 pe_to_start_now
= pe
;
2196 mutex_unlock(&snap
->lock
);
2198 if (pe_to_start_now
) {
2199 start_copy(pe_to_start_now
);
2200 pe_to_start_now
= NULL
;
2205 * Submit the exception against which the bio is queued last,
2206 * to give the other exceptions a head start.
2208 if (pe_to_start_last
)
2209 start_copy(pe_to_start_last
);
2215 * Called on a write from the origin driver.
2217 static int do_origin(struct dm_dev
*origin
, struct bio
*bio
)
2220 int r
= DM_MAPIO_REMAPPED
;
2222 down_read(&_origins_lock
);
2223 o
= __lookup_origin(origin
->bdev
);
2225 r
= __origin_write(&o
->snapshots
, bio
->bi_iter
.bi_sector
, bio
);
2226 up_read(&_origins_lock
);
2232 * Trigger exceptions in all non-merging snapshots.
2234 * The chunk size of the merging snapshot may be larger than the chunk
2235 * size of some other snapshot so we may need to reallocate multiple
2236 * chunks in other snapshots.
2238 * We scan all the overlapping exceptions in the other snapshots.
2239 * Returns 1 if anything was reallocated and must be waited for,
2240 * otherwise returns 0.
2242 * size must be a multiple of merging_snap's chunk_size.
2244 static int origin_write_extent(struct dm_snapshot
*merging_snap
,
2245 sector_t sector
, unsigned size
)
2252 * The origin's __minimum_chunk_size() got stored in max_io_len
2253 * by snapshot_merge_resume().
2255 down_read(&_origins_lock
);
2256 o
= __lookup_origin(merging_snap
->origin
->bdev
);
2257 for (n
= 0; n
< size
; n
+= merging_snap
->ti
->max_io_len
)
2258 if (__origin_write(&o
->snapshots
, sector
+ n
, NULL
) ==
2261 up_read(&_origins_lock
);
2267 * Origin: maps a linear range of a device, with hooks for snapshotting.
2271 * Construct an origin mapping: <dev_path>
2272 * The context for an origin is merely a 'struct dm_dev *'
2273 * pointing to the real device.
2275 static int origin_ctr(struct dm_target
*ti
, unsigned int argc
, char **argv
)
2278 struct dm_origin
*o
;
2281 ti
->error
= "origin: incorrect number of arguments";
2285 o
= kmalloc(sizeof(struct dm_origin
), GFP_KERNEL
);
2287 ti
->error
= "Cannot allocate private origin structure";
2292 r
= dm_get_device(ti
, argv
[0], dm_table_get_mode(ti
->table
), &o
->dev
);
2294 ti
->error
= "Cannot get target device";
2300 ti
->num_flush_bios
= 1;
2310 static void origin_dtr(struct dm_target
*ti
)
2312 struct dm_origin
*o
= ti
->private;
2314 dm_put_device(ti
, o
->dev
);
2318 static int origin_map(struct dm_target
*ti
, struct bio
*bio
)
2320 struct dm_origin
*o
= ti
->private;
2321 unsigned available_sectors
;
2323 bio_set_dev(bio
, o
->dev
->bdev
);
2325 if (unlikely(bio
->bi_opf
& REQ_PREFLUSH
))
2326 return DM_MAPIO_REMAPPED
;
2328 if (bio_data_dir(bio
) != WRITE
)
2329 return DM_MAPIO_REMAPPED
;
2331 available_sectors
= o
->split_boundary
-
2332 ((unsigned)bio
->bi_iter
.bi_sector
& (o
->split_boundary
- 1));
2334 if (bio_sectors(bio
) > available_sectors
)
2335 dm_accept_partial_bio(bio
, available_sectors
);
2337 /* Only tell snapshots if this is a write */
2338 return do_origin(o
->dev
, bio
);
2341 static long origin_dax_direct_access(struct dm_target
*ti
, pgoff_t pgoff
,
2342 long nr_pages
, void **kaddr
, pfn_t
*pfn
)
2344 DMWARN("device does not support dax.");
2349 * Set the target "max_io_len" field to the minimum of all the snapshots'
2352 static void origin_resume(struct dm_target
*ti
)
2354 struct dm_origin
*o
= ti
->private;
2356 o
->split_boundary
= get_origin_minimum_chunksize(o
->dev
->bdev
);
2358 down_write(&_origins_lock
);
2359 __insert_dm_origin(o
);
2360 up_write(&_origins_lock
);
2363 static void origin_postsuspend(struct dm_target
*ti
)
2365 struct dm_origin
*o
= ti
->private;
2367 down_write(&_origins_lock
);
2368 __remove_dm_origin(o
);
2369 up_write(&_origins_lock
);
2372 static void origin_status(struct dm_target
*ti
, status_type_t type
,
2373 unsigned status_flags
, char *result
, unsigned maxlen
)
2375 struct dm_origin
*o
= ti
->private;
2378 case STATUSTYPE_INFO
:
2382 case STATUSTYPE_TABLE
:
2383 snprintf(result
, maxlen
, "%s", o
->dev
->name
);
2388 static int origin_iterate_devices(struct dm_target
*ti
,
2389 iterate_devices_callout_fn fn
, void *data
)
2391 struct dm_origin
*o
= ti
->private;
2393 return fn(ti
, o
->dev
, 0, ti
->len
, data
);
2396 static struct target_type origin_target
= {
2397 .name
= "snapshot-origin",
2398 .version
= {1, 9, 0},
2399 .module
= THIS_MODULE
,
2403 .resume
= origin_resume
,
2404 .postsuspend
= origin_postsuspend
,
2405 .status
= origin_status
,
2406 .iterate_devices
= origin_iterate_devices
,
2407 .direct_access
= origin_dax_direct_access
,
2410 static struct target_type snapshot_target
= {
2412 .version
= {1, 15, 0},
2413 .module
= THIS_MODULE
,
2414 .ctr
= snapshot_ctr
,
2415 .dtr
= snapshot_dtr
,
2416 .map
= snapshot_map
,
2417 .end_io
= snapshot_end_io
,
2418 .preresume
= snapshot_preresume
,
2419 .resume
= snapshot_resume
,
2420 .status
= snapshot_status
,
2421 .iterate_devices
= snapshot_iterate_devices
,
2424 static struct target_type merge_target
= {
2425 .name
= dm_snapshot_merge_target_name
,
2426 .version
= {1, 4, 0},
2427 .module
= THIS_MODULE
,
2428 .ctr
= snapshot_ctr
,
2429 .dtr
= snapshot_dtr
,
2430 .map
= snapshot_merge_map
,
2431 .end_io
= snapshot_end_io
,
2432 .presuspend
= snapshot_merge_presuspend
,
2433 .preresume
= snapshot_preresume
,
2434 .resume
= snapshot_merge_resume
,
2435 .status
= snapshot_status
,
2436 .iterate_devices
= snapshot_iterate_devices
,
2439 static int __init
dm_snapshot_init(void)
2443 r
= dm_exception_store_init();
2445 DMERR("Failed to initialize exception stores");
2449 r
= init_origin_hash();
2451 DMERR("init_origin_hash failed.");
2452 goto bad_origin_hash
;
2455 exception_cache
= KMEM_CACHE(dm_exception
, 0);
2456 if (!exception_cache
) {
2457 DMERR("Couldn't create exception cache.");
2459 goto bad_exception_cache
;
2462 pending_cache
= KMEM_CACHE(dm_snap_pending_exception
, 0);
2463 if (!pending_cache
) {
2464 DMERR("Couldn't create pending cache.");
2466 goto bad_pending_cache
;
2469 r
= dm_register_target(&snapshot_target
);
2471 DMERR("snapshot target register failed %d", r
);
2472 goto bad_register_snapshot_target
;
2475 r
= dm_register_target(&origin_target
);
2477 DMERR("Origin target register failed %d", r
);
2478 goto bad_register_origin_target
;
2481 r
= dm_register_target(&merge_target
);
2483 DMERR("Merge target register failed %d", r
);
2484 goto bad_register_merge_target
;
2489 bad_register_merge_target
:
2490 dm_unregister_target(&origin_target
);
2491 bad_register_origin_target
:
2492 dm_unregister_target(&snapshot_target
);
2493 bad_register_snapshot_target
:
2494 kmem_cache_destroy(pending_cache
);
2496 kmem_cache_destroy(exception_cache
);
2497 bad_exception_cache
:
2500 dm_exception_store_exit();
2505 static void __exit
dm_snapshot_exit(void)
2507 dm_unregister_target(&snapshot_target
);
2508 dm_unregister_target(&origin_target
);
2509 dm_unregister_target(&merge_target
);
2512 kmem_cache_destroy(pending_cache
);
2513 kmem_cache_destroy(exception_cache
);
2515 dm_exception_store_exit();
2519 module_init(dm_snapshot_init
);
2520 module_exit(dm_snapshot_exit
);
2522 MODULE_DESCRIPTION(DM_NAME
" snapshot target");
2523 MODULE_AUTHOR("Joe Thornber");
2524 MODULE_LICENSE("GPL");
2525 MODULE_ALIAS("dm-snapshot-origin");
2526 MODULE_ALIAS("dm-snapshot-merge");