Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * dm-snapshot.c | |
3 | * | |
4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
5 | * | |
6 | * This file is released under the GPL. | |
7 | */ | |
8 | ||
9 | #include <linux/blkdev.h> | |
1da177e4 | 10 | #include <linux/device-mapper.h> |
90fa1527 | 11 | #include <linux/delay.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/init.h> | |
14 | #include <linux/kdev_t.h> | |
15 | #include <linux/list.h> | |
f79ae415 | 16 | #include <linux/list_bl.h> |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/vmalloc.h> | |
6f3c3f0a | 21 | #include <linux/log2.h> |
a765e20e | 22 | #include <linux/dm-kcopyd.h> |
721b1d98 | 23 | #include <linux/semaphore.h> |
1da177e4 | 24 | |
b735fede MP |
25 | #include "dm.h" |
26 | ||
aea53d92 | 27 | #include "dm-exception-store.h" |
1da177e4 | 28 | |
72d94861 AK |
29 | #define DM_MSG_PREFIX "snapshots" |
30 | ||
d698aa45 MP |
31 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
32 | ||
33 | #define dm_target_is_snapshot_merge(ti) \ | |
34 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
35 | ||
cd45daff MP |
36 | /* |
37 | * The size of the mempool used to track chunks in use. | |
38 | */ | |
39 | #define MIN_IOS 256 | |
40 | ||
ccc45ea8 JB |
41 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
42 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
43 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
44 | ||
191437a5 | 45 | struct dm_exception_table { |
ccc45ea8 JB |
46 | uint32_t hash_mask; |
47 | unsigned hash_shift; | |
f79ae415 | 48 | struct hlist_bl_head *table; |
ccc45ea8 JB |
49 | }; |
50 | ||
51 | struct dm_snapshot { | |
4ad8d880 | 52 | struct rw_semaphore lock; |
ccc45ea8 JB |
53 | |
54 | struct dm_dev *origin; | |
fc56f6fb MS |
55 | struct dm_dev *cow; |
56 | ||
57 | struct dm_target *ti; | |
ccc45ea8 JB |
58 | |
59 | /* List of snapshots per Origin */ | |
60 | struct list_head list; | |
61 | ||
d8ddb1cf MS |
62 | /* |
63 | * You can't use a snapshot if this is 0 (e.g. if full). | |
64 | * A snapshot-merge target never clears this. | |
65 | */ | |
ccc45ea8 JB |
66 | int valid; |
67 | ||
76c44f6d MP |
68 | /* |
69 | * The snapshot overflowed because of a write to the snapshot device. | |
70 | * We don't have to invalidate the snapshot in this case, but we need | |
71 | * to prevent further writes. | |
72 | */ | |
73 | int snapshot_overflowed; | |
74 | ||
ccc45ea8 JB |
75 | /* Origin writes don't trigger exceptions until this is set */ |
76 | int active; | |
77 | ||
ccc45ea8 JB |
78 | atomic_t pending_exceptions_count; |
79 | ||
3f1637f2 NT |
80 | spinlock_t pe_allocation_lock; |
81 | ||
82 | /* Protected by "pe_allocation_lock" */ | |
230c83af MP |
83 | sector_t exception_start_sequence; |
84 | ||
85 | /* Protected by kcopyd single-threaded callback */ | |
86 | sector_t exception_complete_sequence; | |
87 | ||
88 | /* | |
89 | * A list of pending exceptions that completed out of order. | |
90 | * Protected by kcopyd single-threaded callback. | |
91 | */ | |
3db2776d | 92 | struct rb_root out_of_order_tree; |
230c83af | 93 | |
6f1c819c | 94 | mempool_t pending_pool; |
924e600d | 95 | |
191437a5 JB |
96 | struct dm_exception_table pending; |
97 | struct dm_exception_table complete; | |
ccc45ea8 JB |
98 | |
99 | /* | |
100 | * pe_lock protects all pending_exception operations and access | |
101 | * as well as the snapshot_bios list. | |
102 | */ | |
103 | spinlock_t pe_lock; | |
104 | ||
924e600d MS |
105 | /* Chunks with outstanding reads */ |
106 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
107 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
108 | ||
ccc45ea8 JB |
109 | /* The on disk metadata handler */ |
110 | struct dm_exception_store *store; | |
111 | ||
721b1d98 NT |
112 | /* Maximum number of in-flight COW jobs. */ |
113 | struct semaphore cow_count; | |
114 | ||
ccc45ea8 JB |
115 | struct dm_kcopyd_client *kcopyd_client; |
116 | ||
924e600d MS |
117 | /* Wait for events based on state_bits */ |
118 | unsigned long state_bits; | |
119 | ||
120 | /* Range of chunks currently being merged. */ | |
121 | chunk_t first_merging_chunk; | |
122 | int num_merging_chunks; | |
1e03f97e | 123 | |
d8ddb1cf MS |
124 | /* |
125 | * The merge operation failed if this flag is set. | |
126 | * Failure modes are handled as follows: | |
127 | * - I/O error reading the header | |
128 | * => don't load the target; abort. | |
129 | * - Header does not have "valid" flag set | |
130 | * => use the origin; forget about the snapshot. | |
131 | * - I/O error when reading exceptions | |
132 | * => don't load the target; abort. | |
133 | * (We can't use the intermediate origin state.) | |
134 | * - I/O error while merging | |
135 | * => stop merging; set merge_failed; process I/O normally. | |
136 | */ | |
137 | int merge_failed; | |
138 | ||
9fe86254 MP |
139 | /* |
140 | * Incoming bios that overlap with chunks being merged must wait | |
141 | * for them to be committed. | |
142 | */ | |
143 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
144 | }; |
145 | ||
1e03f97e MP |
146 | /* |
147 | * state_bits: | |
148 | * RUNNING_MERGE - Merge operation is in progress. | |
149 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
150 | * cleared afterwards. | |
151 | */ | |
152 | #define RUNNING_MERGE 0 | |
153 | #define SHUTDOWN_MERGE 1 | |
154 | ||
721b1d98 NT |
155 | /* |
156 | * Maximum number of chunks being copied on write. | |
157 | * | |
158 | * The value was decided experimentally as a trade-off between memory | |
159 | * consumption, stalling the kernel's workqueues and maintaining a high enough | |
160 | * throughput. | |
161 | */ | |
162 | #define DEFAULT_COW_THRESHOLD 2048 | |
163 | ||
164 | static int cow_threshold = DEFAULT_COW_THRESHOLD; | |
165 | module_param_named(snapshot_cow_threshold, cow_threshold, int, 0644); | |
166 | MODULE_PARM_DESC(snapshot_cow_threshold, "Maximum number of chunks being copied on write"); | |
167 | ||
df5d2e90 MP |
168 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
169 | "A percentage of time allocated for copy on write"); | |
170 | ||
c2411045 MP |
171 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
172 | { | |
173 | return s->origin; | |
174 | } | |
175 | EXPORT_SYMBOL(dm_snap_origin); | |
176 | ||
fc56f6fb MS |
177 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
178 | { | |
179 | return s->cow; | |
180 | } | |
181 | EXPORT_SYMBOL(dm_snap_cow); | |
182 | ||
ccc45ea8 JB |
183 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
184 | chunk_t chunk) | |
185 | { | |
186 | return chunk << store->chunk_shift; | |
187 | } | |
188 | ||
189 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
190 | { | |
191 | /* | |
192 | * There is only ever one instance of a particular block | |
193 | * device so we can compare pointers safely. | |
194 | */ | |
195 | return lhs == rhs; | |
196 | } | |
197 | ||
028867ac | 198 | struct dm_snap_pending_exception { |
1d4989c8 | 199 | struct dm_exception e; |
1da177e4 LT |
200 | |
201 | /* | |
202 | * Origin buffers waiting for this to complete are held | |
203 | * in a bio list | |
204 | */ | |
205 | struct bio_list origin_bios; | |
206 | struct bio_list snapshot_bios; | |
207 | ||
1da177e4 LT |
208 | /* Pointer back to snapshot context */ |
209 | struct dm_snapshot *snap; | |
210 | ||
211 | /* | |
212 | * 1 indicates the exception has already been sent to | |
213 | * kcopyd. | |
214 | */ | |
215 | int started; | |
a6e50b40 | 216 | |
230c83af MP |
217 | /* There was copying error. */ |
218 | int copy_error; | |
219 | ||
220 | /* A sequence number, it is used for in-order completion. */ | |
221 | sector_t exception_sequence; | |
222 | ||
3db2776d | 223 | struct rb_node out_of_order_node; |
230c83af | 224 | |
a6e50b40 MP |
225 | /* |
226 | * For writing a complete chunk, bypassing the copy. | |
227 | */ | |
228 | struct bio *full_bio; | |
229 | bio_end_io_t *full_bio_end_io; | |
1da177e4 LT |
230 | }; |
231 | ||
232 | /* | |
233 | * Hash table mapping origin volumes to lists of snapshots and | |
234 | * a lock to protect it | |
235 | */ | |
e18b890b CL |
236 | static struct kmem_cache *exception_cache; |
237 | static struct kmem_cache *pending_cache; | |
1da177e4 | 238 | |
cd45daff MP |
239 | struct dm_snap_tracked_chunk { |
240 | struct hlist_node node; | |
241 | chunk_t chunk; | |
242 | }; | |
243 | ||
ee18026a MP |
244 | static void init_tracked_chunk(struct bio *bio) |
245 | { | |
246 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
247 | INIT_HLIST_NODE(&c->node); | |
248 | } | |
249 | ||
250 | static bool is_bio_tracked(struct bio *bio) | |
251 | { | |
252 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
253 | return !hlist_unhashed(&c->node); | |
254 | } | |
255 | ||
256 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 257 | { |
42bc954f | 258 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
259 | |
260 | c->chunk = chunk; | |
261 | ||
9aa0c0e6 | 262 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
263 | hlist_add_head(&c->node, |
264 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 265 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
266 | } |
267 | ||
ee18026a | 268 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 269 | { |
ee18026a | 270 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
271 | unsigned long flags; |
272 | ||
273 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
274 | hlist_del(&c->node); | |
275 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
276 | } |
277 | ||
a8d41b59 MP |
278 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
279 | { | |
280 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
281 | int found = 0; |
282 | ||
283 | spin_lock_irq(&s->tracked_chunk_lock); | |
284 | ||
b67bfe0d | 285 | hlist_for_each_entry(c, |
a8d41b59 MP |
286 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
287 | if (c->chunk == chunk) { | |
288 | found = 1; | |
289 | break; | |
290 | } | |
291 | } | |
292 | ||
293 | spin_unlock_irq(&s->tracked_chunk_lock); | |
294 | ||
295 | return found; | |
296 | } | |
297 | ||
615d1eb9 MS |
298 | /* |
299 | * This conflicting I/O is extremely improbable in the caller, | |
300 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
301 | */ | |
302 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
303 | { | |
304 | while (__chunk_is_tracked(s, chunk)) | |
305 | msleep(1); | |
306 | } | |
307 | ||
1da177e4 LT |
308 | /* |
309 | * One of these per registered origin, held in the snapshot_origins hash | |
310 | */ | |
311 | struct origin { | |
312 | /* The origin device */ | |
313 | struct block_device *bdev; | |
314 | ||
315 | struct list_head hash_list; | |
316 | ||
317 | /* List of snapshots for this origin */ | |
318 | struct list_head snapshots; | |
319 | }; | |
320 | ||
b735fede MP |
321 | /* |
322 | * This structure is allocated for each origin target | |
323 | */ | |
324 | struct dm_origin { | |
325 | struct dm_dev *dev; | |
326 | struct dm_target *ti; | |
327 | unsigned split_boundary; | |
328 | struct list_head hash_list; | |
329 | }; | |
330 | ||
1da177e4 LT |
331 | /* |
332 | * Size of the hash table for origin volumes. If we make this | |
333 | * the size of the minors list then it should be nearly perfect | |
334 | */ | |
335 | #define ORIGIN_HASH_SIZE 256 | |
336 | #define ORIGIN_MASK 0xFF | |
337 | static struct list_head *_origins; | |
b735fede | 338 | static struct list_head *_dm_origins; |
1da177e4 LT |
339 | static struct rw_semaphore _origins_lock; |
340 | ||
73dfd078 MP |
341 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
342 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
343 | static uint64_t _pending_exceptions_done_count; | |
344 | ||
1da177e4 LT |
345 | static int init_origin_hash(void) |
346 | { | |
347 | int i; | |
348 | ||
6da2ec56 KC |
349 | _origins = kmalloc_array(ORIGIN_HASH_SIZE, sizeof(struct list_head), |
350 | GFP_KERNEL); | |
1da177e4 | 351 | if (!_origins) { |
b735fede | 352 | DMERR("unable to allocate memory for _origins"); |
1da177e4 LT |
353 | return -ENOMEM; |
354 | } | |
1da177e4 LT |
355 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
356 | INIT_LIST_HEAD(_origins + i); | |
b735fede | 357 | |
6da2ec56 KC |
358 | _dm_origins = kmalloc_array(ORIGIN_HASH_SIZE, |
359 | sizeof(struct list_head), | |
360 | GFP_KERNEL); | |
b735fede MP |
361 | if (!_dm_origins) { |
362 | DMERR("unable to allocate memory for _dm_origins"); | |
363 | kfree(_origins); | |
364 | return -ENOMEM; | |
365 | } | |
366 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
367 | INIT_LIST_HEAD(_dm_origins + i); | |
368 | ||
1da177e4 LT |
369 | init_rwsem(&_origins_lock); |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void exit_origin_hash(void) | |
375 | { | |
376 | kfree(_origins); | |
b735fede | 377 | kfree(_dm_origins); |
1da177e4 LT |
378 | } |
379 | ||
028867ac | 380 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
381 | { |
382 | return bdev->bd_dev & ORIGIN_MASK; | |
383 | } | |
384 | ||
385 | static struct origin *__lookup_origin(struct block_device *origin) | |
386 | { | |
387 | struct list_head *ol; | |
388 | struct origin *o; | |
389 | ||
390 | ol = &_origins[origin_hash(origin)]; | |
391 | list_for_each_entry (o, ol, hash_list) | |
392 | if (bdev_equal(o->bdev, origin)) | |
393 | return o; | |
394 | ||
395 | return NULL; | |
396 | } | |
397 | ||
398 | static void __insert_origin(struct origin *o) | |
399 | { | |
400 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
401 | list_add_tail(&o->hash_list, sl); | |
402 | } | |
403 | ||
b735fede MP |
404 | static struct dm_origin *__lookup_dm_origin(struct block_device *origin) |
405 | { | |
406 | struct list_head *ol; | |
407 | struct dm_origin *o; | |
408 | ||
409 | ol = &_dm_origins[origin_hash(origin)]; | |
410 | list_for_each_entry (o, ol, hash_list) | |
411 | if (bdev_equal(o->dev->bdev, origin)) | |
412 | return o; | |
413 | ||
414 | return NULL; | |
415 | } | |
416 | ||
417 | static void __insert_dm_origin(struct dm_origin *o) | |
418 | { | |
419 | struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)]; | |
420 | list_add_tail(&o->hash_list, sl); | |
421 | } | |
422 | ||
423 | static void __remove_dm_origin(struct dm_origin *o) | |
424 | { | |
425 | list_del(&o->hash_list); | |
426 | } | |
427 | ||
c1f0c183 MS |
428 | /* |
429 | * _origins_lock must be held when calling this function. | |
430 | * Returns number of snapshots registered using the supplied cow device, plus: | |
431 | * snap_src - a snapshot suitable for use as a source of exception handover | |
432 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
433 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
434 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 435 | * |
9d3b15c4 | 436 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
437 | * 0: NULL, NULL - first new snapshot |
438 | * 1: snap_src, NULL - normal snapshot | |
439 | * 2: snap_src, snap_dest - waiting for handover | |
440 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
441 | * 1: NULL, snap_dest - source got destroyed without handover | |
442 | */ | |
443 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
444 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
445 | struct dm_snapshot **snap_dest, |
446 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
447 | { |
448 | struct dm_snapshot *s; | |
449 | struct origin *o; | |
450 | int count = 0; | |
451 | int active; | |
452 | ||
453 | o = __lookup_origin(snap->origin->bdev); | |
454 | if (!o) | |
455 | goto out; | |
456 | ||
457 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
458 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
459 | *snap_merge = s; | |
c1f0c183 MS |
460 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
461 | continue; | |
462 | ||
4ad8d880 | 463 | down_read(&s->lock); |
c1f0c183 | 464 | active = s->active; |
4ad8d880 | 465 | up_read(&s->lock); |
c1f0c183 MS |
466 | |
467 | if (active) { | |
468 | if (snap_src) | |
469 | *snap_src = s; | |
470 | } else if (snap_dest) | |
471 | *snap_dest = s; | |
472 | ||
473 | count++; | |
474 | } | |
475 | ||
476 | out: | |
477 | return count; | |
478 | } | |
479 | ||
480 | /* | |
481 | * On success, returns 1 if this snapshot is a handover destination, | |
482 | * otherwise returns 0. | |
483 | */ | |
484 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
485 | { | |
486 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 487 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
488 | |
489 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
490 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
491 | &snap_merge) == 2) || | |
c1f0c183 MS |
492 | snap_dest) { |
493 | snap->ti->error = "Snapshot cow pairing for exception " | |
494 | "table handover failed"; | |
495 | return -EINVAL; | |
496 | } | |
497 | ||
498 | /* | |
499 | * If no snap_src was found, snap cannot become a handover | |
500 | * destination. | |
501 | */ | |
502 | if (!snap_src) | |
503 | return 0; | |
504 | ||
9d3b15c4 MP |
505 | /* |
506 | * Non-snapshot-merge handover? | |
507 | */ | |
508 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
509 | return 1; | |
510 | ||
511 | /* | |
512 | * Do not allow more than one merging snapshot. | |
513 | */ | |
514 | if (snap_merge) { | |
515 | snap->ti->error = "A snapshot is already merging."; | |
516 | return -EINVAL; | |
517 | } | |
518 | ||
1e03f97e MP |
519 | if (!snap_src->store->type->prepare_merge || |
520 | !snap_src->store->type->commit_merge) { | |
521 | snap->ti->error = "Snapshot exception store does not " | |
522 | "support snapshot-merge."; | |
523 | return -EINVAL; | |
524 | } | |
525 | ||
c1f0c183 MS |
526 | return 1; |
527 | } | |
528 | ||
529 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
530 | { | |
531 | struct dm_snapshot *l; | |
532 | ||
533 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
534 | list_for_each_entry(l, &o->snapshots, list) | |
535 | if (l->store->chunk_size < s->store->chunk_size) | |
536 | break; | |
537 | list_add_tail(&s->list, &l->list); | |
538 | } | |
539 | ||
1da177e4 LT |
540 | /* |
541 | * Make a note of the snapshot and its origin so we can look it | |
542 | * up when the origin has a write on it. | |
c1f0c183 MS |
543 | * |
544 | * Also validate snapshot exception store handovers. | |
545 | * On success, returns 1 if this registration is a handover destination, | |
546 | * otherwise returns 0. | |
1da177e4 LT |
547 | */ |
548 | static int register_snapshot(struct dm_snapshot *snap) | |
549 | { | |
c1f0c183 | 550 | struct origin *o, *new_o = NULL; |
1da177e4 | 551 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 552 | int r = 0; |
1da177e4 | 553 | |
60c856c8 MP |
554 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
555 | if (!new_o) | |
556 | return -ENOMEM; | |
557 | ||
1da177e4 | 558 | down_write(&_origins_lock); |
1da177e4 | 559 | |
c1f0c183 MS |
560 | r = __validate_exception_handover(snap); |
561 | if (r < 0) { | |
562 | kfree(new_o); | |
563 | goto out; | |
564 | } | |
565 | ||
566 | o = __lookup_origin(bdev); | |
60c856c8 MP |
567 | if (o) |
568 | kfree(new_o); | |
569 | else { | |
1da177e4 | 570 | /* New origin */ |
60c856c8 | 571 | o = new_o; |
1da177e4 LT |
572 | |
573 | /* Initialise the struct */ | |
574 | INIT_LIST_HEAD(&o->snapshots); | |
575 | o->bdev = bdev; | |
576 | ||
577 | __insert_origin(o); | |
578 | } | |
579 | ||
c1f0c183 MS |
580 | __insert_snapshot(o, snap); |
581 | ||
582 | out: | |
583 | up_write(&_origins_lock); | |
584 | ||
585 | return r; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Move snapshot to correct place in list according to chunk size. | |
590 | */ | |
591 | static void reregister_snapshot(struct dm_snapshot *s) | |
592 | { | |
593 | struct block_device *bdev = s->origin->bdev; | |
594 | ||
595 | down_write(&_origins_lock); | |
596 | ||
597 | list_del(&s->list); | |
598 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
599 | |
600 | up_write(&_origins_lock); | |
1da177e4 LT |
601 | } |
602 | ||
603 | static void unregister_snapshot(struct dm_snapshot *s) | |
604 | { | |
605 | struct origin *o; | |
606 | ||
607 | down_write(&_origins_lock); | |
608 | o = __lookup_origin(s->origin->bdev); | |
609 | ||
610 | list_del(&s->list); | |
c1f0c183 | 611 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
612 | list_del(&o->hash_list); |
613 | kfree(o); | |
614 | } | |
615 | ||
616 | up_write(&_origins_lock); | |
617 | } | |
618 | ||
619 | /* | |
620 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
621 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
622 | * some consecutive chunks to be grouped together. | |
1da177e4 | 623 | */ |
f79ae415 NT |
624 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk); |
625 | ||
626 | /* Lock to protect access to the completed and pending exception hash tables. */ | |
627 | struct dm_exception_table_lock { | |
628 | struct hlist_bl_head *complete_slot; | |
629 | struct hlist_bl_head *pending_slot; | |
630 | }; | |
631 | ||
632 | static void dm_exception_table_lock_init(struct dm_snapshot *s, chunk_t chunk, | |
633 | struct dm_exception_table_lock *lock) | |
634 | { | |
635 | struct dm_exception_table *complete = &s->complete; | |
636 | struct dm_exception_table *pending = &s->pending; | |
637 | ||
638 | lock->complete_slot = &complete->table[exception_hash(complete, chunk)]; | |
639 | lock->pending_slot = &pending->table[exception_hash(pending, chunk)]; | |
640 | } | |
641 | ||
642 | static void dm_exception_table_lock(struct dm_exception_table_lock *lock) | |
643 | { | |
644 | hlist_bl_lock(lock->complete_slot); | |
645 | hlist_bl_lock(lock->pending_slot); | |
646 | } | |
647 | ||
648 | static void dm_exception_table_unlock(struct dm_exception_table_lock *lock) | |
649 | { | |
650 | hlist_bl_unlock(lock->pending_slot); | |
651 | hlist_bl_unlock(lock->complete_slot); | |
652 | } | |
653 | ||
3510cb94 JB |
654 | static int dm_exception_table_init(struct dm_exception_table *et, |
655 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
656 | { |
657 | unsigned int i; | |
658 | ||
d74f81f8 | 659 | et->hash_shift = hash_shift; |
1da177e4 | 660 | et->hash_mask = size - 1; |
f79ae415 | 661 | et->table = dm_vcalloc(size, sizeof(struct hlist_bl_head)); |
1da177e4 LT |
662 | if (!et->table) |
663 | return -ENOMEM; | |
664 | ||
665 | for (i = 0; i < size; i++) | |
f79ae415 | 666 | INIT_HLIST_BL_HEAD(et->table + i); |
1da177e4 LT |
667 | |
668 | return 0; | |
669 | } | |
670 | ||
3510cb94 JB |
671 | static void dm_exception_table_exit(struct dm_exception_table *et, |
672 | struct kmem_cache *mem) | |
1da177e4 | 673 | { |
f79ae415 NT |
674 | struct hlist_bl_head *slot; |
675 | struct dm_exception *ex; | |
676 | struct hlist_bl_node *pos, *n; | |
1da177e4 LT |
677 | int i, size; |
678 | ||
679 | size = et->hash_mask + 1; | |
680 | for (i = 0; i < size; i++) { | |
681 | slot = et->table + i; | |
682 | ||
f79ae415 | 683 | hlist_bl_for_each_entry_safe(ex, pos, n, slot, hash_list) |
1da177e4 LT |
684 | kmem_cache_free(mem, ex); |
685 | } | |
686 | ||
687 | vfree(et->table); | |
688 | } | |
689 | ||
191437a5 | 690 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 691 | { |
d74f81f8 | 692 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
693 | } |
694 | ||
3510cb94 | 695 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 | 696 | { |
f79ae415 | 697 | hlist_bl_del(&e->hash_list); |
1da177e4 LT |
698 | } |
699 | ||
700 | /* | |
701 | * Return the exception data for a sector, or NULL if not | |
702 | * remapped. | |
703 | */ | |
3510cb94 JB |
704 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
705 | chunk_t chunk) | |
1da177e4 | 706 | { |
f79ae415 NT |
707 | struct hlist_bl_head *slot; |
708 | struct hlist_bl_node *pos; | |
1d4989c8 | 709 | struct dm_exception *e; |
1da177e4 LT |
710 | |
711 | slot = &et->table[exception_hash(et, chunk)]; | |
f79ae415 | 712 | hlist_bl_for_each_entry(e, pos, slot, hash_list) |
d74f81f8 MB |
713 | if (chunk >= e->old_chunk && |
714 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
715 | return e; |
716 | ||
717 | return NULL; | |
718 | } | |
719 | ||
119bc547 | 720 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 721 | { |
1d4989c8 | 722 | struct dm_exception *e; |
1da177e4 | 723 | |
119bc547 MP |
724 | e = kmem_cache_alloc(exception_cache, gfp); |
725 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
726 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
727 | ||
728 | return e; | |
729 | } | |
730 | ||
3510cb94 | 731 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
732 | { |
733 | kmem_cache_free(exception_cache, e); | |
734 | } | |
735 | ||
92e86812 | 736 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 737 | { |
6f1c819c | 738 | struct dm_snap_pending_exception *pe = mempool_alloc(&s->pending_pool, |
92e86812 MP |
739 | GFP_NOIO); |
740 | ||
879129d2 | 741 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
742 | pe->snap = s; |
743 | ||
744 | return pe; | |
1da177e4 LT |
745 | } |
746 | ||
028867ac | 747 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 748 | { |
879129d2 MP |
749 | struct dm_snapshot *s = pe->snap; |
750 | ||
6f1c819c | 751 | mempool_free(pe, &s->pending_pool); |
4e857c58 | 752 | smp_mb__before_atomic(); |
879129d2 | 753 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
754 | } |
755 | ||
3510cb94 JB |
756 | static void dm_insert_exception(struct dm_exception_table *eh, |
757 | struct dm_exception *new_e) | |
d74f81f8 | 758 | { |
f79ae415 NT |
759 | struct hlist_bl_head *l; |
760 | struct hlist_bl_node *pos; | |
1d4989c8 | 761 | struct dm_exception *e = NULL; |
d74f81f8 MB |
762 | |
763 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
764 | ||
765 | /* Add immediately if this table doesn't support consecutive chunks */ | |
766 | if (!eh->hash_shift) | |
767 | goto out; | |
768 | ||
769 | /* List is ordered by old_chunk */ | |
f79ae415 | 770 | hlist_bl_for_each_entry(e, pos, l, hash_list) { |
d74f81f8 MB |
771 | /* Insert after an existing chunk? */ |
772 | if (new_e->old_chunk == (e->old_chunk + | |
773 | dm_consecutive_chunk_count(e) + 1) && | |
774 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
775 | dm_consecutive_chunk_count(e) + 1)) { | |
776 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 777 | free_completed_exception(new_e); |
d74f81f8 MB |
778 | return; |
779 | } | |
780 | ||
781 | /* Insert before an existing chunk? */ | |
782 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
783 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
784 | dm_consecutive_chunk_count_inc(e); | |
785 | e->old_chunk--; | |
786 | e->new_chunk--; | |
3510cb94 | 787 | free_completed_exception(new_e); |
d74f81f8 MB |
788 | return; |
789 | } | |
790 | ||
f79ae415 | 791 | if (new_e->old_chunk < e->old_chunk) |
d74f81f8 MB |
792 | break; |
793 | } | |
794 | ||
795 | out: | |
f79ae415 NT |
796 | if (!e) { |
797 | /* | |
798 | * Either the table doesn't support consecutive chunks or slot | |
799 | * l is empty. | |
800 | */ | |
801 | hlist_bl_add_head(&new_e->hash_list, l); | |
802 | } else if (new_e->old_chunk < e->old_chunk) { | |
803 | /* Add before an existing exception */ | |
804 | hlist_bl_add_before(&new_e->hash_list, &e->hash_list); | |
805 | } else { | |
806 | /* Add to l's tail: e is the last exception in this slot */ | |
807 | hlist_bl_add_behind(&new_e->hash_list, &e->hash_list); | |
808 | } | |
d74f81f8 MB |
809 | } |
810 | ||
a159c1ac JB |
811 | /* |
812 | * Callback used by the exception stores to load exceptions when | |
813 | * initialising. | |
814 | */ | |
815 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 816 | { |
f79ae415 | 817 | struct dm_exception_table_lock lock; |
a159c1ac | 818 | struct dm_snapshot *s = context; |
1d4989c8 | 819 | struct dm_exception *e; |
1da177e4 | 820 | |
119bc547 | 821 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
822 | if (!e) |
823 | return -ENOMEM; | |
824 | ||
825 | e->old_chunk = old; | |
d74f81f8 MB |
826 | |
827 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 828 | e->new_chunk = new; |
d74f81f8 | 829 | |
f79ae415 NT |
830 | /* |
831 | * Although there is no need to lock access to the exception tables | |
832 | * here, if we don't then hlist_bl_add_head(), called by | |
833 | * dm_insert_exception(), will complain about accessing the | |
834 | * corresponding list without locking it first. | |
835 | */ | |
836 | dm_exception_table_lock_init(s, old, &lock); | |
837 | ||
838 | dm_exception_table_lock(&lock); | |
3510cb94 | 839 | dm_insert_exception(&s->complete, e); |
f79ae415 | 840 | dm_exception_table_unlock(&lock); |
d74f81f8 | 841 | |
1da177e4 LT |
842 | return 0; |
843 | } | |
844 | ||
7e201b35 MP |
845 | /* |
846 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
847 | * Return zero if the origin has no snapshots. | |
848 | */ | |
542f9038 | 849 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
850 | { |
851 | struct dm_snapshot *snap; | |
852 | unsigned chunk_size = 0; | |
853 | ||
854 | if (o) | |
855 | list_for_each_entry(snap, &o->snapshots, list) | |
856 | chunk_size = min_not_zero(chunk_size, | |
857 | snap->store->chunk_size); | |
858 | ||
542f9038 | 859 | return (uint32_t) chunk_size; |
7e201b35 MP |
860 | } |
861 | ||
1da177e4 LT |
862 | /* |
863 | * Hard coded magic. | |
864 | */ | |
865 | static int calc_max_buckets(void) | |
866 | { | |
867 | /* use a fixed size of 2MB */ | |
868 | unsigned long mem = 2 * 1024 * 1024; | |
f79ae415 | 869 | mem /= sizeof(struct hlist_bl_head); |
1da177e4 LT |
870 | |
871 | return mem; | |
872 | } | |
873 | ||
1da177e4 LT |
874 | /* |
875 | * Allocate room for a suitable hash table. | |
876 | */ | |
fee1998e | 877 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 878 | { |
60e356f3 | 879 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
880 | |
881 | /* | |
882 | * Calculate based on the size of the original volume or | |
883 | * the COW volume... | |
884 | */ | |
fc56f6fb | 885 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
886 | max_buckets = calc_max_buckets(); |
887 | ||
60e356f3 | 888 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
889 | hash_size = min(hash_size, max_buckets); |
890 | ||
8e87b9b8 MP |
891 | if (hash_size < 64) |
892 | hash_size = 64; | |
8defd830 | 893 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
894 | if (dm_exception_table_init(&s->complete, hash_size, |
895 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
896 | return -ENOMEM; |
897 | ||
898 | /* | |
899 | * Allocate hash table for in-flight exceptions | |
900 | * Make this smaller than the real hash table | |
901 | */ | |
902 | hash_size >>= 3; | |
903 | if (hash_size < 64) | |
904 | hash_size = 64; | |
905 | ||
3510cb94 JB |
906 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
907 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
908 | return -ENOMEM; |
909 | } | |
910 | ||
911 | return 0; | |
912 | } | |
913 | ||
1e03f97e MP |
914 | static void merge_shutdown(struct dm_snapshot *s) |
915 | { | |
916 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 917 | smp_mb__after_atomic(); |
1e03f97e MP |
918 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
919 | } | |
920 | ||
9fe86254 MP |
921 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
922 | { | |
923 | s->first_merging_chunk = 0; | |
924 | s->num_merging_chunks = 0; | |
925 | ||
926 | return bio_list_get(&s->bios_queued_during_merge); | |
927 | } | |
928 | ||
1e03f97e MP |
929 | /* |
930 | * Remove one chunk from the index of completed exceptions. | |
931 | */ | |
932 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
933 | chunk_t old_chunk) | |
934 | { | |
935 | struct dm_exception *e; | |
936 | ||
1e03f97e MP |
937 | e = dm_lookup_exception(&s->complete, old_chunk); |
938 | if (!e) { | |
939 | DMERR("Corruption detected: exception for block %llu is " | |
940 | "on disk but not in memory", | |
941 | (unsigned long long)old_chunk); | |
942 | return -EINVAL; | |
943 | } | |
944 | ||
945 | /* | |
946 | * If this is the only chunk using this exception, remove exception. | |
947 | */ | |
948 | if (!dm_consecutive_chunk_count(e)) { | |
949 | dm_remove_exception(e); | |
950 | free_completed_exception(e); | |
951 | return 0; | |
952 | } | |
953 | ||
954 | /* | |
955 | * The chunk may be either at the beginning or the end of a | |
956 | * group of consecutive chunks - never in the middle. We are | |
957 | * removing chunks in the opposite order to that in which they | |
958 | * were added, so this should always be true. | |
959 | * Decrement the consecutive chunk counter and adjust the | |
960 | * starting point if necessary. | |
961 | */ | |
962 | if (old_chunk == e->old_chunk) { | |
963 | e->old_chunk++; | |
964 | e->new_chunk++; | |
965 | } else if (old_chunk != e->old_chunk + | |
966 | dm_consecutive_chunk_count(e)) { | |
967 | DMERR("Attempt to merge block %llu from the " | |
968 | "middle of a chunk range [%llu - %llu]", | |
969 | (unsigned long long)old_chunk, | |
970 | (unsigned long long)e->old_chunk, | |
971 | (unsigned long long) | |
972 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
973 | return -EINVAL; | |
974 | } | |
975 | ||
976 | dm_consecutive_chunk_count_dec(e); | |
977 | ||
978 | return 0; | |
979 | } | |
980 | ||
9fe86254 MP |
981 | static void flush_bios(struct bio *bio); |
982 | ||
983 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 984 | { |
9fe86254 MP |
985 | struct bio *b = NULL; |
986 | int r; | |
987 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e | 988 | |
4ad8d880 | 989 | down_write(&s->lock); |
9fe86254 MP |
990 | |
991 | /* | |
992 | * Process chunks (and associated exceptions) in reverse order | |
993 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
994 | */ | |
995 | do { | |
996 | r = __remove_single_exception_chunk(s, old_chunk); | |
997 | if (r) | |
998 | goto out; | |
999 | } while (old_chunk-- > s->first_merging_chunk); | |
1000 | ||
1001 | b = __release_queued_bios_after_merge(s); | |
1002 | ||
1003 | out: | |
4ad8d880 | 1004 | up_write(&s->lock); |
9fe86254 MP |
1005 | if (b) |
1006 | flush_bios(b); | |
1e03f97e MP |
1007 | |
1008 | return r; | |
1009 | } | |
1010 | ||
73dfd078 MP |
1011 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
1012 | sector_t sector, unsigned chunk_size); | |
1013 | ||
1e03f97e MP |
1014 | static void merge_callback(int read_err, unsigned long write_err, |
1015 | void *context); | |
1016 | ||
73dfd078 MP |
1017 | static uint64_t read_pending_exceptions_done_count(void) |
1018 | { | |
1019 | uint64_t pending_exceptions_done; | |
1020 | ||
1021 | spin_lock(&_pending_exceptions_done_spinlock); | |
1022 | pending_exceptions_done = _pending_exceptions_done_count; | |
1023 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1024 | ||
1025 | return pending_exceptions_done; | |
1026 | } | |
1027 | ||
1028 | static void increment_pending_exceptions_done_count(void) | |
1029 | { | |
1030 | spin_lock(&_pending_exceptions_done_spinlock); | |
1031 | _pending_exceptions_done_count++; | |
1032 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1033 | ||
1034 | wake_up_all(&_pending_exceptions_done); | |
1035 | } | |
1036 | ||
1e03f97e MP |
1037 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
1038 | { | |
8a2d5286 | 1039 | int i, linear_chunks; |
1e03f97e MP |
1040 | chunk_t old_chunk, new_chunk; |
1041 | struct dm_io_region src, dest; | |
8a2d5286 | 1042 | sector_t io_size; |
73dfd078 | 1043 | uint64_t previous_count; |
1e03f97e MP |
1044 | |
1045 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
1046 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
1047 | goto shut; | |
1048 | ||
1049 | /* | |
1050 | * valid flag never changes during merge, so no lock required. | |
1051 | */ | |
1052 | if (!s->valid) { | |
1053 | DMERR("Snapshot is invalid: can't merge"); | |
1054 | goto shut; | |
1055 | } | |
1056 | ||
8a2d5286 MS |
1057 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
1058 | &new_chunk); | |
1059 | if (linear_chunks <= 0) { | |
d8ddb1cf | 1060 | if (linear_chunks < 0) { |
1e03f97e MP |
1061 | DMERR("Read error in exception store: " |
1062 | "shutting down merge"); | |
4ad8d880 | 1063 | down_write(&s->lock); |
d8ddb1cf | 1064 | s->merge_failed = 1; |
4ad8d880 | 1065 | up_write(&s->lock); |
d8ddb1cf | 1066 | } |
1e03f97e MP |
1067 | goto shut; |
1068 | } | |
1069 | ||
8a2d5286 MS |
1070 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
1071 | old_chunk = old_chunk + 1 - linear_chunks; | |
1072 | new_chunk = new_chunk + 1 - linear_chunks; | |
1073 | ||
1074 | /* | |
1075 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
1076 | * from the exception store to the origin | |
1077 | */ | |
1078 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 1079 | |
1e03f97e MP |
1080 | dest.bdev = s->origin->bdev; |
1081 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 1082 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
1083 | |
1084 | src.bdev = s->cow->bdev; | |
1085 | src.sector = chunk_to_sector(s->store, new_chunk); | |
1086 | src.count = dest.count; | |
1087 | ||
73dfd078 MP |
1088 | /* |
1089 | * Reallocate any exceptions needed in other snapshots then | |
1090 | * wait for the pending exceptions to complete. | |
1091 | * Each time any pending exception (globally on the system) | |
1092 | * completes we are woken and repeat the process to find out | |
1093 | * if we can proceed. While this may not seem a particularly | |
1094 | * efficient algorithm, it is not expected to have any | |
1095 | * significant impact on performance. | |
1096 | */ | |
1097 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 1098 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
1099 | wait_event(_pending_exceptions_done, |
1100 | (read_pending_exceptions_done_count() != | |
1101 | previous_count)); | |
1102 | /* Retry after the wait, until all exceptions are done. */ | |
1103 | previous_count = read_pending_exceptions_done_count(); | |
1104 | } | |
1105 | ||
4ad8d880 | 1106 | down_write(&s->lock); |
9fe86254 | 1107 | s->first_merging_chunk = old_chunk; |
8a2d5286 | 1108 | s->num_merging_chunks = linear_chunks; |
4ad8d880 | 1109 | up_write(&s->lock); |
9fe86254 | 1110 | |
8a2d5286 MS |
1111 | /* Wait until writes to all 'linear_chunks' drain */ |
1112 | for (i = 0; i < linear_chunks; i++) | |
1113 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 1114 | |
1e03f97e MP |
1115 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
1116 | return; | |
1117 | ||
1118 | shut: | |
1119 | merge_shutdown(s); | |
1120 | } | |
1121 | ||
9fe86254 MP |
1122 | static void error_bios(struct bio *bio); |
1123 | ||
1e03f97e MP |
1124 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
1125 | { | |
1126 | struct dm_snapshot *s = context; | |
9fe86254 | 1127 | struct bio *b = NULL; |
1e03f97e MP |
1128 | |
1129 | if (read_err || write_err) { | |
1130 | if (read_err) | |
1131 | DMERR("Read error: shutting down merge."); | |
1132 | else | |
1133 | DMERR("Write error: shutting down merge."); | |
1134 | goto shut; | |
1135 | } | |
1136 | ||
9fe86254 MP |
1137 | if (s->store->type->commit_merge(s->store, |
1138 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1139 | DMERR("Write error in exception store: shutting down merge"); |
1140 | goto shut; | |
1141 | } | |
1142 | ||
9fe86254 MP |
1143 | if (remove_single_exception_chunk(s) < 0) |
1144 | goto shut; | |
1145 | ||
1e03f97e MP |
1146 | snapshot_merge_next_chunks(s); |
1147 | ||
1148 | return; | |
1149 | ||
1150 | shut: | |
4ad8d880 | 1151 | down_write(&s->lock); |
d8ddb1cf | 1152 | s->merge_failed = 1; |
9fe86254 | 1153 | b = __release_queued_bios_after_merge(s); |
4ad8d880 | 1154 | up_write(&s->lock); |
9fe86254 MP |
1155 | error_bios(b); |
1156 | ||
1e03f97e MP |
1157 | merge_shutdown(s); |
1158 | } | |
1159 | ||
1160 | static void start_merge(struct dm_snapshot *s) | |
1161 | { | |
1162 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1163 | snapshot_merge_next_chunks(s); | |
1164 | } | |
1165 | ||
1e03f97e MP |
1166 | /* |
1167 | * Stop the merging process and wait until it finishes. | |
1168 | */ | |
1169 | static void stop_merge(struct dm_snapshot *s) | |
1170 | { | |
1171 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1172 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1173 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1174 | } | |
1175 | ||
1da177e4 | 1176 | /* |
b0d3cc01 | 1177 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p|po|n> <chunk-size> |
1da177e4 LT |
1178 | */ |
1179 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1180 | { | |
1181 | struct dm_snapshot *s; | |
cd45daff | 1182 | int i; |
1da177e4 | 1183 | int r = -EINVAL; |
fc56f6fb | 1184 | char *origin_path, *cow_path; |
4df2bf46 | 1185 | dev_t origin_dev, cow_dev; |
55a62eef | 1186 | unsigned args_used, num_flush_bios = 1; |
10b8106a | 1187 | fmode_t origin_mode = FMODE_READ; |
1da177e4 | 1188 | |
4c7e3bf4 | 1189 | if (argc != 4) { |
72d94861 | 1190 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1191 | r = -EINVAL; |
fc56f6fb | 1192 | goto bad; |
1da177e4 LT |
1193 | } |
1194 | ||
10b8106a | 1195 | if (dm_target_is_snapshot_merge(ti)) { |
55a62eef | 1196 | num_flush_bios = 2; |
10b8106a MS |
1197 | origin_mode = FMODE_WRITE; |
1198 | } | |
1199 | ||
d3775354 | 1200 | s = kzalloc(sizeof(*s), GFP_KERNEL); |
fc56f6fb | 1201 | if (!s) { |
a2d2b034 | 1202 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1203 | r = -ENOMEM; |
1204 | goto bad; | |
1205 | } | |
1206 | ||
c2411045 MP |
1207 | origin_path = argv[0]; |
1208 | argv++; | |
1209 | argc--; | |
1210 | ||
1211 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1212 | if (r) { | |
1213 | ti->error = "Cannot get origin device"; | |
1214 | goto bad_origin; | |
1215 | } | |
4df2bf46 | 1216 | origin_dev = s->origin->bdev->bd_dev; |
c2411045 | 1217 | |
fc56f6fb MS |
1218 | cow_path = argv[0]; |
1219 | argv++; | |
1220 | argc--; | |
1221 | ||
4df2bf46 D |
1222 | cow_dev = dm_get_dev_t(cow_path); |
1223 | if (cow_dev && cow_dev == origin_dev) { | |
1224 | ti->error = "COW device cannot be the same as origin device"; | |
1225 | r = -EINVAL; | |
1226 | goto bad_cow; | |
1227 | } | |
1228 | ||
024d37e9 | 1229 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1230 | if (r) { |
1231 | ti->error = "Cannot get COW device"; | |
1232 | goto bad_cow; | |
1233 | } | |
1234 | ||
1235 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1236 | if (r) { |
1237 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1238 | r = -EINVAL; |
fc56f6fb | 1239 | goto bad_store; |
1da177e4 LT |
1240 | } |
1241 | ||
fee1998e JB |
1242 | argv += args_used; |
1243 | argc -= args_used; | |
1244 | ||
fc56f6fb | 1245 | s->ti = ti; |
1da177e4 | 1246 | s->valid = 1; |
76c44f6d | 1247 | s->snapshot_overflowed = 0; |
aa14edeb | 1248 | s->active = 0; |
879129d2 | 1249 | atomic_set(&s->pending_exceptions_count, 0); |
3f1637f2 | 1250 | spin_lock_init(&s->pe_allocation_lock); |
230c83af MP |
1251 | s->exception_start_sequence = 0; |
1252 | s->exception_complete_sequence = 0; | |
3db2776d | 1253 | s->out_of_order_tree = RB_ROOT; |
4ad8d880 | 1254 | init_rwsem(&s->lock); |
c1f0c183 | 1255 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1256 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1257 | s->state_bits = 0; |
d8ddb1cf | 1258 | s->merge_failed = 0; |
9fe86254 MP |
1259 | s->first_merging_chunk = 0; |
1260 | s->num_merging_chunks = 0; | |
1261 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1262 | |
1263 | /* Allocate hash table for COW data */ | |
fee1998e | 1264 | if (init_hash_tables(s)) { |
1da177e4 LT |
1265 | ti->error = "Unable to allocate hash table space"; |
1266 | r = -ENOMEM; | |
fee1998e | 1267 | goto bad_hash_tables; |
1da177e4 LT |
1268 | } |
1269 | ||
721b1d98 NT |
1270 | sema_init(&s->cow_count, (cow_threshold > 0) ? cow_threshold : INT_MAX); |
1271 | ||
df5d2e90 | 1272 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1273 | if (IS_ERR(s->kcopyd_client)) { |
1274 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1275 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1276 | goto bad_kcopyd; |
1da177e4 LT |
1277 | } |
1278 | ||
6f1c819c KO |
1279 | r = mempool_init_slab_pool(&s->pending_pool, MIN_IOS, pending_cache); |
1280 | if (r) { | |
92e86812 | 1281 | ti->error = "Could not allocate mempool for pending exceptions"; |
fee1998e | 1282 | goto bad_pending_pool; |
92e86812 MP |
1283 | } |
1284 | ||
cd45daff MP |
1285 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1286 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1287 | ||
1288 | spin_lock_init(&s->tracked_chunk_lock); | |
1289 | ||
c1f0c183 | 1290 | ti->private = s; |
55a62eef | 1291 | ti->num_flush_bios = num_flush_bios; |
30187e1d | 1292 | ti->per_io_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1293 | |
1294 | /* Add snapshot to the list of snapshots for this origin */ | |
1295 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1296 | r = register_snapshot(s); | |
1297 | if (r == -ENOMEM) { | |
1298 | ti->error = "Snapshot origin struct allocation failed"; | |
1299 | goto bad_load_and_register; | |
1300 | } else if (r < 0) { | |
1301 | /* invalid handover, register_snapshot has set ti->error */ | |
1302 | goto bad_load_and_register; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Metadata must only be loaded into one table at once, so skip this | |
1307 | * if metadata will be handed over during resume. | |
1308 | * Chunk size will be set during the handover - set it to zero to | |
1309 | * ensure it's ignored. | |
1310 | */ | |
1311 | if (r > 0) { | |
1312 | s->store->chunk_size = 0; | |
1313 | return 0; | |
1314 | } | |
1315 | ||
493df71c JB |
1316 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1317 | (void *)s); | |
0764147b | 1318 | if (r < 0) { |
f9cea4f7 | 1319 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1320 | goto bad_read_metadata; |
0764147b MB |
1321 | } else if (r > 0) { |
1322 | s->valid = 0; | |
1323 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1324 | } |
aa14edeb | 1325 | |
3f2412dc MP |
1326 | if (!s->store->chunk_size) { |
1327 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1328 | goto bad_read_metadata; |
1da177e4 | 1329 | } |
542f9038 MS |
1330 | |
1331 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1332 | if (r) | |
1333 | goto bad_read_metadata; | |
1da177e4 LT |
1334 | |
1335 | return 0; | |
1336 | ||
c1f0c183 MS |
1337 | bad_read_metadata: |
1338 | unregister_snapshot(s); | |
1339 | ||
fee1998e | 1340 | bad_load_and_register: |
6f1c819c | 1341 | mempool_exit(&s->pending_pool); |
92e86812 | 1342 | |
fee1998e | 1343 | bad_pending_pool: |
eb69aca5 | 1344 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1345 | |
fee1998e | 1346 | bad_kcopyd: |
3510cb94 JB |
1347 | dm_exception_table_exit(&s->pending, pending_cache); |
1348 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1349 | |
fee1998e | 1350 | bad_hash_tables: |
fc56f6fb | 1351 | dm_exception_store_destroy(s->store); |
1da177e4 | 1352 | |
fc56f6fb MS |
1353 | bad_store: |
1354 | dm_put_device(ti, s->cow); | |
fee1998e | 1355 | |
fc56f6fb | 1356 | bad_cow: |
c2411045 MP |
1357 | dm_put_device(ti, s->origin); |
1358 | ||
1359 | bad_origin: | |
fc56f6fb MS |
1360 | kfree(s); |
1361 | ||
1362 | bad: | |
1da177e4 LT |
1363 | return r; |
1364 | } | |
1365 | ||
31c93a0c MB |
1366 | static void __free_exceptions(struct dm_snapshot *s) |
1367 | { | |
eb69aca5 | 1368 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1369 | s->kcopyd_client = NULL; |
1370 | ||
3510cb94 JB |
1371 | dm_exception_table_exit(&s->pending, pending_cache); |
1372 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1373 | } |
1374 | ||
c1f0c183 MS |
1375 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1376 | struct dm_snapshot *snap_dest) | |
1377 | { | |
1378 | union { | |
1379 | struct dm_exception_table table_swap; | |
1380 | struct dm_exception_store *store_swap; | |
1381 | } u; | |
1382 | ||
1383 | /* | |
1384 | * Swap all snapshot context information between the two instances. | |
1385 | */ | |
1386 | u.table_swap = snap_dest->complete; | |
1387 | snap_dest->complete = snap_src->complete; | |
1388 | snap_src->complete = u.table_swap; | |
1389 | ||
1390 | u.store_swap = snap_dest->store; | |
1391 | snap_dest->store = snap_src->store; | |
b0d3cc01 | 1392 | snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow; |
c1f0c183 MS |
1393 | snap_src->store = u.store_swap; |
1394 | ||
1395 | snap_dest->store->snap = snap_dest; | |
1396 | snap_src->store->snap = snap_src; | |
1397 | ||
542f9038 | 1398 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 | 1399 | snap_dest->valid = snap_src->valid; |
76c44f6d | 1400 | snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed; |
c1f0c183 MS |
1401 | |
1402 | /* | |
1403 | * Set source invalid to ensure it receives no further I/O. | |
1404 | */ | |
1405 | snap_src->valid = 0; | |
1406 | } | |
1407 | ||
1da177e4 LT |
1408 | static void snapshot_dtr(struct dm_target *ti) |
1409 | { | |
cd45daff MP |
1410 | #ifdef CONFIG_DM_DEBUG |
1411 | int i; | |
1412 | #endif | |
028867ac | 1413 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1414 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1415 | |
c1f0c183 MS |
1416 | down_read(&_origins_lock); |
1417 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1418 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 1419 | if (snap_src && snap_dest && (s == snap_src)) { |
4ad8d880 | 1420 | down_write(&snap_dest->lock); |
c1f0c183 | 1421 | snap_dest->valid = 0; |
4ad8d880 | 1422 | up_write(&snap_dest->lock); |
c1f0c183 MS |
1423 | DMERR("Cancelling snapshot handover."); |
1424 | } | |
1425 | up_read(&_origins_lock); | |
1426 | ||
1e03f97e MP |
1427 | if (dm_target_is_snapshot_merge(ti)) |
1428 | stop_merge(s); | |
1429 | ||
138728dc AK |
1430 | /* Prevent further origin writes from using this snapshot. */ |
1431 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1432 | unregister_snapshot(s); |
1433 | ||
879129d2 | 1434 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1435 | msleep(1); |
879129d2 | 1436 | /* |
6f1c819c | 1437 | * Ensure instructions in mempool_exit aren't reordered |
879129d2 MP |
1438 | * before atomic_read. |
1439 | */ | |
1440 | smp_mb(); | |
1441 | ||
cd45daff MP |
1442 | #ifdef CONFIG_DM_DEBUG |
1443 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1444 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1445 | #endif | |
1446 | ||
31c93a0c | 1447 | __free_exceptions(s); |
1da177e4 | 1448 | |
6f1c819c | 1449 | mempool_exit(&s->pending_pool); |
92e86812 | 1450 | |
fee1998e | 1451 | dm_exception_store_destroy(s->store); |
138728dc | 1452 | |
fc56f6fb MS |
1453 | dm_put_device(ti, s->cow); |
1454 | ||
c2411045 MP |
1455 | dm_put_device(ti, s->origin); |
1456 | ||
1da177e4 LT |
1457 | kfree(s); |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Flush a list of buffers. | |
1462 | */ | |
1463 | static void flush_bios(struct bio *bio) | |
1464 | { | |
1465 | struct bio *n; | |
1466 | ||
1467 | while (bio) { | |
1468 | n = bio->bi_next; | |
1469 | bio->bi_next = NULL; | |
1470 | generic_make_request(bio); | |
1471 | bio = n; | |
1472 | } | |
1473 | } | |
1474 | ||
515ad66c MP |
1475 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1476 | ||
1477 | /* | |
1478 | * Flush a list of buffers. | |
1479 | */ | |
1480 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1481 | { | |
1482 | struct bio *n; | |
1483 | int r; | |
1484 | ||
1485 | while (bio) { | |
1486 | n = bio->bi_next; | |
1487 | bio->bi_next = NULL; | |
1488 | r = do_origin(s->origin, bio); | |
1489 | if (r == DM_MAPIO_REMAPPED) | |
1490 | generic_make_request(bio); | |
1491 | bio = n; | |
1492 | } | |
1493 | } | |
1494 | ||
1da177e4 LT |
1495 | /* |
1496 | * Error a list of buffers. | |
1497 | */ | |
1498 | static void error_bios(struct bio *bio) | |
1499 | { | |
1500 | struct bio *n; | |
1501 | ||
1502 | while (bio) { | |
1503 | n = bio->bi_next; | |
1504 | bio->bi_next = NULL; | |
6712ecf8 | 1505 | bio_io_error(bio); |
1da177e4 LT |
1506 | bio = n; |
1507 | } | |
1508 | } | |
1509 | ||
695368ac | 1510 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1511 | { |
1512 | if (!s->valid) | |
1513 | return; | |
1514 | ||
1515 | if (err == -EIO) | |
1516 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1517 | else if (err == -ENOMEM) | |
1518 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1519 | ||
493df71c JB |
1520 | if (s->store->type->drop_snapshot) |
1521 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1522 | |
1523 | s->valid = 0; | |
1524 | ||
fc56f6fb | 1525 | dm_table_event(s->ti->table); |
76df1c65 AK |
1526 | } |
1527 | ||
3f1637f2 NT |
1528 | static void invalidate_snapshot(struct dm_snapshot *s, int err) |
1529 | { | |
1530 | down_write(&s->lock); | |
1531 | __invalidate_snapshot(s, err); | |
1532 | up_write(&s->lock); | |
1533 | } | |
1534 | ||
385277bf | 1535 | static void pending_complete(void *context, int success) |
1da177e4 | 1536 | { |
385277bf | 1537 | struct dm_snap_pending_exception *pe = context; |
1d4989c8 | 1538 | struct dm_exception *e; |
1da177e4 | 1539 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1540 | struct bio *origin_bios = NULL; |
1541 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1542 | struct bio *full_bio = NULL; |
f79ae415 | 1543 | struct dm_exception_table_lock lock; |
9d493fa8 | 1544 | int error = 0; |
1da177e4 | 1545 | |
f79ae415 NT |
1546 | dm_exception_table_lock_init(s, pe->e.old_chunk, &lock); |
1547 | ||
76df1c65 AK |
1548 | if (!success) { |
1549 | /* Read/write error - snapshot is unusable */ | |
3f1637f2 | 1550 | invalidate_snapshot(s, -EIO); |
9d493fa8 | 1551 | error = 1; |
f79ae415 NT |
1552 | |
1553 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1554 | goto out; |
1555 | } | |
1556 | ||
119bc547 | 1557 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1558 | if (!e) { |
3f1637f2 | 1559 | invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1560 | error = 1; |
f79ae415 NT |
1561 | |
1562 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1563 | goto out; |
1564 | } | |
1565 | *e = pe->e; | |
1da177e4 | 1566 | |
3f1637f2 | 1567 | down_read(&s->lock); |
f79ae415 | 1568 | dm_exception_table_lock(&lock); |
76df1c65 | 1569 | if (!s->valid) { |
3f1637f2 | 1570 | up_read(&s->lock); |
3510cb94 | 1571 | free_completed_exception(e); |
9d493fa8 | 1572 | error = 1; |
3f1637f2 | 1573 | |
76df1c65 | 1574 | goto out; |
1da177e4 LT |
1575 | } |
1576 | ||
9d493fa8 | 1577 | /* |
65fc7c37 NT |
1578 | * Add a proper exception. After inserting the completed exception all |
1579 | * subsequent snapshot reads to this chunk will be redirected to the | |
1580 | * COW device. This ensures that we do not starve. Moreover, as long | |
1581 | * as the pending exception exists, neither origin writes nor snapshot | |
1582 | * merging can overwrite the chunk in origin. | |
9d493fa8 | 1583 | */ |
3510cb94 | 1584 | dm_insert_exception(&s->complete, e); |
3f1637f2 | 1585 | up_read(&s->lock); |
76df1c65 | 1586 | |
65fc7c37 NT |
1587 | /* Wait for conflicting reads to drain */ |
1588 | if (__chunk_is_tracked(s, pe->e.old_chunk)) { | |
f79ae415 | 1589 | dm_exception_table_unlock(&lock); |
65fc7c37 | 1590 | __check_for_conflicting_io(s, pe->e.old_chunk); |
f79ae415 | 1591 | dm_exception_table_lock(&lock); |
65fc7c37 NT |
1592 | } |
1593 | ||
a2d2b034 | 1594 | out: |
65fc7c37 | 1595 | /* Remove the in-flight exception from the list */ |
3510cb94 | 1596 | dm_remove_exception(&pe->e); |
f79ae415 NT |
1597 | |
1598 | dm_exception_table_unlock(&lock); | |
1599 | ||
9d493fa8 | 1600 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1601 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 | 1602 | full_bio = pe->full_bio; |
fe3265b1 | 1603 | if (full_bio) |
a6e50b40 | 1604 | full_bio->bi_end_io = pe->full_bio_end_io; |
73dfd078 MP |
1605 | increment_pending_exceptions_done_count(); |
1606 | ||
9d493fa8 | 1607 | /* Submit any pending write bios */ |
a6e50b40 MP |
1608 | if (error) { |
1609 | if (full_bio) | |
1610 | bio_io_error(full_bio); | |
9d493fa8 | 1611 | error_bios(snapshot_bios); |
a6e50b40 MP |
1612 | } else { |
1613 | if (full_bio) | |
4246a0b6 | 1614 | bio_endio(full_bio); |
9d493fa8 | 1615 | flush_bios(snapshot_bios); |
a6e50b40 | 1616 | } |
9d493fa8 | 1617 | |
515ad66c | 1618 | retry_origin_bios(s, origin_bios); |
22aa66a3 MP |
1619 | |
1620 | free_pending_exception(pe); | |
1da177e4 LT |
1621 | } |
1622 | ||
230c83af MP |
1623 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1624 | { | |
1625 | struct dm_snapshot *s = pe->snap; | |
1626 | ||
385277bf MP |
1627 | /* Update the metadata if we are persistent */ |
1628 | s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error, | |
1629 | pending_complete, pe); | |
230c83af MP |
1630 | } |
1631 | ||
1da177e4 LT |
1632 | /* |
1633 | * Called when the copy I/O has finished. kcopyd actually runs | |
1634 | * this code so don't block. | |
1635 | */ | |
4cdc1d1f | 1636 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1637 | { |
028867ac | 1638 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1639 | struct dm_snapshot *s = pe->snap; |
1640 | ||
230c83af | 1641 | pe->copy_error = read_err || write_err; |
1da177e4 | 1642 | |
230c83af | 1643 | if (pe->exception_sequence == s->exception_complete_sequence) { |
3db2776d DJ |
1644 | struct rb_node *next; |
1645 | ||
230c83af MP |
1646 | s->exception_complete_sequence++; |
1647 | complete_exception(pe); | |
1648 | ||
3db2776d DJ |
1649 | next = rb_first(&s->out_of_order_tree); |
1650 | while (next) { | |
1651 | pe = rb_entry(next, struct dm_snap_pending_exception, | |
1652 | out_of_order_node); | |
230c83af MP |
1653 | if (pe->exception_sequence != s->exception_complete_sequence) |
1654 | break; | |
3db2776d | 1655 | next = rb_next(next); |
230c83af | 1656 | s->exception_complete_sequence++; |
3db2776d | 1657 | rb_erase(&pe->out_of_order_node, &s->out_of_order_tree); |
230c83af | 1658 | complete_exception(pe); |
3db2776d | 1659 | cond_resched(); |
230c83af MP |
1660 | } |
1661 | } else { | |
3db2776d DJ |
1662 | struct rb_node *parent = NULL; |
1663 | struct rb_node **p = &s->out_of_order_tree.rb_node; | |
230c83af MP |
1664 | struct dm_snap_pending_exception *pe2; |
1665 | ||
3db2776d DJ |
1666 | while (*p) { |
1667 | pe2 = rb_entry(*p, struct dm_snap_pending_exception, out_of_order_node); | |
1668 | parent = *p; | |
1669 | ||
1670 | BUG_ON(pe->exception_sequence == pe2->exception_sequence); | |
1671 | if (pe->exception_sequence < pe2->exception_sequence) | |
1672 | p = &((*p)->rb_left); | |
1673 | else | |
1674 | p = &((*p)->rb_right); | |
230c83af | 1675 | } |
3db2776d DJ |
1676 | |
1677 | rb_link_node(&pe->out_of_order_node, parent, p); | |
1678 | rb_insert_color(&pe->out_of_order_node, &s->out_of_order_tree); | |
230c83af | 1679 | } |
721b1d98 | 1680 | up(&s->cow_count); |
1da177e4 LT |
1681 | } |
1682 | ||
1683 | /* | |
1684 | * Dispatches the copy operation to kcopyd. | |
1685 | */ | |
028867ac | 1686 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1687 | { |
1688 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1689 | struct dm_io_region src, dest; |
1da177e4 LT |
1690 | struct block_device *bdev = s->origin->bdev; |
1691 | sector_t dev_size; | |
1692 | ||
1693 | dev_size = get_dev_size(bdev); | |
1694 | ||
1695 | src.bdev = bdev; | |
71fab00a | 1696 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1697 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1698 | |
fc56f6fb | 1699 | dest.bdev = s->cow->bdev; |
71fab00a | 1700 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1701 | dest.count = src.count; |
1702 | ||
1703 | /* Hand over to kcopyd */ | |
721b1d98 | 1704 | down(&s->cow_count); |
a2d2b034 | 1705 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1706 | } |
1707 | ||
4246a0b6 | 1708 | static void full_bio_end_io(struct bio *bio) |
a6e50b40 MP |
1709 | { |
1710 | void *callback_data = bio->bi_private; | |
1711 | ||
4e4cbee9 | 1712 | dm_kcopyd_do_callback(callback_data, 0, bio->bi_status ? 1 : 0); |
a6e50b40 MP |
1713 | } |
1714 | ||
1715 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1716 | struct bio *bio) | |
1717 | { | |
1718 | struct dm_snapshot *s = pe->snap; | |
1719 | void *callback_data; | |
1720 | ||
1721 | pe->full_bio = bio; | |
1722 | pe->full_bio_end_io = bio->bi_end_io; | |
a6e50b40 | 1723 | |
721b1d98 | 1724 | down(&s->cow_count); |
a6e50b40 MP |
1725 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, |
1726 | copy_callback, pe); | |
1727 | ||
1728 | bio->bi_end_io = full_bio_end_io; | |
1729 | bio->bi_private = callback_data; | |
1730 | ||
1731 | generic_make_request(bio); | |
1732 | } | |
1733 | ||
2913808e MP |
1734 | static struct dm_snap_pending_exception * |
1735 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1736 | { | |
3510cb94 | 1737 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1738 | |
1739 | if (!e) | |
1740 | return NULL; | |
1741 | ||
1742 | return container_of(e, struct dm_snap_pending_exception, e); | |
1743 | } | |
1744 | ||
1da177e4 | 1745 | /* |
65fc7c37 | 1746 | * Inserts a pending exception into the pending table. |
1da177e4 | 1747 | * |
3f1637f2 NT |
1748 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1749 | * before calling this. | |
1da177e4 | 1750 | */ |
028867ac | 1751 | static struct dm_snap_pending_exception * |
65fc7c37 NT |
1752 | __insert_pending_exception(struct dm_snapshot *s, |
1753 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1754 | { |
76df1c65 AK |
1755 | pe->e.old_chunk = chunk; |
1756 | bio_list_init(&pe->origin_bios); | |
1757 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1758 | pe->started = 0; |
a6e50b40 | 1759 | pe->full_bio = NULL; |
76df1c65 | 1760 | |
3f1637f2 | 1761 | spin_lock(&s->pe_allocation_lock); |
493df71c | 1762 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
3f1637f2 | 1763 | spin_unlock(&s->pe_allocation_lock); |
76df1c65 AK |
1764 | free_pending_exception(pe); |
1765 | return NULL; | |
1766 | } | |
1767 | ||
230c83af | 1768 | pe->exception_sequence = s->exception_start_sequence++; |
3f1637f2 | 1769 | spin_unlock(&s->pe_allocation_lock); |
230c83af | 1770 | |
3510cb94 | 1771 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1772 | |
1da177e4 LT |
1773 | return pe; |
1774 | } | |
1775 | ||
65fc7c37 NT |
1776 | /* |
1777 | * Looks to see if this snapshot already has a pending exception | |
1778 | * for this chunk, otherwise it allocates a new one and inserts | |
1779 | * it into the pending table. | |
1780 | * | |
3f1637f2 NT |
1781 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1782 | * before calling this. | |
65fc7c37 NT |
1783 | */ |
1784 | static struct dm_snap_pending_exception * | |
1785 | __find_pending_exception(struct dm_snapshot *s, | |
1786 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1787 | { | |
1788 | struct dm_snap_pending_exception *pe2; | |
1789 | ||
1790 | pe2 = __lookup_pending_exception(s, chunk); | |
1791 | if (pe2) { | |
1792 | free_pending_exception(pe); | |
1793 | return pe2; | |
1794 | } | |
1795 | ||
1796 | return __insert_pending_exception(s, pe, chunk); | |
1797 | } | |
1798 | ||
1d4989c8 | 1799 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1800 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1801 | { |
74d46992 | 1802 | bio_set_dev(bio, s->cow->bdev); |
4f024f37 KO |
1803 | bio->bi_iter.bi_sector = |
1804 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1805 | (chunk - e->old_chunk)) + | |
1806 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1807 | } |
1808 | ||
7de3ee57 | 1809 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1810 | { |
1d4989c8 | 1811 | struct dm_exception *e; |
028867ac | 1812 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1813 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1814 | chunk_t chunk; |
028867ac | 1815 | struct dm_snap_pending_exception *pe = NULL; |
f79ae415 | 1816 | struct dm_exception_table_lock lock; |
1da177e4 | 1817 | |
ee18026a MP |
1818 | init_tracked_chunk(bio); |
1819 | ||
1eff9d32 | 1820 | if (bio->bi_opf & REQ_PREFLUSH) { |
74d46992 | 1821 | bio_set_dev(bio, s->cow->bdev); |
494b3ee7 MP |
1822 | return DM_MAPIO_REMAPPED; |
1823 | } | |
1824 | ||
4f024f37 | 1825 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
f79ae415 | 1826 | dm_exception_table_lock_init(s, chunk, &lock); |
1da177e4 LT |
1827 | |
1828 | /* Full snapshots are not usable */ | |
76df1c65 | 1829 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1830 | if (!s->valid) |
846785e6 | 1831 | return DM_MAPIO_KILL; |
1da177e4 | 1832 | |
3f1637f2 | 1833 | down_read(&s->lock); |
f79ae415 | 1834 | dm_exception_table_lock(&lock); |
ba40a2aa | 1835 | |
70246286 CH |
1836 | if (!s->valid || (unlikely(s->snapshot_overflowed) && |
1837 | bio_data_dir(bio) == WRITE)) { | |
846785e6 | 1838 | r = DM_MAPIO_KILL; |
ba40a2aa AK |
1839 | goto out_unlock; |
1840 | } | |
1841 | ||
1842 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1843 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1844 | if (e) { |
d74f81f8 | 1845 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1846 | goto out_unlock; |
1847 | } | |
1848 | ||
1da177e4 LT |
1849 | /* |
1850 | * Write to snapshot - higher level takes care of RW/RO | |
1851 | * flags so we should only get this if we are | |
1852 | * writeable. | |
1853 | */ | |
70246286 | 1854 | if (bio_data_dir(bio) == WRITE) { |
2913808e | 1855 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1856 | if (!pe) { |
f79ae415 | 1857 | dm_exception_table_unlock(&lock); |
c6621392 | 1858 | pe = alloc_pending_exception(s); |
f79ae415 | 1859 | dm_exception_table_lock(&lock); |
c6621392 | 1860 | |
3510cb94 | 1861 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1862 | if (e) { |
1863 | free_pending_exception(pe); | |
1864 | remap_exception(s, e, bio, chunk); | |
1865 | goto out_unlock; | |
1866 | } | |
1867 | ||
c6621392 | 1868 | pe = __find_pending_exception(s, pe, chunk); |
2913808e | 1869 | if (!pe) { |
f79ae415 | 1870 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
1871 | up_read(&s->lock); |
1872 | ||
1873 | down_write(&s->lock); | |
f79ae415 | 1874 | |
b0d3cc01 | 1875 | if (s->store->userspace_supports_overflow) { |
3f1637f2 NT |
1876 | if (s->valid && !s->snapshot_overflowed) { |
1877 | s->snapshot_overflowed = 1; | |
1878 | DMERR("Snapshot overflowed: Unable to allocate exception."); | |
1879 | } | |
b0d3cc01 MS |
1880 | } else |
1881 | __invalidate_snapshot(s, -ENOMEM); | |
f79ae415 NT |
1882 | up_write(&s->lock); |
1883 | ||
846785e6 | 1884 | r = DM_MAPIO_KILL; |
f79ae415 | 1885 | goto out; |
2913808e | 1886 | } |
1da177e4 LT |
1887 | } |
1888 | ||
d74f81f8 | 1889 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 1890 | |
d2a7ad29 | 1891 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1892 | |
a6e50b40 | 1893 | if (!pe->started && |
4f024f37 KO |
1894 | bio->bi_iter.bi_size == |
1895 | (s->store->chunk_size << SECTOR_SHIFT)) { | |
a6e50b40 | 1896 | pe->started = 1; |
3f1637f2 | 1897 | |
f79ae415 | 1898 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
1899 | up_read(&s->lock); |
1900 | ||
a6e50b40 MP |
1901 | start_full_bio(pe, bio); |
1902 | goto out; | |
1903 | } | |
1904 | ||
1905 | bio_list_add(&pe->snapshot_bios, bio); | |
1906 | ||
76df1c65 | 1907 | if (!pe->started) { |
3f1637f2 | 1908 | /* this is protected by the exception table lock */ |
76df1c65 | 1909 | pe->started = 1; |
3f1637f2 | 1910 | |
f79ae415 | 1911 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
1912 | up_read(&s->lock); |
1913 | ||
76df1c65 | 1914 | start_copy(pe); |
ba40a2aa AK |
1915 | goto out; |
1916 | } | |
cd45daff | 1917 | } else { |
74d46992 | 1918 | bio_set_dev(bio, s->origin->bdev); |
ee18026a | 1919 | track_chunk(s, bio, chunk); |
cd45daff | 1920 | } |
1da177e4 | 1921 | |
a2d2b034 | 1922 | out_unlock: |
f79ae415 | 1923 | dm_exception_table_unlock(&lock); |
3f1637f2 | 1924 | up_read(&s->lock); |
a2d2b034 | 1925 | out: |
1da177e4 LT |
1926 | return r; |
1927 | } | |
1928 | ||
3452c2a1 MP |
1929 | /* |
1930 | * A snapshot-merge target behaves like a combination of a snapshot | |
1931 | * target and a snapshot-origin target. It only generates new | |
1932 | * exceptions in other snapshots and not in the one that is being | |
1933 | * merged. | |
1934 | * | |
1935 | * For each chunk, if there is an existing exception, it is used to | |
1936 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1937 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1938 | * If merging is currently taking place on the chunk in question, the |
1939 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 1940 | */ |
7de3ee57 | 1941 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
1942 | { |
1943 | struct dm_exception *e; | |
1944 | struct dm_snapshot *s = ti->private; | |
1945 | int r = DM_MAPIO_REMAPPED; | |
1946 | chunk_t chunk; | |
1947 | ||
ee18026a MP |
1948 | init_tracked_chunk(bio); |
1949 | ||
1eff9d32 | 1950 | if (bio->bi_opf & REQ_PREFLUSH) { |
55a62eef | 1951 | if (!dm_bio_get_target_bio_nr(bio)) |
74d46992 | 1952 | bio_set_dev(bio, s->origin->bdev); |
10b8106a | 1953 | else |
74d46992 | 1954 | bio_set_dev(bio, s->cow->bdev); |
10b8106a MS |
1955 | return DM_MAPIO_REMAPPED; |
1956 | } | |
1957 | ||
4f024f37 | 1958 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 1959 | |
4ad8d880 | 1960 | down_write(&s->lock); |
3452c2a1 | 1961 | |
d2fdb776 MP |
1962 | /* Full merging snapshots are redirected to the origin */ |
1963 | if (!s->valid) | |
1964 | goto redirect_to_origin; | |
3452c2a1 MP |
1965 | |
1966 | /* If the block is already remapped - use that */ | |
1967 | e = dm_lookup_exception(&s->complete, chunk); | |
1968 | if (e) { | |
9fe86254 | 1969 | /* Queue writes overlapping with chunks being merged */ |
70246286 | 1970 | if (bio_data_dir(bio) == WRITE && |
9fe86254 MP |
1971 | chunk >= s->first_merging_chunk && |
1972 | chunk < (s->first_merging_chunk + | |
1973 | s->num_merging_chunks)) { | |
74d46992 | 1974 | bio_set_dev(bio, s->origin->bdev); |
9fe86254 MP |
1975 | bio_list_add(&s->bios_queued_during_merge, bio); |
1976 | r = DM_MAPIO_SUBMITTED; | |
1977 | goto out_unlock; | |
1978 | } | |
17aa0332 | 1979 | |
3452c2a1 | 1980 | remap_exception(s, e, bio, chunk); |
17aa0332 | 1981 | |
70246286 | 1982 | if (bio_data_dir(bio) == WRITE) |
ee18026a | 1983 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
1984 | goto out_unlock; |
1985 | } | |
1986 | ||
d2fdb776 | 1987 | redirect_to_origin: |
74d46992 | 1988 | bio_set_dev(bio, s->origin->bdev); |
3452c2a1 | 1989 | |
70246286 | 1990 | if (bio_data_dir(bio) == WRITE) { |
4ad8d880 | 1991 | up_write(&s->lock); |
3452c2a1 MP |
1992 | return do_origin(s->origin, bio); |
1993 | } | |
1994 | ||
1995 | out_unlock: | |
4ad8d880 | 1996 | up_write(&s->lock); |
3452c2a1 MP |
1997 | |
1998 | return r; | |
1999 | } | |
2000 | ||
4e4cbee9 CH |
2001 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
2002 | blk_status_t *error) | |
cd45daff MP |
2003 | { |
2004 | struct dm_snapshot *s = ti->private; | |
cd45daff | 2005 | |
ee18026a MP |
2006 | if (is_bio_tracked(bio)) |
2007 | stop_tracking_chunk(s, bio); | |
cd45daff | 2008 | |
1be56909 | 2009 | return DM_ENDIO_DONE; |
cd45daff MP |
2010 | } |
2011 | ||
1e03f97e MP |
2012 | static void snapshot_merge_presuspend(struct dm_target *ti) |
2013 | { | |
2014 | struct dm_snapshot *s = ti->private; | |
2015 | ||
2016 | stop_merge(s); | |
2017 | } | |
2018 | ||
c1f0c183 MS |
2019 | static int snapshot_preresume(struct dm_target *ti) |
2020 | { | |
2021 | int r = 0; | |
2022 | struct dm_snapshot *s = ti->private; | |
2023 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
2024 | ||
2025 | down_read(&_origins_lock); | |
9d3b15c4 | 2026 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2027 | if (snap_src && snap_dest) { |
4ad8d880 | 2028 | down_read(&snap_src->lock); |
c1f0c183 MS |
2029 | if (s == snap_src) { |
2030 | DMERR("Unable to resume snapshot source until " | |
2031 | "handover completes."); | |
2032 | r = -EINVAL; | |
b83b2f29 | 2033 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
2034 | DMERR("Unable to perform snapshot handover until " |
2035 | "source is suspended."); | |
2036 | r = -EINVAL; | |
2037 | } | |
4ad8d880 | 2038 | up_read(&snap_src->lock); |
c1f0c183 MS |
2039 | } |
2040 | up_read(&_origins_lock); | |
2041 | ||
2042 | return r; | |
2043 | } | |
2044 | ||
1da177e4 LT |
2045 | static void snapshot_resume(struct dm_target *ti) |
2046 | { | |
028867ac | 2047 | struct dm_snapshot *s = ti->private; |
09ee96b2 | 2048 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL; |
b735fede MP |
2049 | struct dm_origin *o; |
2050 | struct mapped_device *origin_md = NULL; | |
09ee96b2 | 2051 | bool must_restart_merging = false; |
c1f0c183 MS |
2052 | |
2053 | down_read(&_origins_lock); | |
b735fede MP |
2054 | |
2055 | o = __lookup_dm_origin(s->origin->bdev); | |
2056 | if (o) | |
2057 | origin_md = dm_table_get_md(o->ti->table); | |
09ee96b2 MP |
2058 | if (!origin_md) { |
2059 | (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging); | |
2060 | if (snap_merging) | |
2061 | origin_md = dm_table_get_md(snap_merging->ti->table); | |
2062 | } | |
b735fede MP |
2063 | if (origin_md == dm_table_get_md(ti->table)) |
2064 | origin_md = NULL; | |
09ee96b2 MP |
2065 | if (origin_md) { |
2066 | if (dm_hold(origin_md)) | |
2067 | origin_md = NULL; | |
2068 | } | |
b735fede | 2069 | |
09ee96b2 MP |
2070 | up_read(&_origins_lock); |
2071 | ||
2072 | if (origin_md) { | |
b735fede | 2073 | dm_internal_suspend_fast(origin_md); |
09ee96b2 MP |
2074 | if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) { |
2075 | must_restart_merging = true; | |
2076 | stop_merge(snap_merging); | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | down_read(&_origins_lock); | |
b735fede | 2081 | |
9d3b15c4 | 2082 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2083 | if (snap_src && snap_dest) { |
4ad8d880 NT |
2084 | down_write(&snap_src->lock); |
2085 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
c1f0c183 | 2086 | __handover_exceptions(snap_src, snap_dest); |
4ad8d880 NT |
2087 | up_write(&snap_dest->lock); |
2088 | up_write(&snap_src->lock); | |
c1f0c183 | 2089 | } |
b735fede | 2090 | |
c1f0c183 MS |
2091 | up_read(&_origins_lock); |
2092 | ||
09ee96b2 MP |
2093 | if (origin_md) { |
2094 | if (must_restart_merging) | |
2095 | start_merge(snap_merging); | |
2096 | dm_internal_resume_fast(origin_md); | |
2097 | dm_put(origin_md); | |
2098 | } | |
2099 | ||
c1f0c183 MS |
2100 | /* Now we have correct chunk size, reregister */ |
2101 | reregister_snapshot(s); | |
1da177e4 | 2102 | |
4ad8d880 | 2103 | down_write(&s->lock); |
aa14edeb | 2104 | s->active = 1; |
4ad8d880 | 2105 | up_write(&s->lock); |
1da177e4 LT |
2106 | } |
2107 | ||
542f9038 | 2108 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 2109 | { |
542f9038 | 2110 | uint32_t min_chunksize; |
1e03f97e MP |
2111 | |
2112 | down_read(&_origins_lock); | |
2113 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
2114 | up_read(&_origins_lock); | |
2115 | ||
2116 | return min_chunksize; | |
2117 | } | |
2118 | ||
2119 | static void snapshot_merge_resume(struct dm_target *ti) | |
2120 | { | |
2121 | struct dm_snapshot *s = ti->private; | |
2122 | ||
2123 | /* | |
2124 | * Handover exceptions from existing snapshot. | |
2125 | */ | |
2126 | snapshot_resume(ti); | |
2127 | ||
2128 | /* | |
542f9038 | 2129 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 2130 | */ |
542f9038 | 2131 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
2132 | |
2133 | start_merge(s); | |
2134 | } | |
2135 | ||
fd7c092e MP |
2136 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
2137 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2138 | { |
2e4a31df | 2139 | unsigned sz = 0; |
028867ac | 2140 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
2141 | |
2142 | switch (type) { | |
2143 | case STATUSTYPE_INFO: | |
94e76572 | 2144 | |
4ad8d880 | 2145 | down_write(&snap->lock); |
94e76572 | 2146 | |
1da177e4 | 2147 | if (!snap->valid) |
2e4a31df | 2148 | DMEMIT("Invalid"); |
d8ddb1cf MS |
2149 | else if (snap->merge_failed) |
2150 | DMEMIT("Merge failed"); | |
76c44f6d MP |
2151 | else if (snap->snapshot_overflowed) |
2152 | DMEMIT("Overflow"); | |
1da177e4 | 2153 | else { |
985903bb MS |
2154 | if (snap->store->type->usage) { |
2155 | sector_t total_sectors, sectors_allocated, | |
2156 | metadata_sectors; | |
2157 | snap->store->type->usage(snap->store, | |
2158 | &total_sectors, | |
2159 | §ors_allocated, | |
2160 | &metadata_sectors); | |
2161 | DMEMIT("%llu/%llu %llu", | |
2162 | (unsigned long long)sectors_allocated, | |
2163 | (unsigned long long)total_sectors, | |
2164 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
2165 | } |
2166 | else | |
2e4a31df | 2167 | DMEMIT("Unknown"); |
1da177e4 | 2168 | } |
94e76572 | 2169 | |
4ad8d880 | 2170 | up_write(&snap->lock); |
94e76572 | 2171 | |
1da177e4 LT |
2172 | break; |
2173 | ||
2174 | case STATUSTYPE_TABLE: | |
2175 | /* | |
2176 | * kdevname returns a static pointer so we need | |
2177 | * to make private copies if the output is to | |
2178 | * make sense. | |
2179 | */ | |
fc56f6fb | 2180 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
2181 | snap->store->type->status(snap->store, type, result + sz, |
2182 | maxlen - sz); | |
1da177e4 LT |
2183 | break; |
2184 | } | |
1da177e4 LT |
2185 | } |
2186 | ||
8811f46c MS |
2187 | static int snapshot_iterate_devices(struct dm_target *ti, |
2188 | iterate_devices_callout_fn fn, void *data) | |
2189 | { | |
2190 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
2191 | int r; |
2192 | ||
2193 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 2194 | |
1e5554c8 MP |
2195 | if (!r) |
2196 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
2197 | ||
2198 | return r; | |
8811f46c MS |
2199 | } |
2200 | ||
2201 | ||
1da177e4 LT |
2202 | /*----------------------------------------------------------------- |
2203 | * Origin methods | |
2204 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
2205 | |
2206 | /* | |
2207 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
2208 | * supplied bio was ignored. The caller may submit it immediately. | |
2209 | * (No remapping actually occurs as the origin is always a direct linear | |
2210 | * map.) | |
2211 | * | |
2212 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
2213 | * and any supplied bio is added to a list to be submitted once all | |
2214 | * the necessary exceptions exist. | |
2215 | */ | |
2216 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
2217 | struct bio *bio) | |
1da177e4 | 2218 | { |
515ad66c | 2219 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2220 | struct dm_snapshot *snap; |
1d4989c8 | 2221 | struct dm_exception *e; |
65fc7c37 | 2222 | struct dm_snap_pending_exception *pe, *pe2; |
515ad66c MP |
2223 | struct dm_snap_pending_exception *pe_to_start_now = NULL; |
2224 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
f79ae415 | 2225 | struct dm_exception_table_lock lock; |
1da177e4 LT |
2226 | chunk_t chunk; |
2227 | ||
2228 | /* Do all the snapshots on this origin */ | |
2229 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
2230 | /* |
2231 | * Don't make new exceptions in a merging snapshot | |
2232 | * because it has effectively been deleted | |
2233 | */ | |
2234 | if (dm_target_is_snapshot_merge(snap->ti)) | |
2235 | continue; | |
2236 | ||
d5e404c1 | 2237 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 2238 | if (sector >= dm_table_get_size(snap->ti->table)) |
f79ae415 | 2239 | continue; |
1da177e4 LT |
2240 | |
2241 | /* | |
2242 | * Remember, different snapshots can have | |
2243 | * different chunk sizes. | |
2244 | */ | |
9eaae8ff | 2245 | chunk = sector_to_chunk(snap->store, sector); |
f79ae415 NT |
2246 | dm_exception_table_lock_init(snap, chunk, &lock); |
2247 | ||
3f1637f2 | 2248 | down_read(&snap->lock); |
f79ae415 NT |
2249 | dm_exception_table_lock(&lock); |
2250 | ||
2251 | /* Only deal with valid and active snapshots */ | |
2252 | if (!snap->valid || !snap->active) | |
2253 | goto next_snapshot; | |
1da177e4 | 2254 | |
2913808e | 2255 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2256 | if (!pe) { |
65fc7c37 NT |
2257 | /* |
2258 | * Check exception table to see if block is already | |
2259 | * remapped in this snapshot and trigger an exception | |
2260 | * if not. | |
2261 | */ | |
2262 | e = dm_lookup_exception(&snap->complete, chunk); | |
2263 | if (e) | |
2264 | goto next_snapshot; | |
2265 | ||
f79ae415 | 2266 | dm_exception_table_unlock(&lock); |
c6621392 | 2267 | pe = alloc_pending_exception(snap); |
f79ae415 | 2268 | dm_exception_table_lock(&lock); |
c6621392 | 2269 | |
65fc7c37 NT |
2270 | pe2 = __lookup_pending_exception(snap, chunk); |
2271 | ||
2272 | if (!pe2) { | |
2273 | e = dm_lookup_exception(&snap->complete, chunk); | |
2274 | if (e) { | |
2275 | free_pending_exception(pe); | |
2276 | goto next_snapshot; | |
2277 | } | |
2278 | ||
2279 | pe = __insert_pending_exception(snap, pe, chunk); | |
2280 | if (!pe) { | |
f79ae415 | 2281 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2282 | up_read(&snap->lock); |
f79ae415 | 2283 | |
3f1637f2 | 2284 | invalidate_snapshot(snap, -ENOMEM); |
f79ae415 | 2285 | continue; |
65fc7c37 NT |
2286 | } |
2287 | } else { | |
35bf659b | 2288 | free_pending_exception(pe); |
65fc7c37 | 2289 | pe = pe2; |
2913808e | 2290 | } |
76df1c65 AK |
2291 | } |
2292 | ||
515ad66c | 2293 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2294 | |
515ad66c MP |
2295 | /* |
2296 | * If an origin bio was supplied, queue it to wait for the | |
2297 | * completion of this exception, and start this one last, | |
2298 | * at the end of the function. | |
2299 | */ | |
2300 | if (bio) { | |
2301 | bio_list_add(&pe->origin_bios, bio); | |
2302 | bio = NULL; | |
76df1c65 | 2303 | |
515ad66c MP |
2304 | if (!pe->started) { |
2305 | pe->started = 1; | |
2306 | pe_to_start_last = pe; | |
2307 | } | |
76df1c65 AK |
2308 | } |
2309 | ||
2310 | if (!pe->started) { | |
2311 | pe->started = 1; | |
515ad66c | 2312 | pe_to_start_now = pe; |
1da177e4 LT |
2313 | } |
2314 | ||
a2d2b034 | 2315 | next_snapshot: |
f79ae415 | 2316 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2317 | up_read(&snap->lock); |
1da177e4 | 2318 | |
515ad66c MP |
2319 | if (pe_to_start_now) { |
2320 | start_copy(pe_to_start_now); | |
2321 | pe_to_start_now = NULL; | |
2322 | } | |
b4b610f6 AK |
2323 | } |
2324 | ||
1da177e4 | 2325 | /* |
515ad66c MP |
2326 | * Submit the exception against which the bio is queued last, |
2327 | * to give the other exceptions a head start. | |
1da177e4 | 2328 | */ |
515ad66c MP |
2329 | if (pe_to_start_last) |
2330 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2331 | |
2332 | return r; | |
2333 | } | |
2334 | ||
2335 | /* | |
2336 | * Called on a write from the origin driver. | |
2337 | */ | |
2338 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2339 | { | |
2340 | struct origin *o; | |
d2a7ad29 | 2341 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2342 | |
2343 | down_read(&_origins_lock); | |
2344 | o = __lookup_origin(origin->bdev); | |
2345 | if (o) | |
4f024f37 | 2346 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
1da177e4 LT |
2347 | up_read(&_origins_lock); |
2348 | ||
2349 | return r; | |
2350 | } | |
2351 | ||
73dfd078 MP |
2352 | /* |
2353 | * Trigger exceptions in all non-merging snapshots. | |
2354 | * | |
2355 | * The chunk size of the merging snapshot may be larger than the chunk | |
2356 | * size of some other snapshot so we may need to reallocate multiple | |
2357 | * chunks in other snapshots. | |
2358 | * | |
2359 | * We scan all the overlapping exceptions in the other snapshots. | |
2360 | * Returns 1 if anything was reallocated and must be waited for, | |
2361 | * otherwise returns 0. | |
2362 | * | |
2363 | * size must be a multiple of merging_snap's chunk_size. | |
2364 | */ | |
2365 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2366 | sector_t sector, unsigned size) | |
2367 | { | |
2368 | int must_wait = 0; | |
2369 | sector_t n; | |
2370 | struct origin *o; | |
2371 | ||
2372 | /* | |
542f9038 | 2373 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2374 | * by snapshot_merge_resume(). |
2375 | */ | |
2376 | down_read(&_origins_lock); | |
2377 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2378 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2379 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2380 | DM_MAPIO_SUBMITTED) | |
2381 | must_wait = 1; | |
2382 | up_read(&_origins_lock); | |
2383 | ||
2384 | return must_wait; | |
2385 | } | |
2386 | ||
1da177e4 LT |
2387 | /* |
2388 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2389 | */ | |
2390 | ||
2391 | /* | |
2392 | * Construct an origin mapping: <dev_path> | |
2393 | * The context for an origin is merely a 'struct dm_dev *' | |
2394 | * pointing to the real device. | |
2395 | */ | |
2396 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2397 | { | |
2398 | int r; | |
599cdf3b | 2399 | struct dm_origin *o; |
1da177e4 LT |
2400 | |
2401 | if (argc != 1) { | |
72d94861 | 2402 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2403 | return -EINVAL; |
2404 | } | |
2405 | ||
599cdf3b MP |
2406 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2407 | if (!o) { | |
2408 | ti->error = "Cannot allocate private origin structure"; | |
2409 | r = -ENOMEM; | |
2410 | goto bad_alloc; | |
2411 | } | |
2412 | ||
2413 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2414 | if (r) { |
2415 | ti->error = "Cannot get target device"; | |
599cdf3b | 2416 | goto bad_open; |
1da177e4 LT |
2417 | } |
2418 | ||
b735fede | 2419 | o->ti = ti; |
599cdf3b | 2420 | ti->private = o; |
55a62eef | 2421 | ti->num_flush_bios = 1; |
494b3ee7 | 2422 | |
1da177e4 | 2423 | return 0; |
599cdf3b MP |
2424 | |
2425 | bad_open: | |
2426 | kfree(o); | |
2427 | bad_alloc: | |
2428 | return r; | |
1da177e4 LT |
2429 | } |
2430 | ||
2431 | static void origin_dtr(struct dm_target *ti) | |
2432 | { | |
599cdf3b | 2433 | struct dm_origin *o = ti->private; |
b735fede | 2434 | |
599cdf3b MP |
2435 | dm_put_device(ti, o->dev); |
2436 | kfree(o); | |
1da177e4 LT |
2437 | } |
2438 | ||
7de3ee57 | 2439 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2440 | { |
599cdf3b | 2441 | struct dm_origin *o = ti->private; |
298eaa89 | 2442 | unsigned available_sectors; |
1da177e4 | 2443 | |
74d46992 | 2444 | bio_set_dev(bio, o->dev->bdev); |
1da177e4 | 2445 | |
1eff9d32 | 2446 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) |
494b3ee7 MP |
2447 | return DM_MAPIO_REMAPPED; |
2448 | ||
70246286 | 2449 | if (bio_data_dir(bio) != WRITE) |
494b3ee7 MP |
2450 | return DM_MAPIO_REMAPPED; |
2451 | ||
298eaa89 MP |
2452 | available_sectors = o->split_boundary - |
2453 | ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1)); | |
2454 | ||
2455 | if (bio_sectors(bio) > available_sectors) | |
2456 | dm_accept_partial_bio(bio, available_sectors); | |
2457 | ||
1da177e4 | 2458 | /* Only tell snapshots if this is a write */ |
298eaa89 | 2459 | return do_origin(o->dev, bio); |
1da177e4 LT |
2460 | } |
2461 | ||
1da177e4 | 2462 | /* |
542f9038 | 2463 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2464 | * chunk sizes. |
2465 | */ | |
2466 | static void origin_resume(struct dm_target *ti) | |
2467 | { | |
599cdf3b | 2468 | struct dm_origin *o = ti->private; |
1da177e4 | 2469 | |
298eaa89 | 2470 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
b735fede MP |
2471 | |
2472 | down_write(&_origins_lock); | |
2473 | __insert_dm_origin(o); | |
2474 | up_write(&_origins_lock); | |
2475 | } | |
2476 | ||
2477 | static void origin_postsuspend(struct dm_target *ti) | |
2478 | { | |
2479 | struct dm_origin *o = ti->private; | |
2480 | ||
2481 | down_write(&_origins_lock); | |
2482 | __remove_dm_origin(o); | |
2483 | up_write(&_origins_lock); | |
1da177e4 LT |
2484 | } |
2485 | ||
fd7c092e MP |
2486 | static void origin_status(struct dm_target *ti, status_type_t type, |
2487 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2488 | { |
599cdf3b | 2489 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2490 | |
2491 | switch (type) { | |
2492 | case STATUSTYPE_INFO: | |
2493 | result[0] = '\0'; | |
2494 | break; | |
2495 | ||
2496 | case STATUSTYPE_TABLE: | |
599cdf3b | 2497 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 LT |
2498 | break; |
2499 | } | |
1da177e4 LT |
2500 | } |
2501 | ||
8811f46c MS |
2502 | static int origin_iterate_devices(struct dm_target *ti, |
2503 | iterate_devices_callout_fn fn, void *data) | |
2504 | { | |
599cdf3b | 2505 | struct dm_origin *o = ti->private; |
8811f46c | 2506 | |
599cdf3b | 2507 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2508 | } |
2509 | ||
1da177e4 LT |
2510 | static struct target_type origin_target = { |
2511 | .name = "snapshot-origin", | |
b735fede | 2512 | .version = {1, 9, 0}, |
1da177e4 LT |
2513 | .module = THIS_MODULE, |
2514 | .ctr = origin_ctr, | |
2515 | .dtr = origin_dtr, | |
2516 | .map = origin_map, | |
2517 | .resume = origin_resume, | |
b735fede | 2518 | .postsuspend = origin_postsuspend, |
1da177e4 | 2519 | .status = origin_status, |
8811f46c | 2520 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2521 | }; |
2522 | ||
2523 | static struct target_type snapshot_target = { | |
2524 | .name = "snapshot", | |
b0d3cc01 | 2525 | .version = {1, 15, 0}, |
1da177e4 LT |
2526 | .module = THIS_MODULE, |
2527 | .ctr = snapshot_ctr, | |
2528 | .dtr = snapshot_dtr, | |
2529 | .map = snapshot_map, | |
cd45daff | 2530 | .end_io = snapshot_end_io, |
c1f0c183 | 2531 | .preresume = snapshot_preresume, |
1da177e4 LT |
2532 | .resume = snapshot_resume, |
2533 | .status = snapshot_status, | |
8811f46c | 2534 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2535 | }; |
2536 | ||
d698aa45 MP |
2537 | static struct target_type merge_target = { |
2538 | .name = dm_snapshot_merge_target_name, | |
b0d3cc01 | 2539 | .version = {1, 4, 0}, |
d698aa45 MP |
2540 | .module = THIS_MODULE, |
2541 | .ctr = snapshot_ctr, | |
2542 | .dtr = snapshot_dtr, | |
3452c2a1 | 2543 | .map = snapshot_merge_map, |
d698aa45 | 2544 | .end_io = snapshot_end_io, |
1e03f97e | 2545 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2546 | .preresume = snapshot_preresume, |
1e03f97e | 2547 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2548 | .status = snapshot_status, |
2549 | .iterate_devices = snapshot_iterate_devices, | |
2550 | }; | |
2551 | ||
1da177e4 LT |
2552 | static int __init dm_snapshot_init(void) |
2553 | { | |
2554 | int r; | |
2555 | ||
4db6bfe0 AK |
2556 | r = dm_exception_store_init(); |
2557 | if (r) { | |
2558 | DMERR("Failed to initialize exception stores"); | |
2559 | return r; | |
2560 | } | |
2561 | ||
1da177e4 LT |
2562 | r = init_origin_hash(); |
2563 | if (r) { | |
2564 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2565 | goto bad_origin_hash; |
1da177e4 LT |
2566 | } |
2567 | ||
1d4989c8 | 2568 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2569 | if (!exception_cache) { |
2570 | DMERR("Couldn't create exception cache."); | |
2571 | r = -ENOMEM; | |
d698aa45 | 2572 | goto bad_exception_cache; |
1da177e4 LT |
2573 | } |
2574 | ||
028867ac | 2575 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2576 | if (!pending_cache) { |
2577 | DMERR("Couldn't create pending cache."); | |
2578 | r = -ENOMEM; | |
d698aa45 | 2579 | goto bad_pending_cache; |
1da177e4 LT |
2580 | } |
2581 | ||
7e6358d2 | 2582 | r = dm_register_target(&snapshot_target); |
2583 | if (r < 0) { | |
2584 | DMERR("snapshot target register failed %d", r); | |
2585 | goto bad_register_snapshot_target; | |
2586 | } | |
2587 | ||
2588 | r = dm_register_target(&origin_target); | |
2589 | if (r < 0) { | |
2590 | DMERR("Origin target register failed %d", r); | |
2591 | goto bad_register_origin_target; | |
2592 | } | |
2593 | ||
2594 | r = dm_register_target(&merge_target); | |
2595 | if (r < 0) { | |
2596 | DMERR("Merge target register failed %d", r); | |
2597 | goto bad_register_merge_target; | |
2598 | } | |
2599 | ||
1da177e4 LT |
2600 | return 0; |
2601 | ||
d698aa45 | 2602 | bad_register_merge_target: |
1da177e4 | 2603 | dm_unregister_target(&origin_target); |
d698aa45 | 2604 | bad_register_origin_target: |
1da177e4 | 2605 | dm_unregister_target(&snapshot_target); |
034a186d | 2606 | bad_register_snapshot_target: |
7e6358d2 | 2607 | kmem_cache_destroy(pending_cache); |
2608 | bad_pending_cache: | |
2609 | kmem_cache_destroy(exception_cache); | |
2610 | bad_exception_cache: | |
2611 | exit_origin_hash(); | |
2612 | bad_origin_hash: | |
034a186d | 2613 | dm_exception_store_exit(); |
d698aa45 | 2614 | |
1da177e4 LT |
2615 | return r; |
2616 | } | |
2617 | ||
2618 | static void __exit dm_snapshot_exit(void) | |
2619 | { | |
10d3bd09 MP |
2620 | dm_unregister_target(&snapshot_target); |
2621 | dm_unregister_target(&origin_target); | |
d698aa45 | 2622 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2623 | |
2624 | exit_origin_hash(); | |
1da177e4 LT |
2625 | kmem_cache_destroy(pending_cache); |
2626 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2627 | |
2628 | dm_exception_store_exit(); | |
1da177e4 LT |
2629 | } |
2630 | ||
2631 | /* Module hooks */ | |
2632 | module_init(dm_snapshot_init); | |
2633 | module_exit(dm_snapshot_exit); | |
2634 | ||
2635 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2636 | MODULE_AUTHOR("Joe Thornber"); | |
2637 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2638 | MODULE_ALIAS("dm-snapshot-origin"); |
2639 | MODULE_ALIAS("dm-snapshot-merge"); |