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> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
a765e20e | 21 | #include <linux/dm-kcopyd.h> |
1da177e4 | 22 | |
aea53d92 | 23 | #include "dm-exception-store.h" |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "snapshots" |
26 | ||
d698aa45 MP |
27 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
28 | ||
29 | #define dm_target_is_snapshot_merge(ti) \ | |
30 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
31 | ||
cd45daff MP |
32 | /* |
33 | * The size of the mempool used to track chunks in use. | |
34 | */ | |
35 | #define MIN_IOS 256 | |
36 | ||
ccc45ea8 JB |
37 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
38 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
39 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
40 | ||
191437a5 | 41 | struct dm_exception_table { |
ccc45ea8 JB |
42 | uint32_t hash_mask; |
43 | unsigned hash_shift; | |
44 | struct list_head *table; | |
45 | }; | |
46 | ||
47 | struct dm_snapshot { | |
48 | struct rw_semaphore lock; | |
49 | ||
50 | struct dm_dev *origin; | |
fc56f6fb MS |
51 | struct dm_dev *cow; |
52 | ||
53 | struct dm_target *ti; | |
ccc45ea8 JB |
54 | |
55 | /* List of snapshots per Origin */ | |
56 | struct list_head list; | |
57 | ||
d8ddb1cf MS |
58 | /* |
59 | * You can't use a snapshot if this is 0 (e.g. if full). | |
60 | * A snapshot-merge target never clears this. | |
61 | */ | |
ccc45ea8 JB |
62 | int valid; |
63 | ||
64 | /* Origin writes don't trigger exceptions until this is set */ | |
65 | int active; | |
66 | ||
ccc45ea8 JB |
67 | atomic_t pending_exceptions_count; |
68 | ||
924e600d MS |
69 | mempool_t *pending_pool; |
70 | ||
191437a5 JB |
71 | struct dm_exception_table pending; |
72 | struct dm_exception_table complete; | |
ccc45ea8 JB |
73 | |
74 | /* | |
75 | * pe_lock protects all pending_exception operations and access | |
76 | * as well as the snapshot_bios list. | |
77 | */ | |
78 | spinlock_t pe_lock; | |
79 | ||
924e600d MS |
80 | /* Chunks with outstanding reads */ |
81 | spinlock_t tracked_chunk_lock; | |
82 | mempool_t *tracked_chunk_pool; | |
83 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
84 | ||
ccc45ea8 JB |
85 | /* The on disk metadata handler */ |
86 | struct dm_exception_store *store; | |
87 | ||
88 | struct dm_kcopyd_client *kcopyd_client; | |
89 | ||
924e600d MS |
90 | /* Wait for events based on state_bits */ |
91 | unsigned long state_bits; | |
92 | ||
93 | /* Range of chunks currently being merged. */ | |
94 | chunk_t first_merging_chunk; | |
95 | int num_merging_chunks; | |
1e03f97e | 96 | |
d8ddb1cf MS |
97 | /* |
98 | * The merge operation failed if this flag is set. | |
99 | * Failure modes are handled as follows: | |
100 | * - I/O error reading the header | |
101 | * => don't load the target; abort. | |
102 | * - Header does not have "valid" flag set | |
103 | * => use the origin; forget about the snapshot. | |
104 | * - I/O error when reading exceptions | |
105 | * => don't load the target; abort. | |
106 | * (We can't use the intermediate origin state.) | |
107 | * - I/O error while merging | |
108 | * => stop merging; set merge_failed; process I/O normally. | |
109 | */ | |
110 | int merge_failed; | |
111 | ||
9fe86254 MP |
112 | /* |
113 | * Incoming bios that overlap with chunks being merged must wait | |
114 | * for them to be committed. | |
115 | */ | |
116 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
117 | }; |
118 | ||
1e03f97e MP |
119 | /* |
120 | * state_bits: | |
121 | * RUNNING_MERGE - Merge operation is in progress. | |
122 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
123 | * cleared afterwards. | |
124 | */ | |
125 | #define RUNNING_MERGE 0 | |
126 | #define SHUTDOWN_MERGE 1 | |
127 | ||
c2411045 MP |
128 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
129 | { | |
130 | return s->origin; | |
131 | } | |
132 | EXPORT_SYMBOL(dm_snap_origin); | |
133 | ||
fc56f6fb MS |
134 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
135 | { | |
136 | return s->cow; | |
137 | } | |
138 | EXPORT_SYMBOL(dm_snap_cow); | |
139 | ||
ccc45ea8 JB |
140 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
141 | chunk_t chunk) | |
142 | { | |
143 | return chunk << store->chunk_shift; | |
144 | } | |
145 | ||
146 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
147 | { | |
148 | /* | |
149 | * There is only ever one instance of a particular block | |
150 | * device so we can compare pointers safely. | |
151 | */ | |
152 | return lhs == rhs; | |
153 | } | |
154 | ||
028867ac | 155 | struct dm_snap_pending_exception { |
1d4989c8 | 156 | struct dm_exception e; |
1da177e4 LT |
157 | |
158 | /* | |
159 | * Origin buffers waiting for this to complete are held | |
160 | * in a bio list | |
161 | */ | |
162 | struct bio_list origin_bios; | |
163 | struct bio_list snapshot_bios; | |
164 | ||
1da177e4 LT |
165 | /* Pointer back to snapshot context */ |
166 | struct dm_snapshot *snap; | |
167 | ||
168 | /* | |
169 | * 1 indicates the exception has already been sent to | |
170 | * kcopyd. | |
171 | */ | |
172 | int started; | |
173 | }; | |
174 | ||
175 | /* | |
176 | * Hash table mapping origin volumes to lists of snapshots and | |
177 | * a lock to protect it | |
178 | */ | |
e18b890b CL |
179 | static struct kmem_cache *exception_cache; |
180 | static struct kmem_cache *pending_cache; | |
1da177e4 | 181 | |
cd45daff MP |
182 | struct dm_snap_tracked_chunk { |
183 | struct hlist_node node; | |
184 | chunk_t chunk; | |
185 | }; | |
186 | ||
187 | static struct kmem_cache *tracked_chunk_cache; | |
188 | ||
189 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
190 | chunk_t chunk) | |
191 | { | |
192 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
193 | GFP_NOIO); | |
194 | unsigned long flags; | |
195 | ||
196 | c->chunk = chunk; | |
197 | ||
198 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
199 | hlist_add_head(&c->node, | |
200 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
201 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
202 | ||
203 | return c; | |
204 | } | |
205 | ||
206 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
207 | struct dm_snap_tracked_chunk *c) | |
208 | { | |
209 | unsigned long flags; | |
210 | ||
211 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
212 | hlist_del(&c->node); | |
213 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
214 | ||
215 | mempool_free(c, s->tracked_chunk_pool); | |
216 | } | |
217 | ||
a8d41b59 MP |
218 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
219 | { | |
220 | struct dm_snap_tracked_chunk *c; | |
221 | struct hlist_node *hn; | |
222 | int found = 0; | |
223 | ||
224 | spin_lock_irq(&s->tracked_chunk_lock); | |
225 | ||
226 | hlist_for_each_entry(c, hn, | |
227 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
228 | if (c->chunk == chunk) { | |
229 | found = 1; | |
230 | break; | |
231 | } | |
232 | } | |
233 | ||
234 | spin_unlock_irq(&s->tracked_chunk_lock); | |
235 | ||
236 | return found; | |
237 | } | |
238 | ||
615d1eb9 MS |
239 | /* |
240 | * This conflicting I/O is extremely improbable in the caller, | |
241 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
242 | */ | |
243 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
244 | { | |
245 | while (__chunk_is_tracked(s, chunk)) | |
246 | msleep(1); | |
247 | } | |
248 | ||
1da177e4 LT |
249 | /* |
250 | * One of these per registered origin, held in the snapshot_origins hash | |
251 | */ | |
252 | struct origin { | |
253 | /* The origin device */ | |
254 | struct block_device *bdev; | |
255 | ||
256 | struct list_head hash_list; | |
257 | ||
258 | /* List of snapshots for this origin */ | |
259 | struct list_head snapshots; | |
260 | }; | |
261 | ||
262 | /* | |
263 | * Size of the hash table for origin volumes. If we make this | |
264 | * the size of the minors list then it should be nearly perfect | |
265 | */ | |
266 | #define ORIGIN_HASH_SIZE 256 | |
267 | #define ORIGIN_MASK 0xFF | |
268 | static struct list_head *_origins; | |
269 | static struct rw_semaphore _origins_lock; | |
270 | ||
73dfd078 MP |
271 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
272 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
273 | static uint64_t _pending_exceptions_done_count; | |
274 | ||
1da177e4 LT |
275 | static int init_origin_hash(void) |
276 | { | |
277 | int i; | |
278 | ||
279 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
280 | GFP_KERNEL); | |
281 | if (!_origins) { | |
72d94861 | 282 | DMERR("unable to allocate memory"); |
1da177e4 LT |
283 | return -ENOMEM; |
284 | } | |
285 | ||
286 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
287 | INIT_LIST_HEAD(_origins + i); | |
288 | init_rwsem(&_origins_lock); | |
289 | ||
290 | return 0; | |
291 | } | |
292 | ||
293 | static void exit_origin_hash(void) | |
294 | { | |
295 | kfree(_origins); | |
296 | } | |
297 | ||
028867ac | 298 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
299 | { |
300 | return bdev->bd_dev & ORIGIN_MASK; | |
301 | } | |
302 | ||
303 | static struct origin *__lookup_origin(struct block_device *origin) | |
304 | { | |
305 | struct list_head *ol; | |
306 | struct origin *o; | |
307 | ||
308 | ol = &_origins[origin_hash(origin)]; | |
309 | list_for_each_entry (o, ol, hash_list) | |
310 | if (bdev_equal(o->bdev, origin)) | |
311 | return o; | |
312 | ||
313 | return NULL; | |
314 | } | |
315 | ||
316 | static void __insert_origin(struct origin *o) | |
317 | { | |
318 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
319 | list_add_tail(&o->hash_list, sl); | |
320 | } | |
321 | ||
c1f0c183 MS |
322 | /* |
323 | * _origins_lock must be held when calling this function. | |
324 | * Returns number of snapshots registered using the supplied cow device, plus: | |
325 | * snap_src - a snapshot suitable for use as a source of exception handover | |
326 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
327 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
328 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 329 | * |
9d3b15c4 | 330 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
331 | * 0: NULL, NULL - first new snapshot |
332 | * 1: snap_src, NULL - normal snapshot | |
333 | * 2: snap_src, snap_dest - waiting for handover | |
334 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
335 | * 1: NULL, snap_dest - source got destroyed without handover | |
336 | */ | |
337 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
338 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
339 | struct dm_snapshot **snap_dest, |
340 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
341 | { |
342 | struct dm_snapshot *s; | |
343 | struct origin *o; | |
344 | int count = 0; | |
345 | int active; | |
346 | ||
347 | o = __lookup_origin(snap->origin->bdev); | |
348 | if (!o) | |
349 | goto out; | |
350 | ||
351 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
352 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
353 | *snap_merge = s; | |
c1f0c183 MS |
354 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
355 | continue; | |
356 | ||
357 | down_read(&s->lock); | |
358 | active = s->active; | |
359 | up_read(&s->lock); | |
360 | ||
361 | if (active) { | |
362 | if (snap_src) | |
363 | *snap_src = s; | |
364 | } else if (snap_dest) | |
365 | *snap_dest = s; | |
366 | ||
367 | count++; | |
368 | } | |
369 | ||
370 | out: | |
371 | return count; | |
372 | } | |
373 | ||
374 | /* | |
375 | * On success, returns 1 if this snapshot is a handover destination, | |
376 | * otherwise returns 0. | |
377 | */ | |
378 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
379 | { | |
380 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 381 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
382 | |
383 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
384 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
385 | &snap_merge) == 2) || | |
c1f0c183 MS |
386 | snap_dest) { |
387 | snap->ti->error = "Snapshot cow pairing for exception " | |
388 | "table handover failed"; | |
389 | return -EINVAL; | |
390 | } | |
391 | ||
392 | /* | |
393 | * If no snap_src was found, snap cannot become a handover | |
394 | * destination. | |
395 | */ | |
396 | if (!snap_src) | |
397 | return 0; | |
398 | ||
9d3b15c4 MP |
399 | /* |
400 | * Non-snapshot-merge handover? | |
401 | */ | |
402 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
403 | return 1; | |
404 | ||
405 | /* | |
406 | * Do not allow more than one merging snapshot. | |
407 | */ | |
408 | if (snap_merge) { | |
409 | snap->ti->error = "A snapshot is already merging."; | |
410 | return -EINVAL; | |
411 | } | |
412 | ||
1e03f97e MP |
413 | if (!snap_src->store->type->prepare_merge || |
414 | !snap_src->store->type->commit_merge) { | |
415 | snap->ti->error = "Snapshot exception store does not " | |
416 | "support snapshot-merge."; | |
417 | return -EINVAL; | |
418 | } | |
419 | ||
c1f0c183 MS |
420 | return 1; |
421 | } | |
422 | ||
423 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
424 | { | |
425 | struct dm_snapshot *l; | |
426 | ||
427 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
428 | list_for_each_entry(l, &o->snapshots, list) | |
429 | if (l->store->chunk_size < s->store->chunk_size) | |
430 | break; | |
431 | list_add_tail(&s->list, &l->list); | |
432 | } | |
433 | ||
1da177e4 LT |
434 | /* |
435 | * Make a note of the snapshot and its origin so we can look it | |
436 | * up when the origin has a write on it. | |
c1f0c183 MS |
437 | * |
438 | * Also validate snapshot exception store handovers. | |
439 | * On success, returns 1 if this registration is a handover destination, | |
440 | * otherwise returns 0. | |
1da177e4 LT |
441 | */ |
442 | static int register_snapshot(struct dm_snapshot *snap) | |
443 | { | |
c1f0c183 | 444 | struct origin *o, *new_o = NULL; |
1da177e4 | 445 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 446 | int r = 0; |
1da177e4 | 447 | |
60c856c8 MP |
448 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
449 | if (!new_o) | |
450 | return -ENOMEM; | |
451 | ||
1da177e4 | 452 | down_write(&_origins_lock); |
1da177e4 | 453 | |
c1f0c183 MS |
454 | r = __validate_exception_handover(snap); |
455 | if (r < 0) { | |
456 | kfree(new_o); | |
457 | goto out; | |
458 | } | |
459 | ||
460 | o = __lookup_origin(bdev); | |
60c856c8 MP |
461 | if (o) |
462 | kfree(new_o); | |
463 | else { | |
1da177e4 | 464 | /* New origin */ |
60c856c8 | 465 | o = new_o; |
1da177e4 LT |
466 | |
467 | /* Initialise the struct */ | |
468 | INIT_LIST_HEAD(&o->snapshots); | |
469 | o->bdev = bdev; | |
470 | ||
471 | __insert_origin(o); | |
472 | } | |
473 | ||
c1f0c183 MS |
474 | __insert_snapshot(o, snap); |
475 | ||
476 | out: | |
477 | up_write(&_origins_lock); | |
478 | ||
479 | return r; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Move snapshot to correct place in list according to chunk size. | |
484 | */ | |
485 | static void reregister_snapshot(struct dm_snapshot *s) | |
486 | { | |
487 | struct block_device *bdev = s->origin->bdev; | |
488 | ||
489 | down_write(&_origins_lock); | |
490 | ||
491 | list_del(&s->list); | |
492 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
493 | |
494 | up_write(&_origins_lock); | |
1da177e4 LT |
495 | } |
496 | ||
497 | static void unregister_snapshot(struct dm_snapshot *s) | |
498 | { | |
499 | struct origin *o; | |
500 | ||
501 | down_write(&_origins_lock); | |
502 | o = __lookup_origin(s->origin->bdev); | |
503 | ||
504 | list_del(&s->list); | |
c1f0c183 | 505 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
506 | list_del(&o->hash_list); |
507 | kfree(o); | |
508 | } | |
509 | ||
510 | up_write(&_origins_lock); | |
511 | } | |
512 | ||
513 | /* | |
514 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
515 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
516 | * some consecutive chunks to be grouped together. | |
1da177e4 | 517 | */ |
3510cb94 JB |
518 | static int dm_exception_table_init(struct dm_exception_table *et, |
519 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
520 | { |
521 | unsigned int i; | |
522 | ||
d74f81f8 | 523 | et->hash_shift = hash_shift; |
1da177e4 LT |
524 | et->hash_mask = size - 1; |
525 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
526 | if (!et->table) | |
527 | return -ENOMEM; | |
528 | ||
529 | for (i = 0; i < size; i++) | |
530 | INIT_LIST_HEAD(et->table + i); | |
531 | ||
532 | return 0; | |
533 | } | |
534 | ||
3510cb94 JB |
535 | static void dm_exception_table_exit(struct dm_exception_table *et, |
536 | struct kmem_cache *mem) | |
1da177e4 LT |
537 | { |
538 | struct list_head *slot; | |
1d4989c8 | 539 | struct dm_exception *ex, *next; |
1da177e4 LT |
540 | int i, size; |
541 | ||
542 | size = et->hash_mask + 1; | |
543 | for (i = 0; i < size; i++) { | |
544 | slot = et->table + i; | |
545 | ||
546 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
547 | kmem_cache_free(mem, ex); | |
548 | } | |
549 | ||
550 | vfree(et->table); | |
551 | } | |
552 | ||
191437a5 | 553 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 554 | { |
d74f81f8 | 555 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
556 | } |
557 | ||
3510cb94 | 558 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
559 | { |
560 | list_del(&e->hash_list); | |
561 | } | |
562 | ||
563 | /* | |
564 | * Return the exception data for a sector, or NULL if not | |
565 | * remapped. | |
566 | */ | |
3510cb94 JB |
567 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
568 | chunk_t chunk) | |
1da177e4 LT |
569 | { |
570 | struct list_head *slot; | |
1d4989c8 | 571 | struct dm_exception *e; |
1da177e4 LT |
572 | |
573 | slot = &et->table[exception_hash(et, chunk)]; | |
574 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
575 | if (chunk >= e->old_chunk && |
576 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
577 | return e; |
578 | ||
579 | return NULL; | |
580 | } | |
581 | ||
3510cb94 | 582 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 583 | { |
1d4989c8 | 584 | struct dm_exception *e; |
1da177e4 LT |
585 | |
586 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
587 | if (!e) | |
588 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
589 | ||
590 | return e; | |
591 | } | |
592 | ||
3510cb94 | 593 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
594 | { |
595 | kmem_cache_free(exception_cache, e); | |
596 | } | |
597 | ||
92e86812 | 598 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 599 | { |
92e86812 MP |
600 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
601 | GFP_NOIO); | |
602 | ||
879129d2 | 603 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
604 | pe->snap = s; |
605 | ||
606 | return pe; | |
1da177e4 LT |
607 | } |
608 | ||
028867ac | 609 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 610 | { |
879129d2 MP |
611 | struct dm_snapshot *s = pe->snap; |
612 | ||
613 | mempool_free(pe, s->pending_pool); | |
614 | smp_mb__before_atomic_dec(); | |
615 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
616 | } |
617 | ||
3510cb94 JB |
618 | static void dm_insert_exception(struct dm_exception_table *eh, |
619 | struct dm_exception *new_e) | |
d74f81f8 | 620 | { |
d74f81f8 | 621 | struct list_head *l; |
1d4989c8 | 622 | struct dm_exception *e = NULL; |
d74f81f8 MB |
623 | |
624 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
625 | ||
626 | /* Add immediately if this table doesn't support consecutive chunks */ | |
627 | if (!eh->hash_shift) | |
628 | goto out; | |
629 | ||
630 | /* List is ordered by old_chunk */ | |
631 | list_for_each_entry_reverse(e, l, hash_list) { | |
632 | /* Insert after an existing chunk? */ | |
633 | if (new_e->old_chunk == (e->old_chunk + | |
634 | dm_consecutive_chunk_count(e) + 1) && | |
635 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
636 | dm_consecutive_chunk_count(e) + 1)) { | |
637 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 638 | free_completed_exception(new_e); |
d74f81f8 MB |
639 | return; |
640 | } | |
641 | ||
642 | /* Insert before an existing chunk? */ | |
643 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
644 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
645 | dm_consecutive_chunk_count_inc(e); | |
646 | e->old_chunk--; | |
647 | e->new_chunk--; | |
3510cb94 | 648 | free_completed_exception(new_e); |
d74f81f8 MB |
649 | return; |
650 | } | |
651 | ||
652 | if (new_e->old_chunk > e->old_chunk) | |
653 | break; | |
654 | } | |
655 | ||
656 | out: | |
657 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
658 | } | |
659 | ||
a159c1ac JB |
660 | /* |
661 | * Callback used by the exception stores to load exceptions when | |
662 | * initialising. | |
663 | */ | |
664 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 665 | { |
a159c1ac | 666 | struct dm_snapshot *s = context; |
1d4989c8 | 667 | struct dm_exception *e; |
1da177e4 | 668 | |
3510cb94 | 669 | e = alloc_completed_exception(); |
1da177e4 LT |
670 | if (!e) |
671 | return -ENOMEM; | |
672 | ||
673 | e->old_chunk = old; | |
d74f81f8 MB |
674 | |
675 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 676 | e->new_chunk = new; |
d74f81f8 | 677 | |
3510cb94 | 678 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 679 | |
1da177e4 LT |
680 | return 0; |
681 | } | |
682 | ||
7e201b35 MP |
683 | /* |
684 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
685 | * Return zero if the origin has no snapshots. | |
686 | */ | |
687 | static sector_t __minimum_chunk_size(struct origin *o) | |
688 | { | |
689 | struct dm_snapshot *snap; | |
690 | unsigned chunk_size = 0; | |
691 | ||
692 | if (o) | |
693 | list_for_each_entry(snap, &o->snapshots, list) | |
694 | chunk_size = min_not_zero(chunk_size, | |
695 | snap->store->chunk_size); | |
696 | ||
697 | return chunk_size; | |
698 | } | |
699 | ||
1da177e4 LT |
700 | /* |
701 | * Hard coded magic. | |
702 | */ | |
703 | static int calc_max_buckets(void) | |
704 | { | |
705 | /* use a fixed size of 2MB */ | |
706 | unsigned long mem = 2 * 1024 * 1024; | |
707 | mem /= sizeof(struct list_head); | |
708 | ||
709 | return mem; | |
710 | } | |
711 | ||
1da177e4 LT |
712 | /* |
713 | * Allocate room for a suitable hash table. | |
714 | */ | |
fee1998e | 715 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
716 | { |
717 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
718 | ||
719 | /* | |
720 | * Calculate based on the size of the original volume or | |
721 | * the COW volume... | |
722 | */ | |
fc56f6fb | 723 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
724 | origin_dev_size = get_dev_size(s->origin->bdev); |
725 | max_buckets = calc_max_buckets(); | |
726 | ||
fee1998e | 727 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
728 | hash_size = min(hash_size, max_buckets); |
729 | ||
8e87b9b8 MP |
730 | if (hash_size < 64) |
731 | hash_size = 64; | |
8defd830 | 732 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
733 | if (dm_exception_table_init(&s->complete, hash_size, |
734 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
735 | return -ENOMEM; |
736 | ||
737 | /* | |
738 | * Allocate hash table for in-flight exceptions | |
739 | * Make this smaller than the real hash table | |
740 | */ | |
741 | hash_size >>= 3; | |
742 | if (hash_size < 64) | |
743 | hash_size = 64; | |
744 | ||
3510cb94 JB |
745 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
746 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
747 | return -ENOMEM; |
748 | } | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
1e03f97e MP |
753 | static void merge_shutdown(struct dm_snapshot *s) |
754 | { | |
755 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
756 | smp_mb__after_clear_bit(); | |
757 | wake_up_bit(&s->state_bits, RUNNING_MERGE); | |
758 | } | |
759 | ||
9fe86254 MP |
760 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
761 | { | |
762 | s->first_merging_chunk = 0; | |
763 | s->num_merging_chunks = 0; | |
764 | ||
765 | return bio_list_get(&s->bios_queued_during_merge); | |
766 | } | |
767 | ||
1e03f97e MP |
768 | /* |
769 | * Remove one chunk from the index of completed exceptions. | |
770 | */ | |
771 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
772 | chunk_t old_chunk) | |
773 | { | |
774 | struct dm_exception *e; | |
775 | ||
1e03f97e MP |
776 | e = dm_lookup_exception(&s->complete, old_chunk); |
777 | if (!e) { | |
778 | DMERR("Corruption detected: exception for block %llu is " | |
779 | "on disk but not in memory", | |
780 | (unsigned long long)old_chunk); | |
781 | return -EINVAL; | |
782 | } | |
783 | ||
784 | /* | |
785 | * If this is the only chunk using this exception, remove exception. | |
786 | */ | |
787 | if (!dm_consecutive_chunk_count(e)) { | |
788 | dm_remove_exception(e); | |
789 | free_completed_exception(e); | |
790 | return 0; | |
791 | } | |
792 | ||
793 | /* | |
794 | * The chunk may be either at the beginning or the end of a | |
795 | * group of consecutive chunks - never in the middle. We are | |
796 | * removing chunks in the opposite order to that in which they | |
797 | * were added, so this should always be true. | |
798 | * Decrement the consecutive chunk counter and adjust the | |
799 | * starting point if necessary. | |
800 | */ | |
801 | if (old_chunk == e->old_chunk) { | |
802 | e->old_chunk++; | |
803 | e->new_chunk++; | |
804 | } else if (old_chunk != e->old_chunk + | |
805 | dm_consecutive_chunk_count(e)) { | |
806 | DMERR("Attempt to merge block %llu from the " | |
807 | "middle of a chunk range [%llu - %llu]", | |
808 | (unsigned long long)old_chunk, | |
809 | (unsigned long long)e->old_chunk, | |
810 | (unsigned long long) | |
811 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
812 | return -EINVAL; | |
813 | } | |
814 | ||
815 | dm_consecutive_chunk_count_dec(e); | |
816 | ||
817 | return 0; | |
818 | } | |
819 | ||
9fe86254 MP |
820 | static void flush_bios(struct bio *bio); |
821 | ||
822 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 823 | { |
9fe86254 MP |
824 | struct bio *b = NULL; |
825 | int r; | |
826 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
827 | |
828 | down_write(&s->lock); | |
9fe86254 MP |
829 | |
830 | /* | |
831 | * Process chunks (and associated exceptions) in reverse order | |
832 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
833 | */ | |
834 | do { | |
835 | r = __remove_single_exception_chunk(s, old_chunk); | |
836 | if (r) | |
837 | goto out; | |
838 | } while (old_chunk-- > s->first_merging_chunk); | |
839 | ||
840 | b = __release_queued_bios_after_merge(s); | |
841 | ||
842 | out: | |
1e03f97e | 843 | up_write(&s->lock); |
9fe86254 MP |
844 | if (b) |
845 | flush_bios(b); | |
1e03f97e MP |
846 | |
847 | return r; | |
848 | } | |
849 | ||
73dfd078 MP |
850 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
851 | sector_t sector, unsigned chunk_size); | |
852 | ||
1e03f97e MP |
853 | static void merge_callback(int read_err, unsigned long write_err, |
854 | void *context); | |
855 | ||
73dfd078 MP |
856 | static uint64_t read_pending_exceptions_done_count(void) |
857 | { | |
858 | uint64_t pending_exceptions_done; | |
859 | ||
860 | spin_lock(&_pending_exceptions_done_spinlock); | |
861 | pending_exceptions_done = _pending_exceptions_done_count; | |
862 | spin_unlock(&_pending_exceptions_done_spinlock); | |
863 | ||
864 | return pending_exceptions_done; | |
865 | } | |
866 | ||
867 | static void increment_pending_exceptions_done_count(void) | |
868 | { | |
869 | spin_lock(&_pending_exceptions_done_spinlock); | |
870 | _pending_exceptions_done_count++; | |
871 | spin_unlock(&_pending_exceptions_done_spinlock); | |
872 | ||
873 | wake_up_all(&_pending_exceptions_done); | |
874 | } | |
875 | ||
1e03f97e MP |
876 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
877 | { | |
8a2d5286 | 878 | int i, linear_chunks; |
1e03f97e MP |
879 | chunk_t old_chunk, new_chunk; |
880 | struct dm_io_region src, dest; | |
8a2d5286 | 881 | sector_t io_size; |
73dfd078 | 882 | uint64_t previous_count; |
1e03f97e MP |
883 | |
884 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
885 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
886 | goto shut; | |
887 | ||
888 | /* | |
889 | * valid flag never changes during merge, so no lock required. | |
890 | */ | |
891 | if (!s->valid) { | |
892 | DMERR("Snapshot is invalid: can't merge"); | |
893 | goto shut; | |
894 | } | |
895 | ||
8a2d5286 MS |
896 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
897 | &new_chunk); | |
898 | if (linear_chunks <= 0) { | |
d8ddb1cf | 899 | if (linear_chunks < 0) { |
1e03f97e MP |
900 | DMERR("Read error in exception store: " |
901 | "shutting down merge"); | |
d8ddb1cf MS |
902 | down_write(&s->lock); |
903 | s->merge_failed = 1; | |
904 | up_write(&s->lock); | |
905 | } | |
1e03f97e MP |
906 | goto shut; |
907 | } | |
908 | ||
8a2d5286 MS |
909 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
910 | old_chunk = old_chunk + 1 - linear_chunks; | |
911 | new_chunk = new_chunk + 1 - linear_chunks; | |
912 | ||
913 | /* | |
914 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
915 | * from the exception store to the origin | |
916 | */ | |
917 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 918 | |
1e03f97e MP |
919 | dest.bdev = s->origin->bdev; |
920 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 921 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
922 | |
923 | src.bdev = s->cow->bdev; | |
924 | src.sector = chunk_to_sector(s->store, new_chunk); | |
925 | src.count = dest.count; | |
926 | ||
73dfd078 MP |
927 | /* |
928 | * Reallocate any exceptions needed in other snapshots then | |
929 | * wait for the pending exceptions to complete. | |
930 | * Each time any pending exception (globally on the system) | |
931 | * completes we are woken and repeat the process to find out | |
932 | * if we can proceed. While this may not seem a particularly | |
933 | * efficient algorithm, it is not expected to have any | |
934 | * significant impact on performance. | |
935 | */ | |
936 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 937 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
938 | wait_event(_pending_exceptions_done, |
939 | (read_pending_exceptions_done_count() != | |
940 | previous_count)); | |
941 | /* Retry after the wait, until all exceptions are done. */ | |
942 | previous_count = read_pending_exceptions_done_count(); | |
943 | } | |
944 | ||
9fe86254 MP |
945 | down_write(&s->lock); |
946 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 947 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
948 | up_write(&s->lock); |
949 | ||
8a2d5286 MS |
950 | /* Wait until writes to all 'linear_chunks' drain */ |
951 | for (i = 0; i < linear_chunks; i++) | |
952 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 953 | |
1e03f97e MP |
954 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
955 | return; | |
956 | ||
957 | shut: | |
958 | merge_shutdown(s); | |
959 | } | |
960 | ||
9fe86254 MP |
961 | static void error_bios(struct bio *bio); |
962 | ||
1e03f97e MP |
963 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
964 | { | |
965 | struct dm_snapshot *s = context; | |
9fe86254 | 966 | struct bio *b = NULL; |
1e03f97e MP |
967 | |
968 | if (read_err || write_err) { | |
969 | if (read_err) | |
970 | DMERR("Read error: shutting down merge."); | |
971 | else | |
972 | DMERR("Write error: shutting down merge."); | |
973 | goto shut; | |
974 | } | |
975 | ||
9fe86254 MP |
976 | if (s->store->type->commit_merge(s->store, |
977 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
978 | DMERR("Write error in exception store: shutting down merge"); |
979 | goto shut; | |
980 | } | |
981 | ||
9fe86254 MP |
982 | if (remove_single_exception_chunk(s) < 0) |
983 | goto shut; | |
984 | ||
1e03f97e MP |
985 | snapshot_merge_next_chunks(s); |
986 | ||
987 | return; | |
988 | ||
989 | shut: | |
9fe86254 | 990 | down_write(&s->lock); |
d8ddb1cf | 991 | s->merge_failed = 1; |
9fe86254 MP |
992 | b = __release_queued_bios_after_merge(s); |
993 | up_write(&s->lock); | |
994 | error_bios(b); | |
995 | ||
1e03f97e MP |
996 | merge_shutdown(s); |
997 | } | |
998 | ||
999 | static void start_merge(struct dm_snapshot *s) | |
1000 | { | |
1001 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1002 | snapshot_merge_next_chunks(s); | |
1003 | } | |
1004 | ||
1005 | static int wait_schedule(void *ptr) | |
1006 | { | |
1007 | schedule(); | |
1008 | ||
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Stop the merging process and wait until it finishes. | |
1014 | */ | |
1015 | static void stop_merge(struct dm_snapshot *s) | |
1016 | { | |
1017 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1018 | wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule, | |
1019 | TASK_UNINTERRUPTIBLE); | |
1020 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1021 | } | |
1022 | ||
1da177e4 LT |
1023 | /* |
1024 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
1025 | */ | |
1026 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1027 | { | |
1028 | struct dm_snapshot *s; | |
cd45daff | 1029 | int i; |
1da177e4 | 1030 | int r = -EINVAL; |
fc56f6fb | 1031 | char *origin_path, *cow_path; |
10b8106a MS |
1032 | unsigned args_used, num_flush_requests = 1; |
1033 | fmode_t origin_mode = FMODE_READ; | |
1da177e4 | 1034 | |
4c7e3bf4 | 1035 | if (argc != 4) { |
72d94861 | 1036 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1037 | r = -EINVAL; |
fc56f6fb | 1038 | goto bad; |
1da177e4 LT |
1039 | } |
1040 | ||
10b8106a MS |
1041 | if (dm_target_is_snapshot_merge(ti)) { |
1042 | num_flush_requests = 2; | |
1043 | origin_mode = FMODE_WRITE; | |
1044 | } | |
1045 | ||
fc56f6fb MS |
1046 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1047 | if (!s) { | |
a2d2b034 | 1048 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1049 | r = -ENOMEM; |
1050 | goto bad; | |
1051 | } | |
1052 | ||
c2411045 MP |
1053 | origin_path = argv[0]; |
1054 | argv++; | |
1055 | argc--; | |
1056 | ||
1057 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1058 | if (r) { | |
1059 | ti->error = "Cannot get origin device"; | |
1060 | goto bad_origin; | |
1061 | } | |
1062 | ||
fc56f6fb MS |
1063 | cow_path = argv[0]; |
1064 | argv++; | |
1065 | argc--; | |
1066 | ||
024d37e9 | 1067 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1068 | if (r) { |
1069 | ti->error = "Cannot get COW device"; | |
1070 | goto bad_cow; | |
1071 | } | |
1072 | ||
1073 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1074 | if (r) { |
1075 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1076 | r = -EINVAL; |
fc56f6fb | 1077 | goto bad_store; |
1da177e4 LT |
1078 | } |
1079 | ||
fee1998e JB |
1080 | argv += args_used; |
1081 | argc -= args_used; | |
1082 | ||
fc56f6fb | 1083 | s->ti = ti; |
1da177e4 | 1084 | s->valid = 1; |
aa14edeb | 1085 | s->active = 0; |
879129d2 | 1086 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 1087 | init_rwsem(&s->lock); |
c1f0c183 | 1088 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1089 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1090 | s->state_bits = 0; |
d8ddb1cf | 1091 | s->merge_failed = 0; |
9fe86254 MP |
1092 | s->first_merging_chunk = 0; |
1093 | s->num_merging_chunks = 0; | |
1094 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1095 | |
1096 | /* Allocate hash table for COW data */ | |
fee1998e | 1097 | if (init_hash_tables(s)) { |
1da177e4 LT |
1098 | ti->error = "Unable to allocate hash table space"; |
1099 | r = -ENOMEM; | |
fee1998e | 1100 | goto bad_hash_tables; |
1da177e4 LT |
1101 | } |
1102 | ||
fa34ce73 MP |
1103 | s->kcopyd_client = dm_kcopyd_client_create(); |
1104 | if (IS_ERR(s->kcopyd_client)) { | |
1105 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1106 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1107 | goto bad_kcopyd; |
1da177e4 LT |
1108 | } |
1109 | ||
92e86812 MP |
1110 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1111 | if (!s->pending_pool) { | |
1112 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 1113 | goto bad_pending_pool; |
92e86812 MP |
1114 | } |
1115 | ||
cd45daff MP |
1116 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
1117 | tracked_chunk_cache); | |
1118 | if (!s->tracked_chunk_pool) { | |
1119 | ti->error = "Could not allocate tracked_chunk mempool for " | |
1120 | "tracking reads"; | |
92e86812 | 1121 | goto bad_tracked_chunk_pool; |
cd45daff MP |
1122 | } |
1123 | ||
1124 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1125 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1126 | ||
1127 | spin_lock_init(&s->tracked_chunk_lock); | |
1128 | ||
c1f0c183 | 1129 | ti->private = s; |
10b8106a | 1130 | ti->num_flush_requests = num_flush_requests; |
c1f0c183 MS |
1131 | |
1132 | /* Add snapshot to the list of snapshots for this origin */ | |
1133 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1134 | r = register_snapshot(s); | |
1135 | if (r == -ENOMEM) { | |
1136 | ti->error = "Snapshot origin struct allocation failed"; | |
1137 | goto bad_load_and_register; | |
1138 | } else if (r < 0) { | |
1139 | /* invalid handover, register_snapshot has set ti->error */ | |
1140 | goto bad_load_and_register; | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Metadata must only be loaded into one table at once, so skip this | |
1145 | * if metadata will be handed over during resume. | |
1146 | * Chunk size will be set during the handover - set it to zero to | |
1147 | * ensure it's ignored. | |
1148 | */ | |
1149 | if (r > 0) { | |
1150 | s->store->chunk_size = 0; | |
1151 | return 0; | |
1152 | } | |
1153 | ||
493df71c JB |
1154 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1155 | (void *)s); | |
0764147b | 1156 | if (r < 0) { |
f9cea4f7 | 1157 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1158 | goto bad_read_metadata; |
0764147b MB |
1159 | } else if (r > 0) { |
1160 | s->valid = 0; | |
1161 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1162 | } |
aa14edeb | 1163 | |
3f2412dc MP |
1164 | if (!s->store->chunk_size) { |
1165 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1166 | goto bad_read_metadata; |
1da177e4 | 1167 | } |
d0216849 | 1168 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
1169 | |
1170 | return 0; | |
1171 | ||
c1f0c183 MS |
1172 | bad_read_metadata: |
1173 | unregister_snapshot(s); | |
1174 | ||
fee1998e | 1175 | bad_load_and_register: |
cd45daff MP |
1176 | mempool_destroy(s->tracked_chunk_pool); |
1177 | ||
fee1998e | 1178 | bad_tracked_chunk_pool: |
92e86812 MP |
1179 | mempool_destroy(s->pending_pool); |
1180 | ||
fee1998e | 1181 | bad_pending_pool: |
eb69aca5 | 1182 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1183 | |
fee1998e | 1184 | bad_kcopyd: |
3510cb94 JB |
1185 | dm_exception_table_exit(&s->pending, pending_cache); |
1186 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1187 | |
fee1998e | 1188 | bad_hash_tables: |
fc56f6fb | 1189 | dm_exception_store_destroy(s->store); |
1da177e4 | 1190 | |
fc56f6fb MS |
1191 | bad_store: |
1192 | dm_put_device(ti, s->cow); | |
fee1998e | 1193 | |
fc56f6fb | 1194 | bad_cow: |
c2411045 MP |
1195 | dm_put_device(ti, s->origin); |
1196 | ||
1197 | bad_origin: | |
fc56f6fb MS |
1198 | kfree(s); |
1199 | ||
1200 | bad: | |
1da177e4 LT |
1201 | return r; |
1202 | } | |
1203 | ||
31c93a0c MB |
1204 | static void __free_exceptions(struct dm_snapshot *s) |
1205 | { | |
eb69aca5 | 1206 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1207 | s->kcopyd_client = NULL; |
1208 | ||
3510cb94 JB |
1209 | dm_exception_table_exit(&s->pending, pending_cache); |
1210 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1211 | } |
1212 | ||
c1f0c183 MS |
1213 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1214 | struct dm_snapshot *snap_dest) | |
1215 | { | |
1216 | union { | |
1217 | struct dm_exception_table table_swap; | |
1218 | struct dm_exception_store *store_swap; | |
1219 | } u; | |
1220 | ||
1221 | /* | |
1222 | * Swap all snapshot context information between the two instances. | |
1223 | */ | |
1224 | u.table_swap = snap_dest->complete; | |
1225 | snap_dest->complete = snap_src->complete; | |
1226 | snap_src->complete = u.table_swap; | |
1227 | ||
1228 | u.store_swap = snap_dest->store; | |
1229 | snap_dest->store = snap_src->store; | |
1230 | snap_src->store = u.store_swap; | |
1231 | ||
1232 | snap_dest->store->snap = snap_dest; | |
1233 | snap_src->store->snap = snap_src; | |
1234 | ||
1235 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
1236 | snap_dest->valid = snap_src->valid; | |
1237 | ||
1238 | /* | |
1239 | * Set source invalid to ensure it receives no further I/O. | |
1240 | */ | |
1241 | snap_src->valid = 0; | |
1242 | } | |
1243 | ||
1da177e4 LT |
1244 | static void snapshot_dtr(struct dm_target *ti) |
1245 | { | |
cd45daff MP |
1246 | #ifdef CONFIG_DM_DEBUG |
1247 | int i; | |
1248 | #endif | |
028867ac | 1249 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1250 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1251 | |
c1f0c183 MS |
1252 | down_read(&_origins_lock); |
1253 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1254 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1255 | if (snap_src && snap_dest && (s == snap_src)) { |
1256 | down_write(&snap_dest->lock); | |
1257 | snap_dest->valid = 0; | |
1258 | up_write(&snap_dest->lock); | |
1259 | DMERR("Cancelling snapshot handover."); | |
1260 | } | |
1261 | up_read(&_origins_lock); | |
1262 | ||
1e03f97e MP |
1263 | if (dm_target_is_snapshot_merge(ti)) |
1264 | stop_merge(s); | |
1265 | ||
138728dc AK |
1266 | /* Prevent further origin writes from using this snapshot. */ |
1267 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1268 | unregister_snapshot(s); |
1269 | ||
879129d2 | 1270 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1271 | msleep(1); |
879129d2 MP |
1272 | /* |
1273 | * Ensure instructions in mempool_destroy aren't reordered | |
1274 | * before atomic_read. | |
1275 | */ | |
1276 | smp_mb(); | |
1277 | ||
cd45daff MP |
1278 | #ifdef CONFIG_DM_DEBUG |
1279 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1280 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1281 | #endif | |
1282 | ||
1283 | mempool_destroy(s->tracked_chunk_pool); | |
1284 | ||
31c93a0c | 1285 | __free_exceptions(s); |
1da177e4 | 1286 | |
92e86812 MP |
1287 | mempool_destroy(s->pending_pool); |
1288 | ||
fee1998e | 1289 | dm_exception_store_destroy(s->store); |
138728dc | 1290 | |
fc56f6fb MS |
1291 | dm_put_device(ti, s->cow); |
1292 | ||
c2411045 MP |
1293 | dm_put_device(ti, s->origin); |
1294 | ||
1da177e4 LT |
1295 | kfree(s); |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Flush a list of buffers. | |
1300 | */ | |
1301 | static void flush_bios(struct bio *bio) | |
1302 | { | |
1303 | struct bio *n; | |
1304 | ||
1305 | while (bio) { | |
1306 | n = bio->bi_next; | |
1307 | bio->bi_next = NULL; | |
1308 | generic_make_request(bio); | |
1309 | bio = n; | |
1310 | } | |
1311 | } | |
1312 | ||
515ad66c MP |
1313 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1314 | ||
1315 | /* | |
1316 | * Flush a list of buffers. | |
1317 | */ | |
1318 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1319 | { | |
1320 | struct bio *n; | |
1321 | int r; | |
1322 | ||
1323 | while (bio) { | |
1324 | n = bio->bi_next; | |
1325 | bio->bi_next = NULL; | |
1326 | r = do_origin(s->origin, bio); | |
1327 | if (r == DM_MAPIO_REMAPPED) | |
1328 | generic_make_request(bio); | |
1329 | bio = n; | |
1330 | } | |
1331 | } | |
1332 | ||
1da177e4 LT |
1333 | /* |
1334 | * Error a list of buffers. | |
1335 | */ | |
1336 | static void error_bios(struct bio *bio) | |
1337 | { | |
1338 | struct bio *n; | |
1339 | ||
1340 | while (bio) { | |
1341 | n = bio->bi_next; | |
1342 | bio->bi_next = NULL; | |
6712ecf8 | 1343 | bio_io_error(bio); |
1da177e4 LT |
1344 | bio = n; |
1345 | } | |
1346 | } | |
1347 | ||
695368ac | 1348 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1349 | { |
1350 | if (!s->valid) | |
1351 | return; | |
1352 | ||
1353 | if (err == -EIO) | |
1354 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1355 | else if (err == -ENOMEM) | |
1356 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1357 | ||
493df71c JB |
1358 | if (s->store->type->drop_snapshot) |
1359 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1360 | |
1361 | s->valid = 0; | |
1362 | ||
fc56f6fb | 1363 | dm_table_event(s->ti->table); |
76df1c65 AK |
1364 | } |
1365 | ||
028867ac | 1366 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1367 | { |
1d4989c8 | 1368 | struct dm_exception *e; |
1da177e4 | 1369 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1370 | struct bio *origin_bios = NULL; |
1371 | struct bio *snapshot_bios = NULL; | |
1372 | int error = 0; | |
1da177e4 | 1373 | |
76df1c65 AK |
1374 | if (!success) { |
1375 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1376 | down_write(&s->lock); |
695368ac | 1377 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1378 | error = 1; |
76df1c65 AK |
1379 | goto out; |
1380 | } | |
1381 | ||
3510cb94 | 1382 | e = alloc_completed_exception(); |
76df1c65 | 1383 | if (!e) { |
1da177e4 | 1384 | down_write(&s->lock); |
695368ac | 1385 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1386 | error = 1; |
76df1c65 AK |
1387 | goto out; |
1388 | } | |
1389 | *e = pe->e; | |
1da177e4 | 1390 | |
76df1c65 AK |
1391 | down_write(&s->lock); |
1392 | if (!s->valid) { | |
3510cb94 | 1393 | free_completed_exception(e); |
9d493fa8 | 1394 | error = 1; |
76df1c65 | 1395 | goto out; |
1da177e4 LT |
1396 | } |
1397 | ||
615d1eb9 MS |
1398 | /* Check for conflicting reads */ |
1399 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1400 | |
9d493fa8 AK |
1401 | /* |
1402 | * Add a proper exception, and remove the | |
1403 | * in-flight exception from the list. | |
1404 | */ | |
3510cb94 | 1405 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1406 | |
a2d2b034 | 1407 | out: |
3510cb94 | 1408 | dm_remove_exception(&pe->e); |
9d493fa8 | 1409 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c MP |
1410 | origin_bios = bio_list_get(&pe->origin_bios); |
1411 | free_pending_exception(pe); | |
1da177e4 | 1412 | |
73dfd078 MP |
1413 | increment_pending_exceptions_done_count(); |
1414 | ||
9d493fa8 AK |
1415 | up_write(&s->lock); |
1416 | ||
1417 | /* Submit any pending write bios */ | |
1418 | if (error) | |
1419 | error_bios(snapshot_bios); | |
1420 | else | |
1421 | flush_bios(snapshot_bios); | |
1422 | ||
515ad66c | 1423 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1424 | } |
1425 | ||
1426 | static void commit_callback(void *context, int success) | |
1427 | { | |
028867ac AK |
1428 | struct dm_snap_pending_exception *pe = context; |
1429 | ||
1da177e4 LT |
1430 | pending_complete(pe, success); |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * Called when the copy I/O has finished. kcopyd actually runs | |
1435 | * this code so don't block. | |
1436 | */ | |
4cdc1d1f | 1437 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1438 | { |
028867ac | 1439 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1440 | struct dm_snapshot *s = pe->snap; |
1441 | ||
1442 | if (read_err || write_err) | |
1443 | pending_complete(pe, 0); | |
1444 | ||
1445 | else | |
1446 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1447 | s->store->type->commit_exception(s->store, &pe->e, |
1448 | commit_callback, pe); | |
1da177e4 LT |
1449 | } |
1450 | ||
1451 | /* | |
1452 | * Dispatches the copy operation to kcopyd. | |
1453 | */ | |
028867ac | 1454 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1455 | { |
1456 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1457 | struct dm_io_region src, dest; |
1da177e4 LT |
1458 | struct block_device *bdev = s->origin->bdev; |
1459 | sector_t dev_size; | |
1460 | ||
1461 | dev_size = get_dev_size(bdev); | |
1462 | ||
1463 | src.bdev = bdev; | |
71fab00a | 1464 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1465 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1466 | |
fc56f6fb | 1467 | dest.bdev = s->cow->bdev; |
71fab00a | 1468 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1469 | dest.count = src.count; |
1470 | ||
1471 | /* Hand over to kcopyd */ | |
a2d2b034 | 1472 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1473 | } |
1474 | ||
2913808e MP |
1475 | static struct dm_snap_pending_exception * |
1476 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1477 | { | |
3510cb94 | 1478 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1479 | |
1480 | if (!e) | |
1481 | return NULL; | |
1482 | ||
1483 | return container_of(e, struct dm_snap_pending_exception, e); | |
1484 | } | |
1485 | ||
1da177e4 LT |
1486 | /* |
1487 | * Looks to see if this snapshot already has a pending exception | |
1488 | * for this chunk, otherwise it allocates a new one and inserts | |
1489 | * it into the pending table. | |
1490 | * | |
1491 | * NOTE: a write lock must be held on snap->lock before calling | |
1492 | * this. | |
1493 | */ | |
028867ac | 1494 | static struct dm_snap_pending_exception * |
c6621392 MP |
1495 | __find_pending_exception(struct dm_snapshot *s, |
1496 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1497 | { |
c6621392 | 1498 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1499 | |
2913808e MP |
1500 | pe2 = __lookup_pending_exception(s, chunk); |
1501 | if (pe2) { | |
76df1c65 | 1502 | free_pending_exception(pe); |
2913808e | 1503 | return pe2; |
1da177e4 LT |
1504 | } |
1505 | ||
76df1c65 AK |
1506 | pe->e.old_chunk = chunk; |
1507 | bio_list_init(&pe->origin_bios); | |
1508 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 AK |
1509 | pe->started = 0; |
1510 | ||
493df71c | 1511 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1512 | free_pending_exception(pe); |
1513 | return NULL; | |
1514 | } | |
1515 | ||
3510cb94 | 1516 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1517 | |
1da177e4 LT |
1518 | return pe; |
1519 | } | |
1520 | ||
1d4989c8 | 1521 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1522 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1523 | { |
fc56f6fb | 1524 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1525 | bio->bi_sector = chunk_to_sector(s->store, |
1526 | dm_chunk_number(e->new_chunk) + | |
1527 | (chunk - e->old_chunk)) + | |
1528 | (bio->bi_sector & | |
1529 | s->store->chunk_mask); | |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1533 | union map_info *map_context) | |
1534 | { | |
1d4989c8 | 1535 | struct dm_exception *e; |
028867ac | 1536 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1537 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1538 | chunk_t chunk; |
028867ac | 1539 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1540 | |
d87f4c14 | 1541 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1542 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1543 | return DM_MAPIO_REMAPPED; |
1544 | } | |
1545 | ||
71fab00a | 1546 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1547 | |
1548 | /* Full snapshots are not usable */ | |
76df1c65 | 1549 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1550 | if (!s->valid) |
f6a80ea8 | 1551 | return -EIO; |
1da177e4 | 1552 | |
ba40a2aa AK |
1553 | /* FIXME: should only take write lock if we need |
1554 | * to copy an exception */ | |
1555 | down_write(&s->lock); | |
1556 | ||
1557 | if (!s->valid) { | |
1558 | r = -EIO; | |
1559 | goto out_unlock; | |
1560 | } | |
1561 | ||
1562 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1563 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1564 | if (e) { |
d74f81f8 | 1565 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1566 | goto out_unlock; |
1567 | } | |
1568 | ||
1da177e4 LT |
1569 | /* |
1570 | * Write to snapshot - higher level takes care of RW/RO | |
1571 | * flags so we should only get this if we are | |
1572 | * writeable. | |
1573 | */ | |
1574 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1575 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1576 | if (!pe) { |
c6621392 MP |
1577 | up_write(&s->lock); |
1578 | pe = alloc_pending_exception(s); | |
1579 | down_write(&s->lock); | |
1580 | ||
1581 | if (!s->valid) { | |
1582 | free_pending_exception(pe); | |
1583 | r = -EIO; | |
1584 | goto out_unlock; | |
1585 | } | |
1586 | ||
3510cb94 | 1587 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1588 | if (e) { |
1589 | free_pending_exception(pe); | |
1590 | remap_exception(s, e, bio, chunk); | |
1591 | goto out_unlock; | |
1592 | } | |
1593 | ||
c6621392 | 1594 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1595 | if (!pe) { |
1596 | __invalidate_snapshot(s, -ENOMEM); | |
1597 | r = -EIO; | |
1598 | goto out_unlock; | |
1599 | } | |
1da177e4 LT |
1600 | } |
1601 | ||
d74f81f8 | 1602 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1603 | bio_list_add(&pe->snapshot_bios, bio); |
1604 | ||
d2a7ad29 | 1605 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1606 | |
76df1c65 AK |
1607 | if (!pe->started) { |
1608 | /* this is protected by snap->lock */ | |
1609 | pe->started = 1; | |
ba40a2aa | 1610 | up_write(&s->lock); |
76df1c65 | 1611 | start_copy(pe); |
ba40a2aa AK |
1612 | goto out; |
1613 | } | |
cd45daff | 1614 | } else { |
ba40a2aa | 1615 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1616 | map_context->ptr = track_chunk(s, chunk); |
1617 | } | |
1da177e4 | 1618 | |
a2d2b034 | 1619 | out_unlock: |
ba40a2aa | 1620 | up_write(&s->lock); |
a2d2b034 | 1621 | out: |
1da177e4 LT |
1622 | return r; |
1623 | } | |
1624 | ||
3452c2a1 MP |
1625 | /* |
1626 | * A snapshot-merge target behaves like a combination of a snapshot | |
1627 | * target and a snapshot-origin target. It only generates new | |
1628 | * exceptions in other snapshots and not in the one that is being | |
1629 | * merged. | |
1630 | * | |
1631 | * For each chunk, if there is an existing exception, it is used to | |
1632 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1633 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1634 | * If merging is currently taking place on the chunk in question, the |
1635 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 MP |
1636 | */ |
1637 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio, | |
1638 | union map_info *map_context) | |
1639 | { | |
1640 | struct dm_exception *e; | |
1641 | struct dm_snapshot *s = ti->private; | |
1642 | int r = DM_MAPIO_REMAPPED; | |
1643 | chunk_t chunk; | |
1644 | ||
d87f4c14 | 1645 | if (bio->bi_rw & REQ_FLUSH) { |
57cba5d3 | 1646 | if (!map_context->target_request_nr) |
10b8106a MS |
1647 | bio->bi_bdev = s->origin->bdev; |
1648 | else | |
1649 | bio->bi_bdev = s->cow->bdev; | |
1650 | map_context->ptr = NULL; | |
1651 | return DM_MAPIO_REMAPPED; | |
1652 | } | |
1653 | ||
3452c2a1 MP |
1654 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1655 | ||
9fe86254 | 1656 | down_write(&s->lock); |
3452c2a1 | 1657 | |
d2fdb776 MP |
1658 | /* Full merging snapshots are redirected to the origin */ |
1659 | if (!s->valid) | |
1660 | goto redirect_to_origin; | |
3452c2a1 MP |
1661 | |
1662 | /* If the block is already remapped - use that */ | |
1663 | e = dm_lookup_exception(&s->complete, chunk); | |
1664 | if (e) { | |
9fe86254 MP |
1665 | /* Queue writes overlapping with chunks being merged */ |
1666 | if (bio_rw(bio) == WRITE && | |
1667 | chunk >= s->first_merging_chunk && | |
1668 | chunk < (s->first_merging_chunk + | |
1669 | s->num_merging_chunks)) { | |
1670 | bio->bi_bdev = s->origin->bdev; | |
1671 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1672 | r = DM_MAPIO_SUBMITTED; | |
1673 | goto out_unlock; | |
1674 | } | |
17aa0332 | 1675 | |
3452c2a1 | 1676 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1677 | |
1678 | if (bio_rw(bio) == WRITE) | |
1679 | map_context->ptr = track_chunk(s, chunk); | |
3452c2a1 MP |
1680 | goto out_unlock; |
1681 | } | |
1682 | ||
d2fdb776 | 1683 | redirect_to_origin: |
3452c2a1 MP |
1684 | bio->bi_bdev = s->origin->bdev; |
1685 | ||
1686 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1687 | up_write(&s->lock); |
3452c2a1 MP |
1688 | return do_origin(s->origin, bio); |
1689 | } | |
1690 | ||
1691 | out_unlock: | |
9fe86254 | 1692 | up_write(&s->lock); |
3452c2a1 MP |
1693 | |
1694 | return r; | |
1695 | } | |
1696 | ||
cd45daff MP |
1697 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1698 | int error, union map_info *map_context) | |
1699 | { | |
1700 | struct dm_snapshot *s = ti->private; | |
1701 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1702 | ||
1703 | if (c) | |
1704 | stop_tracking_chunk(s, c); | |
1705 | ||
1706 | return 0; | |
1707 | } | |
1708 | ||
1e03f97e MP |
1709 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1710 | { | |
1711 | struct dm_snapshot *s = ti->private; | |
1712 | ||
1713 | stop_merge(s); | |
1714 | } | |
1715 | ||
c1f0c183 MS |
1716 | static int snapshot_preresume(struct dm_target *ti) |
1717 | { | |
1718 | int r = 0; | |
1719 | struct dm_snapshot *s = ti->private; | |
1720 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1721 | ||
1722 | down_read(&_origins_lock); | |
9d3b15c4 | 1723 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1724 | if (snap_src && snap_dest) { |
1725 | down_read(&snap_src->lock); | |
1726 | if (s == snap_src) { | |
1727 | DMERR("Unable to resume snapshot source until " | |
1728 | "handover completes."); | |
1729 | r = -EINVAL; | |
b83b2f29 | 1730 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
1731 | DMERR("Unable to perform snapshot handover until " |
1732 | "source is suspended."); | |
1733 | r = -EINVAL; | |
1734 | } | |
1735 | up_read(&snap_src->lock); | |
1736 | } | |
1737 | up_read(&_origins_lock); | |
1738 | ||
1739 | return r; | |
1740 | } | |
1741 | ||
1da177e4 LT |
1742 | static void snapshot_resume(struct dm_target *ti) |
1743 | { | |
028867ac | 1744 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1745 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1746 | ||
1747 | down_read(&_origins_lock); | |
9d3b15c4 | 1748 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1749 | if (snap_src && snap_dest) { |
1750 | down_write(&snap_src->lock); | |
1751 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1752 | __handover_exceptions(snap_src, snap_dest); | |
1753 | up_write(&snap_dest->lock); | |
1754 | up_write(&snap_src->lock); | |
1755 | } | |
1756 | up_read(&_origins_lock); | |
1757 | ||
1758 | /* Now we have correct chunk size, reregister */ | |
1759 | reregister_snapshot(s); | |
1da177e4 | 1760 | |
aa14edeb AK |
1761 | down_write(&s->lock); |
1762 | s->active = 1; | |
1763 | up_write(&s->lock); | |
1da177e4 LT |
1764 | } |
1765 | ||
1e03f97e MP |
1766 | static sector_t get_origin_minimum_chunksize(struct block_device *bdev) |
1767 | { | |
1768 | sector_t min_chunksize; | |
1769 | ||
1770 | down_read(&_origins_lock); | |
1771 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1772 | up_read(&_origins_lock); | |
1773 | ||
1774 | return min_chunksize; | |
1775 | } | |
1776 | ||
1777 | static void snapshot_merge_resume(struct dm_target *ti) | |
1778 | { | |
1779 | struct dm_snapshot *s = ti->private; | |
1780 | ||
1781 | /* | |
1782 | * Handover exceptions from existing snapshot. | |
1783 | */ | |
1784 | snapshot_resume(ti); | |
1785 | ||
1786 | /* | |
1787 | * snapshot-merge acts as an origin, so set ti->split_io | |
1788 | */ | |
1789 | ti->split_io = get_origin_minimum_chunksize(s->origin->bdev); | |
1790 | ||
1791 | start_merge(s); | |
1792 | } | |
1793 | ||
1da177e4 LT |
1794 | static int snapshot_status(struct dm_target *ti, status_type_t type, |
1795 | char *result, unsigned int maxlen) | |
1796 | { | |
2e4a31df | 1797 | unsigned sz = 0; |
028867ac | 1798 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1799 | |
1800 | switch (type) { | |
1801 | case STATUSTYPE_INFO: | |
94e76572 MP |
1802 | |
1803 | down_write(&snap->lock); | |
1804 | ||
1da177e4 | 1805 | if (!snap->valid) |
2e4a31df | 1806 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1807 | else if (snap->merge_failed) |
1808 | DMEMIT("Merge failed"); | |
1da177e4 | 1809 | else { |
985903bb MS |
1810 | if (snap->store->type->usage) { |
1811 | sector_t total_sectors, sectors_allocated, | |
1812 | metadata_sectors; | |
1813 | snap->store->type->usage(snap->store, | |
1814 | &total_sectors, | |
1815 | §ors_allocated, | |
1816 | &metadata_sectors); | |
1817 | DMEMIT("%llu/%llu %llu", | |
1818 | (unsigned long long)sectors_allocated, | |
1819 | (unsigned long long)total_sectors, | |
1820 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1821 | } |
1822 | else | |
2e4a31df | 1823 | DMEMIT("Unknown"); |
1da177e4 | 1824 | } |
94e76572 MP |
1825 | |
1826 | up_write(&snap->lock); | |
1827 | ||
1da177e4 LT |
1828 | break; |
1829 | ||
1830 | case STATUSTYPE_TABLE: | |
1831 | /* | |
1832 | * kdevname returns a static pointer so we need | |
1833 | * to make private copies if the output is to | |
1834 | * make sense. | |
1835 | */ | |
fc56f6fb | 1836 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1837 | snap->store->type->status(snap->store, type, result + sz, |
1838 | maxlen - sz); | |
1da177e4 LT |
1839 | break; |
1840 | } | |
1841 | ||
1842 | return 0; | |
1843 | } | |
1844 | ||
8811f46c MS |
1845 | static int snapshot_iterate_devices(struct dm_target *ti, |
1846 | iterate_devices_callout_fn fn, void *data) | |
1847 | { | |
1848 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
1849 | int r; |
1850 | ||
1851 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 1852 | |
1e5554c8 MP |
1853 | if (!r) |
1854 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
1855 | ||
1856 | return r; | |
8811f46c MS |
1857 | } |
1858 | ||
1859 | ||
1da177e4 LT |
1860 | /*----------------------------------------------------------------- |
1861 | * Origin methods | |
1862 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1863 | |
1864 | /* | |
1865 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1866 | * supplied bio was ignored. The caller may submit it immediately. | |
1867 | * (No remapping actually occurs as the origin is always a direct linear | |
1868 | * map.) | |
1869 | * | |
1870 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1871 | * and any supplied bio is added to a list to be submitted once all | |
1872 | * the necessary exceptions exist. | |
1873 | */ | |
1874 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1875 | struct bio *bio) | |
1da177e4 | 1876 | { |
515ad66c | 1877 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1878 | struct dm_snapshot *snap; |
1d4989c8 | 1879 | struct dm_exception *e; |
515ad66c MP |
1880 | struct dm_snap_pending_exception *pe; |
1881 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1882 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1883 | chunk_t chunk; |
1884 | ||
1885 | /* Do all the snapshots on this origin */ | |
1886 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
1887 | /* |
1888 | * Don't make new exceptions in a merging snapshot | |
1889 | * because it has effectively been deleted | |
1890 | */ | |
1891 | if (dm_target_is_snapshot_merge(snap->ti)) | |
1892 | continue; | |
1893 | ||
76df1c65 AK |
1894 | down_write(&snap->lock); |
1895 | ||
aa14edeb AK |
1896 | /* Only deal with valid and active snapshots */ |
1897 | if (!snap->valid || !snap->active) | |
76df1c65 | 1898 | goto next_snapshot; |
1da177e4 | 1899 | |
d5e404c1 | 1900 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1901 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1902 | goto next_snapshot; |
1da177e4 LT |
1903 | |
1904 | /* | |
1905 | * Remember, different snapshots can have | |
1906 | * different chunk sizes. | |
1907 | */ | |
9eaae8ff | 1908 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1909 | |
1910 | /* | |
1911 | * Check exception table to see if block | |
1912 | * is already remapped in this snapshot | |
1913 | * and trigger an exception if not. | |
1914 | */ | |
3510cb94 | 1915 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1916 | if (e) |
1917 | goto next_snapshot; | |
1918 | ||
2913808e | 1919 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1920 | if (!pe) { |
c6621392 MP |
1921 | up_write(&snap->lock); |
1922 | pe = alloc_pending_exception(snap); | |
1923 | down_write(&snap->lock); | |
1924 | ||
1925 | if (!snap->valid) { | |
1926 | free_pending_exception(pe); | |
1927 | goto next_snapshot; | |
1928 | } | |
1929 | ||
3510cb94 | 1930 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1931 | if (e) { |
1932 | free_pending_exception(pe); | |
1933 | goto next_snapshot; | |
1934 | } | |
1935 | ||
c6621392 | 1936 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1937 | if (!pe) { |
1938 | __invalidate_snapshot(snap, -ENOMEM); | |
1939 | goto next_snapshot; | |
1940 | } | |
76df1c65 AK |
1941 | } |
1942 | ||
515ad66c | 1943 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 1944 | |
515ad66c MP |
1945 | /* |
1946 | * If an origin bio was supplied, queue it to wait for the | |
1947 | * completion of this exception, and start this one last, | |
1948 | * at the end of the function. | |
1949 | */ | |
1950 | if (bio) { | |
1951 | bio_list_add(&pe->origin_bios, bio); | |
1952 | bio = NULL; | |
76df1c65 | 1953 | |
515ad66c MP |
1954 | if (!pe->started) { |
1955 | pe->started = 1; | |
1956 | pe_to_start_last = pe; | |
1957 | } | |
76df1c65 AK |
1958 | } |
1959 | ||
1960 | if (!pe->started) { | |
1961 | pe->started = 1; | |
515ad66c | 1962 | pe_to_start_now = pe; |
1da177e4 LT |
1963 | } |
1964 | ||
a2d2b034 | 1965 | next_snapshot: |
1da177e4 | 1966 | up_write(&snap->lock); |
1da177e4 | 1967 | |
515ad66c MP |
1968 | if (pe_to_start_now) { |
1969 | start_copy(pe_to_start_now); | |
1970 | pe_to_start_now = NULL; | |
1971 | } | |
b4b610f6 AK |
1972 | } |
1973 | ||
1da177e4 | 1974 | /* |
515ad66c MP |
1975 | * Submit the exception against which the bio is queued last, |
1976 | * to give the other exceptions a head start. | |
1da177e4 | 1977 | */ |
515ad66c MP |
1978 | if (pe_to_start_last) |
1979 | start_copy(pe_to_start_last); | |
1da177e4 LT |
1980 | |
1981 | return r; | |
1982 | } | |
1983 | ||
1984 | /* | |
1985 | * Called on a write from the origin driver. | |
1986 | */ | |
1987 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1988 | { | |
1989 | struct origin *o; | |
d2a7ad29 | 1990 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1991 | |
1992 | down_read(&_origins_lock); | |
1993 | o = __lookup_origin(origin->bdev); | |
1994 | if (o) | |
9eaae8ff | 1995 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
1996 | up_read(&_origins_lock); |
1997 | ||
1998 | return r; | |
1999 | } | |
2000 | ||
73dfd078 MP |
2001 | /* |
2002 | * Trigger exceptions in all non-merging snapshots. | |
2003 | * | |
2004 | * The chunk size of the merging snapshot may be larger than the chunk | |
2005 | * size of some other snapshot so we may need to reallocate multiple | |
2006 | * chunks in other snapshots. | |
2007 | * | |
2008 | * We scan all the overlapping exceptions in the other snapshots. | |
2009 | * Returns 1 if anything was reallocated and must be waited for, | |
2010 | * otherwise returns 0. | |
2011 | * | |
2012 | * size must be a multiple of merging_snap's chunk_size. | |
2013 | */ | |
2014 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2015 | sector_t sector, unsigned size) | |
2016 | { | |
2017 | int must_wait = 0; | |
2018 | sector_t n; | |
2019 | struct origin *o; | |
2020 | ||
2021 | /* | |
2022 | * The origin's __minimum_chunk_size() got stored in split_io | |
2023 | * by snapshot_merge_resume(). | |
2024 | */ | |
2025 | down_read(&_origins_lock); | |
2026 | o = __lookup_origin(merging_snap->origin->bdev); | |
2027 | for (n = 0; n < size; n += merging_snap->ti->split_io) | |
2028 | if (__origin_write(&o->snapshots, sector + n, NULL) == | |
2029 | DM_MAPIO_SUBMITTED) | |
2030 | must_wait = 1; | |
2031 | up_read(&_origins_lock); | |
2032 | ||
2033 | return must_wait; | |
2034 | } | |
2035 | ||
1da177e4 LT |
2036 | /* |
2037 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2038 | */ | |
2039 | ||
2040 | /* | |
2041 | * Construct an origin mapping: <dev_path> | |
2042 | * The context for an origin is merely a 'struct dm_dev *' | |
2043 | * pointing to the real device. | |
2044 | */ | |
2045 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2046 | { | |
2047 | int r; | |
2048 | struct dm_dev *dev; | |
2049 | ||
2050 | if (argc != 1) { | |
72d94861 | 2051 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2052 | return -EINVAL; |
2053 | } | |
2054 | ||
8215d6ec | 2055 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev); |
1da177e4 LT |
2056 | if (r) { |
2057 | ti->error = "Cannot get target device"; | |
2058 | return r; | |
2059 | } | |
2060 | ||
2061 | ti->private = dev; | |
494b3ee7 MP |
2062 | ti->num_flush_requests = 1; |
2063 | ||
1da177e4 LT |
2064 | return 0; |
2065 | } | |
2066 | ||
2067 | static void origin_dtr(struct dm_target *ti) | |
2068 | { | |
028867ac | 2069 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2070 | dm_put_device(ti, dev); |
2071 | } | |
2072 | ||
2073 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
2074 | union map_info *map_context) | |
2075 | { | |
028867ac | 2076 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2077 | bio->bi_bdev = dev->bdev; |
2078 | ||
d87f4c14 | 2079 | if (bio->bi_rw & REQ_FLUSH) |
494b3ee7 MP |
2080 | return DM_MAPIO_REMAPPED; |
2081 | ||
1da177e4 | 2082 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 2083 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
2084 | } |
2085 | ||
1da177e4 LT |
2086 | /* |
2087 | * Set the target "split_io" field to the minimum of all the snapshots' | |
2088 | * chunk sizes. | |
2089 | */ | |
2090 | static void origin_resume(struct dm_target *ti) | |
2091 | { | |
028867ac | 2092 | struct dm_dev *dev = ti->private; |
1da177e4 | 2093 | |
1e03f97e | 2094 | ti->split_io = get_origin_minimum_chunksize(dev->bdev); |
1da177e4 LT |
2095 | } |
2096 | ||
2097 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
2098 | unsigned int maxlen) | |
2099 | { | |
028867ac | 2100 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2101 | |
2102 | switch (type) { | |
2103 | case STATUSTYPE_INFO: | |
2104 | result[0] = '\0'; | |
2105 | break; | |
2106 | ||
2107 | case STATUSTYPE_TABLE: | |
2108 | snprintf(result, maxlen, "%s", dev->name); | |
2109 | break; | |
2110 | } | |
2111 | ||
2112 | return 0; | |
2113 | } | |
2114 | ||
b1d55528 MP |
2115 | static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
2116 | struct bio_vec *biovec, int max_size) | |
2117 | { | |
2118 | struct dm_dev *dev = ti->private; | |
2119 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2120 | ||
2121 | if (!q->merge_bvec_fn) | |
2122 | return max_size; | |
2123 | ||
2124 | bvm->bi_bdev = dev->bdev; | |
2125 | bvm->bi_sector = bvm->bi_sector; | |
2126 | ||
2127 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2128 | } | |
2129 | ||
8811f46c MS |
2130 | static int origin_iterate_devices(struct dm_target *ti, |
2131 | iterate_devices_callout_fn fn, void *data) | |
2132 | { | |
2133 | struct dm_dev *dev = ti->private; | |
2134 | ||
2135 | return fn(ti, dev, 0, ti->len, data); | |
2136 | } | |
2137 | ||
1da177e4 LT |
2138 | static struct target_type origin_target = { |
2139 | .name = "snapshot-origin", | |
b83b2f29 | 2140 | .version = {1, 7, 1}, |
1da177e4 LT |
2141 | .module = THIS_MODULE, |
2142 | .ctr = origin_ctr, | |
2143 | .dtr = origin_dtr, | |
2144 | .map = origin_map, | |
2145 | .resume = origin_resume, | |
2146 | .status = origin_status, | |
b1d55528 | 2147 | .merge = origin_merge, |
8811f46c | 2148 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2149 | }; |
2150 | ||
2151 | static struct target_type snapshot_target = { | |
2152 | .name = "snapshot", | |
b83b2f29 | 2153 | .version = {1, 10, 0}, |
1da177e4 LT |
2154 | .module = THIS_MODULE, |
2155 | .ctr = snapshot_ctr, | |
2156 | .dtr = snapshot_dtr, | |
2157 | .map = snapshot_map, | |
cd45daff | 2158 | .end_io = snapshot_end_io, |
c1f0c183 | 2159 | .preresume = snapshot_preresume, |
1da177e4 LT |
2160 | .resume = snapshot_resume, |
2161 | .status = snapshot_status, | |
8811f46c | 2162 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2163 | }; |
2164 | ||
d698aa45 MP |
2165 | static struct target_type merge_target = { |
2166 | .name = dm_snapshot_merge_target_name, | |
b83b2f29 | 2167 | .version = {1, 1, 0}, |
d698aa45 MP |
2168 | .module = THIS_MODULE, |
2169 | .ctr = snapshot_ctr, | |
2170 | .dtr = snapshot_dtr, | |
3452c2a1 | 2171 | .map = snapshot_merge_map, |
d698aa45 | 2172 | .end_io = snapshot_end_io, |
1e03f97e | 2173 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2174 | .preresume = snapshot_preresume, |
1e03f97e | 2175 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2176 | .status = snapshot_status, |
2177 | .iterate_devices = snapshot_iterate_devices, | |
2178 | }; | |
2179 | ||
1da177e4 LT |
2180 | static int __init dm_snapshot_init(void) |
2181 | { | |
2182 | int r; | |
2183 | ||
4db6bfe0 AK |
2184 | r = dm_exception_store_init(); |
2185 | if (r) { | |
2186 | DMERR("Failed to initialize exception stores"); | |
2187 | return r; | |
2188 | } | |
2189 | ||
1da177e4 | 2190 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2191 | if (r < 0) { |
1da177e4 | 2192 | DMERR("snapshot target register failed %d", r); |
034a186d | 2193 | goto bad_register_snapshot_target; |
1da177e4 LT |
2194 | } |
2195 | ||
2196 | r = dm_register_target(&origin_target); | |
2197 | if (r < 0) { | |
72d94861 | 2198 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2199 | goto bad_register_origin_target; |
2200 | } | |
2201 | ||
2202 | r = dm_register_target(&merge_target); | |
2203 | if (r < 0) { | |
2204 | DMERR("Merge target register failed %d", r); | |
2205 | goto bad_register_merge_target; | |
1da177e4 LT |
2206 | } |
2207 | ||
2208 | r = init_origin_hash(); | |
2209 | if (r) { | |
2210 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2211 | goto bad_origin_hash; |
1da177e4 LT |
2212 | } |
2213 | ||
1d4989c8 | 2214 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2215 | if (!exception_cache) { |
2216 | DMERR("Couldn't create exception cache."); | |
2217 | r = -ENOMEM; | |
d698aa45 | 2218 | goto bad_exception_cache; |
1da177e4 LT |
2219 | } |
2220 | ||
028867ac | 2221 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2222 | if (!pending_cache) { |
2223 | DMERR("Couldn't create pending cache."); | |
2224 | r = -ENOMEM; | |
d698aa45 | 2225 | goto bad_pending_cache; |
1da177e4 LT |
2226 | } |
2227 | ||
cd45daff MP |
2228 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
2229 | if (!tracked_chunk_cache) { | |
2230 | DMERR("Couldn't create cache to track chunks in use."); | |
2231 | r = -ENOMEM; | |
d698aa45 | 2232 | goto bad_tracked_chunk_cache; |
cd45daff MP |
2233 | } |
2234 | ||
1da177e4 LT |
2235 | return 0; |
2236 | ||
d698aa45 | 2237 | bad_tracked_chunk_cache: |
1da177e4 | 2238 | kmem_cache_destroy(pending_cache); |
d698aa45 | 2239 | bad_pending_cache: |
1da177e4 | 2240 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2241 | bad_exception_cache: |
1da177e4 | 2242 | exit_origin_hash(); |
d698aa45 MP |
2243 | bad_origin_hash: |
2244 | dm_unregister_target(&merge_target); | |
2245 | bad_register_merge_target: | |
1da177e4 | 2246 | dm_unregister_target(&origin_target); |
d698aa45 | 2247 | bad_register_origin_target: |
1da177e4 | 2248 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2249 | bad_register_snapshot_target: |
2250 | dm_exception_store_exit(); | |
d698aa45 | 2251 | |
1da177e4 LT |
2252 | return r; |
2253 | } | |
2254 | ||
2255 | static void __exit dm_snapshot_exit(void) | |
2256 | { | |
10d3bd09 MP |
2257 | dm_unregister_target(&snapshot_target); |
2258 | dm_unregister_target(&origin_target); | |
d698aa45 | 2259 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2260 | |
2261 | exit_origin_hash(); | |
1da177e4 LT |
2262 | kmem_cache_destroy(pending_cache); |
2263 | kmem_cache_destroy(exception_cache); | |
cd45daff | 2264 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
2265 | |
2266 | dm_exception_store_exit(); | |
1da177e4 LT |
2267 | } |
2268 | ||
2269 | /* Module hooks */ | |
2270 | module_init(dm_snapshot_init); | |
2271 | module_exit(dm_snapshot_exit); | |
2272 | ||
2273 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2274 | MODULE_AUTHOR("Joe Thornber"); | |
2275 | MODULE_LICENSE("GPL"); |