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 | ||
1da177e4 LT |
32 | /* |
33 | * The percentage increment we will wake up users at | |
34 | */ | |
35 | #define WAKE_UP_PERCENT 5 | |
36 | ||
37 | /* | |
38 | * kcopyd priority of snapshot operations | |
39 | */ | |
40 | #define SNAPSHOT_COPY_PRIORITY 2 | |
41 | ||
42 | /* | |
8ee2767a | 43 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 44 | */ |
8ee2767a | 45 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 46 | |
cd45daff MP |
47 | /* |
48 | * The size of the mempool used to track chunks in use. | |
49 | */ | |
50 | #define MIN_IOS 256 | |
51 | ||
ccc45ea8 JB |
52 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
53 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
54 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
55 | ||
191437a5 | 56 | struct dm_exception_table { |
ccc45ea8 JB |
57 | uint32_t hash_mask; |
58 | unsigned hash_shift; | |
59 | struct list_head *table; | |
60 | }; | |
61 | ||
62 | struct dm_snapshot { | |
63 | struct rw_semaphore lock; | |
64 | ||
65 | struct dm_dev *origin; | |
fc56f6fb MS |
66 | struct dm_dev *cow; |
67 | ||
68 | struct dm_target *ti; | |
ccc45ea8 JB |
69 | |
70 | /* List of snapshots per Origin */ | |
71 | struct list_head list; | |
72 | ||
d8ddb1cf MS |
73 | /* |
74 | * You can't use a snapshot if this is 0 (e.g. if full). | |
75 | * A snapshot-merge target never clears this. | |
76 | */ | |
ccc45ea8 JB |
77 | int valid; |
78 | ||
79 | /* Origin writes don't trigger exceptions until this is set */ | |
80 | int active; | |
81 | ||
c26655ca MS |
82 | /* Whether or not owning mapped_device is suspended */ |
83 | int suspended; | |
84 | ||
ccc45ea8 JB |
85 | atomic_t pending_exceptions_count; |
86 | ||
924e600d MS |
87 | mempool_t *pending_pool; |
88 | ||
191437a5 JB |
89 | struct dm_exception_table pending; |
90 | struct dm_exception_table complete; | |
ccc45ea8 JB |
91 | |
92 | /* | |
93 | * pe_lock protects all pending_exception operations and access | |
94 | * as well as the snapshot_bios list. | |
95 | */ | |
96 | spinlock_t pe_lock; | |
97 | ||
924e600d MS |
98 | /* Chunks with outstanding reads */ |
99 | spinlock_t tracked_chunk_lock; | |
100 | mempool_t *tracked_chunk_pool; | |
101 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
102 | ||
ccc45ea8 JB |
103 | /* The on disk metadata handler */ |
104 | struct dm_exception_store *store; | |
105 | ||
106 | struct dm_kcopyd_client *kcopyd_client; | |
107 | ||
924e600d MS |
108 | /* Wait for events based on state_bits */ |
109 | unsigned long state_bits; | |
110 | ||
111 | /* Range of chunks currently being merged. */ | |
112 | chunk_t first_merging_chunk; | |
113 | int num_merging_chunks; | |
1e03f97e | 114 | |
d8ddb1cf MS |
115 | /* |
116 | * The merge operation failed if this flag is set. | |
117 | * Failure modes are handled as follows: | |
118 | * - I/O error reading the header | |
119 | * => don't load the target; abort. | |
120 | * - Header does not have "valid" flag set | |
121 | * => use the origin; forget about the snapshot. | |
122 | * - I/O error when reading exceptions | |
123 | * => don't load the target; abort. | |
124 | * (We can't use the intermediate origin state.) | |
125 | * - I/O error while merging | |
126 | * => stop merging; set merge_failed; process I/O normally. | |
127 | */ | |
128 | int merge_failed; | |
129 | ||
9fe86254 MP |
130 | /* |
131 | * Incoming bios that overlap with chunks being merged must wait | |
132 | * for them to be committed. | |
133 | */ | |
134 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
135 | }; |
136 | ||
1e03f97e MP |
137 | /* |
138 | * state_bits: | |
139 | * RUNNING_MERGE - Merge operation is in progress. | |
140 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
141 | * cleared afterwards. | |
142 | */ | |
143 | #define RUNNING_MERGE 0 | |
144 | #define SHUTDOWN_MERGE 1 | |
145 | ||
c2411045 MP |
146 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
147 | { | |
148 | return s->origin; | |
149 | } | |
150 | EXPORT_SYMBOL(dm_snap_origin); | |
151 | ||
fc56f6fb MS |
152 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
153 | { | |
154 | return s->cow; | |
155 | } | |
156 | EXPORT_SYMBOL(dm_snap_cow); | |
157 | ||
ccc45ea8 JB |
158 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
159 | chunk_t chunk) | |
160 | { | |
161 | return chunk << store->chunk_shift; | |
162 | } | |
163 | ||
164 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
165 | { | |
166 | /* | |
167 | * There is only ever one instance of a particular block | |
168 | * device so we can compare pointers safely. | |
169 | */ | |
170 | return lhs == rhs; | |
171 | } | |
172 | ||
028867ac | 173 | struct dm_snap_pending_exception { |
1d4989c8 | 174 | struct dm_exception e; |
1da177e4 LT |
175 | |
176 | /* | |
177 | * Origin buffers waiting for this to complete are held | |
178 | * in a bio list | |
179 | */ | |
180 | struct bio_list origin_bios; | |
181 | struct bio_list snapshot_bios; | |
182 | ||
1da177e4 LT |
183 | /* Pointer back to snapshot context */ |
184 | struct dm_snapshot *snap; | |
185 | ||
186 | /* | |
187 | * 1 indicates the exception has already been sent to | |
188 | * kcopyd. | |
189 | */ | |
190 | int started; | |
191 | }; | |
192 | ||
193 | /* | |
194 | * Hash table mapping origin volumes to lists of snapshots and | |
195 | * a lock to protect it | |
196 | */ | |
e18b890b CL |
197 | static struct kmem_cache *exception_cache; |
198 | static struct kmem_cache *pending_cache; | |
1da177e4 | 199 | |
cd45daff MP |
200 | struct dm_snap_tracked_chunk { |
201 | struct hlist_node node; | |
202 | chunk_t chunk; | |
203 | }; | |
204 | ||
205 | static struct kmem_cache *tracked_chunk_cache; | |
206 | ||
207 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
208 | chunk_t chunk) | |
209 | { | |
210 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
211 | GFP_NOIO); | |
212 | unsigned long flags; | |
213 | ||
214 | c->chunk = chunk; | |
215 | ||
216 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
217 | hlist_add_head(&c->node, | |
218 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
219 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
220 | ||
221 | return c; | |
222 | } | |
223 | ||
224 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
225 | struct dm_snap_tracked_chunk *c) | |
226 | { | |
227 | unsigned long flags; | |
228 | ||
229 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
230 | hlist_del(&c->node); | |
231 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
232 | ||
233 | mempool_free(c, s->tracked_chunk_pool); | |
234 | } | |
235 | ||
a8d41b59 MP |
236 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
237 | { | |
238 | struct dm_snap_tracked_chunk *c; | |
239 | struct hlist_node *hn; | |
240 | int found = 0; | |
241 | ||
242 | spin_lock_irq(&s->tracked_chunk_lock); | |
243 | ||
244 | hlist_for_each_entry(c, hn, | |
245 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
246 | if (c->chunk == chunk) { | |
247 | found = 1; | |
248 | break; | |
249 | } | |
250 | } | |
251 | ||
252 | spin_unlock_irq(&s->tracked_chunk_lock); | |
253 | ||
254 | return found; | |
255 | } | |
256 | ||
615d1eb9 MS |
257 | /* |
258 | * This conflicting I/O is extremely improbable in the caller, | |
259 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
260 | */ | |
261 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
262 | { | |
263 | while (__chunk_is_tracked(s, chunk)) | |
264 | msleep(1); | |
265 | } | |
266 | ||
1da177e4 LT |
267 | /* |
268 | * One of these per registered origin, held in the snapshot_origins hash | |
269 | */ | |
270 | struct origin { | |
271 | /* The origin device */ | |
272 | struct block_device *bdev; | |
273 | ||
274 | struct list_head hash_list; | |
275 | ||
276 | /* List of snapshots for this origin */ | |
277 | struct list_head snapshots; | |
278 | }; | |
279 | ||
280 | /* | |
281 | * Size of the hash table for origin volumes. If we make this | |
282 | * the size of the minors list then it should be nearly perfect | |
283 | */ | |
284 | #define ORIGIN_HASH_SIZE 256 | |
285 | #define ORIGIN_MASK 0xFF | |
286 | static struct list_head *_origins; | |
287 | static struct rw_semaphore _origins_lock; | |
288 | ||
73dfd078 MP |
289 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
290 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
291 | static uint64_t _pending_exceptions_done_count; | |
292 | ||
1da177e4 LT |
293 | static int init_origin_hash(void) |
294 | { | |
295 | int i; | |
296 | ||
297 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
298 | GFP_KERNEL); | |
299 | if (!_origins) { | |
72d94861 | 300 | DMERR("unable to allocate memory"); |
1da177e4 LT |
301 | return -ENOMEM; |
302 | } | |
303 | ||
304 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
305 | INIT_LIST_HEAD(_origins + i); | |
306 | init_rwsem(&_origins_lock); | |
307 | ||
308 | return 0; | |
309 | } | |
310 | ||
311 | static void exit_origin_hash(void) | |
312 | { | |
313 | kfree(_origins); | |
314 | } | |
315 | ||
028867ac | 316 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
317 | { |
318 | return bdev->bd_dev & ORIGIN_MASK; | |
319 | } | |
320 | ||
321 | static struct origin *__lookup_origin(struct block_device *origin) | |
322 | { | |
323 | struct list_head *ol; | |
324 | struct origin *o; | |
325 | ||
326 | ol = &_origins[origin_hash(origin)]; | |
327 | list_for_each_entry (o, ol, hash_list) | |
328 | if (bdev_equal(o->bdev, origin)) | |
329 | return o; | |
330 | ||
331 | return NULL; | |
332 | } | |
333 | ||
334 | static void __insert_origin(struct origin *o) | |
335 | { | |
336 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
337 | list_add_tail(&o->hash_list, sl); | |
338 | } | |
339 | ||
c1f0c183 MS |
340 | /* |
341 | * _origins_lock must be held when calling this function. | |
342 | * Returns number of snapshots registered using the supplied cow device, plus: | |
343 | * snap_src - a snapshot suitable for use as a source of exception handover | |
344 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
345 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
346 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 347 | * |
9d3b15c4 | 348 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
349 | * 0: NULL, NULL - first new snapshot |
350 | * 1: snap_src, NULL - normal snapshot | |
351 | * 2: snap_src, snap_dest - waiting for handover | |
352 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
353 | * 1: NULL, snap_dest - source got destroyed without handover | |
354 | */ | |
355 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
356 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
357 | struct dm_snapshot **snap_dest, |
358 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
359 | { |
360 | struct dm_snapshot *s; | |
361 | struct origin *o; | |
362 | int count = 0; | |
363 | int active; | |
364 | ||
365 | o = __lookup_origin(snap->origin->bdev); | |
366 | if (!o) | |
367 | goto out; | |
368 | ||
369 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
370 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
371 | *snap_merge = s; | |
c1f0c183 MS |
372 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
373 | continue; | |
374 | ||
375 | down_read(&s->lock); | |
376 | active = s->active; | |
377 | up_read(&s->lock); | |
378 | ||
379 | if (active) { | |
380 | if (snap_src) | |
381 | *snap_src = s; | |
382 | } else if (snap_dest) | |
383 | *snap_dest = s; | |
384 | ||
385 | count++; | |
386 | } | |
387 | ||
388 | out: | |
389 | return count; | |
390 | } | |
391 | ||
392 | /* | |
393 | * On success, returns 1 if this snapshot is a handover destination, | |
394 | * otherwise returns 0. | |
395 | */ | |
396 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
397 | { | |
398 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 399 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
400 | |
401 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
402 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
403 | &snap_merge) == 2) || | |
c1f0c183 MS |
404 | snap_dest) { |
405 | snap->ti->error = "Snapshot cow pairing for exception " | |
406 | "table handover failed"; | |
407 | return -EINVAL; | |
408 | } | |
409 | ||
410 | /* | |
411 | * If no snap_src was found, snap cannot become a handover | |
412 | * destination. | |
413 | */ | |
414 | if (!snap_src) | |
415 | return 0; | |
416 | ||
9d3b15c4 MP |
417 | /* |
418 | * Non-snapshot-merge handover? | |
419 | */ | |
420 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
421 | return 1; | |
422 | ||
423 | /* | |
424 | * Do not allow more than one merging snapshot. | |
425 | */ | |
426 | if (snap_merge) { | |
427 | snap->ti->error = "A snapshot is already merging."; | |
428 | return -EINVAL; | |
429 | } | |
430 | ||
1e03f97e MP |
431 | if (!snap_src->store->type->prepare_merge || |
432 | !snap_src->store->type->commit_merge) { | |
433 | snap->ti->error = "Snapshot exception store does not " | |
434 | "support snapshot-merge."; | |
435 | return -EINVAL; | |
436 | } | |
437 | ||
c1f0c183 MS |
438 | return 1; |
439 | } | |
440 | ||
441 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
442 | { | |
443 | struct dm_snapshot *l; | |
444 | ||
445 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
446 | list_for_each_entry(l, &o->snapshots, list) | |
447 | if (l->store->chunk_size < s->store->chunk_size) | |
448 | break; | |
449 | list_add_tail(&s->list, &l->list); | |
450 | } | |
451 | ||
1da177e4 LT |
452 | /* |
453 | * Make a note of the snapshot and its origin so we can look it | |
454 | * up when the origin has a write on it. | |
c1f0c183 MS |
455 | * |
456 | * Also validate snapshot exception store handovers. | |
457 | * On success, returns 1 if this registration is a handover destination, | |
458 | * otherwise returns 0. | |
1da177e4 LT |
459 | */ |
460 | static int register_snapshot(struct dm_snapshot *snap) | |
461 | { | |
c1f0c183 | 462 | struct origin *o, *new_o = NULL; |
1da177e4 | 463 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 464 | int r = 0; |
1da177e4 | 465 | |
60c856c8 MP |
466 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
467 | if (!new_o) | |
468 | return -ENOMEM; | |
469 | ||
1da177e4 | 470 | down_write(&_origins_lock); |
1da177e4 | 471 | |
c1f0c183 MS |
472 | r = __validate_exception_handover(snap); |
473 | if (r < 0) { | |
474 | kfree(new_o); | |
475 | goto out; | |
476 | } | |
477 | ||
478 | o = __lookup_origin(bdev); | |
60c856c8 MP |
479 | if (o) |
480 | kfree(new_o); | |
481 | else { | |
1da177e4 | 482 | /* New origin */ |
60c856c8 | 483 | o = new_o; |
1da177e4 LT |
484 | |
485 | /* Initialise the struct */ | |
486 | INIT_LIST_HEAD(&o->snapshots); | |
487 | o->bdev = bdev; | |
488 | ||
489 | __insert_origin(o); | |
490 | } | |
491 | ||
c1f0c183 MS |
492 | __insert_snapshot(o, snap); |
493 | ||
494 | out: | |
495 | up_write(&_origins_lock); | |
496 | ||
497 | return r; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Move snapshot to correct place in list according to chunk size. | |
502 | */ | |
503 | static void reregister_snapshot(struct dm_snapshot *s) | |
504 | { | |
505 | struct block_device *bdev = s->origin->bdev; | |
506 | ||
507 | down_write(&_origins_lock); | |
508 | ||
509 | list_del(&s->list); | |
510 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
511 | |
512 | up_write(&_origins_lock); | |
1da177e4 LT |
513 | } |
514 | ||
515 | static void unregister_snapshot(struct dm_snapshot *s) | |
516 | { | |
517 | struct origin *o; | |
518 | ||
519 | down_write(&_origins_lock); | |
520 | o = __lookup_origin(s->origin->bdev); | |
521 | ||
522 | list_del(&s->list); | |
c1f0c183 | 523 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
524 | list_del(&o->hash_list); |
525 | kfree(o); | |
526 | } | |
527 | ||
528 | up_write(&_origins_lock); | |
529 | } | |
530 | ||
531 | /* | |
532 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
533 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
534 | * some consecutive chunks to be grouped together. | |
1da177e4 | 535 | */ |
3510cb94 JB |
536 | static int dm_exception_table_init(struct dm_exception_table *et, |
537 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
538 | { |
539 | unsigned int i; | |
540 | ||
d74f81f8 | 541 | et->hash_shift = hash_shift; |
1da177e4 LT |
542 | et->hash_mask = size - 1; |
543 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
544 | if (!et->table) | |
545 | return -ENOMEM; | |
546 | ||
547 | for (i = 0; i < size; i++) | |
548 | INIT_LIST_HEAD(et->table + i); | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
3510cb94 JB |
553 | static void dm_exception_table_exit(struct dm_exception_table *et, |
554 | struct kmem_cache *mem) | |
1da177e4 LT |
555 | { |
556 | struct list_head *slot; | |
1d4989c8 | 557 | struct dm_exception *ex, *next; |
1da177e4 LT |
558 | int i, size; |
559 | ||
560 | size = et->hash_mask + 1; | |
561 | for (i = 0; i < size; i++) { | |
562 | slot = et->table + i; | |
563 | ||
564 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
565 | kmem_cache_free(mem, ex); | |
566 | } | |
567 | ||
568 | vfree(et->table); | |
569 | } | |
570 | ||
191437a5 | 571 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 572 | { |
d74f81f8 | 573 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
574 | } |
575 | ||
3510cb94 | 576 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
577 | { |
578 | list_del(&e->hash_list); | |
579 | } | |
580 | ||
581 | /* | |
582 | * Return the exception data for a sector, or NULL if not | |
583 | * remapped. | |
584 | */ | |
3510cb94 JB |
585 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
586 | chunk_t chunk) | |
1da177e4 LT |
587 | { |
588 | struct list_head *slot; | |
1d4989c8 | 589 | struct dm_exception *e; |
1da177e4 LT |
590 | |
591 | slot = &et->table[exception_hash(et, chunk)]; | |
592 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
593 | if (chunk >= e->old_chunk && |
594 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
595 | return e; |
596 | ||
597 | return NULL; | |
598 | } | |
599 | ||
3510cb94 | 600 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 601 | { |
1d4989c8 | 602 | struct dm_exception *e; |
1da177e4 LT |
603 | |
604 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
605 | if (!e) | |
606 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
607 | ||
608 | return e; | |
609 | } | |
610 | ||
3510cb94 | 611 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
612 | { |
613 | kmem_cache_free(exception_cache, e); | |
614 | } | |
615 | ||
92e86812 | 616 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 617 | { |
92e86812 MP |
618 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
619 | GFP_NOIO); | |
620 | ||
879129d2 | 621 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
622 | pe->snap = s; |
623 | ||
624 | return pe; | |
1da177e4 LT |
625 | } |
626 | ||
028867ac | 627 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 628 | { |
879129d2 MP |
629 | struct dm_snapshot *s = pe->snap; |
630 | ||
631 | mempool_free(pe, s->pending_pool); | |
632 | smp_mb__before_atomic_dec(); | |
633 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
634 | } |
635 | ||
3510cb94 JB |
636 | static void dm_insert_exception(struct dm_exception_table *eh, |
637 | struct dm_exception *new_e) | |
d74f81f8 | 638 | { |
d74f81f8 | 639 | struct list_head *l; |
1d4989c8 | 640 | struct dm_exception *e = NULL; |
d74f81f8 MB |
641 | |
642 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
643 | ||
644 | /* Add immediately if this table doesn't support consecutive chunks */ | |
645 | if (!eh->hash_shift) | |
646 | goto out; | |
647 | ||
648 | /* List is ordered by old_chunk */ | |
649 | list_for_each_entry_reverse(e, l, hash_list) { | |
650 | /* Insert after an existing chunk? */ | |
651 | if (new_e->old_chunk == (e->old_chunk + | |
652 | dm_consecutive_chunk_count(e) + 1) && | |
653 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
654 | dm_consecutive_chunk_count(e) + 1)) { | |
655 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 656 | free_completed_exception(new_e); |
d74f81f8 MB |
657 | return; |
658 | } | |
659 | ||
660 | /* Insert before an existing chunk? */ | |
661 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
662 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
663 | dm_consecutive_chunk_count_inc(e); | |
664 | e->old_chunk--; | |
665 | e->new_chunk--; | |
3510cb94 | 666 | free_completed_exception(new_e); |
d74f81f8 MB |
667 | return; |
668 | } | |
669 | ||
670 | if (new_e->old_chunk > e->old_chunk) | |
671 | break; | |
672 | } | |
673 | ||
674 | out: | |
675 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
676 | } | |
677 | ||
a159c1ac JB |
678 | /* |
679 | * Callback used by the exception stores to load exceptions when | |
680 | * initialising. | |
681 | */ | |
682 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 683 | { |
a159c1ac | 684 | struct dm_snapshot *s = context; |
1d4989c8 | 685 | struct dm_exception *e; |
1da177e4 | 686 | |
3510cb94 | 687 | e = alloc_completed_exception(); |
1da177e4 LT |
688 | if (!e) |
689 | return -ENOMEM; | |
690 | ||
691 | e->old_chunk = old; | |
d74f81f8 MB |
692 | |
693 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 694 | e->new_chunk = new; |
d74f81f8 | 695 | |
3510cb94 | 696 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 697 | |
1da177e4 LT |
698 | return 0; |
699 | } | |
700 | ||
7e201b35 MP |
701 | /* |
702 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
703 | * Return zero if the origin has no snapshots. | |
704 | */ | |
705 | static sector_t __minimum_chunk_size(struct origin *o) | |
706 | { | |
707 | struct dm_snapshot *snap; | |
708 | unsigned chunk_size = 0; | |
709 | ||
710 | if (o) | |
711 | list_for_each_entry(snap, &o->snapshots, list) | |
712 | chunk_size = min_not_zero(chunk_size, | |
713 | snap->store->chunk_size); | |
714 | ||
715 | return chunk_size; | |
716 | } | |
717 | ||
1da177e4 LT |
718 | /* |
719 | * Hard coded magic. | |
720 | */ | |
721 | static int calc_max_buckets(void) | |
722 | { | |
723 | /* use a fixed size of 2MB */ | |
724 | unsigned long mem = 2 * 1024 * 1024; | |
725 | mem /= sizeof(struct list_head); | |
726 | ||
727 | return mem; | |
728 | } | |
729 | ||
1da177e4 LT |
730 | /* |
731 | * Allocate room for a suitable hash table. | |
732 | */ | |
fee1998e | 733 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
734 | { |
735 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
736 | ||
737 | /* | |
738 | * Calculate based on the size of the original volume or | |
739 | * the COW volume... | |
740 | */ | |
fc56f6fb | 741 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
742 | origin_dev_size = get_dev_size(s->origin->bdev); |
743 | max_buckets = calc_max_buckets(); | |
744 | ||
fee1998e | 745 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
746 | hash_size = min(hash_size, max_buckets); |
747 | ||
8e87b9b8 MP |
748 | if (hash_size < 64) |
749 | hash_size = 64; | |
8defd830 | 750 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
751 | if (dm_exception_table_init(&s->complete, hash_size, |
752 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
753 | return -ENOMEM; |
754 | ||
755 | /* | |
756 | * Allocate hash table for in-flight exceptions | |
757 | * Make this smaller than the real hash table | |
758 | */ | |
759 | hash_size >>= 3; | |
760 | if (hash_size < 64) | |
761 | hash_size = 64; | |
762 | ||
3510cb94 JB |
763 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
764 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
765 | return -ENOMEM; |
766 | } | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
1e03f97e MP |
771 | static void merge_shutdown(struct dm_snapshot *s) |
772 | { | |
773 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
774 | smp_mb__after_clear_bit(); | |
775 | wake_up_bit(&s->state_bits, RUNNING_MERGE); | |
776 | } | |
777 | ||
9fe86254 MP |
778 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
779 | { | |
780 | s->first_merging_chunk = 0; | |
781 | s->num_merging_chunks = 0; | |
782 | ||
783 | return bio_list_get(&s->bios_queued_during_merge); | |
784 | } | |
785 | ||
1e03f97e MP |
786 | /* |
787 | * Remove one chunk from the index of completed exceptions. | |
788 | */ | |
789 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
790 | chunk_t old_chunk) | |
791 | { | |
792 | struct dm_exception *e; | |
793 | ||
1e03f97e MP |
794 | e = dm_lookup_exception(&s->complete, old_chunk); |
795 | if (!e) { | |
796 | DMERR("Corruption detected: exception for block %llu is " | |
797 | "on disk but not in memory", | |
798 | (unsigned long long)old_chunk); | |
799 | return -EINVAL; | |
800 | } | |
801 | ||
802 | /* | |
803 | * If this is the only chunk using this exception, remove exception. | |
804 | */ | |
805 | if (!dm_consecutive_chunk_count(e)) { | |
806 | dm_remove_exception(e); | |
807 | free_completed_exception(e); | |
808 | return 0; | |
809 | } | |
810 | ||
811 | /* | |
812 | * The chunk may be either at the beginning or the end of a | |
813 | * group of consecutive chunks - never in the middle. We are | |
814 | * removing chunks in the opposite order to that in which they | |
815 | * were added, so this should always be true. | |
816 | * Decrement the consecutive chunk counter and adjust the | |
817 | * starting point if necessary. | |
818 | */ | |
819 | if (old_chunk == e->old_chunk) { | |
820 | e->old_chunk++; | |
821 | e->new_chunk++; | |
822 | } else if (old_chunk != e->old_chunk + | |
823 | dm_consecutive_chunk_count(e)) { | |
824 | DMERR("Attempt to merge block %llu from the " | |
825 | "middle of a chunk range [%llu - %llu]", | |
826 | (unsigned long long)old_chunk, | |
827 | (unsigned long long)e->old_chunk, | |
828 | (unsigned long long) | |
829 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
830 | return -EINVAL; | |
831 | } | |
832 | ||
833 | dm_consecutive_chunk_count_dec(e); | |
834 | ||
835 | return 0; | |
836 | } | |
837 | ||
9fe86254 MP |
838 | static void flush_bios(struct bio *bio); |
839 | ||
840 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 841 | { |
9fe86254 MP |
842 | struct bio *b = NULL; |
843 | int r; | |
844 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
845 | |
846 | down_write(&s->lock); | |
9fe86254 MP |
847 | |
848 | /* | |
849 | * Process chunks (and associated exceptions) in reverse order | |
850 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
851 | */ | |
852 | do { | |
853 | r = __remove_single_exception_chunk(s, old_chunk); | |
854 | if (r) | |
855 | goto out; | |
856 | } while (old_chunk-- > s->first_merging_chunk); | |
857 | ||
858 | b = __release_queued_bios_after_merge(s); | |
859 | ||
860 | out: | |
1e03f97e | 861 | up_write(&s->lock); |
9fe86254 MP |
862 | if (b) |
863 | flush_bios(b); | |
1e03f97e MP |
864 | |
865 | return r; | |
866 | } | |
867 | ||
73dfd078 MP |
868 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
869 | sector_t sector, unsigned chunk_size); | |
870 | ||
1e03f97e MP |
871 | static void merge_callback(int read_err, unsigned long write_err, |
872 | void *context); | |
873 | ||
73dfd078 MP |
874 | static uint64_t read_pending_exceptions_done_count(void) |
875 | { | |
876 | uint64_t pending_exceptions_done; | |
877 | ||
878 | spin_lock(&_pending_exceptions_done_spinlock); | |
879 | pending_exceptions_done = _pending_exceptions_done_count; | |
880 | spin_unlock(&_pending_exceptions_done_spinlock); | |
881 | ||
882 | return pending_exceptions_done; | |
883 | } | |
884 | ||
885 | static void increment_pending_exceptions_done_count(void) | |
886 | { | |
887 | spin_lock(&_pending_exceptions_done_spinlock); | |
888 | _pending_exceptions_done_count++; | |
889 | spin_unlock(&_pending_exceptions_done_spinlock); | |
890 | ||
891 | wake_up_all(&_pending_exceptions_done); | |
892 | } | |
893 | ||
1e03f97e MP |
894 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
895 | { | |
8a2d5286 | 896 | int i, linear_chunks; |
1e03f97e MP |
897 | chunk_t old_chunk, new_chunk; |
898 | struct dm_io_region src, dest; | |
8a2d5286 | 899 | sector_t io_size; |
73dfd078 | 900 | uint64_t previous_count; |
1e03f97e MP |
901 | |
902 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
903 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
904 | goto shut; | |
905 | ||
906 | /* | |
907 | * valid flag never changes during merge, so no lock required. | |
908 | */ | |
909 | if (!s->valid) { | |
910 | DMERR("Snapshot is invalid: can't merge"); | |
911 | goto shut; | |
912 | } | |
913 | ||
8a2d5286 MS |
914 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
915 | &new_chunk); | |
916 | if (linear_chunks <= 0) { | |
d8ddb1cf | 917 | if (linear_chunks < 0) { |
1e03f97e MP |
918 | DMERR("Read error in exception store: " |
919 | "shutting down merge"); | |
d8ddb1cf MS |
920 | down_write(&s->lock); |
921 | s->merge_failed = 1; | |
922 | up_write(&s->lock); | |
923 | } | |
1e03f97e MP |
924 | goto shut; |
925 | } | |
926 | ||
8a2d5286 MS |
927 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
928 | old_chunk = old_chunk + 1 - linear_chunks; | |
929 | new_chunk = new_chunk + 1 - linear_chunks; | |
930 | ||
931 | /* | |
932 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
933 | * from the exception store to the origin | |
934 | */ | |
935 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 936 | |
1e03f97e MP |
937 | dest.bdev = s->origin->bdev; |
938 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 939 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
940 | |
941 | src.bdev = s->cow->bdev; | |
942 | src.sector = chunk_to_sector(s->store, new_chunk); | |
943 | src.count = dest.count; | |
944 | ||
73dfd078 MP |
945 | /* |
946 | * Reallocate any exceptions needed in other snapshots then | |
947 | * wait for the pending exceptions to complete. | |
948 | * Each time any pending exception (globally on the system) | |
949 | * completes we are woken and repeat the process to find out | |
950 | * if we can proceed. While this may not seem a particularly | |
951 | * efficient algorithm, it is not expected to have any | |
952 | * significant impact on performance. | |
953 | */ | |
954 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 955 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
956 | wait_event(_pending_exceptions_done, |
957 | (read_pending_exceptions_done_count() != | |
958 | previous_count)); | |
959 | /* Retry after the wait, until all exceptions are done. */ | |
960 | previous_count = read_pending_exceptions_done_count(); | |
961 | } | |
962 | ||
9fe86254 MP |
963 | down_write(&s->lock); |
964 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 965 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
966 | up_write(&s->lock); |
967 | ||
8a2d5286 MS |
968 | /* Wait until writes to all 'linear_chunks' drain */ |
969 | for (i = 0; i < linear_chunks; i++) | |
970 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 971 | |
1e03f97e MP |
972 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
973 | return; | |
974 | ||
975 | shut: | |
976 | merge_shutdown(s); | |
977 | } | |
978 | ||
9fe86254 MP |
979 | static void error_bios(struct bio *bio); |
980 | ||
1e03f97e MP |
981 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
982 | { | |
983 | struct dm_snapshot *s = context; | |
9fe86254 | 984 | struct bio *b = NULL; |
1e03f97e MP |
985 | |
986 | if (read_err || write_err) { | |
987 | if (read_err) | |
988 | DMERR("Read error: shutting down merge."); | |
989 | else | |
990 | DMERR("Write error: shutting down merge."); | |
991 | goto shut; | |
992 | } | |
993 | ||
9fe86254 MP |
994 | if (s->store->type->commit_merge(s->store, |
995 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
996 | DMERR("Write error in exception store: shutting down merge"); |
997 | goto shut; | |
998 | } | |
999 | ||
9fe86254 MP |
1000 | if (remove_single_exception_chunk(s) < 0) |
1001 | goto shut; | |
1002 | ||
1e03f97e MP |
1003 | snapshot_merge_next_chunks(s); |
1004 | ||
1005 | return; | |
1006 | ||
1007 | shut: | |
9fe86254 | 1008 | down_write(&s->lock); |
d8ddb1cf | 1009 | s->merge_failed = 1; |
9fe86254 MP |
1010 | b = __release_queued_bios_after_merge(s); |
1011 | up_write(&s->lock); | |
1012 | error_bios(b); | |
1013 | ||
1e03f97e MP |
1014 | merge_shutdown(s); |
1015 | } | |
1016 | ||
1017 | static void start_merge(struct dm_snapshot *s) | |
1018 | { | |
1019 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1020 | snapshot_merge_next_chunks(s); | |
1021 | } | |
1022 | ||
1023 | static int wait_schedule(void *ptr) | |
1024 | { | |
1025 | schedule(); | |
1026 | ||
1027 | return 0; | |
1028 | } | |
1029 | ||
1030 | /* | |
1031 | * Stop the merging process and wait until it finishes. | |
1032 | */ | |
1033 | static void stop_merge(struct dm_snapshot *s) | |
1034 | { | |
1035 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1036 | wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule, | |
1037 | TASK_UNINTERRUPTIBLE); | |
1038 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1039 | } | |
1040 | ||
1da177e4 LT |
1041 | /* |
1042 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
1043 | */ | |
1044 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1045 | { | |
1046 | struct dm_snapshot *s; | |
cd45daff | 1047 | int i; |
1da177e4 | 1048 | int r = -EINVAL; |
fc56f6fb | 1049 | char *origin_path, *cow_path; |
10b8106a MS |
1050 | unsigned args_used, num_flush_requests = 1; |
1051 | fmode_t origin_mode = FMODE_READ; | |
1da177e4 | 1052 | |
4c7e3bf4 | 1053 | if (argc != 4) { |
72d94861 | 1054 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1055 | r = -EINVAL; |
fc56f6fb | 1056 | goto bad; |
1da177e4 LT |
1057 | } |
1058 | ||
10b8106a MS |
1059 | if (dm_target_is_snapshot_merge(ti)) { |
1060 | num_flush_requests = 2; | |
1061 | origin_mode = FMODE_WRITE; | |
1062 | } | |
1063 | ||
fc56f6fb MS |
1064 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1065 | if (!s) { | |
1066 | ti->error = "Cannot allocate snapshot context private " | |
1067 | "structure"; | |
1068 | r = -ENOMEM; | |
1069 | goto bad; | |
1070 | } | |
1071 | ||
c2411045 MP |
1072 | origin_path = argv[0]; |
1073 | argv++; | |
1074 | argc--; | |
1075 | ||
1076 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1077 | if (r) { | |
1078 | ti->error = "Cannot get origin device"; | |
1079 | goto bad_origin; | |
1080 | } | |
1081 | ||
fc56f6fb MS |
1082 | cow_path = argv[0]; |
1083 | argv++; | |
1084 | argc--; | |
1085 | ||
8215d6ec | 1086 | r = dm_get_device(ti, cow_path, FMODE_READ | FMODE_WRITE, &s->cow); |
fc56f6fb MS |
1087 | if (r) { |
1088 | ti->error = "Cannot get COW device"; | |
1089 | goto bad_cow; | |
1090 | } | |
1091 | ||
1092 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1093 | if (r) { |
1094 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1095 | r = -EINVAL; |
fc56f6fb | 1096 | goto bad_store; |
1da177e4 LT |
1097 | } |
1098 | ||
fee1998e JB |
1099 | argv += args_used; |
1100 | argc -= args_used; | |
1101 | ||
fc56f6fb | 1102 | s->ti = ti; |
1da177e4 | 1103 | s->valid = 1; |
aa14edeb | 1104 | s->active = 0; |
c26655ca | 1105 | s->suspended = 0; |
879129d2 | 1106 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 1107 | init_rwsem(&s->lock); |
c1f0c183 | 1108 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1109 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1110 | s->state_bits = 0; |
d8ddb1cf | 1111 | s->merge_failed = 0; |
9fe86254 MP |
1112 | s->first_merging_chunk = 0; |
1113 | s->num_merging_chunks = 0; | |
1114 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1115 | |
1116 | /* Allocate hash table for COW data */ | |
fee1998e | 1117 | if (init_hash_tables(s)) { |
1da177e4 LT |
1118 | ti->error = "Unable to allocate hash table space"; |
1119 | r = -ENOMEM; | |
fee1998e | 1120 | goto bad_hash_tables; |
1da177e4 LT |
1121 | } |
1122 | ||
eb69aca5 | 1123 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
1124 | if (r) { |
1125 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 1126 | goto bad_kcopyd; |
1da177e4 LT |
1127 | } |
1128 | ||
92e86812 MP |
1129 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1130 | if (!s->pending_pool) { | |
1131 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 1132 | goto bad_pending_pool; |
92e86812 MP |
1133 | } |
1134 | ||
cd45daff MP |
1135 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
1136 | tracked_chunk_cache); | |
1137 | if (!s->tracked_chunk_pool) { | |
1138 | ti->error = "Could not allocate tracked_chunk mempool for " | |
1139 | "tracking reads"; | |
92e86812 | 1140 | goto bad_tracked_chunk_pool; |
cd45daff MP |
1141 | } |
1142 | ||
1143 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1144 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1145 | ||
1146 | spin_lock_init(&s->tracked_chunk_lock); | |
1147 | ||
c1f0c183 | 1148 | ti->private = s; |
10b8106a | 1149 | ti->num_flush_requests = num_flush_requests; |
c1f0c183 MS |
1150 | |
1151 | /* Add snapshot to the list of snapshots for this origin */ | |
1152 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1153 | r = register_snapshot(s); | |
1154 | if (r == -ENOMEM) { | |
1155 | ti->error = "Snapshot origin struct allocation failed"; | |
1156 | goto bad_load_and_register; | |
1157 | } else if (r < 0) { | |
1158 | /* invalid handover, register_snapshot has set ti->error */ | |
1159 | goto bad_load_and_register; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Metadata must only be loaded into one table at once, so skip this | |
1164 | * if metadata will be handed over during resume. | |
1165 | * Chunk size will be set during the handover - set it to zero to | |
1166 | * ensure it's ignored. | |
1167 | */ | |
1168 | if (r > 0) { | |
1169 | s->store->chunk_size = 0; | |
1170 | return 0; | |
1171 | } | |
1172 | ||
493df71c JB |
1173 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1174 | (void *)s); | |
0764147b | 1175 | if (r < 0) { |
f9cea4f7 | 1176 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1177 | goto bad_read_metadata; |
0764147b MB |
1178 | } else if (r > 0) { |
1179 | s->valid = 0; | |
1180 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1181 | } |
aa14edeb | 1182 | |
3f2412dc MP |
1183 | if (!s->store->chunk_size) { |
1184 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1185 | goto bad_read_metadata; |
1da177e4 | 1186 | } |
d0216849 | 1187 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
1188 | |
1189 | return 0; | |
1190 | ||
c1f0c183 MS |
1191 | bad_read_metadata: |
1192 | unregister_snapshot(s); | |
1193 | ||
fee1998e | 1194 | bad_load_and_register: |
cd45daff MP |
1195 | mempool_destroy(s->tracked_chunk_pool); |
1196 | ||
fee1998e | 1197 | bad_tracked_chunk_pool: |
92e86812 MP |
1198 | mempool_destroy(s->pending_pool); |
1199 | ||
fee1998e | 1200 | bad_pending_pool: |
eb69aca5 | 1201 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1202 | |
fee1998e | 1203 | bad_kcopyd: |
3510cb94 JB |
1204 | dm_exception_table_exit(&s->pending, pending_cache); |
1205 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1206 | |
fee1998e | 1207 | bad_hash_tables: |
fc56f6fb | 1208 | dm_exception_store_destroy(s->store); |
1da177e4 | 1209 | |
fc56f6fb MS |
1210 | bad_store: |
1211 | dm_put_device(ti, s->cow); | |
fee1998e | 1212 | |
fc56f6fb | 1213 | bad_cow: |
c2411045 MP |
1214 | dm_put_device(ti, s->origin); |
1215 | ||
1216 | bad_origin: | |
fc56f6fb MS |
1217 | kfree(s); |
1218 | ||
1219 | bad: | |
1da177e4 LT |
1220 | return r; |
1221 | } | |
1222 | ||
31c93a0c MB |
1223 | static void __free_exceptions(struct dm_snapshot *s) |
1224 | { | |
eb69aca5 | 1225 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1226 | s->kcopyd_client = NULL; |
1227 | ||
3510cb94 JB |
1228 | dm_exception_table_exit(&s->pending, pending_cache); |
1229 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1230 | } |
1231 | ||
c1f0c183 MS |
1232 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1233 | struct dm_snapshot *snap_dest) | |
1234 | { | |
1235 | union { | |
1236 | struct dm_exception_table table_swap; | |
1237 | struct dm_exception_store *store_swap; | |
1238 | } u; | |
1239 | ||
1240 | /* | |
1241 | * Swap all snapshot context information between the two instances. | |
1242 | */ | |
1243 | u.table_swap = snap_dest->complete; | |
1244 | snap_dest->complete = snap_src->complete; | |
1245 | snap_src->complete = u.table_swap; | |
1246 | ||
1247 | u.store_swap = snap_dest->store; | |
1248 | snap_dest->store = snap_src->store; | |
1249 | snap_src->store = u.store_swap; | |
1250 | ||
1251 | snap_dest->store->snap = snap_dest; | |
1252 | snap_src->store->snap = snap_src; | |
1253 | ||
1254 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
1255 | snap_dest->valid = snap_src->valid; | |
1256 | ||
1257 | /* | |
1258 | * Set source invalid to ensure it receives no further I/O. | |
1259 | */ | |
1260 | snap_src->valid = 0; | |
1261 | } | |
1262 | ||
1da177e4 LT |
1263 | static void snapshot_dtr(struct dm_target *ti) |
1264 | { | |
cd45daff MP |
1265 | #ifdef CONFIG_DM_DEBUG |
1266 | int i; | |
1267 | #endif | |
028867ac | 1268 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1269 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1270 | |
c1f0c183 MS |
1271 | down_read(&_origins_lock); |
1272 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1273 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1274 | if (snap_src && snap_dest && (s == snap_src)) { |
1275 | down_write(&snap_dest->lock); | |
1276 | snap_dest->valid = 0; | |
1277 | up_write(&snap_dest->lock); | |
1278 | DMERR("Cancelling snapshot handover."); | |
1279 | } | |
1280 | up_read(&_origins_lock); | |
1281 | ||
1e03f97e MP |
1282 | if (dm_target_is_snapshot_merge(ti)) |
1283 | stop_merge(s); | |
1284 | ||
138728dc AK |
1285 | /* Prevent further origin writes from using this snapshot. */ |
1286 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1287 | unregister_snapshot(s); |
1288 | ||
879129d2 | 1289 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1290 | msleep(1); |
879129d2 MP |
1291 | /* |
1292 | * Ensure instructions in mempool_destroy aren't reordered | |
1293 | * before atomic_read. | |
1294 | */ | |
1295 | smp_mb(); | |
1296 | ||
cd45daff MP |
1297 | #ifdef CONFIG_DM_DEBUG |
1298 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1299 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1300 | #endif | |
1301 | ||
1302 | mempool_destroy(s->tracked_chunk_pool); | |
1303 | ||
31c93a0c | 1304 | __free_exceptions(s); |
1da177e4 | 1305 | |
92e86812 MP |
1306 | mempool_destroy(s->pending_pool); |
1307 | ||
fee1998e | 1308 | dm_exception_store_destroy(s->store); |
138728dc | 1309 | |
fc56f6fb MS |
1310 | dm_put_device(ti, s->cow); |
1311 | ||
c2411045 MP |
1312 | dm_put_device(ti, s->origin); |
1313 | ||
1da177e4 LT |
1314 | kfree(s); |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Flush a list of buffers. | |
1319 | */ | |
1320 | static void flush_bios(struct bio *bio) | |
1321 | { | |
1322 | struct bio *n; | |
1323 | ||
1324 | while (bio) { | |
1325 | n = bio->bi_next; | |
1326 | bio->bi_next = NULL; | |
1327 | generic_make_request(bio); | |
1328 | bio = n; | |
1329 | } | |
1330 | } | |
1331 | ||
515ad66c MP |
1332 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1333 | ||
1334 | /* | |
1335 | * Flush a list of buffers. | |
1336 | */ | |
1337 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1338 | { | |
1339 | struct bio *n; | |
1340 | int r; | |
1341 | ||
1342 | while (bio) { | |
1343 | n = bio->bi_next; | |
1344 | bio->bi_next = NULL; | |
1345 | r = do_origin(s->origin, bio); | |
1346 | if (r == DM_MAPIO_REMAPPED) | |
1347 | generic_make_request(bio); | |
1348 | bio = n; | |
1349 | } | |
1350 | } | |
1351 | ||
1da177e4 LT |
1352 | /* |
1353 | * Error a list of buffers. | |
1354 | */ | |
1355 | static void error_bios(struct bio *bio) | |
1356 | { | |
1357 | struct bio *n; | |
1358 | ||
1359 | while (bio) { | |
1360 | n = bio->bi_next; | |
1361 | bio->bi_next = NULL; | |
6712ecf8 | 1362 | bio_io_error(bio); |
1da177e4 LT |
1363 | bio = n; |
1364 | } | |
1365 | } | |
1366 | ||
695368ac | 1367 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1368 | { |
1369 | if (!s->valid) | |
1370 | return; | |
1371 | ||
1372 | if (err == -EIO) | |
1373 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1374 | else if (err == -ENOMEM) | |
1375 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1376 | ||
493df71c JB |
1377 | if (s->store->type->drop_snapshot) |
1378 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1379 | |
1380 | s->valid = 0; | |
1381 | ||
fc56f6fb | 1382 | dm_table_event(s->ti->table); |
76df1c65 AK |
1383 | } |
1384 | ||
028867ac | 1385 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1386 | { |
1d4989c8 | 1387 | struct dm_exception *e; |
1da177e4 | 1388 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1389 | struct bio *origin_bios = NULL; |
1390 | struct bio *snapshot_bios = NULL; | |
1391 | int error = 0; | |
1da177e4 | 1392 | |
76df1c65 AK |
1393 | if (!success) { |
1394 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1395 | down_write(&s->lock); |
695368ac | 1396 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1397 | error = 1; |
76df1c65 AK |
1398 | goto out; |
1399 | } | |
1400 | ||
3510cb94 | 1401 | e = alloc_completed_exception(); |
76df1c65 | 1402 | if (!e) { |
1da177e4 | 1403 | down_write(&s->lock); |
695368ac | 1404 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1405 | error = 1; |
76df1c65 AK |
1406 | goto out; |
1407 | } | |
1408 | *e = pe->e; | |
1da177e4 | 1409 | |
76df1c65 AK |
1410 | down_write(&s->lock); |
1411 | if (!s->valid) { | |
3510cb94 | 1412 | free_completed_exception(e); |
9d493fa8 | 1413 | error = 1; |
76df1c65 | 1414 | goto out; |
1da177e4 LT |
1415 | } |
1416 | ||
615d1eb9 MS |
1417 | /* Check for conflicting reads */ |
1418 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1419 | |
9d493fa8 AK |
1420 | /* |
1421 | * Add a proper exception, and remove the | |
1422 | * in-flight exception from the list. | |
1423 | */ | |
3510cb94 | 1424 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1425 | |
1da177e4 | 1426 | out: |
3510cb94 | 1427 | dm_remove_exception(&pe->e); |
9d493fa8 | 1428 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c MP |
1429 | origin_bios = bio_list_get(&pe->origin_bios); |
1430 | free_pending_exception(pe); | |
1da177e4 | 1431 | |
73dfd078 MP |
1432 | increment_pending_exceptions_done_count(); |
1433 | ||
9d493fa8 AK |
1434 | up_write(&s->lock); |
1435 | ||
1436 | /* Submit any pending write bios */ | |
1437 | if (error) | |
1438 | error_bios(snapshot_bios); | |
1439 | else | |
1440 | flush_bios(snapshot_bios); | |
1441 | ||
515ad66c | 1442 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | static void commit_callback(void *context, int success) | |
1446 | { | |
028867ac AK |
1447 | struct dm_snap_pending_exception *pe = context; |
1448 | ||
1da177e4 LT |
1449 | pending_complete(pe, success); |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * Called when the copy I/O has finished. kcopyd actually runs | |
1454 | * this code so don't block. | |
1455 | */ | |
4cdc1d1f | 1456 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1457 | { |
028867ac | 1458 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1459 | struct dm_snapshot *s = pe->snap; |
1460 | ||
1461 | if (read_err || write_err) | |
1462 | pending_complete(pe, 0); | |
1463 | ||
1464 | else | |
1465 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1466 | s->store->type->commit_exception(s->store, &pe->e, |
1467 | commit_callback, pe); | |
1da177e4 LT |
1468 | } |
1469 | ||
1470 | /* | |
1471 | * Dispatches the copy operation to kcopyd. | |
1472 | */ | |
028867ac | 1473 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1474 | { |
1475 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1476 | struct dm_io_region src, dest; |
1da177e4 LT |
1477 | struct block_device *bdev = s->origin->bdev; |
1478 | sector_t dev_size; | |
1479 | ||
1480 | dev_size = get_dev_size(bdev); | |
1481 | ||
1482 | src.bdev = bdev; | |
71fab00a | 1483 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1484 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1485 | |
fc56f6fb | 1486 | dest.bdev = s->cow->bdev; |
71fab00a | 1487 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1488 | dest.count = src.count; |
1489 | ||
1490 | /* Hand over to kcopyd */ | |
eb69aca5 | 1491 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1492 | &src, 1, &dest, 0, copy_callback, pe); |
1493 | } | |
1494 | ||
2913808e MP |
1495 | static struct dm_snap_pending_exception * |
1496 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1497 | { | |
3510cb94 | 1498 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1499 | |
1500 | if (!e) | |
1501 | return NULL; | |
1502 | ||
1503 | return container_of(e, struct dm_snap_pending_exception, e); | |
1504 | } | |
1505 | ||
1da177e4 LT |
1506 | /* |
1507 | * Looks to see if this snapshot already has a pending exception | |
1508 | * for this chunk, otherwise it allocates a new one and inserts | |
1509 | * it into the pending table. | |
1510 | * | |
1511 | * NOTE: a write lock must be held on snap->lock before calling | |
1512 | * this. | |
1513 | */ | |
028867ac | 1514 | static struct dm_snap_pending_exception * |
c6621392 MP |
1515 | __find_pending_exception(struct dm_snapshot *s, |
1516 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1517 | { |
c6621392 | 1518 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1519 | |
2913808e MP |
1520 | pe2 = __lookup_pending_exception(s, chunk); |
1521 | if (pe2) { | |
76df1c65 | 1522 | free_pending_exception(pe); |
2913808e | 1523 | return pe2; |
1da177e4 LT |
1524 | } |
1525 | ||
76df1c65 AK |
1526 | pe->e.old_chunk = chunk; |
1527 | bio_list_init(&pe->origin_bios); | |
1528 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 AK |
1529 | pe->started = 0; |
1530 | ||
493df71c | 1531 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1532 | free_pending_exception(pe); |
1533 | return NULL; | |
1534 | } | |
1535 | ||
3510cb94 | 1536 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1537 | |
1da177e4 LT |
1538 | return pe; |
1539 | } | |
1540 | ||
1d4989c8 | 1541 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1542 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1543 | { |
fc56f6fb | 1544 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1545 | bio->bi_sector = chunk_to_sector(s->store, |
1546 | dm_chunk_number(e->new_chunk) + | |
1547 | (chunk - e->old_chunk)) + | |
1548 | (bio->bi_sector & | |
1549 | s->store->chunk_mask); | |
1da177e4 LT |
1550 | } |
1551 | ||
1552 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1553 | union map_info *map_context) | |
1554 | { | |
1d4989c8 | 1555 | struct dm_exception *e; |
028867ac | 1556 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1557 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1558 | chunk_t chunk; |
028867ac | 1559 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1560 | |
d87f4c14 | 1561 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1562 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1563 | return DM_MAPIO_REMAPPED; |
1564 | } | |
1565 | ||
71fab00a | 1566 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1567 | |
1568 | /* Full snapshots are not usable */ | |
76df1c65 | 1569 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1570 | if (!s->valid) |
f6a80ea8 | 1571 | return -EIO; |
1da177e4 | 1572 | |
ba40a2aa AK |
1573 | /* FIXME: should only take write lock if we need |
1574 | * to copy an exception */ | |
1575 | down_write(&s->lock); | |
1576 | ||
1577 | if (!s->valid) { | |
1578 | r = -EIO; | |
1579 | goto out_unlock; | |
1580 | } | |
1581 | ||
1582 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1583 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1584 | if (e) { |
d74f81f8 | 1585 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1586 | goto out_unlock; |
1587 | } | |
1588 | ||
1da177e4 LT |
1589 | /* |
1590 | * Write to snapshot - higher level takes care of RW/RO | |
1591 | * flags so we should only get this if we are | |
1592 | * writeable. | |
1593 | */ | |
1594 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1595 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1596 | if (!pe) { |
c6621392 MP |
1597 | up_write(&s->lock); |
1598 | pe = alloc_pending_exception(s); | |
1599 | down_write(&s->lock); | |
1600 | ||
1601 | if (!s->valid) { | |
1602 | free_pending_exception(pe); | |
1603 | r = -EIO; | |
1604 | goto out_unlock; | |
1605 | } | |
1606 | ||
3510cb94 | 1607 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1608 | if (e) { |
1609 | free_pending_exception(pe); | |
1610 | remap_exception(s, e, bio, chunk); | |
1611 | goto out_unlock; | |
1612 | } | |
1613 | ||
c6621392 | 1614 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1615 | if (!pe) { |
1616 | __invalidate_snapshot(s, -ENOMEM); | |
1617 | r = -EIO; | |
1618 | goto out_unlock; | |
1619 | } | |
1da177e4 LT |
1620 | } |
1621 | ||
d74f81f8 | 1622 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1623 | bio_list_add(&pe->snapshot_bios, bio); |
1624 | ||
d2a7ad29 | 1625 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1626 | |
76df1c65 AK |
1627 | if (!pe->started) { |
1628 | /* this is protected by snap->lock */ | |
1629 | pe->started = 1; | |
ba40a2aa | 1630 | up_write(&s->lock); |
76df1c65 | 1631 | start_copy(pe); |
ba40a2aa AK |
1632 | goto out; |
1633 | } | |
cd45daff | 1634 | } else { |
ba40a2aa | 1635 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1636 | map_context->ptr = track_chunk(s, chunk); |
1637 | } | |
1da177e4 | 1638 | |
ba40a2aa AK |
1639 | out_unlock: |
1640 | up_write(&s->lock); | |
1641 | out: | |
1da177e4 LT |
1642 | return r; |
1643 | } | |
1644 | ||
3452c2a1 MP |
1645 | /* |
1646 | * A snapshot-merge target behaves like a combination of a snapshot | |
1647 | * target and a snapshot-origin target. It only generates new | |
1648 | * exceptions in other snapshots and not in the one that is being | |
1649 | * merged. | |
1650 | * | |
1651 | * For each chunk, if there is an existing exception, it is used to | |
1652 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1653 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1654 | * If merging is currently taking place on the chunk in question, the |
1655 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 MP |
1656 | */ |
1657 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio, | |
1658 | union map_info *map_context) | |
1659 | { | |
1660 | struct dm_exception *e; | |
1661 | struct dm_snapshot *s = ti->private; | |
1662 | int r = DM_MAPIO_REMAPPED; | |
1663 | chunk_t chunk; | |
1664 | ||
d87f4c14 | 1665 | if (bio->bi_rw & REQ_FLUSH) { |
57cba5d3 | 1666 | if (!map_context->target_request_nr) |
10b8106a MS |
1667 | bio->bi_bdev = s->origin->bdev; |
1668 | else | |
1669 | bio->bi_bdev = s->cow->bdev; | |
1670 | map_context->ptr = NULL; | |
1671 | return DM_MAPIO_REMAPPED; | |
1672 | } | |
1673 | ||
3452c2a1 MP |
1674 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1675 | ||
9fe86254 | 1676 | down_write(&s->lock); |
3452c2a1 | 1677 | |
d2fdb776 MP |
1678 | /* Full merging snapshots are redirected to the origin */ |
1679 | if (!s->valid) | |
1680 | goto redirect_to_origin; | |
3452c2a1 MP |
1681 | |
1682 | /* If the block is already remapped - use that */ | |
1683 | e = dm_lookup_exception(&s->complete, chunk); | |
1684 | if (e) { | |
9fe86254 MP |
1685 | /* Queue writes overlapping with chunks being merged */ |
1686 | if (bio_rw(bio) == WRITE && | |
1687 | chunk >= s->first_merging_chunk && | |
1688 | chunk < (s->first_merging_chunk + | |
1689 | s->num_merging_chunks)) { | |
1690 | bio->bi_bdev = s->origin->bdev; | |
1691 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1692 | r = DM_MAPIO_SUBMITTED; | |
1693 | goto out_unlock; | |
1694 | } | |
17aa0332 | 1695 | |
3452c2a1 | 1696 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1697 | |
1698 | if (bio_rw(bio) == WRITE) | |
1699 | map_context->ptr = track_chunk(s, chunk); | |
3452c2a1 MP |
1700 | goto out_unlock; |
1701 | } | |
1702 | ||
d2fdb776 | 1703 | redirect_to_origin: |
3452c2a1 MP |
1704 | bio->bi_bdev = s->origin->bdev; |
1705 | ||
1706 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1707 | up_write(&s->lock); |
3452c2a1 MP |
1708 | return do_origin(s->origin, bio); |
1709 | } | |
1710 | ||
1711 | out_unlock: | |
9fe86254 | 1712 | up_write(&s->lock); |
3452c2a1 MP |
1713 | |
1714 | return r; | |
1715 | } | |
1716 | ||
cd45daff MP |
1717 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1718 | int error, union map_info *map_context) | |
1719 | { | |
1720 | struct dm_snapshot *s = ti->private; | |
1721 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1722 | ||
1723 | if (c) | |
1724 | stop_tracking_chunk(s, c); | |
1725 | ||
1726 | return 0; | |
1727 | } | |
1728 | ||
1e03f97e MP |
1729 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1730 | { | |
1731 | struct dm_snapshot *s = ti->private; | |
1732 | ||
1733 | stop_merge(s); | |
1734 | } | |
1735 | ||
c26655ca MS |
1736 | static void snapshot_postsuspend(struct dm_target *ti) |
1737 | { | |
1738 | struct dm_snapshot *s = ti->private; | |
1739 | ||
1740 | down_write(&s->lock); | |
1741 | s->suspended = 1; | |
1742 | up_write(&s->lock); | |
1743 | } | |
1744 | ||
c1f0c183 MS |
1745 | static int snapshot_preresume(struct dm_target *ti) |
1746 | { | |
1747 | int r = 0; | |
1748 | struct dm_snapshot *s = ti->private; | |
1749 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1750 | ||
1751 | down_read(&_origins_lock); | |
9d3b15c4 | 1752 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1753 | if (snap_src && snap_dest) { |
1754 | down_read(&snap_src->lock); | |
1755 | if (s == snap_src) { | |
1756 | DMERR("Unable to resume snapshot source until " | |
1757 | "handover completes."); | |
1758 | r = -EINVAL; | |
1759 | } else if (!snap_src->suspended) { | |
1760 | DMERR("Unable to perform snapshot handover until " | |
1761 | "source is suspended."); | |
1762 | r = -EINVAL; | |
1763 | } | |
1764 | up_read(&snap_src->lock); | |
1765 | } | |
1766 | up_read(&_origins_lock); | |
1767 | ||
1768 | return r; | |
1769 | } | |
1770 | ||
1da177e4 LT |
1771 | static void snapshot_resume(struct dm_target *ti) |
1772 | { | |
028867ac | 1773 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1774 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1775 | ||
1776 | down_read(&_origins_lock); | |
9d3b15c4 | 1777 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1778 | if (snap_src && snap_dest) { |
1779 | down_write(&snap_src->lock); | |
1780 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1781 | __handover_exceptions(snap_src, snap_dest); | |
1782 | up_write(&snap_dest->lock); | |
1783 | up_write(&snap_src->lock); | |
1784 | } | |
1785 | up_read(&_origins_lock); | |
1786 | ||
1787 | /* Now we have correct chunk size, reregister */ | |
1788 | reregister_snapshot(s); | |
1da177e4 | 1789 | |
aa14edeb AK |
1790 | down_write(&s->lock); |
1791 | s->active = 1; | |
c26655ca | 1792 | s->suspended = 0; |
aa14edeb | 1793 | up_write(&s->lock); |
1da177e4 LT |
1794 | } |
1795 | ||
1e03f97e MP |
1796 | static sector_t get_origin_minimum_chunksize(struct block_device *bdev) |
1797 | { | |
1798 | sector_t min_chunksize; | |
1799 | ||
1800 | down_read(&_origins_lock); | |
1801 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1802 | up_read(&_origins_lock); | |
1803 | ||
1804 | return min_chunksize; | |
1805 | } | |
1806 | ||
1807 | static void snapshot_merge_resume(struct dm_target *ti) | |
1808 | { | |
1809 | struct dm_snapshot *s = ti->private; | |
1810 | ||
1811 | /* | |
1812 | * Handover exceptions from existing snapshot. | |
1813 | */ | |
1814 | snapshot_resume(ti); | |
1815 | ||
1816 | /* | |
1817 | * snapshot-merge acts as an origin, so set ti->split_io | |
1818 | */ | |
1819 | ti->split_io = get_origin_minimum_chunksize(s->origin->bdev); | |
1820 | ||
1821 | start_merge(s); | |
1822 | } | |
1823 | ||
1da177e4 LT |
1824 | static int snapshot_status(struct dm_target *ti, status_type_t type, |
1825 | char *result, unsigned int maxlen) | |
1826 | { | |
2e4a31df | 1827 | unsigned sz = 0; |
028867ac | 1828 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1829 | |
1830 | switch (type) { | |
1831 | case STATUSTYPE_INFO: | |
94e76572 MP |
1832 | |
1833 | down_write(&snap->lock); | |
1834 | ||
1da177e4 | 1835 | if (!snap->valid) |
2e4a31df | 1836 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1837 | else if (snap->merge_failed) |
1838 | DMEMIT("Merge failed"); | |
1da177e4 | 1839 | else { |
985903bb MS |
1840 | if (snap->store->type->usage) { |
1841 | sector_t total_sectors, sectors_allocated, | |
1842 | metadata_sectors; | |
1843 | snap->store->type->usage(snap->store, | |
1844 | &total_sectors, | |
1845 | §ors_allocated, | |
1846 | &metadata_sectors); | |
1847 | DMEMIT("%llu/%llu %llu", | |
1848 | (unsigned long long)sectors_allocated, | |
1849 | (unsigned long long)total_sectors, | |
1850 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1851 | } |
1852 | else | |
2e4a31df | 1853 | DMEMIT("Unknown"); |
1da177e4 | 1854 | } |
94e76572 MP |
1855 | |
1856 | up_write(&snap->lock); | |
1857 | ||
1da177e4 LT |
1858 | break; |
1859 | ||
1860 | case STATUSTYPE_TABLE: | |
1861 | /* | |
1862 | * kdevname returns a static pointer so we need | |
1863 | * to make private copies if the output is to | |
1864 | * make sense. | |
1865 | */ | |
fc56f6fb | 1866 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1867 | snap->store->type->status(snap->store, type, result + sz, |
1868 | maxlen - sz); | |
1da177e4 LT |
1869 | break; |
1870 | } | |
1871 | ||
1872 | return 0; | |
1873 | } | |
1874 | ||
8811f46c MS |
1875 | static int snapshot_iterate_devices(struct dm_target *ti, |
1876 | iterate_devices_callout_fn fn, void *data) | |
1877 | { | |
1878 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
1879 | int r; |
1880 | ||
1881 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 1882 | |
1e5554c8 MP |
1883 | if (!r) |
1884 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
1885 | ||
1886 | return r; | |
8811f46c MS |
1887 | } |
1888 | ||
1889 | ||
1da177e4 LT |
1890 | /*----------------------------------------------------------------- |
1891 | * Origin methods | |
1892 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1893 | |
1894 | /* | |
1895 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1896 | * supplied bio was ignored. The caller may submit it immediately. | |
1897 | * (No remapping actually occurs as the origin is always a direct linear | |
1898 | * map.) | |
1899 | * | |
1900 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1901 | * and any supplied bio is added to a list to be submitted once all | |
1902 | * the necessary exceptions exist. | |
1903 | */ | |
1904 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1905 | struct bio *bio) | |
1da177e4 | 1906 | { |
515ad66c | 1907 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1908 | struct dm_snapshot *snap; |
1d4989c8 | 1909 | struct dm_exception *e; |
515ad66c MP |
1910 | struct dm_snap_pending_exception *pe; |
1911 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1912 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1913 | chunk_t chunk; |
1914 | ||
1915 | /* Do all the snapshots on this origin */ | |
1916 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
1917 | /* |
1918 | * Don't make new exceptions in a merging snapshot | |
1919 | * because it has effectively been deleted | |
1920 | */ | |
1921 | if (dm_target_is_snapshot_merge(snap->ti)) | |
1922 | continue; | |
1923 | ||
76df1c65 AK |
1924 | down_write(&snap->lock); |
1925 | ||
aa14edeb AK |
1926 | /* Only deal with valid and active snapshots */ |
1927 | if (!snap->valid || !snap->active) | |
76df1c65 | 1928 | goto next_snapshot; |
1da177e4 | 1929 | |
d5e404c1 | 1930 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1931 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1932 | goto next_snapshot; |
1da177e4 LT |
1933 | |
1934 | /* | |
1935 | * Remember, different snapshots can have | |
1936 | * different chunk sizes. | |
1937 | */ | |
9eaae8ff | 1938 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1939 | |
1940 | /* | |
1941 | * Check exception table to see if block | |
1942 | * is already remapped in this snapshot | |
1943 | * and trigger an exception if not. | |
1944 | */ | |
3510cb94 | 1945 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1946 | if (e) |
1947 | goto next_snapshot; | |
1948 | ||
2913808e | 1949 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1950 | if (!pe) { |
c6621392 MP |
1951 | up_write(&snap->lock); |
1952 | pe = alloc_pending_exception(snap); | |
1953 | down_write(&snap->lock); | |
1954 | ||
1955 | if (!snap->valid) { | |
1956 | free_pending_exception(pe); | |
1957 | goto next_snapshot; | |
1958 | } | |
1959 | ||
3510cb94 | 1960 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1961 | if (e) { |
1962 | free_pending_exception(pe); | |
1963 | goto next_snapshot; | |
1964 | } | |
1965 | ||
c6621392 | 1966 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1967 | if (!pe) { |
1968 | __invalidate_snapshot(snap, -ENOMEM); | |
1969 | goto next_snapshot; | |
1970 | } | |
76df1c65 AK |
1971 | } |
1972 | ||
515ad66c | 1973 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 1974 | |
515ad66c MP |
1975 | /* |
1976 | * If an origin bio was supplied, queue it to wait for the | |
1977 | * completion of this exception, and start this one last, | |
1978 | * at the end of the function. | |
1979 | */ | |
1980 | if (bio) { | |
1981 | bio_list_add(&pe->origin_bios, bio); | |
1982 | bio = NULL; | |
76df1c65 | 1983 | |
515ad66c MP |
1984 | if (!pe->started) { |
1985 | pe->started = 1; | |
1986 | pe_to_start_last = pe; | |
1987 | } | |
76df1c65 AK |
1988 | } |
1989 | ||
1990 | if (!pe->started) { | |
1991 | pe->started = 1; | |
515ad66c | 1992 | pe_to_start_now = pe; |
1da177e4 LT |
1993 | } |
1994 | ||
76df1c65 | 1995 | next_snapshot: |
1da177e4 | 1996 | up_write(&snap->lock); |
1da177e4 | 1997 | |
515ad66c MP |
1998 | if (pe_to_start_now) { |
1999 | start_copy(pe_to_start_now); | |
2000 | pe_to_start_now = NULL; | |
2001 | } | |
b4b610f6 AK |
2002 | } |
2003 | ||
1da177e4 | 2004 | /* |
515ad66c MP |
2005 | * Submit the exception against which the bio is queued last, |
2006 | * to give the other exceptions a head start. | |
1da177e4 | 2007 | */ |
515ad66c MP |
2008 | if (pe_to_start_last) |
2009 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2010 | |
2011 | return r; | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * Called on a write from the origin driver. | |
2016 | */ | |
2017 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2018 | { | |
2019 | struct origin *o; | |
d2a7ad29 | 2020 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2021 | |
2022 | down_read(&_origins_lock); | |
2023 | o = __lookup_origin(origin->bdev); | |
2024 | if (o) | |
9eaae8ff | 2025 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
2026 | up_read(&_origins_lock); |
2027 | ||
2028 | return r; | |
2029 | } | |
2030 | ||
73dfd078 MP |
2031 | /* |
2032 | * Trigger exceptions in all non-merging snapshots. | |
2033 | * | |
2034 | * The chunk size of the merging snapshot may be larger than the chunk | |
2035 | * size of some other snapshot so we may need to reallocate multiple | |
2036 | * chunks in other snapshots. | |
2037 | * | |
2038 | * We scan all the overlapping exceptions in the other snapshots. | |
2039 | * Returns 1 if anything was reallocated and must be waited for, | |
2040 | * otherwise returns 0. | |
2041 | * | |
2042 | * size must be a multiple of merging_snap's chunk_size. | |
2043 | */ | |
2044 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2045 | sector_t sector, unsigned size) | |
2046 | { | |
2047 | int must_wait = 0; | |
2048 | sector_t n; | |
2049 | struct origin *o; | |
2050 | ||
2051 | /* | |
2052 | * The origin's __minimum_chunk_size() got stored in split_io | |
2053 | * by snapshot_merge_resume(). | |
2054 | */ | |
2055 | down_read(&_origins_lock); | |
2056 | o = __lookup_origin(merging_snap->origin->bdev); | |
2057 | for (n = 0; n < size; n += merging_snap->ti->split_io) | |
2058 | if (__origin_write(&o->snapshots, sector + n, NULL) == | |
2059 | DM_MAPIO_SUBMITTED) | |
2060 | must_wait = 1; | |
2061 | up_read(&_origins_lock); | |
2062 | ||
2063 | return must_wait; | |
2064 | } | |
2065 | ||
1da177e4 LT |
2066 | /* |
2067 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2068 | */ | |
2069 | ||
2070 | /* | |
2071 | * Construct an origin mapping: <dev_path> | |
2072 | * The context for an origin is merely a 'struct dm_dev *' | |
2073 | * pointing to the real device. | |
2074 | */ | |
2075 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2076 | { | |
2077 | int r; | |
2078 | struct dm_dev *dev; | |
2079 | ||
2080 | if (argc != 1) { | |
72d94861 | 2081 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2082 | return -EINVAL; |
2083 | } | |
2084 | ||
8215d6ec | 2085 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev); |
1da177e4 LT |
2086 | if (r) { |
2087 | ti->error = "Cannot get target device"; | |
2088 | return r; | |
2089 | } | |
2090 | ||
2091 | ti->private = dev; | |
494b3ee7 MP |
2092 | ti->num_flush_requests = 1; |
2093 | ||
1da177e4 LT |
2094 | return 0; |
2095 | } | |
2096 | ||
2097 | static void origin_dtr(struct dm_target *ti) | |
2098 | { | |
028867ac | 2099 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2100 | dm_put_device(ti, dev); |
2101 | } | |
2102 | ||
2103 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
2104 | union map_info *map_context) | |
2105 | { | |
028867ac | 2106 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2107 | bio->bi_bdev = dev->bdev; |
2108 | ||
d87f4c14 | 2109 | if (bio->bi_rw & REQ_FLUSH) |
494b3ee7 MP |
2110 | return DM_MAPIO_REMAPPED; |
2111 | ||
1da177e4 | 2112 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 2113 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
2114 | } |
2115 | ||
1da177e4 LT |
2116 | /* |
2117 | * Set the target "split_io" field to the minimum of all the snapshots' | |
2118 | * chunk sizes. | |
2119 | */ | |
2120 | static void origin_resume(struct dm_target *ti) | |
2121 | { | |
028867ac | 2122 | struct dm_dev *dev = ti->private; |
1da177e4 | 2123 | |
1e03f97e | 2124 | ti->split_io = get_origin_minimum_chunksize(dev->bdev); |
1da177e4 LT |
2125 | } |
2126 | ||
2127 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
2128 | unsigned int maxlen) | |
2129 | { | |
028867ac | 2130 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2131 | |
2132 | switch (type) { | |
2133 | case STATUSTYPE_INFO: | |
2134 | result[0] = '\0'; | |
2135 | break; | |
2136 | ||
2137 | case STATUSTYPE_TABLE: | |
2138 | snprintf(result, maxlen, "%s", dev->name); | |
2139 | break; | |
2140 | } | |
2141 | ||
2142 | return 0; | |
2143 | } | |
2144 | ||
b1d55528 MP |
2145 | static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
2146 | struct bio_vec *biovec, int max_size) | |
2147 | { | |
2148 | struct dm_dev *dev = ti->private; | |
2149 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2150 | ||
2151 | if (!q->merge_bvec_fn) | |
2152 | return max_size; | |
2153 | ||
2154 | bvm->bi_bdev = dev->bdev; | |
2155 | bvm->bi_sector = bvm->bi_sector; | |
2156 | ||
2157 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2158 | } | |
2159 | ||
8811f46c MS |
2160 | static int origin_iterate_devices(struct dm_target *ti, |
2161 | iterate_devices_callout_fn fn, void *data) | |
2162 | { | |
2163 | struct dm_dev *dev = ti->private; | |
2164 | ||
2165 | return fn(ti, dev, 0, ti->len, data); | |
2166 | } | |
2167 | ||
1da177e4 LT |
2168 | static struct target_type origin_target = { |
2169 | .name = "snapshot-origin", | |
8811f46c | 2170 | .version = {1, 7, 0}, |
1da177e4 LT |
2171 | .module = THIS_MODULE, |
2172 | .ctr = origin_ctr, | |
2173 | .dtr = origin_dtr, | |
2174 | .map = origin_map, | |
2175 | .resume = origin_resume, | |
2176 | .status = origin_status, | |
b1d55528 | 2177 | .merge = origin_merge, |
8811f46c | 2178 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2179 | }; |
2180 | ||
2181 | static struct target_type snapshot_target = { | |
2182 | .name = "snapshot", | |
c26655ca | 2183 | .version = {1, 9, 0}, |
1da177e4 LT |
2184 | .module = THIS_MODULE, |
2185 | .ctr = snapshot_ctr, | |
2186 | .dtr = snapshot_dtr, | |
2187 | .map = snapshot_map, | |
cd45daff | 2188 | .end_io = snapshot_end_io, |
c26655ca | 2189 | .postsuspend = snapshot_postsuspend, |
c1f0c183 | 2190 | .preresume = snapshot_preresume, |
1da177e4 LT |
2191 | .resume = snapshot_resume, |
2192 | .status = snapshot_status, | |
8811f46c | 2193 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2194 | }; |
2195 | ||
d698aa45 MP |
2196 | static struct target_type merge_target = { |
2197 | .name = dm_snapshot_merge_target_name, | |
2198 | .version = {1, 0, 0}, | |
2199 | .module = THIS_MODULE, | |
2200 | .ctr = snapshot_ctr, | |
2201 | .dtr = snapshot_dtr, | |
3452c2a1 | 2202 | .map = snapshot_merge_map, |
d698aa45 | 2203 | .end_io = snapshot_end_io, |
1e03f97e | 2204 | .presuspend = snapshot_merge_presuspend, |
d698aa45 MP |
2205 | .postsuspend = snapshot_postsuspend, |
2206 | .preresume = snapshot_preresume, | |
1e03f97e | 2207 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2208 | .status = snapshot_status, |
2209 | .iterate_devices = snapshot_iterate_devices, | |
2210 | }; | |
2211 | ||
1da177e4 LT |
2212 | static int __init dm_snapshot_init(void) |
2213 | { | |
2214 | int r; | |
2215 | ||
4db6bfe0 AK |
2216 | r = dm_exception_store_init(); |
2217 | if (r) { | |
2218 | DMERR("Failed to initialize exception stores"); | |
2219 | return r; | |
2220 | } | |
2221 | ||
1da177e4 | 2222 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2223 | if (r < 0) { |
1da177e4 | 2224 | DMERR("snapshot target register failed %d", r); |
034a186d | 2225 | goto bad_register_snapshot_target; |
1da177e4 LT |
2226 | } |
2227 | ||
2228 | r = dm_register_target(&origin_target); | |
2229 | if (r < 0) { | |
72d94861 | 2230 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2231 | goto bad_register_origin_target; |
2232 | } | |
2233 | ||
2234 | r = dm_register_target(&merge_target); | |
2235 | if (r < 0) { | |
2236 | DMERR("Merge target register failed %d", r); | |
2237 | goto bad_register_merge_target; | |
1da177e4 LT |
2238 | } |
2239 | ||
2240 | r = init_origin_hash(); | |
2241 | if (r) { | |
2242 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2243 | goto bad_origin_hash; |
1da177e4 LT |
2244 | } |
2245 | ||
1d4989c8 | 2246 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2247 | if (!exception_cache) { |
2248 | DMERR("Couldn't create exception cache."); | |
2249 | r = -ENOMEM; | |
d698aa45 | 2250 | goto bad_exception_cache; |
1da177e4 LT |
2251 | } |
2252 | ||
028867ac | 2253 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2254 | if (!pending_cache) { |
2255 | DMERR("Couldn't create pending cache."); | |
2256 | r = -ENOMEM; | |
d698aa45 | 2257 | goto bad_pending_cache; |
1da177e4 LT |
2258 | } |
2259 | ||
cd45daff MP |
2260 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
2261 | if (!tracked_chunk_cache) { | |
2262 | DMERR("Couldn't create cache to track chunks in use."); | |
2263 | r = -ENOMEM; | |
d698aa45 | 2264 | goto bad_tracked_chunk_cache; |
cd45daff MP |
2265 | } |
2266 | ||
1da177e4 LT |
2267 | return 0; |
2268 | ||
d698aa45 | 2269 | bad_tracked_chunk_cache: |
1da177e4 | 2270 | kmem_cache_destroy(pending_cache); |
d698aa45 | 2271 | bad_pending_cache: |
1da177e4 | 2272 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2273 | bad_exception_cache: |
1da177e4 | 2274 | exit_origin_hash(); |
d698aa45 MP |
2275 | bad_origin_hash: |
2276 | dm_unregister_target(&merge_target); | |
2277 | bad_register_merge_target: | |
1da177e4 | 2278 | dm_unregister_target(&origin_target); |
d698aa45 | 2279 | bad_register_origin_target: |
1da177e4 | 2280 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2281 | bad_register_snapshot_target: |
2282 | dm_exception_store_exit(); | |
d698aa45 | 2283 | |
1da177e4 LT |
2284 | return r; |
2285 | } | |
2286 | ||
2287 | static void __exit dm_snapshot_exit(void) | |
2288 | { | |
10d3bd09 MP |
2289 | dm_unregister_target(&snapshot_target); |
2290 | dm_unregister_target(&origin_target); | |
d698aa45 | 2291 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2292 | |
2293 | exit_origin_hash(); | |
1da177e4 LT |
2294 | kmem_cache_destroy(pending_cache); |
2295 | kmem_cache_destroy(exception_cache); | |
cd45daff | 2296 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
2297 | |
2298 | dm_exception_store_exit(); | |
1da177e4 LT |
2299 | } |
2300 | ||
2301 | /* Module hooks */ | |
2302 | module_init(dm_snapshot_init); | |
2303 | module_exit(dm_snapshot_exit); | |
2304 | ||
2305 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2306 | MODULE_AUTHOR("Joe Thornber"); | |
2307 | MODULE_LICENSE("GPL"); |