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> |
ccc45ea8 | 22 | #include <linux/workqueue.h> |
1da177e4 | 23 | |
aea53d92 | 24 | #include "dm-exception-store.h" |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "snapshots" |
27 | ||
d698aa45 MP |
28 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
29 | ||
30 | #define dm_target_is_snapshot_merge(ti) \ | |
31 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
32 | ||
1da177e4 LT |
33 | /* |
34 | * The percentage increment we will wake up users at | |
35 | */ | |
36 | #define WAKE_UP_PERCENT 5 | |
37 | ||
38 | /* | |
39 | * kcopyd priority of snapshot operations | |
40 | */ | |
41 | #define SNAPSHOT_COPY_PRIORITY 2 | |
42 | ||
43 | /* | |
8ee2767a | 44 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 45 | */ |
8ee2767a | 46 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 47 | |
cd45daff MP |
48 | /* |
49 | * The size of the mempool used to track chunks in use. | |
50 | */ | |
51 | #define MIN_IOS 256 | |
52 | ||
ccc45ea8 JB |
53 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
54 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
55 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
56 | ||
191437a5 | 57 | struct dm_exception_table { |
ccc45ea8 JB |
58 | uint32_t hash_mask; |
59 | unsigned hash_shift; | |
60 | struct list_head *table; | |
61 | }; | |
62 | ||
63 | struct dm_snapshot { | |
64 | struct rw_semaphore lock; | |
65 | ||
66 | struct dm_dev *origin; | |
fc56f6fb MS |
67 | struct dm_dev *cow; |
68 | ||
69 | struct dm_target *ti; | |
ccc45ea8 JB |
70 | |
71 | /* List of snapshots per Origin */ | |
72 | struct list_head list; | |
73 | ||
74 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
75 | int valid; | |
76 | ||
77 | /* Origin writes don't trigger exceptions until this is set */ | |
78 | int active; | |
79 | ||
c26655ca MS |
80 | /* Whether or not owning mapped_device is suspended */ |
81 | int suspended; | |
82 | ||
ccc45ea8 JB |
83 | mempool_t *pending_pool; |
84 | ||
85 | atomic_t pending_exceptions_count; | |
86 | ||
191437a5 JB |
87 | struct dm_exception_table pending; |
88 | struct dm_exception_table complete; | |
ccc45ea8 JB |
89 | |
90 | /* | |
91 | * pe_lock protects all pending_exception operations and access | |
92 | * as well as the snapshot_bios list. | |
93 | */ | |
94 | spinlock_t pe_lock; | |
95 | ||
96 | /* The on disk metadata handler */ | |
97 | struct dm_exception_store *store; | |
98 | ||
99 | struct dm_kcopyd_client *kcopyd_client; | |
100 | ||
101 | /* Queue of snapshot writes for ksnapd to flush */ | |
102 | struct bio_list queued_bios; | |
103 | struct work_struct queued_bios_work; | |
104 | ||
105 | /* Chunks with outstanding reads */ | |
106 | mempool_t *tracked_chunk_pool; | |
107 | spinlock_t tracked_chunk_lock; | |
108 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
109 | }; | |
110 | ||
fc56f6fb MS |
111 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
112 | { | |
113 | return s->cow; | |
114 | } | |
115 | EXPORT_SYMBOL(dm_snap_cow); | |
116 | ||
c642f9e0 | 117 | static struct workqueue_struct *ksnapd; |
c4028958 | 118 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 119 | |
ccc45ea8 JB |
120 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
121 | chunk_t chunk) | |
122 | { | |
123 | return chunk << store->chunk_shift; | |
124 | } | |
125 | ||
126 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
127 | { | |
128 | /* | |
129 | * There is only ever one instance of a particular block | |
130 | * device so we can compare pointers safely. | |
131 | */ | |
132 | return lhs == rhs; | |
133 | } | |
134 | ||
028867ac | 135 | struct dm_snap_pending_exception { |
1d4989c8 | 136 | struct dm_exception e; |
1da177e4 LT |
137 | |
138 | /* | |
139 | * Origin buffers waiting for this to complete are held | |
140 | * in a bio list | |
141 | */ | |
142 | struct bio_list origin_bios; | |
143 | struct bio_list snapshot_bios; | |
144 | ||
1da177e4 LT |
145 | /* Pointer back to snapshot context */ |
146 | struct dm_snapshot *snap; | |
147 | ||
148 | /* | |
149 | * 1 indicates the exception has already been sent to | |
150 | * kcopyd. | |
151 | */ | |
152 | int started; | |
153 | }; | |
154 | ||
155 | /* | |
156 | * Hash table mapping origin volumes to lists of snapshots and | |
157 | * a lock to protect it | |
158 | */ | |
e18b890b CL |
159 | static struct kmem_cache *exception_cache; |
160 | static struct kmem_cache *pending_cache; | |
1da177e4 | 161 | |
cd45daff MP |
162 | struct dm_snap_tracked_chunk { |
163 | struct hlist_node node; | |
164 | chunk_t chunk; | |
165 | }; | |
166 | ||
167 | static struct kmem_cache *tracked_chunk_cache; | |
168 | ||
169 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
170 | chunk_t chunk) | |
171 | { | |
172 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
173 | GFP_NOIO); | |
174 | unsigned long flags; | |
175 | ||
176 | c->chunk = chunk; | |
177 | ||
178 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
179 | hlist_add_head(&c->node, | |
180 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
181 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
182 | ||
183 | return c; | |
184 | } | |
185 | ||
186 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
187 | struct dm_snap_tracked_chunk *c) | |
188 | { | |
189 | unsigned long flags; | |
190 | ||
191 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
192 | hlist_del(&c->node); | |
193 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
194 | ||
195 | mempool_free(c, s->tracked_chunk_pool); | |
196 | } | |
197 | ||
a8d41b59 MP |
198 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
199 | { | |
200 | struct dm_snap_tracked_chunk *c; | |
201 | struct hlist_node *hn; | |
202 | int found = 0; | |
203 | ||
204 | spin_lock_irq(&s->tracked_chunk_lock); | |
205 | ||
206 | hlist_for_each_entry(c, hn, | |
207 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
208 | if (c->chunk == chunk) { | |
209 | found = 1; | |
210 | break; | |
211 | } | |
212 | } | |
213 | ||
214 | spin_unlock_irq(&s->tracked_chunk_lock); | |
215 | ||
216 | return found; | |
217 | } | |
218 | ||
615d1eb9 MS |
219 | /* |
220 | * This conflicting I/O is extremely improbable in the caller, | |
221 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
222 | */ | |
223 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
224 | { | |
225 | while (__chunk_is_tracked(s, chunk)) | |
226 | msleep(1); | |
227 | } | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * One of these per registered origin, held in the snapshot_origins hash | |
231 | */ | |
232 | struct origin { | |
233 | /* The origin device */ | |
234 | struct block_device *bdev; | |
235 | ||
236 | struct list_head hash_list; | |
237 | ||
238 | /* List of snapshots for this origin */ | |
239 | struct list_head snapshots; | |
240 | }; | |
241 | ||
242 | /* | |
243 | * Size of the hash table for origin volumes. If we make this | |
244 | * the size of the minors list then it should be nearly perfect | |
245 | */ | |
246 | #define ORIGIN_HASH_SIZE 256 | |
247 | #define ORIGIN_MASK 0xFF | |
248 | static struct list_head *_origins; | |
249 | static struct rw_semaphore _origins_lock; | |
250 | ||
251 | static int init_origin_hash(void) | |
252 | { | |
253 | int i; | |
254 | ||
255 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
256 | GFP_KERNEL); | |
257 | if (!_origins) { | |
72d94861 | 258 | DMERR("unable to allocate memory"); |
1da177e4 LT |
259 | return -ENOMEM; |
260 | } | |
261 | ||
262 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
263 | INIT_LIST_HEAD(_origins + i); | |
264 | init_rwsem(&_origins_lock); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | static void exit_origin_hash(void) | |
270 | { | |
271 | kfree(_origins); | |
272 | } | |
273 | ||
028867ac | 274 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
275 | { |
276 | return bdev->bd_dev & ORIGIN_MASK; | |
277 | } | |
278 | ||
279 | static struct origin *__lookup_origin(struct block_device *origin) | |
280 | { | |
281 | struct list_head *ol; | |
282 | struct origin *o; | |
283 | ||
284 | ol = &_origins[origin_hash(origin)]; | |
285 | list_for_each_entry (o, ol, hash_list) | |
286 | if (bdev_equal(o->bdev, origin)) | |
287 | return o; | |
288 | ||
289 | return NULL; | |
290 | } | |
291 | ||
292 | static void __insert_origin(struct origin *o) | |
293 | { | |
294 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
295 | list_add_tail(&o->hash_list, sl); | |
296 | } | |
297 | ||
c1f0c183 MS |
298 | /* |
299 | * _origins_lock must be held when calling this function. | |
300 | * Returns number of snapshots registered using the supplied cow device, plus: | |
301 | * snap_src - a snapshot suitable for use as a source of exception handover | |
302 | * snap_dest - a snapshot capable of receiving exception handover. | |
303 | * | |
304 | * Possible return values and states: | |
305 | * 0: NULL, NULL - first new snapshot | |
306 | * 1: snap_src, NULL - normal snapshot | |
307 | * 2: snap_src, snap_dest - waiting for handover | |
308 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
309 | * 1: NULL, snap_dest - source got destroyed without handover | |
310 | */ | |
311 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
312 | struct dm_snapshot **snap_src, | |
313 | struct dm_snapshot **snap_dest) | |
314 | { | |
315 | struct dm_snapshot *s; | |
316 | struct origin *o; | |
317 | int count = 0; | |
318 | int active; | |
319 | ||
320 | o = __lookup_origin(snap->origin->bdev); | |
321 | if (!o) | |
322 | goto out; | |
323 | ||
324 | list_for_each_entry(s, &o->snapshots, list) { | |
325 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) | |
326 | continue; | |
327 | ||
328 | down_read(&s->lock); | |
329 | active = s->active; | |
330 | up_read(&s->lock); | |
331 | ||
332 | if (active) { | |
333 | if (snap_src) | |
334 | *snap_src = s; | |
335 | } else if (snap_dest) | |
336 | *snap_dest = s; | |
337 | ||
338 | count++; | |
339 | } | |
340 | ||
341 | out: | |
342 | return count; | |
343 | } | |
344 | ||
345 | /* | |
346 | * On success, returns 1 if this snapshot is a handover destination, | |
347 | * otherwise returns 0. | |
348 | */ | |
349 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
350 | { | |
351 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
352 | ||
353 | /* Does snapshot need exceptions handed over to it? */ | |
354 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) || | |
355 | snap_dest) { | |
356 | snap->ti->error = "Snapshot cow pairing for exception " | |
357 | "table handover failed"; | |
358 | return -EINVAL; | |
359 | } | |
360 | ||
361 | /* | |
362 | * If no snap_src was found, snap cannot become a handover | |
363 | * destination. | |
364 | */ | |
365 | if (!snap_src) | |
366 | return 0; | |
367 | ||
368 | return 1; | |
369 | } | |
370 | ||
371 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
372 | { | |
373 | struct dm_snapshot *l; | |
374 | ||
375 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
376 | list_for_each_entry(l, &o->snapshots, list) | |
377 | if (l->store->chunk_size < s->store->chunk_size) | |
378 | break; | |
379 | list_add_tail(&s->list, &l->list); | |
380 | } | |
381 | ||
1da177e4 LT |
382 | /* |
383 | * Make a note of the snapshot and its origin so we can look it | |
384 | * up when the origin has a write on it. | |
c1f0c183 MS |
385 | * |
386 | * Also validate snapshot exception store handovers. | |
387 | * On success, returns 1 if this registration is a handover destination, | |
388 | * otherwise returns 0. | |
1da177e4 LT |
389 | */ |
390 | static int register_snapshot(struct dm_snapshot *snap) | |
391 | { | |
c1f0c183 | 392 | struct origin *o, *new_o = NULL; |
1da177e4 | 393 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 394 | int r = 0; |
1da177e4 | 395 | |
60c856c8 MP |
396 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
397 | if (!new_o) | |
398 | return -ENOMEM; | |
399 | ||
1da177e4 | 400 | down_write(&_origins_lock); |
1da177e4 | 401 | |
c1f0c183 MS |
402 | r = __validate_exception_handover(snap); |
403 | if (r < 0) { | |
404 | kfree(new_o); | |
405 | goto out; | |
406 | } | |
407 | ||
408 | o = __lookup_origin(bdev); | |
60c856c8 MP |
409 | if (o) |
410 | kfree(new_o); | |
411 | else { | |
1da177e4 | 412 | /* New origin */ |
60c856c8 | 413 | o = new_o; |
1da177e4 LT |
414 | |
415 | /* Initialise the struct */ | |
416 | INIT_LIST_HEAD(&o->snapshots); | |
417 | o->bdev = bdev; | |
418 | ||
419 | __insert_origin(o); | |
420 | } | |
421 | ||
c1f0c183 MS |
422 | __insert_snapshot(o, snap); |
423 | ||
424 | out: | |
425 | up_write(&_origins_lock); | |
426 | ||
427 | return r; | |
428 | } | |
429 | ||
430 | /* | |
431 | * Move snapshot to correct place in list according to chunk size. | |
432 | */ | |
433 | static void reregister_snapshot(struct dm_snapshot *s) | |
434 | { | |
435 | struct block_device *bdev = s->origin->bdev; | |
436 | ||
437 | down_write(&_origins_lock); | |
438 | ||
439 | list_del(&s->list); | |
440 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
441 | |
442 | up_write(&_origins_lock); | |
1da177e4 LT |
443 | } |
444 | ||
445 | static void unregister_snapshot(struct dm_snapshot *s) | |
446 | { | |
447 | struct origin *o; | |
448 | ||
449 | down_write(&_origins_lock); | |
450 | o = __lookup_origin(s->origin->bdev); | |
451 | ||
452 | list_del(&s->list); | |
c1f0c183 | 453 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
454 | list_del(&o->hash_list); |
455 | kfree(o); | |
456 | } | |
457 | ||
458 | up_write(&_origins_lock); | |
459 | } | |
460 | ||
461 | /* | |
462 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
463 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
464 | * some consecutive chunks to be grouped together. | |
1da177e4 | 465 | */ |
3510cb94 JB |
466 | static int dm_exception_table_init(struct dm_exception_table *et, |
467 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
468 | { |
469 | unsigned int i; | |
470 | ||
d74f81f8 | 471 | et->hash_shift = hash_shift; |
1da177e4 LT |
472 | et->hash_mask = size - 1; |
473 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
474 | if (!et->table) | |
475 | return -ENOMEM; | |
476 | ||
477 | for (i = 0; i < size; i++) | |
478 | INIT_LIST_HEAD(et->table + i); | |
479 | ||
480 | return 0; | |
481 | } | |
482 | ||
3510cb94 JB |
483 | static void dm_exception_table_exit(struct dm_exception_table *et, |
484 | struct kmem_cache *mem) | |
1da177e4 LT |
485 | { |
486 | struct list_head *slot; | |
1d4989c8 | 487 | struct dm_exception *ex, *next; |
1da177e4 LT |
488 | int i, size; |
489 | ||
490 | size = et->hash_mask + 1; | |
491 | for (i = 0; i < size; i++) { | |
492 | slot = et->table + i; | |
493 | ||
494 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
495 | kmem_cache_free(mem, ex); | |
496 | } | |
497 | ||
498 | vfree(et->table); | |
499 | } | |
500 | ||
191437a5 | 501 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 502 | { |
d74f81f8 | 503 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
504 | } |
505 | ||
3510cb94 | 506 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
507 | { |
508 | list_del(&e->hash_list); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Return the exception data for a sector, or NULL if not | |
513 | * remapped. | |
514 | */ | |
3510cb94 JB |
515 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
516 | chunk_t chunk) | |
1da177e4 LT |
517 | { |
518 | struct list_head *slot; | |
1d4989c8 | 519 | struct dm_exception *e; |
1da177e4 LT |
520 | |
521 | slot = &et->table[exception_hash(et, chunk)]; | |
522 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
523 | if (chunk >= e->old_chunk && |
524 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
525 | return e; |
526 | ||
527 | return NULL; | |
528 | } | |
529 | ||
3510cb94 | 530 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 531 | { |
1d4989c8 | 532 | struct dm_exception *e; |
1da177e4 LT |
533 | |
534 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
535 | if (!e) | |
536 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
537 | ||
538 | return e; | |
539 | } | |
540 | ||
3510cb94 | 541 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
542 | { |
543 | kmem_cache_free(exception_cache, e); | |
544 | } | |
545 | ||
92e86812 | 546 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 547 | { |
92e86812 MP |
548 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
549 | GFP_NOIO); | |
550 | ||
879129d2 | 551 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
552 | pe->snap = s; |
553 | ||
554 | return pe; | |
1da177e4 LT |
555 | } |
556 | ||
028867ac | 557 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 558 | { |
879129d2 MP |
559 | struct dm_snapshot *s = pe->snap; |
560 | ||
561 | mempool_free(pe, s->pending_pool); | |
562 | smp_mb__before_atomic_dec(); | |
563 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
564 | } |
565 | ||
3510cb94 JB |
566 | static void dm_insert_exception(struct dm_exception_table *eh, |
567 | struct dm_exception *new_e) | |
d74f81f8 | 568 | { |
d74f81f8 | 569 | struct list_head *l; |
1d4989c8 | 570 | struct dm_exception *e = NULL; |
d74f81f8 MB |
571 | |
572 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
573 | ||
574 | /* Add immediately if this table doesn't support consecutive chunks */ | |
575 | if (!eh->hash_shift) | |
576 | goto out; | |
577 | ||
578 | /* List is ordered by old_chunk */ | |
579 | list_for_each_entry_reverse(e, l, hash_list) { | |
580 | /* Insert after an existing chunk? */ | |
581 | if (new_e->old_chunk == (e->old_chunk + | |
582 | dm_consecutive_chunk_count(e) + 1) && | |
583 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
584 | dm_consecutive_chunk_count(e) + 1)) { | |
585 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 586 | free_completed_exception(new_e); |
d74f81f8 MB |
587 | return; |
588 | } | |
589 | ||
590 | /* Insert before an existing chunk? */ | |
591 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
592 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
593 | dm_consecutive_chunk_count_inc(e); | |
594 | e->old_chunk--; | |
595 | e->new_chunk--; | |
3510cb94 | 596 | free_completed_exception(new_e); |
d74f81f8 MB |
597 | return; |
598 | } | |
599 | ||
600 | if (new_e->old_chunk > e->old_chunk) | |
601 | break; | |
602 | } | |
603 | ||
604 | out: | |
605 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
606 | } | |
607 | ||
a159c1ac JB |
608 | /* |
609 | * Callback used by the exception stores to load exceptions when | |
610 | * initialising. | |
611 | */ | |
612 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 613 | { |
a159c1ac | 614 | struct dm_snapshot *s = context; |
1d4989c8 | 615 | struct dm_exception *e; |
1da177e4 | 616 | |
3510cb94 | 617 | e = alloc_completed_exception(); |
1da177e4 LT |
618 | if (!e) |
619 | return -ENOMEM; | |
620 | ||
621 | e->old_chunk = old; | |
d74f81f8 MB |
622 | |
623 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 624 | e->new_chunk = new; |
d74f81f8 | 625 | |
3510cb94 | 626 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 627 | |
1da177e4 LT |
628 | return 0; |
629 | } | |
630 | ||
7e201b35 MP |
631 | #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r))) |
632 | ||
633 | /* | |
634 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
635 | * Return zero if the origin has no snapshots. | |
636 | */ | |
637 | static sector_t __minimum_chunk_size(struct origin *o) | |
638 | { | |
639 | struct dm_snapshot *snap; | |
640 | unsigned chunk_size = 0; | |
641 | ||
642 | if (o) | |
643 | list_for_each_entry(snap, &o->snapshots, list) | |
644 | chunk_size = min_not_zero(chunk_size, | |
645 | snap->store->chunk_size); | |
646 | ||
647 | return chunk_size; | |
648 | } | |
649 | ||
1da177e4 LT |
650 | /* |
651 | * Hard coded magic. | |
652 | */ | |
653 | static int calc_max_buckets(void) | |
654 | { | |
655 | /* use a fixed size of 2MB */ | |
656 | unsigned long mem = 2 * 1024 * 1024; | |
657 | mem /= sizeof(struct list_head); | |
658 | ||
659 | return mem; | |
660 | } | |
661 | ||
1da177e4 LT |
662 | /* |
663 | * Allocate room for a suitable hash table. | |
664 | */ | |
fee1998e | 665 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
666 | { |
667 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
668 | ||
669 | /* | |
670 | * Calculate based on the size of the original volume or | |
671 | * the COW volume... | |
672 | */ | |
fc56f6fb | 673 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
674 | origin_dev_size = get_dev_size(s->origin->bdev); |
675 | max_buckets = calc_max_buckets(); | |
676 | ||
fee1998e | 677 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
678 | hash_size = min(hash_size, max_buckets); |
679 | ||
8e87b9b8 MP |
680 | if (hash_size < 64) |
681 | hash_size = 64; | |
8defd830 | 682 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
683 | if (dm_exception_table_init(&s->complete, hash_size, |
684 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
685 | return -ENOMEM; |
686 | ||
687 | /* | |
688 | * Allocate hash table for in-flight exceptions | |
689 | * Make this smaller than the real hash table | |
690 | */ | |
691 | hash_size >>= 3; | |
692 | if (hash_size < 64) | |
693 | hash_size = 64; | |
694 | ||
3510cb94 JB |
695 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
696 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
697 | return -ENOMEM; |
698 | } | |
699 | ||
700 | return 0; | |
701 | } | |
702 | ||
1da177e4 LT |
703 | /* |
704 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
705 | */ | |
706 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
707 | { | |
708 | struct dm_snapshot *s; | |
cd45daff | 709 | int i; |
1da177e4 | 710 | int r = -EINVAL; |
fc56f6fb | 711 | char *origin_path, *cow_path; |
fee1998e | 712 | unsigned args_used; |
1da177e4 | 713 | |
4c7e3bf4 | 714 | if (argc != 4) { |
72d94861 | 715 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 716 | r = -EINVAL; |
fc56f6fb | 717 | goto bad; |
1da177e4 LT |
718 | } |
719 | ||
720 | origin_path = argv[0]; | |
fee1998e JB |
721 | argv++; |
722 | argc--; | |
1da177e4 | 723 | |
fc56f6fb MS |
724 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
725 | if (!s) { | |
726 | ti->error = "Cannot allocate snapshot context private " | |
727 | "structure"; | |
728 | r = -ENOMEM; | |
729 | goto bad; | |
730 | } | |
731 | ||
732 | cow_path = argv[0]; | |
733 | argv++; | |
734 | argc--; | |
735 | ||
736 | r = dm_get_device(ti, cow_path, 0, 0, | |
737 | FMODE_READ | FMODE_WRITE, &s->cow); | |
738 | if (r) { | |
739 | ti->error = "Cannot get COW device"; | |
740 | goto bad_cow; | |
741 | } | |
742 | ||
743 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
744 | if (r) { |
745 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 746 | r = -EINVAL; |
fc56f6fb | 747 | goto bad_store; |
1da177e4 LT |
748 | } |
749 | ||
fee1998e JB |
750 | argv += args_used; |
751 | argc -= args_used; | |
752 | ||
1da177e4 LT |
753 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); |
754 | if (r) { | |
755 | ti->error = "Cannot get origin device"; | |
fee1998e | 756 | goto bad_origin; |
1da177e4 LT |
757 | } |
758 | ||
fc56f6fb | 759 | s->ti = ti; |
1da177e4 | 760 | s->valid = 1; |
aa14edeb | 761 | s->active = 0; |
c26655ca | 762 | s->suspended = 0; |
879129d2 | 763 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 764 | init_rwsem(&s->lock); |
c1f0c183 | 765 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 766 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
767 | |
768 | /* Allocate hash table for COW data */ | |
fee1998e | 769 | if (init_hash_tables(s)) { |
1da177e4 LT |
770 | ti->error = "Unable to allocate hash table space"; |
771 | r = -ENOMEM; | |
fee1998e | 772 | goto bad_hash_tables; |
1da177e4 LT |
773 | } |
774 | ||
eb69aca5 | 775 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
776 | if (r) { |
777 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 778 | goto bad_kcopyd; |
1da177e4 LT |
779 | } |
780 | ||
92e86812 MP |
781 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
782 | if (!s->pending_pool) { | |
783 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 784 | goto bad_pending_pool; |
92e86812 MP |
785 | } |
786 | ||
cd45daff MP |
787 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
788 | tracked_chunk_cache); | |
789 | if (!s->tracked_chunk_pool) { | |
790 | ti->error = "Could not allocate tracked_chunk mempool for " | |
791 | "tracking reads"; | |
92e86812 | 792 | goto bad_tracked_chunk_pool; |
cd45daff MP |
793 | } |
794 | ||
795 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
796 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
797 | ||
798 | spin_lock_init(&s->tracked_chunk_lock); | |
799 | ||
c1f0c183 MS |
800 | bio_list_init(&s->queued_bios); |
801 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); | |
802 | ||
803 | ti->private = s; | |
804 | ti->num_flush_requests = 1; | |
805 | ||
806 | /* Add snapshot to the list of snapshots for this origin */ | |
807 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
808 | r = register_snapshot(s); | |
809 | if (r == -ENOMEM) { | |
810 | ti->error = "Snapshot origin struct allocation failed"; | |
811 | goto bad_load_and_register; | |
812 | } else if (r < 0) { | |
813 | /* invalid handover, register_snapshot has set ti->error */ | |
814 | goto bad_load_and_register; | |
815 | } | |
816 | ||
817 | /* | |
818 | * Metadata must only be loaded into one table at once, so skip this | |
819 | * if metadata will be handed over during resume. | |
820 | * Chunk size will be set during the handover - set it to zero to | |
821 | * ensure it's ignored. | |
822 | */ | |
823 | if (r > 0) { | |
824 | s->store->chunk_size = 0; | |
825 | return 0; | |
826 | } | |
827 | ||
493df71c JB |
828 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
829 | (void *)s); | |
0764147b | 830 | if (r < 0) { |
f9cea4f7 | 831 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 832 | goto bad_read_metadata; |
0764147b MB |
833 | } else if (r > 0) { |
834 | s->valid = 0; | |
835 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 836 | } |
aa14edeb | 837 | |
3f2412dc MP |
838 | if (!s->store->chunk_size) { |
839 | ti->error = "Chunk size not set"; | |
c1f0c183 | 840 | goto bad_read_metadata; |
1da177e4 | 841 | } |
d0216849 | 842 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
843 | |
844 | return 0; | |
845 | ||
c1f0c183 MS |
846 | bad_read_metadata: |
847 | unregister_snapshot(s); | |
848 | ||
fee1998e | 849 | bad_load_and_register: |
cd45daff MP |
850 | mempool_destroy(s->tracked_chunk_pool); |
851 | ||
fee1998e | 852 | bad_tracked_chunk_pool: |
92e86812 MP |
853 | mempool_destroy(s->pending_pool); |
854 | ||
fee1998e | 855 | bad_pending_pool: |
eb69aca5 | 856 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 857 | |
fee1998e | 858 | bad_kcopyd: |
3510cb94 JB |
859 | dm_exception_table_exit(&s->pending, pending_cache); |
860 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 861 | |
fee1998e | 862 | bad_hash_tables: |
1da177e4 LT |
863 | dm_put_device(ti, s->origin); |
864 | ||
fee1998e | 865 | bad_origin: |
fc56f6fb | 866 | dm_exception_store_destroy(s->store); |
1da177e4 | 867 | |
fc56f6fb MS |
868 | bad_store: |
869 | dm_put_device(ti, s->cow); | |
fee1998e | 870 | |
fc56f6fb MS |
871 | bad_cow: |
872 | kfree(s); | |
873 | ||
874 | bad: | |
1da177e4 LT |
875 | return r; |
876 | } | |
877 | ||
31c93a0c MB |
878 | static void __free_exceptions(struct dm_snapshot *s) |
879 | { | |
eb69aca5 | 880 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
881 | s->kcopyd_client = NULL; |
882 | ||
3510cb94 JB |
883 | dm_exception_table_exit(&s->pending, pending_cache); |
884 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
885 | } |
886 | ||
c1f0c183 MS |
887 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
888 | struct dm_snapshot *snap_dest) | |
889 | { | |
890 | union { | |
891 | struct dm_exception_table table_swap; | |
892 | struct dm_exception_store *store_swap; | |
893 | } u; | |
894 | ||
895 | /* | |
896 | * Swap all snapshot context information between the two instances. | |
897 | */ | |
898 | u.table_swap = snap_dest->complete; | |
899 | snap_dest->complete = snap_src->complete; | |
900 | snap_src->complete = u.table_swap; | |
901 | ||
902 | u.store_swap = snap_dest->store; | |
903 | snap_dest->store = snap_src->store; | |
904 | snap_src->store = u.store_swap; | |
905 | ||
906 | snap_dest->store->snap = snap_dest; | |
907 | snap_src->store->snap = snap_src; | |
908 | ||
909 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
910 | snap_dest->valid = snap_src->valid; | |
911 | ||
912 | /* | |
913 | * Set source invalid to ensure it receives no further I/O. | |
914 | */ | |
915 | snap_src->valid = 0; | |
916 | } | |
917 | ||
1da177e4 LT |
918 | static void snapshot_dtr(struct dm_target *ti) |
919 | { | |
cd45daff MP |
920 | #ifdef CONFIG_DM_DEBUG |
921 | int i; | |
922 | #endif | |
028867ac | 923 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 924 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 925 | |
ca3a931f AK |
926 | flush_workqueue(ksnapd); |
927 | ||
c1f0c183 MS |
928 | down_read(&_origins_lock); |
929 | /* Check whether exception handover must be cancelled */ | |
930 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
931 | if (snap_src && snap_dest && (s == snap_src)) { | |
932 | down_write(&snap_dest->lock); | |
933 | snap_dest->valid = 0; | |
934 | up_write(&snap_dest->lock); | |
935 | DMERR("Cancelling snapshot handover."); | |
936 | } | |
937 | up_read(&_origins_lock); | |
938 | ||
138728dc AK |
939 | /* Prevent further origin writes from using this snapshot. */ |
940 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
941 | unregister_snapshot(s); |
942 | ||
879129d2 | 943 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 944 | msleep(1); |
879129d2 MP |
945 | /* |
946 | * Ensure instructions in mempool_destroy aren't reordered | |
947 | * before atomic_read. | |
948 | */ | |
949 | smp_mb(); | |
950 | ||
cd45daff MP |
951 | #ifdef CONFIG_DM_DEBUG |
952 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
953 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
954 | #endif | |
955 | ||
956 | mempool_destroy(s->tracked_chunk_pool); | |
957 | ||
31c93a0c | 958 | __free_exceptions(s); |
1da177e4 | 959 | |
92e86812 MP |
960 | mempool_destroy(s->pending_pool); |
961 | ||
1da177e4 | 962 | dm_put_device(ti, s->origin); |
fee1998e JB |
963 | |
964 | dm_exception_store_destroy(s->store); | |
138728dc | 965 | |
fc56f6fb MS |
966 | dm_put_device(ti, s->cow); |
967 | ||
1da177e4 LT |
968 | kfree(s); |
969 | } | |
970 | ||
971 | /* | |
972 | * Flush a list of buffers. | |
973 | */ | |
974 | static void flush_bios(struct bio *bio) | |
975 | { | |
976 | struct bio *n; | |
977 | ||
978 | while (bio) { | |
979 | n = bio->bi_next; | |
980 | bio->bi_next = NULL; | |
981 | generic_make_request(bio); | |
982 | bio = n; | |
983 | } | |
984 | } | |
985 | ||
c4028958 | 986 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 987 | { |
c4028958 DH |
988 | struct dm_snapshot *s = |
989 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
990 | struct bio *queued_bios; |
991 | unsigned long flags; | |
992 | ||
993 | spin_lock_irqsave(&s->pe_lock, flags); | |
994 | queued_bios = bio_list_get(&s->queued_bios); | |
995 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
996 | ||
997 | flush_bios(queued_bios); | |
998 | } | |
999 | ||
515ad66c MP |
1000 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1001 | ||
1002 | /* | |
1003 | * Flush a list of buffers. | |
1004 | */ | |
1005 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1006 | { | |
1007 | struct bio *n; | |
1008 | int r; | |
1009 | ||
1010 | while (bio) { | |
1011 | n = bio->bi_next; | |
1012 | bio->bi_next = NULL; | |
1013 | r = do_origin(s->origin, bio); | |
1014 | if (r == DM_MAPIO_REMAPPED) | |
1015 | generic_make_request(bio); | |
1016 | bio = n; | |
1017 | } | |
1018 | } | |
1019 | ||
1da177e4 LT |
1020 | /* |
1021 | * Error a list of buffers. | |
1022 | */ | |
1023 | static void error_bios(struct bio *bio) | |
1024 | { | |
1025 | struct bio *n; | |
1026 | ||
1027 | while (bio) { | |
1028 | n = bio->bi_next; | |
1029 | bio->bi_next = NULL; | |
6712ecf8 | 1030 | bio_io_error(bio); |
1da177e4 LT |
1031 | bio = n; |
1032 | } | |
1033 | } | |
1034 | ||
695368ac | 1035 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1036 | { |
1037 | if (!s->valid) | |
1038 | return; | |
1039 | ||
1040 | if (err == -EIO) | |
1041 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1042 | else if (err == -ENOMEM) | |
1043 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1044 | ||
493df71c JB |
1045 | if (s->store->type->drop_snapshot) |
1046 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1047 | |
1048 | s->valid = 0; | |
1049 | ||
fc56f6fb | 1050 | dm_table_event(s->ti->table); |
76df1c65 AK |
1051 | } |
1052 | ||
028867ac | 1053 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1054 | { |
1d4989c8 | 1055 | struct dm_exception *e; |
1da177e4 | 1056 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1057 | struct bio *origin_bios = NULL; |
1058 | struct bio *snapshot_bios = NULL; | |
1059 | int error = 0; | |
1da177e4 | 1060 | |
76df1c65 AK |
1061 | if (!success) { |
1062 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1063 | down_write(&s->lock); |
695368ac | 1064 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1065 | error = 1; |
76df1c65 AK |
1066 | goto out; |
1067 | } | |
1068 | ||
3510cb94 | 1069 | e = alloc_completed_exception(); |
76df1c65 | 1070 | if (!e) { |
1da177e4 | 1071 | down_write(&s->lock); |
695368ac | 1072 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1073 | error = 1; |
76df1c65 AK |
1074 | goto out; |
1075 | } | |
1076 | *e = pe->e; | |
1da177e4 | 1077 | |
76df1c65 AK |
1078 | down_write(&s->lock); |
1079 | if (!s->valid) { | |
3510cb94 | 1080 | free_completed_exception(e); |
9d493fa8 | 1081 | error = 1; |
76df1c65 | 1082 | goto out; |
1da177e4 LT |
1083 | } |
1084 | ||
615d1eb9 MS |
1085 | /* Check for conflicting reads */ |
1086 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1087 | |
9d493fa8 AK |
1088 | /* |
1089 | * Add a proper exception, and remove the | |
1090 | * in-flight exception from the list. | |
1091 | */ | |
3510cb94 | 1092 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1093 | |
1da177e4 | 1094 | out: |
3510cb94 | 1095 | dm_remove_exception(&pe->e); |
9d493fa8 | 1096 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c MP |
1097 | origin_bios = bio_list_get(&pe->origin_bios); |
1098 | free_pending_exception(pe); | |
1da177e4 | 1099 | |
9d493fa8 AK |
1100 | up_write(&s->lock); |
1101 | ||
1102 | /* Submit any pending write bios */ | |
1103 | if (error) | |
1104 | error_bios(snapshot_bios); | |
1105 | else | |
1106 | flush_bios(snapshot_bios); | |
1107 | ||
515ad66c | 1108 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1109 | } |
1110 | ||
1111 | static void commit_callback(void *context, int success) | |
1112 | { | |
028867ac AK |
1113 | struct dm_snap_pending_exception *pe = context; |
1114 | ||
1da177e4 LT |
1115 | pending_complete(pe, success); |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * Called when the copy I/O has finished. kcopyd actually runs | |
1120 | * this code so don't block. | |
1121 | */ | |
4cdc1d1f | 1122 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1123 | { |
028867ac | 1124 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1125 | struct dm_snapshot *s = pe->snap; |
1126 | ||
1127 | if (read_err || write_err) | |
1128 | pending_complete(pe, 0); | |
1129 | ||
1130 | else | |
1131 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1132 | s->store->type->commit_exception(s->store, &pe->e, |
1133 | commit_callback, pe); | |
1da177e4 LT |
1134 | } |
1135 | ||
1136 | /* | |
1137 | * Dispatches the copy operation to kcopyd. | |
1138 | */ | |
028867ac | 1139 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1140 | { |
1141 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1142 | struct dm_io_region src, dest; |
1da177e4 LT |
1143 | struct block_device *bdev = s->origin->bdev; |
1144 | sector_t dev_size; | |
1145 | ||
1146 | dev_size = get_dev_size(bdev); | |
1147 | ||
1148 | src.bdev = bdev; | |
71fab00a | 1149 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1150 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1151 | |
fc56f6fb | 1152 | dest.bdev = s->cow->bdev; |
71fab00a | 1153 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1154 | dest.count = src.count; |
1155 | ||
1156 | /* Hand over to kcopyd */ | |
eb69aca5 | 1157 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1158 | &src, 1, &dest, 0, copy_callback, pe); |
1159 | } | |
1160 | ||
2913808e MP |
1161 | static struct dm_snap_pending_exception * |
1162 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1163 | { | |
3510cb94 | 1164 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1165 | |
1166 | if (!e) | |
1167 | return NULL; | |
1168 | ||
1169 | return container_of(e, struct dm_snap_pending_exception, e); | |
1170 | } | |
1171 | ||
1da177e4 LT |
1172 | /* |
1173 | * Looks to see if this snapshot already has a pending exception | |
1174 | * for this chunk, otherwise it allocates a new one and inserts | |
1175 | * it into the pending table. | |
1176 | * | |
1177 | * NOTE: a write lock must be held on snap->lock before calling | |
1178 | * this. | |
1179 | */ | |
028867ac | 1180 | static struct dm_snap_pending_exception * |
c6621392 MP |
1181 | __find_pending_exception(struct dm_snapshot *s, |
1182 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1183 | { |
c6621392 | 1184 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1185 | |
2913808e MP |
1186 | pe2 = __lookup_pending_exception(s, chunk); |
1187 | if (pe2) { | |
76df1c65 | 1188 | free_pending_exception(pe); |
2913808e | 1189 | return pe2; |
1da177e4 LT |
1190 | } |
1191 | ||
76df1c65 AK |
1192 | pe->e.old_chunk = chunk; |
1193 | bio_list_init(&pe->origin_bios); | |
1194 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 AK |
1195 | pe->started = 0; |
1196 | ||
493df71c | 1197 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1198 | free_pending_exception(pe); |
1199 | return NULL; | |
1200 | } | |
1201 | ||
3510cb94 | 1202 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1203 | |
1da177e4 LT |
1204 | return pe; |
1205 | } | |
1206 | ||
1d4989c8 | 1207 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1208 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1209 | { |
fc56f6fb | 1210 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1211 | bio->bi_sector = chunk_to_sector(s->store, |
1212 | dm_chunk_number(e->new_chunk) + | |
1213 | (chunk - e->old_chunk)) + | |
1214 | (bio->bi_sector & | |
1215 | s->store->chunk_mask); | |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1219 | union map_info *map_context) | |
1220 | { | |
1d4989c8 | 1221 | struct dm_exception *e; |
028867ac | 1222 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1223 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1224 | chunk_t chunk; |
028867ac | 1225 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1226 | |
494b3ee7 | 1227 | if (unlikely(bio_empty_barrier(bio))) { |
fc56f6fb | 1228 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1229 | return DM_MAPIO_REMAPPED; |
1230 | } | |
1231 | ||
71fab00a | 1232 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1233 | |
1234 | /* Full snapshots are not usable */ | |
76df1c65 | 1235 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1236 | if (!s->valid) |
f6a80ea8 | 1237 | return -EIO; |
1da177e4 | 1238 | |
ba40a2aa AK |
1239 | /* FIXME: should only take write lock if we need |
1240 | * to copy an exception */ | |
1241 | down_write(&s->lock); | |
1242 | ||
1243 | if (!s->valid) { | |
1244 | r = -EIO; | |
1245 | goto out_unlock; | |
1246 | } | |
1247 | ||
1248 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1249 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1250 | if (e) { |
d74f81f8 | 1251 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1252 | goto out_unlock; |
1253 | } | |
1254 | ||
1da177e4 LT |
1255 | /* |
1256 | * Write to snapshot - higher level takes care of RW/RO | |
1257 | * flags so we should only get this if we are | |
1258 | * writeable. | |
1259 | */ | |
1260 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1261 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1262 | if (!pe) { |
c6621392 MP |
1263 | up_write(&s->lock); |
1264 | pe = alloc_pending_exception(s); | |
1265 | down_write(&s->lock); | |
1266 | ||
1267 | if (!s->valid) { | |
1268 | free_pending_exception(pe); | |
1269 | r = -EIO; | |
1270 | goto out_unlock; | |
1271 | } | |
1272 | ||
3510cb94 | 1273 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1274 | if (e) { |
1275 | free_pending_exception(pe); | |
1276 | remap_exception(s, e, bio, chunk); | |
1277 | goto out_unlock; | |
1278 | } | |
1279 | ||
c6621392 | 1280 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1281 | if (!pe) { |
1282 | __invalidate_snapshot(s, -ENOMEM); | |
1283 | r = -EIO; | |
1284 | goto out_unlock; | |
1285 | } | |
1da177e4 LT |
1286 | } |
1287 | ||
d74f81f8 | 1288 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1289 | bio_list_add(&pe->snapshot_bios, bio); |
1290 | ||
d2a7ad29 | 1291 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1292 | |
76df1c65 AK |
1293 | if (!pe->started) { |
1294 | /* this is protected by snap->lock */ | |
1295 | pe->started = 1; | |
ba40a2aa | 1296 | up_write(&s->lock); |
76df1c65 | 1297 | start_copy(pe); |
ba40a2aa AK |
1298 | goto out; |
1299 | } | |
cd45daff | 1300 | } else { |
ba40a2aa | 1301 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1302 | map_context->ptr = track_chunk(s, chunk); |
1303 | } | |
1da177e4 | 1304 | |
ba40a2aa AK |
1305 | out_unlock: |
1306 | up_write(&s->lock); | |
1307 | out: | |
1da177e4 LT |
1308 | return r; |
1309 | } | |
1310 | ||
cd45daff MP |
1311 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1312 | int error, union map_info *map_context) | |
1313 | { | |
1314 | struct dm_snapshot *s = ti->private; | |
1315 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1316 | ||
1317 | if (c) | |
1318 | stop_tracking_chunk(s, c); | |
1319 | ||
1320 | return 0; | |
1321 | } | |
1322 | ||
c26655ca MS |
1323 | static void snapshot_postsuspend(struct dm_target *ti) |
1324 | { | |
1325 | struct dm_snapshot *s = ti->private; | |
1326 | ||
1327 | down_write(&s->lock); | |
1328 | s->suspended = 1; | |
1329 | up_write(&s->lock); | |
1330 | } | |
1331 | ||
c1f0c183 MS |
1332 | static int snapshot_preresume(struct dm_target *ti) |
1333 | { | |
1334 | int r = 0; | |
1335 | struct dm_snapshot *s = ti->private; | |
1336 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1337 | ||
1338 | down_read(&_origins_lock); | |
1339 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1340 | if (snap_src && snap_dest) { | |
1341 | down_read(&snap_src->lock); | |
1342 | if (s == snap_src) { | |
1343 | DMERR("Unable to resume snapshot source until " | |
1344 | "handover completes."); | |
1345 | r = -EINVAL; | |
1346 | } else if (!snap_src->suspended) { | |
1347 | DMERR("Unable to perform snapshot handover until " | |
1348 | "source is suspended."); | |
1349 | r = -EINVAL; | |
1350 | } | |
1351 | up_read(&snap_src->lock); | |
1352 | } | |
1353 | up_read(&_origins_lock); | |
1354 | ||
1355 | return r; | |
1356 | } | |
1357 | ||
1da177e4 LT |
1358 | static void snapshot_resume(struct dm_target *ti) |
1359 | { | |
028867ac | 1360 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1361 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1362 | ||
1363 | down_read(&_origins_lock); | |
1364 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1365 | if (snap_src && snap_dest) { | |
1366 | down_write(&snap_src->lock); | |
1367 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1368 | __handover_exceptions(snap_src, snap_dest); | |
1369 | up_write(&snap_dest->lock); | |
1370 | up_write(&snap_src->lock); | |
1371 | } | |
1372 | up_read(&_origins_lock); | |
1373 | ||
1374 | /* Now we have correct chunk size, reregister */ | |
1375 | reregister_snapshot(s); | |
1da177e4 | 1376 | |
aa14edeb AK |
1377 | down_write(&s->lock); |
1378 | s->active = 1; | |
c26655ca | 1379 | s->suspended = 0; |
aa14edeb | 1380 | up_write(&s->lock); |
1da177e4 LT |
1381 | } |
1382 | ||
1383 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1384 | char *result, unsigned int maxlen) | |
1385 | { | |
2e4a31df | 1386 | unsigned sz = 0; |
028867ac | 1387 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1388 | |
1389 | switch (type) { | |
1390 | case STATUSTYPE_INFO: | |
94e76572 MP |
1391 | |
1392 | down_write(&snap->lock); | |
1393 | ||
1da177e4 | 1394 | if (!snap->valid) |
2e4a31df | 1395 | DMEMIT("Invalid"); |
1da177e4 | 1396 | else { |
985903bb MS |
1397 | if (snap->store->type->usage) { |
1398 | sector_t total_sectors, sectors_allocated, | |
1399 | metadata_sectors; | |
1400 | snap->store->type->usage(snap->store, | |
1401 | &total_sectors, | |
1402 | §ors_allocated, | |
1403 | &metadata_sectors); | |
1404 | DMEMIT("%llu/%llu %llu", | |
1405 | (unsigned long long)sectors_allocated, | |
1406 | (unsigned long long)total_sectors, | |
1407 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1408 | } |
1409 | else | |
2e4a31df | 1410 | DMEMIT("Unknown"); |
1da177e4 | 1411 | } |
94e76572 MP |
1412 | |
1413 | up_write(&snap->lock); | |
1414 | ||
1da177e4 LT |
1415 | break; |
1416 | ||
1417 | case STATUSTYPE_TABLE: | |
1418 | /* | |
1419 | * kdevname returns a static pointer so we need | |
1420 | * to make private copies if the output is to | |
1421 | * make sense. | |
1422 | */ | |
fc56f6fb | 1423 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1424 | snap->store->type->status(snap->store, type, result + sz, |
1425 | maxlen - sz); | |
1da177e4 LT |
1426 | break; |
1427 | } | |
1428 | ||
1429 | return 0; | |
1430 | } | |
1431 | ||
8811f46c MS |
1432 | static int snapshot_iterate_devices(struct dm_target *ti, |
1433 | iterate_devices_callout_fn fn, void *data) | |
1434 | { | |
1435 | struct dm_snapshot *snap = ti->private; | |
1436 | ||
1437 | return fn(ti, snap->origin, 0, ti->len, data); | |
1438 | } | |
1439 | ||
1440 | ||
1da177e4 LT |
1441 | /*----------------------------------------------------------------- |
1442 | * Origin methods | |
1443 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1444 | |
1445 | /* | |
1446 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1447 | * supplied bio was ignored. The caller may submit it immediately. | |
1448 | * (No remapping actually occurs as the origin is always a direct linear | |
1449 | * map.) | |
1450 | * | |
1451 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1452 | * and any supplied bio is added to a list to be submitted once all | |
1453 | * the necessary exceptions exist. | |
1454 | */ | |
1455 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1456 | struct bio *bio) | |
1da177e4 | 1457 | { |
515ad66c | 1458 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1459 | struct dm_snapshot *snap; |
1d4989c8 | 1460 | struct dm_exception *e; |
515ad66c MP |
1461 | struct dm_snap_pending_exception *pe; |
1462 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1463 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1464 | chunk_t chunk; |
1465 | ||
1466 | /* Do all the snapshots on this origin */ | |
1467 | list_for_each_entry (snap, snapshots, list) { | |
76df1c65 AK |
1468 | down_write(&snap->lock); |
1469 | ||
aa14edeb AK |
1470 | /* Only deal with valid and active snapshots */ |
1471 | if (!snap->valid || !snap->active) | |
76df1c65 | 1472 | goto next_snapshot; |
1da177e4 | 1473 | |
d5e404c1 | 1474 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1475 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1476 | goto next_snapshot; |
1da177e4 LT |
1477 | |
1478 | /* | |
1479 | * Remember, different snapshots can have | |
1480 | * different chunk sizes. | |
1481 | */ | |
9eaae8ff | 1482 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1483 | |
1484 | /* | |
1485 | * Check exception table to see if block | |
1486 | * is already remapped in this snapshot | |
1487 | * and trigger an exception if not. | |
1488 | */ | |
3510cb94 | 1489 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1490 | if (e) |
1491 | goto next_snapshot; | |
1492 | ||
2913808e | 1493 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1494 | if (!pe) { |
c6621392 MP |
1495 | up_write(&snap->lock); |
1496 | pe = alloc_pending_exception(snap); | |
1497 | down_write(&snap->lock); | |
1498 | ||
1499 | if (!snap->valid) { | |
1500 | free_pending_exception(pe); | |
1501 | goto next_snapshot; | |
1502 | } | |
1503 | ||
3510cb94 | 1504 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1505 | if (e) { |
1506 | free_pending_exception(pe); | |
1507 | goto next_snapshot; | |
1508 | } | |
1509 | ||
c6621392 | 1510 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1511 | if (!pe) { |
1512 | __invalidate_snapshot(snap, -ENOMEM); | |
1513 | goto next_snapshot; | |
1514 | } | |
76df1c65 AK |
1515 | } |
1516 | ||
515ad66c | 1517 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 1518 | |
515ad66c MP |
1519 | /* |
1520 | * If an origin bio was supplied, queue it to wait for the | |
1521 | * completion of this exception, and start this one last, | |
1522 | * at the end of the function. | |
1523 | */ | |
1524 | if (bio) { | |
1525 | bio_list_add(&pe->origin_bios, bio); | |
1526 | bio = NULL; | |
76df1c65 | 1527 | |
515ad66c MP |
1528 | if (!pe->started) { |
1529 | pe->started = 1; | |
1530 | pe_to_start_last = pe; | |
1531 | } | |
76df1c65 AK |
1532 | } |
1533 | ||
1534 | if (!pe->started) { | |
1535 | pe->started = 1; | |
515ad66c | 1536 | pe_to_start_now = pe; |
1da177e4 LT |
1537 | } |
1538 | ||
76df1c65 | 1539 | next_snapshot: |
1da177e4 | 1540 | up_write(&snap->lock); |
1da177e4 | 1541 | |
515ad66c MP |
1542 | if (pe_to_start_now) { |
1543 | start_copy(pe_to_start_now); | |
1544 | pe_to_start_now = NULL; | |
1545 | } | |
b4b610f6 AK |
1546 | } |
1547 | ||
1da177e4 | 1548 | /* |
515ad66c MP |
1549 | * Submit the exception against which the bio is queued last, |
1550 | * to give the other exceptions a head start. | |
1da177e4 | 1551 | */ |
515ad66c MP |
1552 | if (pe_to_start_last) |
1553 | start_copy(pe_to_start_last); | |
1da177e4 LT |
1554 | |
1555 | return r; | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * Called on a write from the origin driver. | |
1560 | */ | |
1561 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1562 | { | |
1563 | struct origin *o; | |
d2a7ad29 | 1564 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1565 | |
1566 | down_read(&_origins_lock); | |
1567 | o = __lookup_origin(origin->bdev); | |
1568 | if (o) | |
9eaae8ff | 1569 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
1570 | up_read(&_origins_lock); |
1571 | ||
1572 | return r; | |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1577 | */ | |
1578 | ||
1579 | /* | |
1580 | * Construct an origin mapping: <dev_path> | |
1581 | * The context for an origin is merely a 'struct dm_dev *' | |
1582 | * pointing to the real device. | |
1583 | */ | |
1584 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1585 | { | |
1586 | int r; | |
1587 | struct dm_dev *dev; | |
1588 | ||
1589 | if (argc != 1) { | |
72d94861 | 1590 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1591 | return -EINVAL; |
1592 | } | |
1593 | ||
1594 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1595 | dm_table_get_mode(ti->table), &dev); | |
1596 | if (r) { | |
1597 | ti->error = "Cannot get target device"; | |
1598 | return r; | |
1599 | } | |
1600 | ||
1601 | ti->private = dev; | |
494b3ee7 MP |
1602 | ti->num_flush_requests = 1; |
1603 | ||
1da177e4 LT |
1604 | return 0; |
1605 | } | |
1606 | ||
1607 | static void origin_dtr(struct dm_target *ti) | |
1608 | { | |
028867ac | 1609 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1610 | dm_put_device(ti, dev); |
1611 | } | |
1612 | ||
1613 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1614 | union map_info *map_context) | |
1615 | { | |
028867ac | 1616 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1617 | bio->bi_bdev = dev->bdev; |
1618 | ||
494b3ee7 MP |
1619 | if (unlikely(bio_empty_barrier(bio))) |
1620 | return DM_MAPIO_REMAPPED; | |
1621 | ||
1da177e4 | 1622 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 1623 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1624 | } |
1625 | ||
1da177e4 LT |
1626 | /* |
1627 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1628 | * chunk sizes. | |
1629 | */ | |
1630 | static void origin_resume(struct dm_target *ti) | |
1631 | { | |
028867ac | 1632 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1633 | |
1634 | down_read(&_origins_lock); | |
1da177e4 | 1635 | |
7e201b35 MP |
1636 | ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev)); |
1637 | ||
1638 | up_read(&_origins_lock); | |
1da177e4 LT |
1639 | } |
1640 | ||
1641 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1642 | unsigned int maxlen) | |
1643 | { | |
028867ac | 1644 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1645 | |
1646 | switch (type) { | |
1647 | case STATUSTYPE_INFO: | |
1648 | result[0] = '\0'; | |
1649 | break; | |
1650 | ||
1651 | case STATUSTYPE_TABLE: | |
1652 | snprintf(result, maxlen, "%s", dev->name); | |
1653 | break; | |
1654 | } | |
1655 | ||
1656 | return 0; | |
1657 | } | |
1658 | ||
8811f46c MS |
1659 | static int origin_iterate_devices(struct dm_target *ti, |
1660 | iterate_devices_callout_fn fn, void *data) | |
1661 | { | |
1662 | struct dm_dev *dev = ti->private; | |
1663 | ||
1664 | return fn(ti, dev, 0, ti->len, data); | |
1665 | } | |
1666 | ||
1da177e4 LT |
1667 | static struct target_type origin_target = { |
1668 | .name = "snapshot-origin", | |
8811f46c | 1669 | .version = {1, 7, 0}, |
1da177e4 LT |
1670 | .module = THIS_MODULE, |
1671 | .ctr = origin_ctr, | |
1672 | .dtr = origin_dtr, | |
1673 | .map = origin_map, | |
1674 | .resume = origin_resume, | |
1675 | .status = origin_status, | |
8811f46c | 1676 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
1677 | }; |
1678 | ||
1679 | static struct target_type snapshot_target = { | |
1680 | .name = "snapshot", | |
c26655ca | 1681 | .version = {1, 9, 0}, |
1da177e4 LT |
1682 | .module = THIS_MODULE, |
1683 | .ctr = snapshot_ctr, | |
1684 | .dtr = snapshot_dtr, | |
1685 | .map = snapshot_map, | |
cd45daff | 1686 | .end_io = snapshot_end_io, |
c26655ca | 1687 | .postsuspend = snapshot_postsuspend, |
c1f0c183 | 1688 | .preresume = snapshot_preresume, |
1da177e4 LT |
1689 | .resume = snapshot_resume, |
1690 | .status = snapshot_status, | |
8811f46c | 1691 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
1692 | }; |
1693 | ||
d698aa45 MP |
1694 | static struct target_type merge_target = { |
1695 | .name = dm_snapshot_merge_target_name, | |
1696 | .version = {1, 0, 0}, | |
1697 | .module = THIS_MODULE, | |
1698 | .ctr = snapshot_ctr, | |
1699 | .dtr = snapshot_dtr, | |
1700 | .map = snapshot_map, | |
1701 | .end_io = snapshot_end_io, | |
1702 | .postsuspend = snapshot_postsuspend, | |
1703 | .preresume = snapshot_preresume, | |
1704 | .resume = snapshot_resume, | |
1705 | .status = snapshot_status, | |
1706 | .iterate_devices = snapshot_iterate_devices, | |
1707 | }; | |
1708 | ||
1da177e4 LT |
1709 | static int __init dm_snapshot_init(void) |
1710 | { | |
1711 | int r; | |
1712 | ||
4db6bfe0 AK |
1713 | r = dm_exception_store_init(); |
1714 | if (r) { | |
1715 | DMERR("Failed to initialize exception stores"); | |
1716 | return r; | |
1717 | } | |
1718 | ||
1da177e4 | 1719 | r = dm_register_target(&snapshot_target); |
d698aa45 | 1720 | if (r < 0) { |
1da177e4 | 1721 | DMERR("snapshot target register failed %d", r); |
034a186d | 1722 | goto bad_register_snapshot_target; |
1da177e4 LT |
1723 | } |
1724 | ||
1725 | r = dm_register_target(&origin_target); | |
1726 | if (r < 0) { | |
72d94861 | 1727 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
1728 | goto bad_register_origin_target; |
1729 | } | |
1730 | ||
1731 | r = dm_register_target(&merge_target); | |
1732 | if (r < 0) { | |
1733 | DMERR("Merge target register failed %d", r); | |
1734 | goto bad_register_merge_target; | |
1da177e4 LT |
1735 | } |
1736 | ||
1737 | r = init_origin_hash(); | |
1738 | if (r) { | |
1739 | DMERR("init_origin_hash failed."); | |
d698aa45 | 1740 | goto bad_origin_hash; |
1da177e4 LT |
1741 | } |
1742 | ||
1d4989c8 | 1743 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
1744 | if (!exception_cache) { |
1745 | DMERR("Couldn't create exception cache."); | |
1746 | r = -ENOMEM; | |
d698aa45 | 1747 | goto bad_exception_cache; |
1da177e4 LT |
1748 | } |
1749 | ||
028867ac | 1750 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1751 | if (!pending_cache) { |
1752 | DMERR("Couldn't create pending cache."); | |
1753 | r = -ENOMEM; | |
d698aa45 | 1754 | goto bad_pending_cache; |
1da177e4 LT |
1755 | } |
1756 | ||
cd45daff MP |
1757 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1758 | if (!tracked_chunk_cache) { | |
1759 | DMERR("Couldn't create cache to track chunks in use."); | |
1760 | r = -ENOMEM; | |
d698aa45 | 1761 | goto bad_tracked_chunk_cache; |
cd45daff MP |
1762 | } |
1763 | ||
ca3a931f AK |
1764 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1765 | if (!ksnapd) { | |
1766 | DMERR("Failed to create ksnapd workqueue."); | |
1767 | r = -ENOMEM; | |
92e86812 | 1768 | goto bad_pending_pool; |
ca3a931f AK |
1769 | } |
1770 | ||
1da177e4 LT |
1771 | return 0; |
1772 | ||
4db6bfe0 | 1773 | bad_pending_pool: |
cd45daff | 1774 | kmem_cache_destroy(tracked_chunk_cache); |
d698aa45 | 1775 | bad_tracked_chunk_cache: |
1da177e4 | 1776 | kmem_cache_destroy(pending_cache); |
d698aa45 | 1777 | bad_pending_cache: |
1da177e4 | 1778 | kmem_cache_destroy(exception_cache); |
d698aa45 | 1779 | bad_exception_cache: |
1da177e4 | 1780 | exit_origin_hash(); |
d698aa45 MP |
1781 | bad_origin_hash: |
1782 | dm_unregister_target(&merge_target); | |
1783 | bad_register_merge_target: | |
1da177e4 | 1784 | dm_unregister_target(&origin_target); |
d698aa45 | 1785 | bad_register_origin_target: |
1da177e4 | 1786 | dm_unregister_target(&snapshot_target); |
034a186d JB |
1787 | bad_register_snapshot_target: |
1788 | dm_exception_store_exit(); | |
d698aa45 | 1789 | |
1da177e4 LT |
1790 | return r; |
1791 | } | |
1792 | ||
1793 | static void __exit dm_snapshot_exit(void) | |
1794 | { | |
ca3a931f AK |
1795 | destroy_workqueue(ksnapd); |
1796 | ||
10d3bd09 MP |
1797 | dm_unregister_target(&snapshot_target); |
1798 | dm_unregister_target(&origin_target); | |
d698aa45 | 1799 | dm_unregister_target(&merge_target); |
1da177e4 LT |
1800 | |
1801 | exit_origin_hash(); | |
1da177e4 LT |
1802 | kmem_cache_destroy(pending_cache); |
1803 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1804 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1805 | |
1806 | dm_exception_store_exit(); | |
1da177e4 LT |
1807 | } |
1808 | ||
1809 | /* Module hooks */ | |
1810 | module_init(dm_snapshot_init); | |
1811 | module_exit(dm_snapshot_exit); | |
1812 | ||
1813 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1814 | MODULE_AUTHOR("Joe Thornber"); | |
1815 | MODULE_LICENSE("GPL"); |