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 LT |
10 | #include <linux/ctype.h> |
11 | #include <linux/device-mapper.h> | |
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> | |
20 | ||
21 | #include "dm-snap.h" | |
22 | #include "dm-bio-list.h" | |
23 | #include "kcopyd.h" | |
24 | ||
72d94861 AK |
25 | #define DM_MSG_PREFIX "snapshots" |
26 | ||
1da177e4 LT |
27 | /* |
28 | * The percentage increment we will wake up users at | |
29 | */ | |
30 | #define WAKE_UP_PERCENT 5 | |
31 | ||
32 | /* | |
33 | * kcopyd priority of snapshot operations | |
34 | */ | |
35 | #define SNAPSHOT_COPY_PRIORITY 2 | |
36 | ||
37 | /* | |
38 | * Each snapshot reserves this many pages for io | |
39 | */ | |
40 | #define SNAPSHOT_PAGES 256 | |
41 | ||
ca3a931f AK |
42 | struct workqueue_struct *ksnapd; |
43 | static void flush_queued_bios(void *data); | |
44 | ||
1da177e4 LT |
45 | struct pending_exception { |
46 | struct exception e; | |
47 | ||
48 | /* | |
49 | * Origin buffers waiting for this to complete are held | |
50 | * in a bio list | |
51 | */ | |
52 | struct bio_list origin_bios; | |
53 | struct bio_list snapshot_bios; | |
54 | ||
eccf0817 AK |
55 | /* |
56 | * Short-term queue of pending exceptions prior to submission. | |
57 | */ | |
58 | struct list_head list; | |
59 | ||
1da177e4 | 60 | /* |
b4b610f6 | 61 | * The primary pending_exception is the one that holds |
4b832e8d | 62 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
63 | * group of pending_exceptions. It is always last to get freed. |
64 | * These fields get set up when writing to the origin. | |
1da177e4 | 65 | */ |
b4b610f6 AK |
66 | struct pending_exception *primary_pe; |
67 | ||
68 | /* | |
69 | * Number of pending_exceptions processing this chunk. | |
70 | * When this drops to zero we must complete the origin bios. | |
71 | * If incrementing or decrementing this, hold pe->snap->lock for | |
72 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
73 | * they are the same. | |
74 | */ | |
4b832e8d | 75 | atomic_t ref_count; |
1da177e4 LT |
76 | |
77 | /* Pointer back to snapshot context */ | |
78 | struct dm_snapshot *snap; | |
79 | ||
80 | /* | |
81 | * 1 indicates the exception has already been sent to | |
82 | * kcopyd. | |
83 | */ | |
84 | int started; | |
85 | }; | |
86 | ||
87 | /* | |
88 | * Hash table mapping origin volumes to lists of snapshots and | |
89 | * a lock to protect it | |
90 | */ | |
91 | static kmem_cache_t *exception_cache; | |
92 | static kmem_cache_t *pending_cache; | |
93 | static mempool_t *pending_pool; | |
94 | ||
95 | /* | |
96 | * One of these per registered origin, held in the snapshot_origins hash | |
97 | */ | |
98 | struct origin { | |
99 | /* The origin device */ | |
100 | struct block_device *bdev; | |
101 | ||
102 | struct list_head hash_list; | |
103 | ||
104 | /* List of snapshots for this origin */ | |
105 | struct list_head snapshots; | |
106 | }; | |
107 | ||
108 | /* | |
109 | * Size of the hash table for origin volumes. If we make this | |
110 | * the size of the minors list then it should be nearly perfect | |
111 | */ | |
112 | #define ORIGIN_HASH_SIZE 256 | |
113 | #define ORIGIN_MASK 0xFF | |
114 | static struct list_head *_origins; | |
115 | static struct rw_semaphore _origins_lock; | |
116 | ||
117 | static int init_origin_hash(void) | |
118 | { | |
119 | int i; | |
120 | ||
121 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
122 | GFP_KERNEL); | |
123 | if (!_origins) { | |
72d94861 | 124 | DMERR("unable to allocate memory"); |
1da177e4 LT |
125 | return -ENOMEM; |
126 | } | |
127 | ||
128 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
129 | INIT_LIST_HEAD(_origins + i); | |
130 | init_rwsem(&_origins_lock); | |
131 | ||
132 | return 0; | |
133 | } | |
134 | ||
135 | static void exit_origin_hash(void) | |
136 | { | |
137 | kfree(_origins); | |
138 | } | |
139 | ||
140 | static inline unsigned int origin_hash(struct block_device *bdev) | |
141 | { | |
142 | return bdev->bd_dev & ORIGIN_MASK; | |
143 | } | |
144 | ||
145 | static struct origin *__lookup_origin(struct block_device *origin) | |
146 | { | |
147 | struct list_head *ol; | |
148 | struct origin *o; | |
149 | ||
150 | ol = &_origins[origin_hash(origin)]; | |
151 | list_for_each_entry (o, ol, hash_list) | |
152 | if (bdev_equal(o->bdev, origin)) | |
153 | return o; | |
154 | ||
155 | return NULL; | |
156 | } | |
157 | ||
158 | static void __insert_origin(struct origin *o) | |
159 | { | |
160 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
161 | list_add_tail(&o->hash_list, sl); | |
162 | } | |
163 | ||
164 | /* | |
165 | * Make a note of the snapshot and its origin so we can look it | |
166 | * up when the origin has a write on it. | |
167 | */ | |
168 | static int register_snapshot(struct dm_snapshot *snap) | |
169 | { | |
170 | struct origin *o; | |
171 | struct block_device *bdev = snap->origin->bdev; | |
172 | ||
173 | down_write(&_origins_lock); | |
174 | o = __lookup_origin(bdev); | |
175 | ||
176 | if (!o) { | |
177 | /* New origin */ | |
178 | o = kmalloc(sizeof(*o), GFP_KERNEL); | |
179 | if (!o) { | |
180 | up_write(&_origins_lock); | |
181 | return -ENOMEM; | |
182 | } | |
183 | ||
184 | /* Initialise the struct */ | |
185 | INIT_LIST_HEAD(&o->snapshots); | |
186 | o->bdev = bdev; | |
187 | ||
188 | __insert_origin(o); | |
189 | } | |
190 | ||
191 | list_add_tail(&snap->list, &o->snapshots); | |
192 | ||
193 | up_write(&_origins_lock); | |
194 | return 0; | |
195 | } | |
196 | ||
197 | static void unregister_snapshot(struct dm_snapshot *s) | |
198 | { | |
199 | struct origin *o; | |
200 | ||
201 | down_write(&_origins_lock); | |
202 | o = __lookup_origin(s->origin->bdev); | |
203 | ||
204 | list_del(&s->list); | |
205 | if (list_empty(&o->snapshots)) { | |
206 | list_del(&o->hash_list); | |
207 | kfree(o); | |
208 | } | |
209 | ||
210 | up_write(&_origins_lock); | |
211 | } | |
212 | ||
213 | /* | |
214 | * Implementation of the exception hash tables. | |
215 | */ | |
216 | static int init_exception_table(struct exception_table *et, uint32_t size) | |
217 | { | |
218 | unsigned int i; | |
219 | ||
220 | et->hash_mask = size - 1; | |
221 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
222 | if (!et->table) | |
223 | return -ENOMEM; | |
224 | ||
225 | for (i = 0; i < size; i++) | |
226 | INIT_LIST_HEAD(et->table + i); | |
227 | ||
228 | return 0; | |
229 | } | |
230 | ||
231 | static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem) | |
232 | { | |
233 | struct list_head *slot; | |
234 | struct exception *ex, *next; | |
235 | int i, size; | |
236 | ||
237 | size = et->hash_mask + 1; | |
238 | for (i = 0; i < size; i++) { | |
239 | slot = et->table + i; | |
240 | ||
241 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
242 | kmem_cache_free(mem, ex); | |
243 | } | |
244 | ||
245 | vfree(et->table); | |
246 | } | |
247 | ||
248 | static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk) | |
249 | { | |
250 | return chunk & et->hash_mask; | |
251 | } | |
252 | ||
253 | static void insert_exception(struct exception_table *eh, struct exception *e) | |
254 | { | |
255 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
256 | list_add(&e->hash_list, l); | |
257 | } | |
258 | ||
259 | static inline void remove_exception(struct exception *e) | |
260 | { | |
261 | list_del(&e->hash_list); | |
262 | } | |
263 | ||
264 | /* | |
265 | * Return the exception data for a sector, or NULL if not | |
266 | * remapped. | |
267 | */ | |
268 | static struct exception *lookup_exception(struct exception_table *et, | |
269 | chunk_t chunk) | |
270 | { | |
271 | struct list_head *slot; | |
272 | struct exception *e; | |
273 | ||
274 | slot = &et->table[exception_hash(et, chunk)]; | |
275 | list_for_each_entry (e, slot, hash_list) | |
276 | if (e->old_chunk == chunk) | |
277 | return e; | |
278 | ||
279 | return NULL; | |
280 | } | |
281 | ||
282 | static inline struct exception *alloc_exception(void) | |
283 | { | |
284 | struct exception *e; | |
285 | ||
286 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
287 | if (!e) | |
288 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
289 | ||
290 | return e; | |
291 | } | |
292 | ||
293 | static inline void free_exception(struct exception *e) | |
294 | { | |
295 | kmem_cache_free(exception_cache, e); | |
296 | } | |
297 | ||
298 | static inline struct pending_exception *alloc_pending_exception(void) | |
299 | { | |
300 | return mempool_alloc(pending_pool, GFP_NOIO); | |
301 | } | |
302 | ||
303 | static inline void free_pending_exception(struct pending_exception *pe) | |
304 | { | |
305 | mempool_free(pe, pending_pool); | |
306 | } | |
307 | ||
308 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) | |
309 | { | |
310 | struct exception *e; | |
311 | ||
312 | e = alloc_exception(); | |
313 | if (!e) | |
314 | return -ENOMEM; | |
315 | ||
316 | e->old_chunk = old; | |
317 | e->new_chunk = new; | |
318 | insert_exception(&s->complete, e); | |
319 | return 0; | |
320 | } | |
321 | ||
322 | /* | |
323 | * Hard coded magic. | |
324 | */ | |
325 | static int calc_max_buckets(void) | |
326 | { | |
327 | /* use a fixed size of 2MB */ | |
328 | unsigned long mem = 2 * 1024 * 1024; | |
329 | mem /= sizeof(struct list_head); | |
330 | ||
331 | return mem; | |
332 | } | |
333 | ||
334 | /* | |
335 | * Rounds a number down to a power of 2. | |
336 | */ | |
337 | static inline uint32_t round_down(uint32_t n) | |
338 | { | |
339 | while (n & (n - 1)) | |
340 | n &= (n - 1); | |
341 | return n; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Allocate room for a suitable hash table. | |
346 | */ | |
347 | static int init_hash_tables(struct dm_snapshot *s) | |
348 | { | |
349 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
350 | ||
351 | /* | |
352 | * Calculate based on the size of the original volume or | |
353 | * the COW volume... | |
354 | */ | |
355 | cow_dev_size = get_dev_size(s->cow->bdev); | |
356 | origin_dev_size = get_dev_size(s->origin->bdev); | |
357 | max_buckets = calc_max_buckets(); | |
358 | ||
359 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
360 | hash_size = min(hash_size, max_buckets); | |
361 | ||
362 | /* Round it down to a power of 2 */ | |
363 | hash_size = round_down(hash_size); | |
364 | if (init_exception_table(&s->complete, hash_size)) | |
365 | return -ENOMEM; | |
366 | ||
367 | /* | |
368 | * Allocate hash table for in-flight exceptions | |
369 | * Make this smaller than the real hash table | |
370 | */ | |
371 | hash_size >>= 3; | |
372 | if (hash_size < 64) | |
373 | hash_size = 64; | |
374 | ||
375 | if (init_exception_table(&s->pending, hash_size)) { | |
376 | exit_exception_table(&s->complete, exception_cache); | |
377 | return -ENOMEM; | |
378 | } | |
379 | ||
380 | return 0; | |
381 | } | |
382 | ||
383 | /* | |
384 | * Round a number up to the nearest 'size' boundary. size must | |
385 | * be a power of 2. | |
386 | */ | |
387 | static inline ulong round_up(ulong n, ulong size) | |
388 | { | |
389 | size--; | |
390 | return (n + size) & ~size; | |
391 | } | |
392 | ||
4c7e3bf4 MM |
393 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
394 | char **error) | |
395 | { | |
396 | unsigned long chunk_size; | |
397 | char *value; | |
398 | ||
399 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
400 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
401 | *error = "Invalid chunk size"; | |
402 | return -EINVAL; | |
403 | } | |
404 | ||
405 | if (!chunk_size) { | |
406 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
407 | return 0; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Chunk size must be multiple of page size. Silently | |
412 | * round up if it's not. | |
413 | */ | |
414 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
415 | ||
416 | /* Check chunk_size is a power of 2 */ | |
417 | if (chunk_size & (chunk_size - 1)) { | |
418 | *error = "Chunk size is not a power of 2"; | |
419 | return -EINVAL; | |
420 | } | |
421 | ||
422 | /* Validate the chunk size against the device block size */ | |
423 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
424 | *error = "Chunk size is not a multiple of device blocksize"; | |
425 | return -EINVAL; | |
426 | } | |
427 | ||
428 | s->chunk_size = chunk_size; | |
429 | s->chunk_mask = chunk_size - 1; | |
430 | s->chunk_shift = ffs(chunk_size) - 1; | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
1da177e4 LT |
435 | /* |
436 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
437 | */ | |
438 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
439 | { | |
440 | struct dm_snapshot *s; | |
1da177e4 LT |
441 | int r = -EINVAL; |
442 | char persistent; | |
443 | char *origin_path; | |
444 | char *cow_path; | |
1da177e4 | 445 | |
4c7e3bf4 | 446 | if (argc != 4) { |
72d94861 | 447 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
448 | r = -EINVAL; |
449 | goto bad1; | |
450 | } | |
451 | ||
452 | origin_path = argv[0]; | |
453 | cow_path = argv[1]; | |
454 | persistent = toupper(*argv[2]); | |
455 | ||
456 | if (persistent != 'P' && persistent != 'N') { | |
457 | ti->error = "Persistent flag is not P or N"; | |
458 | r = -EINVAL; | |
459 | goto bad1; | |
460 | } | |
461 | ||
1da177e4 LT |
462 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
463 | if (s == NULL) { | |
464 | ti->error = "Cannot allocate snapshot context private " | |
465 | "structure"; | |
466 | r = -ENOMEM; | |
467 | goto bad1; | |
468 | } | |
469 | ||
470 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
471 | if (r) { | |
472 | ti->error = "Cannot get origin device"; | |
473 | goto bad2; | |
474 | } | |
475 | ||
476 | r = dm_get_device(ti, cow_path, 0, 0, | |
477 | FMODE_READ | FMODE_WRITE, &s->cow); | |
478 | if (r) { | |
479 | dm_put_device(ti, s->origin); | |
480 | ti->error = "Cannot get COW device"; | |
481 | goto bad2; | |
482 | } | |
483 | ||
4c7e3bf4 MM |
484 | r = set_chunk_size(s, argv[3], &ti->error); |
485 | if (r) | |
1da177e4 | 486 | goto bad3; |
1da177e4 | 487 | |
1da177e4 | 488 | s->type = persistent; |
1da177e4 LT |
489 | |
490 | s->valid = 1; | |
aa14edeb | 491 | s->active = 0; |
1da177e4 LT |
492 | s->last_percent = 0; |
493 | init_rwsem(&s->lock); | |
ca3a931f | 494 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
495 | s->table = ti->table; |
496 | ||
497 | /* Allocate hash table for COW data */ | |
498 | if (init_hash_tables(s)) { | |
499 | ti->error = "Unable to allocate hash table space"; | |
500 | r = -ENOMEM; | |
501 | goto bad3; | |
502 | } | |
503 | ||
1da177e4 LT |
504 | s->store.snap = s; |
505 | ||
506 | if (persistent == 'P') | |
4c7e3bf4 | 507 | r = dm_create_persistent(&s->store); |
1da177e4 | 508 | else |
4c7e3bf4 | 509 | r = dm_create_transient(&s->store); |
1da177e4 LT |
510 | |
511 | if (r) { | |
512 | ti->error = "Couldn't create exception store"; | |
513 | r = -EINVAL; | |
514 | goto bad4; | |
515 | } | |
516 | ||
517 | r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); | |
518 | if (r) { | |
519 | ti->error = "Could not create kcopyd client"; | |
520 | goto bad5; | |
521 | } | |
522 | ||
aa14edeb | 523 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 MM |
524 | r = s->store.read_metadata(&s->store); |
525 | if (r) { | |
526 | ti->error = "Failed to read snapshot metadata"; | |
527 | goto bad6; | |
528 | } | |
aa14edeb | 529 | |
ca3a931f AK |
530 | bio_list_init(&s->queued_bios); |
531 | INIT_WORK(&s->queued_bios_work, flush_queued_bios, s); | |
532 | ||
1da177e4 | 533 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 534 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
535 | if (register_snapshot(s)) { |
536 | r = -EINVAL; | |
537 | ti->error = "Cannot register snapshot origin"; | |
538 | goto bad6; | |
539 | } | |
540 | ||
541 | ti->private = s; | |
c51c2752 | 542 | ti->split_io = s->chunk_size; |
1da177e4 LT |
543 | |
544 | return 0; | |
545 | ||
546 | bad6: | |
547 | kcopyd_client_destroy(s->kcopyd_client); | |
548 | ||
549 | bad5: | |
550 | s->store.destroy(&s->store); | |
551 | ||
552 | bad4: | |
553 | exit_exception_table(&s->pending, pending_cache); | |
554 | exit_exception_table(&s->complete, exception_cache); | |
555 | ||
556 | bad3: | |
557 | dm_put_device(ti, s->cow); | |
558 | dm_put_device(ti, s->origin); | |
559 | ||
560 | bad2: | |
561 | kfree(s); | |
562 | ||
563 | bad1: | |
564 | return r; | |
565 | } | |
566 | ||
567 | static void snapshot_dtr(struct dm_target *ti) | |
568 | { | |
569 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
570 | ||
ca3a931f AK |
571 | flush_workqueue(ksnapd); |
572 | ||
138728dc AK |
573 | /* Prevent further origin writes from using this snapshot. */ |
574 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
575 | unregister_snapshot(s); |
576 | ||
138728dc AK |
577 | kcopyd_client_destroy(s->kcopyd_client); |
578 | ||
1da177e4 LT |
579 | exit_exception_table(&s->pending, pending_cache); |
580 | exit_exception_table(&s->complete, exception_cache); | |
581 | ||
582 | /* Deallocate memory used */ | |
583 | s->store.destroy(&s->store); | |
584 | ||
585 | dm_put_device(ti, s->origin); | |
586 | dm_put_device(ti, s->cow); | |
138728dc | 587 | |
1da177e4 LT |
588 | kfree(s); |
589 | } | |
590 | ||
591 | /* | |
592 | * Flush a list of buffers. | |
593 | */ | |
594 | static void flush_bios(struct bio *bio) | |
595 | { | |
596 | struct bio *n; | |
597 | ||
598 | while (bio) { | |
599 | n = bio->bi_next; | |
600 | bio->bi_next = NULL; | |
601 | generic_make_request(bio); | |
602 | bio = n; | |
603 | } | |
604 | } | |
605 | ||
ca3a931f AK |
606 | static void flush_queued_bios(void *data) |
607 | { | |
608 | struct dm_snapshot *s = (struct dm_snapshot *) data; | |
609 | struct bio *queued_bios; | |
610 | unsigned long flags; | |
611 | ||
612 | spin_lock_irqsave(&s->pe_lock, flags); | |
613 | queued_bios = bio_list_get(&s->queued_bios); | |
614 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
615 | ||
616 | flush_bios(queued_bios); | |
617 | } | |
618 | ||
1da177e4 LT |
619 | /* |
620 | * Error a list of buffers. | |
621 | */ | |
622 | static void error_bios(struct bio *bio) | |
623 | { | |
624 | struct bio *n; | |
625 | ||
626 | while (bio) { | |
627 | n = bio->bi_next; | |
628 | bio->bi_next = NULL; | |
629 | bio_io_error(bio, bio->bi_size); | |
630 | bio = n; | |
631 | } | |
632 | } | |
633 | ||
695368ac | 634 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
635 | { |
636 | if (!s->valid) | |
637 | return; | |
638 | ||
639 | if (err == -EIO) | |
640 | DMERR("Invalidating snapshot: Error reading/writing."); | |
641 | else if (err == -ENOMEM) | |
642 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
643 | ||
76df1c65 AK |
644 | if (s->store.drop_snapshot) |
645 | s->store.drop_snapshot(&s->store); | |
646 | ||
647 | s->valid = 0; | |
648 | ||
649 | dm_table_event(s->table); | |
650 | } | |
651 | ||
4b832e8d AK |
652 | static void get_pending_exception(struct pending_exception *pe) |
653 | { | |
654 | atomic_inc(&pe->ref_count); | |
655 | } | |
656 | ||
657 | static struct bio *put_pending_exception(struct pending_exception *pe) | |
658 | { | |
659 | struct pending_exception *primary_pe; | |
660 | struct bio *origin_bios = NULL; | |
661 | ||
662 | primary_pe = pe->primary_pe; | |
663 | ||
664 | /* | |
665 | * If this pe is involved in a write to the origin and | |
666 | * it is the last sibling to complete then release | |
667 | * the bios for the original write to the origin. | |
668 | */ | |
669 | if (primary_pe && | |
670 | atomic_dec_and_test(&primary_pe->ref_count)) | |
671 | origin_bios = bio_list_get(&primary_pe->origin_bios); | |
672 | ||
673 | /* | |
674 | * Free the pe if it's not linked to an origin write or if | |
675 | * it's not itself a primary pe. | |
676 | */ | |
677 | if (!primary_pe || primary_pe != pe) | |
678 | free_pending_exception(pe); | |
679 | ||
680 | /* | |
681 | * Free the primary pe if nothing references it. | |
682 | */ | |
683 | if (primary_pe && !atomic_read(&primary_pe->ref_count)) | |
684 | free_pending_exception(primary_pe); | |
685 | ||
686 | return origin_bios; | |
687 | } | |
688 | ||
1da177e4 LT |
689 | static void pending_complete(struct pending_exception *pe, int success) |
690 | { | |
691 | struct exception *e; | |
692 | struct dm_snapshot *s = pe->snap; | |
9d493fa8 AK |
693 | struct bio *origin_bios = NULL; |
694 | struct bio *snapshot_bios = NULL; | |
695 | int error = 0; | |
1da177e4 | 696 | |
76df1c65 AK |
697 | if (!success) { |
698 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 699 | down_write(&s->lock); |
695368ac | 700 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 701 | error = 1; |
76df1c65 AK |
702 | goto out; |
703 | } | |
704 | ||
705 | e = alloc_exception(); | |
706 | if (!e) { | |
1da177e4 | 707 | down_write(&s->lock); |
695368ac | 708 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 709 | error = 1; |
76df1c65 AK |
710 | goto out; |
711 | } | |
712 | *e = pe->e; | |
1da177e4 | 713 | |
76df1c65 AK |
714 | down_write(&s->lock); |
715 | if (!s->valid) { | |
76df1c65 | 716 | free_exception(e); |
9d493fa8 | 717 | error = 1; |
76df1c65 | 718 | goto out; |
1da177e4 LT |
719 | } |
720 | ||
9d493fa8 AK |
721 | /* |
722 | * Add a proper exception, and remove the | |
723 | * in-flight exception from the list. | |
724 | */ | |
76df1c65 | 725 | insert_exception(&s->complete, e); |
76df1c65 | 726 | |
1da177e4 | 727 | out: |
695368ac | 728 | remove_exception(&pe->e); |
9d493fa8 | 729 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 730 | origin_bios = put_pending_exception(pe); |
1da177e4 | 731 | |
9d493fa8 AK |
732 | up_write(&s->lock); |
733 | ||
734 | /* Submit any pending write bios */ | |
735 | if (error) | |
736 | error_bios(snapshot_bios); | |
737 | else | |
738 | flush_bios(snapshot_bios); | |
739 | ||
740 | flush_bios(origin_bios); | |
1da177e4 LT |
741 | } |
742 | ||
743 | static void commit_callback(void *context, int success) | |
744 | { | |
745 | struct pending_exception *pe = (struct pending_exception *) context; | |
746 | pending_complete(pe, success); | |
747 | } | |
748 | ||
749 | /* | |
750 | * Called when the copy I/O has finished. kcopyd actually runs | |
751 | * this code so don't block. | |
752 | */ | |
753 | static void copy_callback(int read_err, unsigned int write_err, void *context) | |
754 | { | |
755 | struct pending_exception *pe = (struct pending_exception *) context; | |
756 | struct dm_snapshot *s = pe->snap; | |
757 | ||
758 | if (read_err || write_err) | |
759 | pending_complete(pe, 0); | |
760 | ||
761 | else | |
762 | /* Update the metadata if we are persistent */ | |
763 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
764 | pe); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Dispatches the copy operation to kcopyd. | |
769 | */ | |
858119e1 | 770 | static void start_copy(struct pending_exception *pe) |
1da177e4 LT |
771 | { |
772 | struct dm_snapshot *s = pe->snap; | |
773 | struct io_region src, dest; | |
774 | struct block_device *bdev = s->origin->bdev; | |
775 | sector_t dev_size; | |
776 | ||
777 | dev_size = get_dev_size(bdev); | |
778 | ||
779 | src.bdev = bdev; | |
780 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
781 | src.count = min(s->chunk_size, dev_size - src.sector); | |
782 | ||
783 | dest.bdev = s->cow->bdev; | |
784 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
785 | dest.count = src.count; | |
786 | ||
787 | /* Hand over to kcopyd */ | |
788 | kcopyd_copy(s->kcopyd_client, | |
789 | &src, 1, &dest, 0, copy_callback, pe); | |
790 | } | |
791 | ||
792 | /* | |
793 | * Looks to see if this snapshot already has a pending exception | |
794 | * for this chunk, otherwise it allocates a new one and inserts | |
795 | * it into the pending table. | |
796 | * | |
797 | * NOTE: a write lock must be held on snap->lock before calling | |
798 | * this. | |
799 | */ | |
800 | static struct pending_exception * | |
801 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) | |
802 | { | |
803 | struct exception *e; | |
804 | struct pending_exception *pe; | |
805 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); | |
806 | ||
807 | /* | |
808 | * Is there a pending exception for this already ? | |
809 | */ | |
810 | e = lookup_exception(&s->pending, chunk); | |
811 | if (e) { | |
812 | /* cast the exception to a pending exception */ | |
813 | pe = container_of(e, struct pending_exception, e); | |
76df1c65 AK |
814 | goto out; |
815 | } | |
1da177e4 | 816 | |
76df1c65 AK |
817 | /* |
818 | * Create a new pending exception, we don't want | |
819 | * to hold the lock while we do this. | |
820 | */ | |
821 | up_write(&s->lock); | |
822 | pe = alloc_pending_exception(); | |
823 | down_write(&s->lock); | |
1da177e4 | 824 | |
76df1c65 AK |
825 | if (!s->valid) { |
826 | free_pending_exception(pe); | |
827 | return NULL; | |
828 | } | |
1da177e4 | 829 | |
76df1c65 AK |
830 | e = lookup_exception(&s->pending, chunk); |
831 | if (e) { | |
832 | free_pending_exception(pe); | |
833 | pe = container_of(e, struct pending_exception, e); | |
834 | goto out; | |
1da177e4 LT |
835 | } |
836 | ||
76df1c65 AK |
837 | pe->e.old_chunk = chunk; |
838 | bio_list_init(&pe->origin_bios); | |
839 | bio_list_init(&pe->snapshot_bios); | |
840 | pe->primary_pe = NULL; | |
4b832e8d | 841 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
842 | pe->snap = s; |
843 | pe->started = 0; | |
844 | ||
845 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
846 | free_pending_exception(pe); | |
847 | return NULL; | |
848 | } | |
849 | ||
4b832e8d | 850 | get_pending_exception(pe); |
76df1c65 AK |
851 | insert_exception(&s->pending, &pe->e); |
852 | ||
853 | out: | |
1da177e4 LT |
854 | return pe; |
855 | } | |
856 | ||
857 | static inline void remap_exception(struct dm_snapshot *s, struct exception *e, | |
858 | struct bio *bio) | |
859 | { | |
860 | bio->bi_bdev = s->cow->bdev; | |
861 | bio->bi_sector = chunk_to_sector(s, e->new_chunk) + | |
862 | (bio->bi_sector & s->chunk_mask); | |
863 | } | |
864 | ||
865 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
866 | union map_info *map_context) | |
867 | { | |
868 | struct exception *e; | |
869 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
870 | int r = 1; | |
871 | chunk_t chunk; | |
76df1c65 | 872 | struct pending_exception *pe = NULL; |
1da177e4 LT |
873 | |
874 | chunk = sector_to_chunk(s, bio->bi_sector); | |
875 | ||
876 | /* Full snapshots are not usable */ | |
76df1c65 | 877 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 878 | if (!s->valid) |
f6a80ea8 | 879 | return -EIO; |
1da177e4 | 880 | |
4aac0a63 AK |
881 | if (unlikely(bio_barrier(bio))) |
882 | return -EOPNOTSUPP; | |
883 | ||
ba40a2aa AK |
884 | /* FIXME: should only take write lock if we need |
885 | * to copy an exception */ | |
886 | down_write(&s->lock); | |
887 | ||
888 | if (!s->valid) { | |
889 | r = -EIO; | |
890 | goto out_unlock; | |
891 | } | |
892 | ||
893 | /* If the block is already remapped - use that, else remap it */ | |
894 | e = lookup_exception(&s->complete, chunk); | |
895 | if (e) { | |
896 | remap_exception(s, e, bio); | |
897 | goto out_unlock; | |
898 | } | |
899 | ||
1da177e4 LT |
900 | /* |
901 | * Write to snapshot - higher level takes care of RW/RO | |
902 | * flags so we should only get this if we are | |
903 | * writeable. | |
904 | */ | |
905 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
906 | pe = __find_pending_exception(s, bio); |
907 | if (!pe) { | |
695368ac | 908 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
909 | r = -EIO; |
910 | goto out_unlock; | |
1da177e4 LT |
911 | } |
912 | ||
76df1c65 AK |
913 | remap_exception(s, &pe->e, bio); |
914 | bio_list_add(&pe->snapshot_bios, bio); | |
915 | ||
ba40a2aa AK |
916 | r = 0; |
917 | ||
76df1c65 AK |
918 | if (!pe->started) { |
919 | /* this is protected by snap->lock */ | |
920 | pe->started = 1; | |
ba40a2aa | 921 | up_write(&s->lock); |
76df1c65 | 922 | start_copy(pe); |
ba40a2aa AK |
923 | goto out; |
924 | } | |
925 | } else | |
1da177e4 LT |
926 | /* |
927 | * FIXME: this read path scares me because we | |
928 | * always use the origin when we have a pending | |
929 | * exception. However I can't think of a | |
930 | * situation where this is wrong - ejt. | |
931 | */ | |
ba40a2aa | 932 | bio->bi_bdev = s->origin->bdev; |
1da177e4 | 933 | |
ba40a2aa AK |
934 | out_unlock: |
935 | up_write(&s->lock); | |
936 | out: | |
1da177e4 LT |
937 | return r; |
938 | } | |
939 | ||
940 | static void snapshot_resume(struct dm_target *ti) | |
941 | { | |
942 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | |
943 | ||
aa14edeb AK |
944 | down_write(&s->lock); |
945 | s->active = 1; | |
946 | up_write(&s->lock); | |
1da177e4 LT |
947 | } |
948 | ||
949 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
950 | char *result, unsigned int maxlen) | |
951 | { | |
952 | struct dm_snapshot *snap = (struct dm_snapshot *) ti->private; | |
953 | ||
954 | switch (type) { | |
955 | case STATUSTYPE_INFO: | |
956 | if (!snap->valid) | |
957 | snprintf(result, maxlen, "Invalid"); | |
958 | else { | |
959 | if (snap->store.fraction_full) { | |
960 | sector_t numerator, denominator; | |
961 | snap->store.fraction_full(&snap->store, | |
962 | &numerator, | |
963 | &denominator); | |
4ee218cd AM |
964 | snprintf(result, maxlen, "%llu/%llu", |
965 | (unsigned long long)numerator, | |
966 | (unsigned long long)denominator); | |
1da177e4 LT |
967 | } |
968 | else | |
969 | snprintf(result, maxlen, "Unknown"); | |
970 | } | |
971 | break; | |
972 | ||
973 | case STATUSTYPE_TABLE: | |
974 | /* | |
975 | * kdevname returns a static pointer so we need | |
976 | * to make private copies if the output is to | |
977 | * make sense. | |
978 | */ | |
4ee218cd | 979 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 980 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
981 | snap->type, |
982 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
983 | break; |
984 | } | |
985 | ||
986 | return 0; | |
987 | } | |
988 | ||
989 | /*----------------------------------------------------------------- | |
990 | * Origin methods | |
991 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
992 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
993 | { | |
b4b610f6 | 994 | int r = 1, first = 0; |
1da177e4 LT |
995 | struct dm_snapshot *snap; |
996 | struct exception *e; | |
b4b610f6 | 997 | struct pending_exception *pe, *next_pe, *primary_pe = NULL; |
1da177e4 | 998 | chunk_t chunk; |
eccf0817 | 999 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1000 | |
1001 | /* Do all the snapshots on this origin */ | |
1002 | list_for_each_entry (snap, snapshots, list) { | |
1003 | ||
76df1c65 AK |
1004 | down_write(&snap->lock); |
1005 | ||
aa14edeb AK |
1006 | /* Only deal with valid and active snapshots */ |
1007 | if (!snap->valid || !snap->active) | |
76df1c65 | 1008 | goto next_snapshot; |
1da177e4 | 1009 | |
d5e404c1 AK |
1010 | /* Nothing to do if writing beyond end of snapshot */ |
1011 | if (bio->bi_sector >= dm_table_get_size(snap->table)) | |
76df1c65 | 1012 | goto next_snapshot; |
1da177e4 LT |
1013 | |
1014 | /* | |
1015 | * Remember, different snapshots can have | |
1016 | * different chunk sizes. | |
1017 | */ | |
1018 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1019 | ||
1020 | /* | |
1021 | * Check exception table to see if block | |
1022 | * is already remapped in this snapshot | |
1023 | * and trigger an exception if not. | |
b4b610f6 | 1024 | * |
4b832e8d | 1025 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1026 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1027 | */ |
1028 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1029 | if (e) |
1030 | goto next_snapshot; | |
1031 | ||
1032 | pe = __find_pending_exception(snap, bio); | |
1033 | if (!pe) { | |
695368ac | 1034 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1035 | goto next_snapshot; |
1036 | } | |
1037 | ||
1038 | if (!primary_pe) { | |
1039 | /* | |
1040 | * Either every pe here has same | |
1041 | * primary_pe or none has one yet. | |
1042 | */ | |
1043 | if (pe->primary_pe) | |
1044 | primary_pe = pe->primary_pe; | |
1045 | else { | |
1046 | primary_pe = pe; | |
1047 | first = 1; | |
1da177e4 | 1048 | } |
76df1c65 AK |
1049 | |
1050 | bio_list_add(&primary_pe->origin_bios, bio); | |
1051 | ||
1052 | r = 0; | |
1053 | } | |
1054 | ||
1055 | if (!pe->primary_pe) { | |
76df1c65 | 1056 | pe->primary_pe = primary_pe; |
4b832e8d | 1057 | get_pending_exception(primary_pe); |
76df1c65 AK |
1058 | } |
1059 | ||
1060 | if (!pe->started) { | |
1061 | pe->started = 1; | |
1062 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1063 | } |
1064 | ||
76df1c65 | 1065 | next_snapshot: |
1da177e4 LT |
1066 | up_write(&snap->lock); |
1067 | } | |
1068 | ||
b4b610f6 | 1069 | if (!primary_pe) |
4b832e8d | 1070 | return r; |
b4b610f6 AK |
1071 | |
1072 | /* | |
1073 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1074 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1075 | * got completed while we were in the loop above, so it falls to |
1076 | * us here to remove the primary_pe and submit any origin_bios. | |
1077 | */ | |
1078 | ||
4b832e8d | 1079 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1080 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1081 | free_pending_exception(primary_pe); | |
1082 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1083 | return r; |
b4b610f6 AK |
1084 | } |
1085 | ||
1da177e4 LT |
1086 | /* |
1087 | * Now that we have a complete pe list we can start the copying. | |
1088 | */ | |
eccf0817 AK |
1089 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1090 | start_copy(pe); | |
1da177e4 LT |
1091 | |
1092 | return r; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Called on a write from the origin driver. | |
1097 | */ | |
1098 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1099 | { | |
1100 | struct origin *o; | |
1101 | int r = 1; | |
1102 | ||
1103 | down_read(&_origins_lock); | |
1104 | o = __lookup_origin(origin->bdev); | |
1105 | if (o) | |
1106 | r = __origin_write(&o->snapshots, bio); | |
1107 | up_read(&_origins_lock); | |
1108 | ||
1109 | return r; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1114 | */ | |
1115 | ||
1116 | /* | |
1117 | * Construct an origin mapping: <dev_path> | |
1118 | * The context for an origin is merely a 'struct dm_dev *' | |
1119 | * pointing to the real device. | |
1120 | */ | |
1121 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1122 | { | |
1123 | int r; | |
1124 | struct dm_dev *dev; | |
1125 | ||
1126 | if (argc != 1) { | |
72d94861 | 1127 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1128 | return -EINVAL; |
1129 | } | |
1130 | ||
1131 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1132 | dm_table_get_mode(ti->table), &dev); | |
1133 | if (r) { | |
1134 | ti->error = "Cannot get target device"; | |
1135 | return r; | |
1136 | } | |
1137 | ||
1138 | ti->private = dev; | |
1139 | return 0; | |
1140 | } | |
1141 | ||
1142 | static void origin_dtr(struct dm_target *ti) | |
1143 | { | |
1144 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1145 | dm_put_device(ti, dev); | |
1146 | } | |
1147 | ||
1148 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1149 | union map_info *map_context) | |
1150 | { | |
1151 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1152 | bio->bi_bdev = dev->bdev; | |
1153 | ||
4aac0a63 AK |
1154 | if (unlikely(bio_barrier(bio))) |
1155 | return -EOPNOTSUPP; | |
1156 | ||
1da177e4 LT |
1157 | /* Only tell snapshots if this is a write */ |
1158 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1; | |
1159 | } | |
1160 | ||
1161 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1162 | ||
1163 | /* | |
1164 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1165 | * chunk sizes. | |
1166 | */ | |
1167 | static void origin_resume(struct dm_target *ti) | |
1168 | { | |
1169 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1170 | struct dm_snapshot *snap; | |
1171 | struct origin *o; | |
1172 | chunk_t chunk_size = 0; | |
1173 | ||
1174 | down_read(&_origins_lock); | |
1175 | o = __lookup_origin(dev->bdev); | |
1176 | if (o) | |
1177 | list_for_each_entry (snap, &o->snapshots, list) | |
1178 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1179 | up_read(&_origins_lock); | |
1180 | ||
1181 | ti->split_io = chunk_size; | |
1182 | } | |
1183 | ||
1184 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1185 | unsigned int maxlen) | |
1186 | { | |
1187 | struct dm_dev *dev = (struct dm_dev *) ti->private; | |
1188 | ||
1189 | switch (type) { | |
1190 | case STATUSTYPE_INFO: | |
1191 | result[0] = '\0'; | |
1192 | break; | |
1193 | ||
1194 | case STATUSTYPE_TABLE: | |
1195 | snprintf(result, maxlen, "%s", dev->name); | |
1196 | break; | |
1197 | } | |
1198 | ||
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | static struct target_type origin_target = { | |
1203 | .name = "snapshot-origin", | |
4c7e3bf4 | 1204 | .version = {1, 5, 0}, |
1da177e4 LT |
1205 | .module = THIS_MODULE, |
1206 | .ctr = origin_ctr, | |
1207 | .dtr = origin_dtr, | |
1208 | .map = origin_map, | |
1209 | .resume = origin_resume, | |
1210 | .status = origin_status, | |
1211 | }; | |
1212 | ||
1213 | static struct target_type snapshot_target = { | |
1214 | .name = "snapshot", | |
4c7e3bf4 | 1215 | .version = {1, 5, 0}, |
1da177e4 LT |
1216 | .module = THIS_MODULE, |
1217 | .ctr = snapshot_ctr, | |
1218 | .dtr = snapshot_dtr, | |
1219 | .map = snapshot_map, | |
1220 | .resume = snapshot_resume, | |
1221 | .status = snapshot_status, | |
1222 | }; | |
1223 | ||
1224 | static int __init dm_snapshot_init(void) | |
1225 | { | |
1226 | int r; | |
1227 | ||
1228 | r = dm_register_target(&snapshot_target); | |
1229 | if (r) { | |
1230 | DMERR("snapshot target register failed %d", r); | |
1231 | return r; | |
1232 | } | |
1233 | ||
1234 | r = dm_register_target(&origin_target); | |
1235 | if (r < 0) { | |
72d94861 | 1236 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1237 | goto bad1; |
1238 | } | |
1239 | ||
1240 | r = init_origin_hash(); | |
1241 | if (r) { | |
1242 | DMERR("init_origin_hash failed."); | |
1243 | goto bad2; | |
1244 | } | |
1245 | ||
1246 | exception_cache = kmem_cache_create("dm-snapshot-ex", | |
1247 | sizeof(struct exception), | |
1248 | __alignof__(struct exception), | |
1249 | 0, NULL, NULL); | |
1250 | if (!exception_cache) { | |
1251 | DMERR("Couldn't create exception cache."); | |
1252 | r = -ENOMEM; | |
1253 | goto bad3; | |
1254 | } | |
1255 | ||
1256 | pending_cache = | |
1257 | kmem_cache_create("dm-snapshot-in", | |
1258 | sizeof(struct pending_exception), | |
1259 | __alignof__(struct pending_exception), | |
1260 | 0, NULL, NULL); | |
1261 | if (!pending_cache) { | |
1262 | DMERR("Couldn't create pending cache."); | |
1263 | r = -ENOMEM; | |
1264 | goto bad4; | |
1265 | } | |
1266 | ||
93d2341c | 1267 | pending_pool = mempool_create_slab_pool(128, pending_cache); |
1da177e4 LT |
1268 | if (!pending_pool) { |
1269 | DMERR("Couldn't create pending pool."); | |
1270 | r = -ENOMEM; | |
1271 | goto bad5; | |
1272 | } | |
1273 | ||
ca3a931f AK |
1274 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1275 | if (!ksnapd) { | |
1276 | DMERR("Failed to create ksnapd workqueue."); | |
1277 | r = -ENOMEM; | |
1278 | goto bad6; | |
1279 | } | |
1280 | ||
1da177e4 LT |
1281 | return 0; |
1282 | ||
ca3a931f AK |
1283 | bad6: |
1284 | mempool_destroy(pending_pool); | |
1da177e4 LT |
1285 | bad5: |
1286 | kmem_cache_destroy(pending_cache); | |
1287 | bad4: | |
1288 | kmem_cache_destroy(exception_cache); | |
1289 | bad3: | |
1290 | exit_origin_hash(); | |
1291 | bad2: | |
1292 | dm_unregister_target(&origin_target); | |
1293 | bad1: | |
1294 | dm_unregister_target(&snapshot_target); | |
1295 | return r; | |
1296 | } | |
1297 | ||
1298 | static void __exit dm_snapshot_exit(void) | |
1299 | { | |
1300 | int r; | |
1301 | ||
ca3a931f AK |
1302 | destroy_workqueue(ksnapd); |
1303 | ||
1da177e4 LT |
1304 | r = dm_unregister_target(&snapshot_target); |
1305 | if (r) | |
1306 | DMERR("snapshot unregister failed %d", r); | |
1307 | ||
1308 | r = dm_unregister_target(&origin_target); | |
1309 | if (r) | |
1310 | DMERR("origin unregister failed %d", r); | |
1311 | ||
1312 | exit_origin_hash(); | |
1313 | mempool_destroy(pending_pool); | |
1314 | kmem_cache_destroy(pending_cache); | |
1315 | kmem_cache_destroy(exception_cache); | |
1316 | } | |
1317 | ||
1318 | /* Module hooks */ | |
1319 | module_init(dm_snapshot_init); | |
1320 | module_exit(dm_snapshot_exit); | |
1321 | ||
1322 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1323 | MODULE_AUTHOR("Joe Thornber"); | |
1324 | MODULE_LICENSE("GPL"); |