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> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
1da177e4 LT |
21 | |
22 | #include "dm-snap.h" | |
23 | #include "dm-bio-list.h" | |
24 | #include "kcopyd.h" | |
25 | ||
72d94861 AK |
26 | #define DM_MSG_PREFIX "snapshots" |
27 | ||
1da177e4 LT |
28 | /* |
29 | * The percentage increment we will wake up users at | |
30 | */ | |
31 | #define WAKE_UP_PERCENT 5 | |
32 | ||
33 | /* | |
34 | * kcopyd priority of snapshot operations | |
35 | */ | |
36 | #define SNAPSHOT_COPY_PRIORITY 2 | |
37 | ||
38 | /* | |
39 | * Each snapshot reserves this many pages for io | |
40 | */ | |
41 | #define SNAPSHOT_PAGES 256 | |
42 | ||
c642f9e0 | 43 | static struct workqueue_struct *ksnapd; |
c4028958 | 44 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 45 | |
028867ac AK |
46 | struct dm_snap_pending_exception { |
47 | struct dm_snap_exception e; | |
1da177e4 LT |
48 | |
49 | /* | |
50 | * Origin buffers waiting for this to complete are held | |
51 | * in a bio list | |
52 | */ | |
53 | struct bio_list origin_bios; | |
54 | struct bio_list snapshot_bios; | |
55 | ||
eccf0817 AK |
56 | /* |
57 | * Short-term queue of pending exceptions prior to submission. | |
58 | */ | |
59 | struct list_head list; | |
60 | ||
1da177e4 | 61 | /* |
b4b610f6 | 62 | * The primary pending_exception is the one that holds |
4b832e8d | 63 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
64 | * group of pending_exceptions. It is always last to get freed. |
65 | * These fields get set up when writing to the origin. | |
1da177e4 | 66 | */ |
028867ac | 67 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
68 | |
69 | /* | |
70 | * Number of pending_exceptions processing this chunk. | |
71 | * When this drops to zero we must complete the origin bios. | |
72 | * If incrementing or decrementing this, hold pe->snap->lock for | |
73 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
74 | * they are the same. | |
75 | */ | |
4b832e8d | 76 | atomic_t ref_count; |
1da177e4 LT |
77 | |
78 | /* Pointer back to snapshot context */ | |
79 | struct dm_snapshot *snap; | |
80 | ||
81 | /* | |
82 | * 1 indicates the exception has already been sent to | |
83 | * kcopyd. | |
84 | */ | |
85 | int started; | |
86 | }; | |
87 | ||
88 | /* | |
89 | * Hash table mapping origin volumes to lists of snapshots and | |
90 | * a lock to protect it | |
91 | */ | |
e18b890b CL |
92 | static struct kmem_cache *exception_cache; |
93 | static struct kmem_cache *pending_cache; | |
1da177e4 LT |
94 | static mempool_t *pending_pool; |
95 | ||
96 | /* | |
97 | * One of these per registered origin, held in the snapshot_origins hash | |
98 | */ | |
99 | struct origin { | |
100 | /* The origin device */ | |
101 | struct block_device *bdev; | |
102 | ||
103 | struct list_head hash_list; | |
104 | ||
105 | /* List of snapshots for this origin */ | |
106 | struct list_head snapshots; | |
107 | }; | |
108 | ||
109 | /* | |
110 | * Size of the hash table for origin volumes. If we make this | |
111 | * the size of the minors list then it should be nearly perfect | |
112 | */ | |
113 | #define ORIGIN_HASH_SIZE 256 | |
114 | #define ORIGIN_MASK 0xFF | |
115 | static struct list_head *_origins; | |
116 | static struct rw_semaphore _origins_lock; | |
117 | ||
118 | static int init_origin_hash(void) | |
119 | { | |
120 | int i; | |
121 | ||
122 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
123 | GFP_KERNEL); | |
124 | if (!_origins) { | |
72d94861 | 125 | DMERR("unable to allocate memory"); |
1da177e4 LT |
126 | return -ENOMEM; |
127 | } | |
128 | ||
129 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
130 | INIT_LIST_HEAD(_origins + i); | |
131 | init_rwsem(&_origins_lock); | |
132 | ||
133 | return 0; | |
134 | } | |
135 | ||
136 | static void exit_origin_hash(void) | |
137 | { | |
138 | kfree(_origins); | |
139 | } | |
140 | ||
028867ac | 141 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
142 | { |
143 | return bdev->bd_dev & ORIGIN_MASK; | |
144 | } | |
145 | ||
146 | static struct origin *__lookup_origin(struct block_device *origin) | |
147 | { | |
148 | struct list_head *ol; | |
149 | struct origin *o; | |
150 | ||
151 | ol = &_origins[origin_hash(origin)]; | |
152 | list_for_each_entry (o, ol, hash_list) | |
153 | if (bdev_equal(o->bdev, origin)) | |
154 | return o; | |
155 | ||
156 | return NULL; | |
157 | } | |
158 | ||
159 | static void __insert_origin(struct origin *o) | |
160 | { | |
161 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
162 | list_add_tail(&o->hash_list, sl); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Make a note of the snapshot and its origin so we can look it | |
167 | * up when the origin has a write on it. | |
168 | */ | |
169 | static int register_snapshot(struct dm_snapshot *snap) | |
170 | { | |
171 | struct origin *o; | |
172 | struct block_device *bdev = snap->origin->bdev; | |
173 | ||
174 | down_write(&_origins_lock); | |
175 | o = __lookup_origin(bdev); | |
176 | ||
177 | if (!o) { | |
178 | /* New origin */ | |
179 | o = kmalloc(sizeof(*o), GFP_KERNEL); | |
180 | if (!o) { | |
181 | up_write(&_origins_lock); | |
182 | return -ENOMEM; | |
183 | } | |
184 | ||
185 | /* Initialise the struct */ | |
186 | INIT_LIST_HEAD(&o->snapshots); | |
187 | o->bdev = bdev; | |
188 | ||
189 | __insert_origin(o); | |
190 | } | |
191 | ||
192 | list_add_tail(&snap->list, &o->snapshots); | |
193 | ||
194 | up_write(&_origins_lock); | |
195 | return 0; | |
196 | } | |
197 | ||
198 | static void unregister_snapshot(struct dm_snapshot *s) | |
199 | { | |
200 | struct origin *o; | |
201 | ||
202 | down_write(&_origins_lock); | |
203 | o = __lookup_origin(s->origin->bdev); | |
204 | ||
205 | list_del(&s->list); | |
206 | if (list_empty(&o->snapshots)) { | |
207 | list_del(&o->hash_list); | |
208 | kfree(o); | |
209 | } | |
210 | ||
211 | up_write(&_origins_lock); | |
212 | } | |
213 | ||
214 | /* | |
215 | * Implementation of the exception hash tables. | |
216 | */ | |
217 | static int init_exception_table(struct exception_table *et, uint32_t size) | |
218 | { | |
219 | unsigned int i; | |
220 | ||
221 | et->hash_mask = size - 1; | |
222 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
223 | if (!et->table) | |
224 | return -ENOMEM; | |
225 | ||
226 | for (i = 0; i < size; i++) | |
227 | INIT_LIST_HEAD(et->table + i); | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
e18b890b | 232 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
233 | { |
234 | struct list_head *slot; | |
028867ac | 235 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
236 | int i, size; |
237 | ||
238 | size = et->hash_mask + 1; | |
239 | for (i = 0; i < size; i++) { | |
240 | slot = et->table + i; | |
241 | ||
242 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
243 | kmem_cache_free(mem, ex); | |
244 | } | |
245 | ||
246 | vfree(et->table); | |
247 | } | |
248 | ||
028867ac | 249 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 LT |
250 | { |
251 | return chunk & et->hash_mask; | |
252 | } | |
253 | ||
028867ac AK |
254 | static void insert_exception(struct exception_table *eh, |
255 | struct dm_snap_exception *e) | |
1da177e4 LT |
256 | { |
257 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
258 | list_add(&e->hash_list, l); | |
259 | } | |
260 | ||
028867ac | 261 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
262 | { |
263 | list_del(&e->hash_list); | |
264 | } | |
265 | ||
266 | /* | |
267 | * Return the exception data for a sector, or NULL if not | |
268 | * remapped. | |
269 | */ | |
028867ac AK |
270 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
271 | chunk_t chunk) | |
1da177e4 LT |
272 | { |
273 | struct list_head *slot; | |
028867ac | 274 | struct dm_snap_exception *e; |
1da177e4 LT |
275 | |
276 | slot = &et->table[exception_hash(et, chunk)]; | |
277 | list_for_each_entry (e, slot, hash_list) | |
278 | if (e->old_chunk == chunk) | |
279 | return e; | |
280 | ||
281 | return NULL; | |
282 | } | |
283 | ||
028867ac | 284 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 285 | { |
028867ac | 286 | struct dm_snap_exception *e; |
1da177e4 LT |
287 | |
288 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
289 | if (!e) | |
290 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
291 | ||
292 | return e; | |
293 | } | |
294 | ||
028867ac | 295 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
296 | { |
297 | kmem_cache_free(exception_cache, e); | |
298 | } | |
299 | ||
028867ac | 300 | static struct dm_snap_pending_exception *alloc_pending_exception(void) |
1da177e4 LT |
301 | { |
302 | return mempool_alloc(pending_pool, GFP_NOIO); | |
303 | } | |
304 | ||
028867ac | 305 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
306 | { |
307 | mempool_free(pe, pending_pool); | |
308 | } | |
309 | ||
310 | int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) | |
311 | { | |
028867ac | 312 | struct dm_snap_exception *e; |
1da177e4 LT |
313 | |
314 | e = alloc_exception(); | |
315 | if (!e) | |
316 | return -ENOMEM; | |
317 | ||
318 | e->old_chunk = old; | |
319 | e->new_chunk = new; | |
320 | insert_exception(&s->complete, e); | |
321 | return 0; | |
322 | } | |
323 | ||
324 | /* | |
325 | * Hard coded magic. | |
326 | */ | |
327 | static int calc_max_buckets(void) | |
328 | { | |
329 | /* use a fixed size of 2MB */ | |
330 | unsigned long mem = 2 * 1024 * 1024; | |
331 | mem /= sizeof(struct list_head); | |
332 | ||
333 | return mem; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Rounds a number down to a power of 2. | |
338 | */ | |
028867ac | 339 | static uint32_t round_down(uint32_t n) |
1da177e4 LT |
340 | { |
341 | while (n & (n - 1)) | |
342 | n &= (n - 1); | |
343 | return n; | |
344 | } | |
345 | ||
346 | /* | |
347 | * Allocate room for a suitable hash table. | |
348 | */ | |
349 | static int init_hash_tables(struct dm_snapshot *s) | |
350 | { | |
351 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
352 | ||
353 | /* | |
354 | * Calculate based on the size of the original volume or | |
355 | * the COW volume... | |
356 | */ | |
357 | cow_dev_size = get_dev_size(s->cow->bdev); | |
358 | origin_dev_size = get_dev_size(s->origin->bdev); | |
359 | max_buckets = calc_max_buckets(); | |
360 | ||
361 | hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; | |
362 | hash_size = min(hash_size, max_buckets); | |
363 | ||
364 | /* Round it down to a power of 2 */ | |
365 | hash_size = round_down(hash_size); | |
366 | if (init_exception_table(&s->complete, hash_size)) | |
367 | return -ENOMEM; | |
368 | ||
369 | /* | |
370 | * Allocate hash table for in-flight exceptions | |
371 | * Make this smaller than the real hash table | |
372 | */ | |
373 | hash_size >>= 3; | |
374 | if (hash_size < 64) | |
375 | hash_size = 64; | |
376 | ||
377 | if (init_exception_table(&s->pending, hash_size)) { | |
378 | exit_exception_table(&s->complete, exception_cache); | |
379 | return -ENOMEM; | |
380 | } | |
381 | ||
382 | return 0; | |
383 | } | |
384 | ||
385 | /* | |
386 | * Round a number up to the nearest 'size' boundary. size must | |
387 | * be a power of 2. | |
388 | */ | |
028867ac | 389 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
390 | { |
391 | size--; | |
392 | return (n + size) & ~size; | |
393 | } | |
394 | ||
4c7e3bf4 MM |
395 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
396 | char **error) | |
397 | { | |
398 | unsigned long chunk_size; | |
399 | char *value; | |
400 | ||
401 | chunk_size = simple_strtoul(chunk_size_arg, &value, 10); | |
402 | if (*chunk_size_arg == '\0' || *value != '\0') { | |
403 | *error = "Invalid chunk size"; | |
404 | return -EINVAL; | |
405 | } | |
406 | ||
407 | if (!chunk_size) { | |
408 | s->chunk_size = s->chunk_mask = s->chunk_shift = 0; | |
409 | return 0; | |
410 | } | |
411 | ||
412 | /* | |
413 | * Chunk size must be multiple of page size. Silently | |
414 | * round up if it's not. | |
415 | */ | |
416 | chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); | |
417 | ||
418 | /* Check chunk_size is a power of 2 */ | |
6f3c3f0a | 419 | if (!is_power_of_2(chunk_size)) { |
4c7e3bf4 MM |
420 | *error = "Chunk size is not a power of 2"; |
421 | return -EINVAL; | |
422 | } | |
423 | ||
424 | /* Validate the chunk size against the device block size */ | |
425 | if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) { | |
426 | *error = "Chunk size is not a multiple of device blocksize"; | |
427 | return -EINVAL; | |
428 | } | |
429 | ||
430 | s->chunk_size = chunk_size; | |
431 | s->chunk_mask = chunk_size - 1; | |
432 | s->chunk_shift = ffs(chunk_size) - 1; | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
1da177e4 LT |
437 | /* |
438 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
439 | */ | |
440 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
441 | { | |
442 | struct dm_snapshot *s; | |
1da177e4 LT |
443 | int r = -EINVAL; |
444 | char persistent; | |
445 | char *origin_path; | |
446 | char *cow_path; | |
1da177e4 | 447 | |
4c7e3bf4 | 448 | if (argc != 4) { |
72d94861 | 449 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
450 | r = -EINVAL; |
451 | goto bad1; | |
452 | } | |
453 | ||
454 | origin_path = argv[0]; | |
455 | cow_path = argv[1]; | |
456 | persistent = toupper(*argv[2]); | |
457 | ||
458 | if (persistent != 'P' && persistent != 'N') { | |
459 | ti->error = "Persistent flag is not P or N"; | |
460 | r = -EINVAL; | |
461 | goto bad1; | |
462 | } | |
463 | ||
1da177e4 LT |
464 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
465 | if (s == NULL) { | |
466 | ti->error = "Cannot allocate snapshot context private " | |
467 | "structure"; | |
468 | r = -ENOMEM; | |
469 | goto bad1; | |
470 | } | |
471 | ||
472 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
473 | if (r) { | |
474 | ti->error = "Cannot get origin device"; | |
475 | goto bad2; | |
476 | } | |
477 | ||
478 | r = dm_get_device(ti, cow_path, 0, 0, | |
479 | FMODE_READ | FMODE_WRITE, &s->cow); | |
480 | if (r) { | |
481 | dm_put_device(ti, s->origin); | |
482 | ti->error = "Cannot get COW device"; | |
483 | goto bad2; | |
484 | } | |
485 | ||
4c7e3bf4 MM |
486 | r = set_chunk_size(s, argv[3], &ti->error); |
487 | if (r) | |
1da177e4 | 488 | goto bad3; |
1da177e4 | 489 | |
1da177e4 | 490 | s->type = persistent; |
1da177e4 LT |
491 | |
492 | s->valid = 1; | |
aa14edeb | 493 | s->active = 0; |
1da177e4 LT |
494 | s->last_percent = 0; |
495 | init_rwsem(&s->lock); | |
ca3a931f | 496 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
497 | s->table = ti->table; |
498 | ||
499 | /* Allocate hash table for COW data */ | |
500 | if (init_hash_tables(s)) { | |
501 | ti->error = "Unable to allocate hash table space"; | |
502 | r = -ENOMEM; | |
503 | goto bad3; | |
504 | } | |
505 | ||
1da177e4 LT |
506 | s->store.snap = s; |
507 | ||
508 | if (persistent == 'P') | |
4c7e3bf4 | 509 | r = dm_create_persistent(&s->store); |
1da177e4 | 510 | else |
4c7e3bf4 | 511 | r = dm_create_transient(&s->store); |
1da177e4 LT |
512 | |
513 | if (r) { | |
514 | ti->error = "Couldn't create exception store"; | |
515 | r = -EINVAL; | |
516 | goto bad4; | |
517 | } | |
518 | ||
519 | r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); | |
520 | if (r) { | |
521 | ti->error = "Could not create kcopyd client"; | |
522 | goto bad5; | |
523 | } | |
524 | ||
aa14edeb | 525 | /* Metadata must only be loaded into one table at once */ |
f9cea4f7 | 526 | r = s->store.read_metadata(&s->store); |
0764147b | 527 | if (r < 0) { |
f9cea4f7 MM |
528 | ti->error = "Failed to read snapshot metadata"; |
529 | goto bad6; | |
0764147b MB |
530 | } else if (r > 0) { |
531 | s->valid = 0; | |
532 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 533 | } |
aa14edeb | 534 | |
ca3a931f | 535 | bio_list_init(&s->queued_bios); |
c4028958 | 536 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 537 | |
1da177e4 | 538 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 539 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
540 | if (register_snapshot(s)) { |
541 | r = -EINVAL; | |
542 | ti->error = "Cannot register snapshot origin"; | |
543 | goto bad6; | |
544 | } | |
545 | ||
546 | ti->private = s; | |
c51c2752 | 547 | ti->split_io = s->chunk_size; |
1da177e4 LT |
548 | |
549 | return 0; | |
550 | ||
551 | bad6: | |
552 | kcopyd_client_destroy(s->kcopyd_client); | |
553 | ||
554 | bad5: | |
555 | s->store.destroy(&s->store); | |
556 | ||
557 | bad4: | |
558 | exit_exception_table(&s->pending, pending_cache); | |
559 | exit_exception_table(&s->complete, exception_cache); | |
560 | ||
561 | bad3: | |
562 | dm_put_device(ti, s->cow); | |
563 | dm_put_device(ti, s->origin); | |
564 | ||
565 | bad2: | |
566 | kfree(s); | |
567 | ||
568 | bad1: | |
569 | return r; | |
570 | } | |
571 | ||
31c93a0c MB |
572 | static void __free_exceptions(struct dm_snapshot *s) |
573 | { | |
574 | kcopyd_client_destroy(s->kcopyd_client); | |
575 | s->kcopyd_client = NULL; | |
576 | ||
577 | exit_exception_table(&s->pending, pending_cache); | |
578 | exit_exception_table(&s->complete, exception_cache); | |
579 | ||
580 | s->store.destroy(&s->store); | |
581 | } | |
582 | ||
1da177e4 LT |
583 | static void snapshot_dtr(struct dm_target *ti) |
584 | { | |
028867ac | 585 | struct dm_snapshot *s = ti->private; |
1da177e4 | 586 | |
ca3a931f AK |
587 | flush_workqueue(ksnapd); |
588 | ||
138728dc AK |
589 | /* Prevent further origin writes from using this snapshot. */ |
590 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
591 | unregister_snapshot(s); |
592 | ||
31c93a0c | 593 | __free_exceptions(s); |
1da177e4 LT |
594 | |
595 | dm_put_device(ti, s->origin); | |
596 | dm_put_device(ti, s->cow); | |
138728dc | 597 | |
1da177e4 LT |
598 | kfree(s); |
599 | } | |
600 | ||
601 | /* | |
602 | * Flush a list of buffers. | |
603 | */ | |
604 | static void flush_bios(struct bio *bio) | |
605 | { | |
606 | struct bio *n; | |
607 | ||
608 | while (bio) { | |
609 | n = bio->bi_next; | |
610 | bio->bi_next = NULL; | |
611 | generic_make_request(bio); | |
612 | bio = n; | |
613 | } | |
614 | } | |
615 | ||
c4028958 | 616 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 617 | { |
c4028958 DH |
618 | struct dm_snapshot *s = |
619 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
620 | struct bio *queued_bios; |
621 | unsigned long flags; | |
622 | ||
623 | spin_lock_irqsave(&s->pe_lock, flags); | |
624 | queued_bios = bio_list_get(&s->queued_bios); | |
625 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
626 | ||
627 | flush_bios(queued_bios); | |
628 | } | |
629 | ||
1da177e4 LT |
630 | /* |
631 | * Error a list of buffers. | |
632 | */ | |
633 | static void error_bios(struct bio *bio) | |
634 | { | |
635 | struct bio *n; | |
636 | ||
637 | while (bio) { | |
638 | n = bio->bi_next; | |
639 | bio->bi_next = NULL; | |
6712ecf8 | 640 | bio_io_error(bio); |
1da177e4 LT |
641 | bio = n; |
642 | } | |
643 | } | |
644 | ||
695368ac | 645 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
646 | { |
647 | if (!s->valid) | |
648 | return; | |
649 | ||
650 | if (err == -EIO) | |
651 | DMERR("Invalidating snapshot: Error reading/writing."); | |
652 | else if (err == -ENOMEM) | |
653 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
654 | ||
76df1c65 AK |
655 | if (s->store.drop_snapshot) |
656 | s->store.drop_snapshot(&s->store); | |
657 | ||
658 | s->valid = 0; | |
659 | ||
660 | dm_table_event(s->table); | |
661 | } | |
662 | ||
028867ac | 663 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
664 | { |
665 | atomic_inc(&pe->ref_count); | |
666 | } | |
667 | ||
028867ac | 668 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 669 | { |
028867ac | 670 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
671 | struct bio *origin_bios = NULL; |
672 | ||
673 | primary_pe = pe->primary_pe; | |
674 | ||
675 | /* | |
676 | * If this pe is involved in a write to the origin and | |
677 | * it is the last sibling to complete then release | |
678 | * the bios for the original write to the origin. | |
679 | */ | |
680 | if (primary_pe && | |
681 | atomic_dec_and_test(&primary_pe->ref_count)) | |
682 | origin_bios = bio_list_get(&primary_pe->origin_bios); | |
683 | ||
684 | /* | |
685 | * Free the pe if it's not linked to an origin write or if | |
686 | * it's not itself a primary pe. | |
687 | */ | |
688 | if (!primary_pe || primary_pe != pe) | |
689 | free_pending_exception(pe); | |
690 | ||
691 | /* | |
692 | * Free the primary pe if nothing references it. | |
693 | */ | |
694 | if (primary_pe && !atomic_read(&primary_pe->ref_count)) | |
695 | free_pending_exception(primary_pe); | |
696 | ||
697 | return origin_bios; | |
698 | } | |
699 | ||
028867ac | 700 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 701 | { |
028867ac | 702 | struct dm_snap_exception *e; |
1da177e4 | 703 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
704 | struct bio *origin_bios = NULL; |
705 | struct bio *snapshot_bios = NULL; | |
706 | int error = 0; | |
1da177e4 | 707 | |
76df1c65 AK |
708 | if (!success) { |
709 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 710 | down_write(&s->lock); |
695368ac | 711 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 712 | error = 1; |
76df1c65 AK |
713 | goto out; |
714 | } | |
715 | ||
716 | e = alloc_exception(); | |
717 | if (!e) { | |
1da177e4 | 718 | down_write(&s->lock); |
695368ac | 719 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 720 | error = 1; |
76df1c65 AK |
721 | goto out; |
722 | } | |
723 | *e = pe->e; | |
1da177e4 | 724 | |
76df1c65 AK |
725 | down_write(&s->lock); |
726 | if (!s->valid) { | |
76df1c65 | 727 | free_exception(e); |
9d493fa8 | 728 | error = 1; |
76df1c65 | 729 | goto out; |
1da177e4 LT |
730 | } |
731 | ||
9d493fa8 AK |
732 | /* |
733 | * Add a proper exception, and remove the | |
734 | * in-flight exception from the list. | |
735 | */ | |
76df1c65 | 736 | insert_exception(&s->complete, e); |
76df1c65 | 737 | |
1da177e4 | 738 | out: |
695368ac | 739 | remove_exception(&pe->e); |
9d493fa8 | 740 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 741 | origin_bios = put_pending_exception(pe); |
1da177e4 | 742 | |
9d493fa8 AK |
743 | up_write(&s->lock); |
744 | ||
745 | /* Submit any pending write bios */ | |
746 | if (error) | |
747 | error_bios(snapshot_bios); | |
748 | else | |
749 | flush_bios(snapshot_bios); | |
750 | ||
751 | flush_bios(origin_bios); | |
1da177e4 LT |
752 | } |
753 | ||
754 | static void commit_callback(void *context, int success) | |
755 | { | |
028867ac AK |
756 | struct dm_snap_pending_exception *pe = context; |
757 | ||
1da177e4 LT |
758 | pending_complete(pe, success); |
759 | } | |
760 | ||
761 | /* | |
762 | * Called when the copy I/O has finished. kcopyd actually runs | |
763 | * this code so don't block. | |
764 | */ | |
765 | static void copy_callback(int read_err, unsigned int write_err, void *context) | |
766 | { | |
028867ac | 767 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
768 | struct dm_snapshot *s = pe->snap; |
769 | ||
770 | if (read_err || write_err) | |
771 | pending_complete(pe, 0); | |
772 | ||
773 | else | |
774 | /* Update the metadata if we are persistent */ | |
775 | s->store.commit_exception(&s->store, &pe->e, commit_callback, | |
776 | pe); | |
777 | } | |
778 | ||
779 | /* | |
780 | * Dispatches the copy operation to kcopyd. | |
781 | */ | |
028867ac | 782 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
783 | { |
784 | struct dm_snapshot *s = pe->snap; | |
785 | struct io_region src, dest; | |
786 | struct block_device *bdev = s->origin->bdev; | |
787 | sector_t dev_size; | |
788 | ||
789 | dev_size = get_dev_size(bdev); | |
790 | ||
791 | src.bdev = bdev; | |
792 | src.sector = chunk_to_sector(s, pe->e.old_chunk); | |
793 | src.count = min(s->chunk_size, dev_size - src.sector); | |
794 | ||
795 | dest.bdev = s->cow->bdev; | |
796 | dest.sector = chunk_to_sector(s, pe->e.new_chunk); | |
797 | dest.count = src.count; | |
798 | ||
799 | /* Hand over to kcopyd */ | |
800 | kcopyd_copy(s->kcopyd_client, | |
801 | &src, 1, &dest, 0, copy_callback, pe); | |
802 | } | |
803 | ||
804 | /* | |
805 | * Looks to see if this snapshot already has a pending exception | |
806 | * for this chunk, otherwise it allocates a new one and inserts | |
807 | * it into the pending table. | |
808 | * | |
809 | * NOTE: a write lock must be held on snap->lock before calling | |
810 | * this. | |
811 | */ | |
028867ac | 812 | static struct dm_snap_pending_exception * |
1da177e4 LT |
813 | __find_pending_exception(struct dm_snapshot *s, struct bio *bio) |
814 | { | |
028867ac AK |
815 | struct dm_snap_exception *e; |
816 | struct dm_snap_pending_exception *pe; | |
1da177e4 LT |
817 | chunk_t chunk = sector_to_chunk(s, bio->bi_sector); |
818 | ||
819 | /* | |
820 | * Is there a pending exception for this already ? | |
821 | */ | |
822 | e = lookup_exception(&s->pending, chunk); | |
823 | if (e) { | |
824 | /* cast the exception to a pending exception */ | |
028867ac | 825 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 AK |
826 | goto out; |
827 | } | |
1da177e4 | 828 | |
76df1c65 AK |
829 | /* |
830 | * Create a new pending exception, we don't want | |
831 | * to hold the lock while we do this. | |
832 | */ | |
833 | up_write(&s->lock); | |
834 | pe = alloc_pending_exception(); | |
835 | down_write(&s->lock); | |
1da177e4 | 836 | |
76df1c65 AK |
837 | if (!s->valid) { |
838 | free_pending_exception(pe); | |
839 | return NULL; | |
840 | } | |
1da177e4 | 841 | |
76df1c65 AK |
842 | e = lookup_exception(&s->pending, chunk); |
843 | if (e) { | |
844 | free_pending_exception(pe); | |
028867ac | 845 | pe = container_of(e, struct dm_snap_pending_exception, e); |
76df1c65 | 846 | goto out; |
1da177e4 LT |
847 | } |
848 | ||
76df1c65 AK |
849 | pe->e.old_chunk = chunk; |
850 | bio_list_init(&pe->origin_bios); | |
851 | bio_list_init(&pe->snapshot_bios); | |
852 | pe->primary_pe = NULL; | |
4b832e8d | 853 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
854 | pe->snap = s; |
855 | pe->started = 0; | |
856 | ||
857 | if (s->store.prepare_exception(&s->store, &pe->e)) { | |
858 | free_pending_exception(pe); | |
859 | return NULL; | |
860 | } | |
861 | ||
4b832e8d | 862 | get_pending_exception(pe); |
76df1c65 AK |
863 | insert_exception(&s->pending, &pe->e); |
864 | ||
865 | out: | |
1da177e4 LT |
866 | return pe; |
867 | } | |
868 | ||
028867ac AK |
869 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
870 | struct bio *bio) | |
1da177e4 LT |
871 | { |
872 | bio->bi_bdev = s->cow->bdev; | |
873 | bio->bi_sector = chunk_to_sector(s, e->new_chunk) + | |
874 | (bio->bi_sector & s->chunk_mask); | |
875 | } | |
876 | ||
877 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
878 | union map_info *map_context) | |
879 | { | |
028867ac AK |
880 | struct dm_snap_exception *e; |
881 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 882 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 883 | chunk_t chunk; |
028867ac | 884 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 LT |
885 | |
886 | chunk = sector_to_chunk(s, bio->bi_sector); | |
887 | ||
888 | /* Full snapshots are not usable */ | |
76df1c65 | 889 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 890 | if (!s->valid) |
f6a80ea8 | 891 | return -EIO; |
1da177e4 | 892 | |
ba40a2aa AK |
893 | /* FIXME: should only take write lock if we need |
894 | * to copy an exception */ | |
895 | down_write(&s->lock); | |
896 | ||
897 | if (!s->valid) { | |
898 | r = -EIO; | |
899 | goto out_unlock; | |
900 | } | |
901 | ||
902 | /* If the block is already remapped - use that, else remap it */ | |
903 | e = lookup_exception(&s->complete, chunk); | |
904 | if (e) { | |
905 | remap_exception(s, e, bio); | |
906 | goto out_unlock; | |
907 | } | |
908 | ||
1da177e4 LT |
909 | /* |
910 | * Write to snapshot - higher level takes care of RW/RO | |
911 | * flags so we should only get this if we are | |
912 | * writeable. | |
913 | */ | |
914 | if (bio_rw(bio) == WRITE) { | |
76df1c65 AK |
915 | pe = __find_pending_exception(s, bio); |
916 | if (!pe) { | |
695368ac | 917 | __invalidate_snapshot(s, -ENOMEM); |
76df1c65 AK |
918 | r = -EIO; |
919 | goto out_unlock; | |
1da177e4 LT |
920 | } |
921 | ||
76df1c65 AK |
922 | remap_exception(s, &pe->e, bio); |
923 | bio_list_add(&pe->snapshot_bios, bio); | |
924 | ||
d2a7ad29 | 925 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 926 | |
76df1c65 AK |
927 | if (!pe->started) { |
928 | /* this is protected by snap->lock */ | |
929 | pe->started = 1; | |
ba40a2aa | 930 | up_write(&s->lock); |
76df1c65 | 931 | start_copy(pe); |
ba40a2aa AK |
932 | goto out; |
933 | } | |
934 | } else | |
1da177e4 LT |
935 | /* |
936 | * FIXME: this read path scares me because we | |
937 | * always use the origin when we have a pending | |
938 | * exception. However I can't think of a | |
939 | * situation where this is wrong - ejt. | |
940 | */ | |
ba40a2aa | 941 | bio->bi_bdev = s->origin->bdev; |
1da177e4 | 942 | |
ba40a2aa AK |
943 | out_unlock: |
944 | up_write(&s->lock); | |
945 | out: | |
1da177e4 LT |
946 | return r; |
947 | } | |
948 | ||
949 | static void snapshot_resume(struct dm_target *ti) | |
950 | { | |
028867ac | 951 | struct dm_snapshot *s = ti->private; |
1da177e4 | 952 | |
aa14edeb AK |
953 | down_write(&s->lock); |
954 | s->active = 1; | |
955 | up_write(&s->lock); | |
1da177e4 LT |
956 | } |
957 | ||
958 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
959 | char *result, unsigned int maxlen) | |
960 | { | |
028867ac | 961 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
962 | |
963 | switch (type) { | |
964 | case STATUSTYPE_INFO: | |
965 | if (!snap->valid) | |
966 | snprintf(result, maxlen, "Invalid"); | |
967 | else { | |
968 | if (snap->store.fraction_full) { | |
969 | sector_t numerator, denominator; | |
970 | snap->store.fraction_full(&snap->store, | |
971 | &numerator, | |
972 | &denominator); | |
4ee218cd AM |
973 | snprintf(result, maxlen, "%llu/%llu", |
974 | (unsigned long long)numerator, | |
975 | (unsigned long long)denominator); | |
1da177e4 LT |
976 | } |
977 | else | |
978 | snprintf(result, maxlen, "Unknown"); | |
979 | } | |
980 | break; | |
981 | ||
982 | case STATUSTYPE_TABLE: | |
983 | /* | |
984 | * kdevname returns a static pointer so we need | |
985 | * to make private copies if the output is to | |
986 | * make sense. | |
987 | */ | |
4ee218cd | 988 | snprintf(result, maxlen, "%s %s %c %llu", |
1da177e4 | 989 | snap->origin->name, snap->cow->name, |
4ee218cd AM |
990 | snap->type, |
991 | (unsigned long long)snap->chunk_size); | |
1da177e4 LT |
992 | break; |
993 | } | |
994 | ||
995 | return 0; | |
996 | } | |
997 | ||
998 | /*----------------------------------------------------------------- | |
999 | * Origin methods | |
1000 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1001 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1002 | { | |
d2a7ad29 | 1003 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1004 | struct dm_snapshot *snap; |
028867ac AK |
1005 | struct dm_snap_exception *e; |
1006 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1007 | chunk_t chunk; |
eccf0817 | 1008 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1009 | |
1010 | /* Do all the snapshots on this origin */ | |
1011 | list_for_each_entry (snap, snapshots, list) { | |
1012 | ||
76df1c65 AK |
1013 | down_write(&snap->lock); |
1014 | ||
aa14edeb AK |
1015 | /* Only deal with valid and active snapshots */ |
1016 | if (!snap->valid || !snap->active) | |
76df1c65 | 1017 | goto next_snapshot; |
1da177e4 | 1018 | |
d5e404c1 AK |
1019 | /* Nothing to do if writing beyond end of snapshot */ |
1020 | if (bio->bi_sector >= dm_table_get_size(snap->table)) | |
76df1c65 | 1021 | goto next_snapshot; |
1da177e4 LT |
1022 | |
1023 | /* | |
1024 | * Remember, different snapshots can have | |
1025 | * different chunk sizes. | |
1026 | */ | |
1027 | chunk = sector_to_chunk(snap, bio->bi_sector); | |
1028 | ||
1029 | /* | |
1030 | * Check exception table to see if block | |
1031 | * is already remapped in this snapshot | |
1032 | * and trigger an exception if not. | |
b4b610f6 | 1033 | * |
4b832e8d | 1034 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1035 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1036 | */ |
1037 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1038 | if (e) |
1039 | goto next_snapshot; | |
1040 | ||
1041 | pe = __find_pending_exception(snap, bio); | |
1042 | if (!pe) { | |
695368ac | 1043 | __invalidate_snapshot(snap, -ENOMEM); |
76df1c65 AK |
1044 | goto next_snapshot; |
1045 | } | |
1046 | ||
1047 | if (!primary_pe) { | |
1048 | /* | |
1049 | * Either every pe here has same | |
1050 | * primary_pe or none has one yet. | |
1051 | */ | |
1052 | if (pe->primary_pe) | |
1053 | primary_pe = pe->primary_pe; | |
1054 | else { | |
1055 | primary_pe = pe; | |
1056 | first = 1; | |
1da177e4 | 1057 | } |
76df1c65 AK |
1058 | |
1059 | bio_list_add(&primary_pe->origin_bios, bio); | |
1060 | ||
d2a7ad29 | 1061 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1062 | } |
1063 | ||
1064 | if (!pe->primary_pe) { | |
76df1c65 | 1065 | pe->primary_pe = primary_pe; |
4b832e8d | 1066 | get_pending_exception(primary_pe); |
76df1c65 AK |
1067 | } |
1068 | ||
1069 | if (!pe->started) { | |
1070 | pe->started = 1; | |
1071 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1072 | } |
1073 | ||
76df1c65 | 1074 | next_snapshot: |
1da177e4 LT |
1075 | up_write(&snap->lock); |
1076 | } | |
1077 | ||
b4b610f6 | 1078 | if (!primary_pe) |
4b832e8d | 1079 | return r; |
b4b610f6 AK |
1080 | |
1081 | /* | |
1082 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1083 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1084 | * got completed while we were in the loop above, so it falls to |
1085 | * us here to remove the primary_pe and submit any origin_bios. | |
1086 | */ | |
1087 | ||
4b832e8d | 1088 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1089 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1090 | free_pending_exception(primary_pe); | |
1091 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1092 | return r; |
b4b610f6 AK |
1093 | } |
1094 | ||
1da177e4 LT |
1095 | /* |
1096 | * Now that we have a complete pe list we can start the copying. | |
1097 | */ | |
eccf0817 AK |
1098 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1099 | start_copy(pe); | |
1da177e4 LT |
1100 | |
1101 | return r; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Called on a write from the origin driver. | |
1106 | */ | |
1107 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1108 | { | |
1109 | struct origin *o; | |
d2a7ad29 | 1110 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1111 | |
1112 | down_read(&_origins_lock); | |
1113 | o = __lookup_origin(origin->bdev); | |
1114 | if (o) | |
1115 | r = __origin_write(&o->snapshots, bio); | |
1116 | up_read(&_origins_lock); | |
1117 | ||
1118 | return r; | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1123 | */ | |
1124 | ||
1125 | /* | |
1126 | * Construct an origin mapping: <dev_path> | |
1127 | * The context for an origin is merely a 'struct dm_dev *' | |
1128 | * pointing to the real device. | |
1129 | */ | |
1130 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1131 | { | |
1132 | int r; | |
1133 | struct dm_dev *dev; | |
1134 | ||
1135 | if (argc != 1) { | |
72d94861 | 1136 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1137 | return -EINVAL; |
1138 | } | |
1139 | ||
1140 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1141 | dm_table_get_mode(ti->table), &dev); | |
1142 | if (r) { | |
1143 | ti->error = "Cannot get target device"; | |
1144 | return r; | |
1145 | } | |
1146 | ||
1147 | ti->private = dev; | |
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | static void origin_dtr(struct dm_target *ti) | |
1152 | { | |
028867ac | 1153 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1154 | dm_put_device(ti, dev); |
1155 | } | |
1156 | ||
1157 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1158 | union map_info *map_context) | |
1159 | { | |
028867ac | 1160 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1161 | bio->bi_bdev = dev->bdev; |
1162 | ||
1163 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1164 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1165 | } |
1166 | ||
1167 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1168 | ||
1169 | /* | |
1170 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1171 | * chunk sizes. | |
1172 | */ | |
1173 | static void origin_resume(struct dm_target *ti) | |
1174 | { | |
028867ac | 1175 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1176 | struct dm_snapshot *snap; |
1177 | struct origin *o; | |
1178 | chunk_t chunk_size = 0; | |
1179 | ||
1180 | down_read(&_origins_lock); | |
1181 | o = __lookup_origin(dev->bdev); | |
1182 | if (o) | |
1183 | list_for_each_entry (snap, &o->snapshots, list) | |
1184 | chunk_size = min_not_zero(chunk_size, snap->chunk_size); | |
1185 | up_read(&_origins_lock); | |
1186 | ||
1187 | ti->split_io = chunk_size; | |
1188 | } | |
1189 | ||
1190 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1191 | unsigned int maxlen) | |
1192 | { | |
028867ac | 1193 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1194 | |
1195 | switch (type) { | |
1196 | case STATUSTYPE_INFO: | |
1197 | result[0] = '\0'; | |
1198 | break; | |
1199 | ||
1200 | case STATUSTYPE_TABLE: | |
1201 | snprintf(result, maxlen, "%s", dev->name); | |
1202 | break; | |
1203 | } | |
1204 | ||
1205 | return 0; | |
1206 | } | |
1207 | ||
1208 | static struct target_type origin_target = { | |
1209 | .name = "snapshot-origin", | |
4c7e3bf4 | 1210 | .version = {1, 5, 0}, |
1da177e4 LT |
1211 | .module = THIS_MODULE, |
1212 | .ctr = origin_ctr, | |
1213 | .dtr = origin_dtr, | |
1214 | .map = origin_map, | |
1215 | .resume = origin_resume, | |
1216 | .status = origin_status, | |
1217 | }; | |
1218 | ||
1219 | static struct target_type snapshot_target = { | |
1220 | .name = "snapshot", | |
4c7e3bf4 | 1221 | .version = {1, 5, 0}, |
1da177e4 LT |
1222 | .module = THIS_MODULE, |
1223 | .ctr = snapshot_ctr, | |
1224 | .dtr = snapshot_dtr, | |
1225 | .map = snapshot_map, | |
1226 | .resume = snapshot_resume, | |
1227 | .status = snapshot_status, | |
1228 | }; | |
1229 | ||
1230 | static int __init dm_snapshot_init(void) | |
1231 | { | |
1232 | int r; | |
1233 | ||
1234 | r = dm_register_target(&snapshot_target); | |
1235 | if (r) { | |
1236 | DMERR("snapshot target register failed %d", r); | |
1237 | return r; | |
1238 | } | |
1239 | ||
1240 | r = dm_register_target(&origin_target); | |
1241 | if (r < 0) { | |
72d94861 | 1242 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1243 | goto bad1; |
1244 | } | |
1245 | ||
1246 | r = init_origin_hash(); | |
1247 | if (r) { | |
1248 | DMERR("init_origin_hash failed."); | |
1249 | goto bad2; | |
1250 | } | |
1251 | ||
028867ac | 1252 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1253 | if (!exception_cache) { |
1254 | DMERR("Couldn't create exception cache."); | |
1255 | r = -ENOMEM; | |
1256 | goto bad3; | |
1257 | } | |
1258 | ||
028867ac | 1259 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1260 | if (!pending_cache) { |
1261 | DMERR("Couldn't create pending cache."); | |
1262 | r = -ENOMEM; | |
1263 | goto bad4; | |
1264 | } | |
1265 | ||
93d2341c | 1266 | pending_pool = mempool_create_slab_pool(128, pending_cache); |
1da177e4 LT |
1267 | if (!pending_pool) { |
1268 | DMERR("Couldn't create pending pool."); | |
1269 | r = -ENOMEM; | |
1270 | goto bad5; | |
1271 | } | |
1272 | ||
ca3a931f AK |
1273 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1274 | if (!ksnapd) { | |
1275 | DMERR("Failed to create ksnapd workqueue."); | |
1276 | r = -ENOMEM; | |
1277 | goto bad6; | |
1278 | } | |
1279 | ||
1da177e4 LT |
1280 | return 0; |
1281 | ||
ca3a931f AK |
1282 | bad6: |
1283 | mempool_destroy(pending_pool); | |
1da177e4 LT |
1284 | bad5: |
1285 | kmem_cache_destroy(pending_cache); | |
1286 | bad4: | |
1287 | kmem_cache_destroy(exception_cache); | |
1288 | bad3: | |
1289 | exit_origin_hash(); | |
1290 | bad2: | |
1291 | dm_unregister_target(&origin_target); | |
1292 | bad1: | |
1293 | dm_unregister_target(&snapshot_target); | |
1294 | return r; | |
1295 | } | |
1296 | ||
1297 | static void __exit dm_snapshot_exit(void) | |
1298 | { | |
1299 | int r; | |
1300 | ||
ca3a931f AK |
1301 | destroy_workqueue(ksnapd); |
1302 | ||
1da177e4 LT |
1303 | r = dm_unregister_target(&snapshot_target); |
1304 | if (r) | |
1305 | DMERR("snapshot unregister failed %d", r); | |
1306 | ||
1307 | r = dm_unregister_target(&origin_target); | |
1308 | if (r) | |
1309 | DMERR("origin unregister failed %d", r); | |
1310 | ||
1311 | exit_origin_hash(); | |
1312 | mempool_destroy(pending_pool); | |
1313 | kmem_cache_destroy(pending_cache); | |
1314 | kmem_cache_destroy(exception_cache); | |
1315 | } | |
1316 | ||
1317 | /* Module hooks */ | |
1318 | module_init(dm_snapshot_init); | |
1319 | module_exit(dm_snapshot_exit); | |
1320 | ||
1321 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1322 | MODULE_AUTHOR("Joe Thornber"); | |
1323 | MODULE_LICENSE("GPL"); |