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 | ||
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 | /* | |
8ee2767a | 39 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 40 | */ |
8ee2767a | 41 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 42 | |
cd45daff MP |
43 | /* |
44 | * The size of the mempool used to track chunks in use. | |
45 | */ | |
46 | #define MIN_IOS 256 | |
47 | ||
ccc45ea8 JB |
48 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
49 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
50 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
51 | ||
191437a5 | 52 | struct dm_exception_table { |
ccc45ea8 JB |
53 | uint32_t hash_mask; |
54 | unsigned hash_shift; | |
55 | struct list_head *table; | |
56 | }; | |
57 | ||
58 | struct dm_snapshot { | |
59 | struct rw_semaphore lock; | |
60 | ||
61 | struct dm_dev *origin; | |
62 | ||
63 | /* List of snapshots per Origin */ | |
64 | struct list_head list; | |
65 | ||
66 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
67 | int valid; | |
68 | ||
69 | /* Origin writes don't trigger exceptions until this is set */ | |
70 | int active; | |
71 | ||
ccc45ea8 JB |
72 | mempool_t *pending_pool; |
73 | ||
74 | atomic_t pending_exceptions_count; | |
75 | ||
191437a5 JB |
76 | struct dm_exception_table pending; |
77 | struct dm_exception_table complete; | |
ccc45ea8 JB |
78 | |
79 | /* | |
80 | * pe_lock protects all pending_exception operations and access | |
81 | * as well as the snapshot_bios list. | |
82 | */ | |
83 | spinlock_t pe_lock; | |
84 | ||
85 | /* The on disk metadata handler */ | |
86 | struct dm_exception_store *store; | |
87 | ||
88 | struct dm_kcopyd_client *kcopyd_client; | |
89 | ||
90 | /* Queue of snapshot writes for ksnapd to flush */ | |
91 | struct bio_list queued_bios; | |
92 | struct work_struct queued_bios_work; | |
93 | ||
94 | /* Chunks with outstanding reads */ | |
95 | mempool_t *tracked_chunk_pool; | |
96 | spinlock_t tracked_chunk_lock; | |
97 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
98 | }; | |
99 | ||
c642f9e0 | 100 | static struct workqueue_struct *ksnapd; |
c4028958 | 101 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 102 | |
ccc45ea8 JB |
103 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
104 | chunk_t chunk) | |
105 | { | |
106 | return chunk << store->chunk_shift; | |
107 | } | |
108 | ||
109 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
110 | { | |
111 | /* | |
112 | * There is only ever one instance of a particular block | |
113 | * device so we can compare pointers safely. | |
114 | */ | |
115 | return lhs == rhs; | |
116 | } | |
117 | ||
028867ac | 118 | struct dm_snap_pending_exception { |
1d4989c8 | 119 | struct dm_exception e; |
1da177e4 LT |
120 | |
121 | /* | |
122 | * Origin buffers waiting for this to complete are held | |
123 | * in a bio list | |
124 | */ | |
125 | struct bio_list origin_bios; | |
126 | struct bio_list snapshot_bios; | |
127 | ||
eccf0817 AK |
128 | /* |
129 | * Short-term queue of pending exceptions prior to submission. | |
130 | */ | |
131 | struct list_head list; | |
132 | ||
1da177e4 | 133 | /* |
b4b610f6 | 134 | * The primary pending_exception is the one that holds |
4b832e8d | 135 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
136 | * group of pending_exceptions. It is always last to get freed. |
137 | * These fields get set up when writing to the origin. | |
1da177e4 | 138 | */ |
028867ac | 139 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
140 | |
141 | /* | |
142 | * Number of pending_exceptions processing this chunk. | |
143 | * When this drops to zero we must complete the origin bios. | |
144 | * If incrementing or decrementing this, hold pe->snap->lock for | |
145 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
146 | * they are the same. | |
147 | */ | |
4b832e8d | 148 | atomic_t ref_count; |
1da177e4 LT |
149 | |
150 | /* Pointer back to snapshot context */ | |
151 | struct dm_snapshot *snap; | |
152 | ||
153 | /* | |
154 | * 1 indicates the exception has already been sent to | |
155 | * kcopyd. | |
156 | */ | |
157 | int started; | |
158 | }; | |
159 | ||
160 | /* | |
161 | * Hash table mapping origin volumes to lists of snapshots and | |
162 | * a lock to protect it | |
163 | */ | |
e18b890b CL |
164 | static struct kmem_cache *exception_cache; |
165 | static struct kmem_cache *pending_cache; | |
1da177e4 | 166 | |
cd45daff MP |
167 | struct dm_snap_tracked_chunk { |
168 | struct hlist_node node; | |
169 | chunk_t chunk; | |
170 | }; | |
171 | ||
172 | static struct kmem_cache *tracked_chunk_cache; | |
173 | ||
174 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
175 | chunk_t chunk) | |
176 | { | |
177 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
178 | GFP_NOIO); | |
179 | unsigned long flags; | |
180 | ||
181 | c->chunk = chunk; | |
182 | ||
183 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
184 | hlist_add_head(&c->node, | |
185 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
186 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
187 | ||
188 | return c; | |
189 | } | |
190 | ||
191 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
192 | struct dm_snap_tracked_chunk *c) | |
193 | { | |
194 | unsigned long flags; | |
195 | ||
196 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
197 | hlist_del(&c->node); | |
198 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
199 | ||
200 | mempool_free(c, s->tracked_chunk_pool); | |
201 | } | |
202 | ||
a8d41b59 MP |
203 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
204 | { | |
205 | struct dm_snap_tracked_chunk *c; | |
206 | struct hlist_node *hn; | |
207 | int found = 0; | |
208 | ||
209 | spin_lock_irq(&s->tracked_chunk_lock); | |
210 | ||
211 | hlist_for_each_entry(c, hn, | |
212 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
213 | if (c->chunk == chunk) { | |
214 | found = 1; | |
215 | break; | |
216 | } | |
217 | } | |
218 | ||
219 | spin_unlock_irq(&s->tracked_chunk_lock); | |
220 | ||
221 | return found; | |
222 | } | |
223 | ||
1da177e4 LT |
224 | /* |
225 | * One of these per registered origin, held in the snapshot_origins hash | |
226 | */ | |
227 | struct origin { | |
228 | /* The origin device */ | |
229 | struct block_device *bdev; | |
230 | ||
231 | struct list_head hash_list; | |
232 | ||
233 | /* List of snapshots for this origin */ | |
234 | struct list_head snapshots; | |
235 | }; | |
236 | ||
237 | /* | |
238 | * Size of the hash table for origin volumes. If we make this | |
239 | * the size of the minors list then it should be nearly perfect | |
240 | */ | |
241 | #define ORIGIN_HASH_SIZE 256 | |
242 | #define ORIGIN_MASK 0xFF | |
243 | static struct list_head *_origins; | |
244 | static struct rw_semaphore _origins_lock; | |
245 | ||
246 | static int init_origin_hash(void) | |
247 | { | |
248 | int i; | |
249 | ||
250 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
251 | GFP_KERNEL); | |
252 | if (!_origins) { | |
72d94861 | 253 | DMERR("unable to allocate memory"); |
1da177e4 LT |
254 | return -ENOMEM; |
255 | } | |
256 | ||
257 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
258 | INIT_LIST_HEAD(_origins + i); | |
259 | init_rwsem(&_origins_lock); | |
260 | ||
261 | return 0; | |
262 | } | |
263 | ||
264 | static void exit_origin_hash(void) | |
265 | { | |
266 | kfree(_origins); | |
267 | } | |
268 | ||
028867ac | 269 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
270 | { |
271 | return bdev->bd_dev & ORIGIN_MASK; | |
272 | } | |
273 | ||
274 | static struct origin *__lookup_origin(struct block_device *origin) | |
275 | { | |
276 | struct list_head *ol; | |
277 | struct origin *o; | |
278 | ||
279 | ol = &_origins[origin_hash(origin)]; | |
280 | list_for_each_entry (o, ol, hash_list) | |
281 | if (bdev_equal(o->bdev, origin)) | |
282 | return o; | |
283 | ||
284 | return NULL; | |
285 | } | |
286 | ||
287 | static void __insert_origin(struct origin *o) | |
288 | { | |
289 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
290 | list_add_tail(&o->hash_list, sl); | |
291 | } | |
292 | ||
293 | /* | |
294 | * Make a note of the snapshot and its origin so we can look it | |
295 | * up when the origin has a write on it. | |
296 | */ | |
297 | static int register_snapshot(struct dm_snapshot *snap) | |
298 | { | |
6d45d93e | 299 | struct dm_snapshot *l; |
60c856c8 | 300 | struct origin *o, *new_o; |
1da177e4 LT |
301 | struct block_device *bdev = snap->origin->bdev; |
302 | ||
60c856c8 MP |
303 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
304 | if (!new_o) | |
305 | return -ENOMEM; | |
306 | ||
1da177e4 LT |
307 | down_write(&_origins_lock); |
308 | o = __lookup_origin(bdev); | |
309 | ||
60c856c8 MP |
310 | if (o) |
311 | kfree(new_o); | |
312 | else { | |
1da177e4 | 313 | /* New origin */ |
60c856c8 | 314 | o = new_o; |
1da177e4 LT |
315 | |
316 | /* Initialise the struct */ | |
317 | INIT_LIST_HEAD(&o->snapshots); | |
318 | o->bdev = bdev; | |
319 | ||
320 | __insert_origin(o); | |
321 | } | |
322 | ||
6d45d93e MP |
323 | /* Sort the list according to chunk size, largest-first smallest-last */ |
324 | list_for_each_entry(l, &o->snapshots, list) | |
325 | if (l->store->chunk_size < snap->store->chunk_size) | |
326 | break; | |
327 | list_add_tail(&snap->list, &l->list); | |
1da177e4 LT |
328 | |
329 | up_write(&_origins_lock); | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static void unregister_snapshot(struct dm_snapshot *s) | |
334 | { | |
335 | struct origin *o; | |
336 | ||
337 | down_write(&_origins_lock); | |
338 | o = __lookup_origin(s->origin->bdev); | |
339 | ||
340 | list_del(&s->list); | |
341 | if (list_empty(&o->snapshots)) { | |
342 | list_del(&o->hash_list); | |
343 | kfree(o); | |
344 | } | |
345 | ||
346 | up_write(&_origins_lock); | |
347 | } | |
348 | ||
349 | /* | |
350 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
351 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
352 | * some consecutive chunks to be grouped together. | |
1da177e4 | 353 | */ |
3510cb94 JB |
354 | static int dm_exception_table_init(struct dm_exception_table *et, |
355 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
356 | { |
357 | unsigned int i; | |
358 | ||
d74f81f8 | 359 | et->hash_shift = hash_shift; |
1da177e4 LT |
360 | et->hash_mask = size - 1; |
361 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
362 | if (!et->table) | |
363 | return -ENOMEM; | |
364 | ||
365 | for (i = 0; i < size; i++) | |
366 | INIT_LIST_HEAD(et->table + i); | |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
3510cb94 JB |
371 | static void dm_exception_table_exit(struct dm_exception_table *et, |
372 | struct kmem_cache *mem) | |
1da177e4 LT |
373 | { |
374 | struct list_head *slot; | |
1d4989c8 | 375 | struct dm_exception *ex, *next; |
1da177e4 LT |
376 | int i, size; |
377 | ||
378 | size = et->hash_mask + 1; | |
379 | for (i = 0; i < size; i++) { | |
380 | slot = et->table + i; | |
381 | ||
382 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
383 | kmem_cache_free(mem, ex); | |
384 | } | |
385 | ||
386 | vfree(et->table); | |
387 | } | |
388 | ||
191437a5 | 389 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 390 | { |
d74f81f8 | 391 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
392 | } |
393 | ||
3510cb94 | 394 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
395 | { |
396 | list_del(&e->hash_list); | |
397 | } | |
398 | ||
399 | /* | |
400 | * Return the exception data for a sector, or NULL if not | |
401 | * remapped. | |
402 | */ | |
3510cb94 JB |
403 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
404 | chunk_t chunk) | |
1da177e4 LT |
405 | { |
406 | struct list_head *slot; | |
1d4989c8 | 407 | struct dm_exception *e; |
1da177e4 LT |
408 | |
409 | slot = &et->table[exception_hash(et, chunk)]; | |
410 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
411 | if (chunk >= e->old_chunk && |
412 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
413 | return e; |
414 | ||
415 | return NULL; | |
416 | } | |
417 | ||
3510cb94 | 418 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 419 | { |
1d4989c8 | 420 | struct dm_exception *e; |
1da177e4 LT |
421 | |
422 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
423 | if (!e) | |
424 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
425 | ||
426 | return e; | |
427 | } | |
428 | ||
3510cb94 | 429 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
430 | { |
431 | kmem_cache_free(exception_cache, e); | |
432 | } | |
433 | ||
92e86812 | 434 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 435 | { |
92e86812 MP |
436 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
437 | GFP_NOIO); | |
438 | ||
879129d2 | 439 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
440 | pe->snap = s; |
441 | ||
442 | return pe; | |
1da177e4 LT |
443 | } |
444 | ||
028867ac | 445 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 446 | { |
879129d2 MP |
447 | struct dm_snapshot *s = pe->snap; |
448 | ||
449 | mempool_free(pe, s->pending_pool); | |
450 | smp_mb__before_atomic_dec(); | |
451 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
452 | } |
453 | ||
3510cb94 JB |
454 | static void dm_insert_exception(struct dm_exception_table *eh, |
455 | struct dm_exception *new_e) | |
d74f81f8 | 456 | { |
d74f81f8 | 457 | struct list_head *l; |
1d4989c8 | 458 | struct dm_exception *e = NULL; |
d74f81f8 MB |
459 | |
460 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
461 | ||
462 | /* Add immediately if this table doesn't support consecutive chunks */ | |
463 | if (!eh->hash_shift) | |
464 | goto out; | |
465 | ||
466 | /* List is ordered by old_chunk */ | |
467 | list_for_each_entry_reverse(e, l, hash_list) { | |
468 | /* Insert after an existing chunk? */ | |
469 | if (new_e->old_chunk == (e->old_chunk + | |
470 | dm_consecutive_chunk_count(e) + 1) && | |
471 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
472 | dm_consecutive_chunk_count(e) + 1)) { | |
473 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 474 | free_completed_exception(new_e); |
d74f81f8 MB |
475 | return; |
476 | } | |
477 | ||
478 | /* Insert before an existing chunk? */ | |
479 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
480 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
481 | dm_consecutive_chunk_count_inc(e); | |
482 | e->old_chunk--; | |
483 | e->new_chunk--; | |
3510cb94 | 484 | free_completed_exception(new_e); |
d74f81f8 MB |
485 | return; |
486 | } | |
487 | ||
488 | if (new_e->old_chunk > e->old_chunk) | |
489 | break; | |
490 | } | |
491 | ||
492 | out: | |
493 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
494 | } | |
495 | ||
a159c1ac JB |
496 | /* |
497 | * Callback used by the exception stores to load exceptions when | |
498 | * initialising. | |
499 | */ | |
500 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 501 | { |
a159c1ac | 502 | struct dm_snapshot *s = context; |
1d4989c8 | 503 | struct dm_exception *e; |
1da177e4 | 504 | |
3510cb94 | 505 | e = alloc_completed_exception(); |
1da177e4 LT |
506 | if (!e) |
507 | return -ENOMEM; | |
508 | ||
509 | e->old_chunk = old; | |
d74f81f8 MB |
510 | |
511 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 512 | e->new_chunk = new; |
d74f81f8 | 513 | |
3510cb94 | 514 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 515 | |
1da177e4 LT |
516 | return 0; |
517 | } | |
518 | ||
7e201b35 MP |
519 | #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r))) |
520 | ||
521 | /* | |
522 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
523 | * Return zero if the origin has no snapshots. | |
524 | */ | |
525 | static sector_t __minimum_chunk_size(struct origin *o) | |
526 | { | |
527 | struct dm_snapshot *snap; | |
528 | unsigned chunk_size = 0; | |
529 | ||
530 | if (o) | |
531 | list_for_each_entry(snap, &o->snapshots, list) | |
532 | chunk_size = min_not_zero(chunk_size, | |
533 | snap->store->chunk_size); | |
534 | ||
535 | return chunk_size; | |
536 | } | |
537 | ||
1da177e4 LT |
538 | /* |
539 | * Hard coded magic. | |
540 | */ | |
541 | static int calc_max_buckets(void) | |
542 | { | |
543 | /* use a fixed size of 2MB */ | |
544 | unsigned long mem = 2 * 1024 * 1024; | |
545 | mem /= sizeof(struct list_head); | |
546 | ||
547 | return mem; | |
548 | } | |
549 | ||
1da177e4 LT |
550 | /* |
551 | * Allocate room for a suitable hash table. | |
552 | */ | |
fee1998e | 553 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
554 | { |
555 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
556 | ||
557 | /* | |
558 | * Calculate based on the size of the original volume or | |
559 | * the COW volume... | |
560 | */ | |
fee1998e | 561 | cow_dev_size = get_dev_size(s->store->cow->bdev); |
1da177e4 LT |
562 | origin_dev_size = get_dev_size(s->origin->bdev); |
563 | max_buckets = calc_max_buckets(); | |
564 | ||
fee1998e | 565 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
566 | hash_size = min(hash_size, max_buckets); |
567 | ||
8e87b9b8 MP |
568 | if (hash_size < 64) |
569 | hash_size = 64; | |
8defd830 | 570 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
571 | if (dm_exception_table_init(&s->complete, hash_size, |
572 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
573 | return -ENOMEM; |
574 | ||
575 | /* | |
576 | * Allocate hash table for in-flight exceptions | |
577 | * Make this smaller than the real hash table | |
578 | */ | |
579 | hash_size >>= 3; | |
580 | if (hash_size < 64) | |
581 | hash_size = 64; | |
582 | ||
3510cb94 JB |
583 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
584 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
585 | return -ENOMEM; |
586 | } | |
587 | ||
588 | return 0; | |
589 | } | |
590 | ||
1da177e4 LT |
591 | /* |
592 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
593 | */ | |
594 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
595 | { | |
596 | struct dm_snapshot *s; | |
cd45daff | 597 | int i; |
1da177e4 | 598 | int r = -EINVAL; |
1da177e4 | 599 | char *origin_path; |
fee1998e JB |
600 | struct dm_exception_store *store; |
601 | unsigned args_used; | |
1da177e4 | 602 | |
4c7e3bf4 | 603 | if (argc != 4) { |
72d94861 | 604 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 605 | r = -EINVAL; |
fee1998e | 606 | goto bad_args; |
1da177e4 LT |
607 | } |
608 | ||
609 | origin_path = argv[0]; | |
fee1998e JB |
610 | argv++; |
611 | argc--; | |
1da177e4 | 612 | |
fee1998e JB |
613 | r = dm_exception_store_create(ti, argc, argv, &args_used, &store); |
614 | if (r) { | |
615 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 616 | r = -EINVAL; |
fee1998e | 617 | goto bad_args; |
1da177e4 LT |
618 | } |
619 | ||
fee1998e JB |
620 | argv += args_used; |
621 | argc -= args_used; | |
622 | ||
1da177e4 | 623 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
fee1998e | 624 | if (!s) { |
1da177e4 LT |
625 | ti->error = "Cannot allocate snapshot context private " |
626 | "structure"; | |
627 | r = -ENOMEM; | |
fee1998e | 628 | goto bad_snap; |
1da177e4 LT |
629 | } |
630 | ||
631 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
632 | if (r) { | |
633 | ti->error = "Cannot get origin device"; | |
fee1998e | 634 | goto bad_origin; |
1da177e4 LT |
635 | } |
636 | ||
fee1998e | 637 | s->store = store; |
1da177e4 | 638 | s->valid = 1; |
aa14edeb | 639 | s->active = 0; |
879129d2 | 640 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 641 | init_rwsem(&s->lock); |
ca3a931f | 642 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
643 | |
644 | /* Allocate hash table for COW data */ | |
fee1998e | 645 | if (init_hash_tables(s)) { |
1da177e4 LT |
646 | ti->error = "Unable to allocate hash table space"; |
647 | r = -ENOMEM; | |
fee1998e | 648 | goto bad_hash_tables; |
1da177e4 LT |
649 | } |
650 | ||
eb69aca5 | 651 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
652 | if (r) { |
653 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 654 | goto bad_kcopyd; |
1da177e4 LT |
655 | } |
656 | ||
92e86812 MP |
657 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
658 | if (!s->pending_pool) { | |
659 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 660 | goto bad_pending_pool; |
92e86812 MP |
661 | } |
662 | ||
cd45daff MP |
663 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
664 | tracked_chunk_cache); | |
665 | if (!s->tracked_chunk_pool) { | |
666 | ti->error = "Could not allocate tracked_chunk mempool for " | |
667 | "tracking reads"; | |
92e86812 | 668 | goto bad_tracked_chunk_pool; |
cd45daff MP |
669 | } |
670 | ||
671 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
672 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
673 | ||
674 | spin_lock_init(&s->tracked_chunk_lock); | |
675 | ||
aa14edeb | 676 | /* Metadata must only be loaded into one table at once */ |
493df71c JB |
677 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
678 | (void *)s); | |
0764147b | 679 | if (r < 0) { |
f9cea4f7 | 680 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 681 | goto bad_load_and_register; |
0764147b MB |
682 | } else if (r > 0) { |
683 | s->valid = 0; | |
684 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 685 | } |
aa14edeb | 686 | |
ca3a931f | 687 | bio_list_init(&s->queued_bios); |
c4028958 | 688 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 689 | |
3f2412dc MP |
690 | if (!s->store->chunk_size) { |
691 | ti->error = "Chunk size not set"; | |
692 | goto bad_load_and_register; | |
693 | } | |
694 | ||
1da177e4 | 695 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 696 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
697 | if (register_snapshot(s)) { |
698 | r = -EINVAL; | |
699 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 700 | goto bad_load_and_register; |
1da177e4 LT |
701 | } |
702 | ||
703 | ti->private = s; | |
d0216849 | 704 | ti->split_io = s->store->chunk_size; |
494b3ee7 | 705 | ti->num_flush_requests = 1; |
1da177e4 LT |
706 | |
707 | return 0; | |
708 | ||
fee1998e | 709 | bad_load_and_register: |
cd45daff MP |
710 | mempool_destroy(s->tracked_chunk_pool); |
711 | ||
fee1998e | 712 | bad_tracked_chunk_pool: |
92e86812 MP |
713 | mempool_destroy(s->pending_pool); |
714 | ||
fee1998e | 715 | bad_pending_pool: |
eb69aca5 | 716 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 717 | |
fee1998e | 718 | bad_kcopyd: |
3510cb94 JB |
719 | dm_exception_table_exit(&s->pending, pending_cache); |
720 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 721 | |
fee1998e | 722 | bad_hash_tables: |
1da177e4 LT |
723 | dm_put_device(ti, s->origin); |
724 | ||
fee1998e | 725 | bad_origin: |
1da177e4 LT |
726 | kfree(s); |
727 | ||
fee1998e JB |
728 | bad_snap: |
729 | dm_exception_store_destroy(store); | |
730 | ||
731 | bad_args: | |
1da177e4 LT |
732 | return r; |
733 | } | |
734 | ||
31c93a0c MB |
735 | static void __free_exceptions(struct dm_snapshot *s) |
736 | { | |
eb69aca5 | 737 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
738 | s->kcopyd_client = NULL; |
739 | ||
3510cb94 JB |
740 | dm_exception_table_exit(&s->pending, pending_cache); |
741 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
742 | } |
743 | ||
1da177e4 LT |
744 | static void snapshot_dtr(struct dm_target *ti) |
745 | { | |
cd45daff MP |
746 | #ifdef CONFIG_DM_DEBUG |
747 | int i; | |
748 | #endif | |
028867ac | 749 | struct dm_snapshot *s = ti->private; |
1da177e4 | 750 | |
ca3a931f AK |
751 | flush_workqueue(ksnapd); |
752 | ||
138728dc AK |
753 | /* Prevent further origin writes from using this snapshot. */ |
754 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
755 | unregister_snapshot(s); |
756 | ||
879129d2 | 757 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 758 | msleep(1); |
879129d2 MP |
759 | /* |
760 | * Ensure instructions in mempool_destroy aren't reordered | |
761 | * before atomic_read. | |
762 | */ | |
763 | smp_mb(); | |
764 | ||
cd45daff MP |
765 | #ifdef CONFIG_DM_DEBUG |
766 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
767 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
768 | #endif | |
769 | ||
770 | mempool_destroy(s->tracked_chunk_pool); | |
771 | ||
31c93a0c | 772 | __free_exceptions(s); |
1da177e4 | 773 | |
92e86812 MP |
774 | mempool_destroy(s->pending_pool); |
775 | ||
1da177e4 | 776 | dm_put_device(ti, s->origin); |
fee1998e JB |
777 | |
778 | dm_exception_store_destroy(s->store); | |
138728dc | 779 | |
1da177e4 LT |
780 | kfree(s); |
781 | } | |
782 | ||
783 | /* | |
784 | * Flush a list of buffers. | |
785 | */ | |
786 | static void flush_bios(struct bio *bio) | |
787 | { | |
788 | struct bio *n; | |
789 | ||
790 | while (bio) { | |
791 | n = bio->bi_next; | |
792 | bio->bi_next = NULL; | |
793 | generic_make_request(bio); | |
794 | bio = n; | |
795 | } | |
796 | } | |
797 | ||
c4028958 | 798 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 799 | { |
c4028958 DH |
800 | struct dm_snapshot *s = |
801 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
802 | struct bio *queued_bios; |
803 | unsigned long flags; | |
804 | ||
805 | spin_lock_irqsave(&s->pe_lock, flags); | |
806 | queued_bios = bio_list_get(&s->queued_bios); | |
807 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
808 | ||
809 | flush_bios(queued_bios); | |
810 | } | |
811 | ||
1da177e4 LT |
812 | /* |
813 | * Error a list of buffers. | |
814 | */ | |
815 | static void error_bios(struct bio *bio) | |
816 | { | |
817 | struct bio *n; | |
818 | ||
819 | while (bio) { | |
820 | n = bio->bi_next; | |
821 | bio->bi_next = NULL; | |
6712ecf8 | 822 | bio_io_error(bio); |
1da177e4 LT |
823 | bio = n; |
824 | } | |
825 | } | |
826 | ||
695368ac | 827 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
828 | { |
829 | if (!s->valid) | |
830 | return; | |
831 | ||
832 | if (err == -EIO) | |
833 | DMERR("Invalidating snapshot: Error reading/writing."); | |
834 | else if (err == -ENOMEM) | |
835 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
836 | ||
493df71c JB |
837 | if (s->store->type->drop_snapshot) |
838 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
839 | |
840 | s->valid = 0; | |
841 | ||
0cea9c78 | 842 | dm_table_event(s->store->ti->table); |
76df1c65 AK |
843 | } |
844 | ||
028867ac | 845 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
846 | { |
847 | atomic_inc(&pe->ref_count); | |
848 | } | |
849 | ||
028867ac | 850 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 851 | { |
028867ac | 852 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
853 | struct bio *origin_bios = NULL; |
854 | ||
855 | primary_pe = pe->primary_pe; | |
856 | ||
857 | /* | |
858 | * If this pe is involved in a write to the origin and | |
859 | * it is the last sibling to complete then release | |
860 | * the bios for the original write to the origin. | |
861 | */ | |
862 | if (primary_pe && | |
7c5f78b9 | 863 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 864 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
865 | free_pending_exception(primary_pe); |
866 | } | |
4b832e8d AK |
867 | |
868 | /* | |
869 | * Free the pe if it's not linked to an origin write or if | |
870 | * it's not itself a primary pe. | |
871 | */ | |
872 | if (!primary_pe || primary_pe != pe) | |
873 | free_pending_exception(pe); | |
874 | ||
4b832e8d AK |
875 | return origin_bios; |
876 | } | |
877 | ||
028867ac | 878 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 879 | { |
1d4989c8 | 880 | struct dm_exception *e; |
1da177e4 | 881 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
882 | struct bio *origin_bios = NULL; |
883 | struct bio *snapshot_bios = NULL; | |
884 | int error = 0; | |
1da177e4 | 885 | |
76df1c65 AK |
886 | if (!success) { |
887 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 888 | down_write(&s->lock); |
695368ac | 889 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 890 | error = 1; |
76df1c65 AK |
891 | goto out; |
892 | } | |
893 | ||
3510cb94 | 894 | e = alloc_completed_exception(); |
76df1c65 | 895 | if (!e) { |
1da177e4 | 896 | down_write(&s->lock); |
695368ac | 897 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 898 | error = 1; |
76df1c65 AK |
899 | goto out; |
900 | } | |
901 | *e = pe->e; | |
1da177e4 | 902 | |
76df1c65 AK |
903 | down_write(&s->lock); |
904 | if (!s->valid) { | |
3510cb94 | 905 | free_completed_exception(e); |
9d493fa8 | 906 | error = 1; |
76df1c65 | 907 | goto out; |
1da177e4 LT |
908 | } |
909 | ||
a8d41b59 MP |
910 | /* |
911 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 912 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
913 | */ |
914 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 915 | msleep(1); |
a8d41b59 | 916 | |
9d493fa8 AK |
917 | /* |
918 | * Add a proper exception, and remove the | |
919 | * in-flight exception from the list. | |
920 | */ | |
3510cb94 | 921 | dm_insert_exception(&s->complete, e); |
76df1c65 | 922 | |
1da177e4 | 923 | out: |
3510cb94 | 924 | dm_remove_exception(&pe->e); |
9d493fa8 | 925 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 926 | origin_bios = put_pending_exception(pe); |
1da177e4 | 927 | |
9d493fa8 AK |
928 | up_write(&s->lock); |
929 | ||
930 | /* Submit any pending write bios */ | |
931 | if (error) | |
932 | error_bios(snapshot_bios); | |
933 | else | |
934 | flush_bios(snapshot_bios); | |
935 | ||
936 | flush_bios(origin_bios); | |
1da177e4 LT |
937 | } |
938 | ||
939 | static void commit_callback(void *context, int success) | |
940 | { | |
028867ac AK |
941 | struct dm_snap_pending_exception *pe = context; |
942 | ||
1da177e4 LT |
943 | pending_complete(pe, success); |
944 | } | |
945 | ||
946 | /* | |
947 | * Called when the copy I/O has finished. kcopyd actually runs | |
948 | * this code so don't block. | |
949 | */ | |
4cdc1d1f | 950 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 951 | { |
028867ac | 952 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
953 | struct dm_snapshot *s = pe->snap; |
954 | ||
955 | if (read_err || write_err) | |
956 | pending_complete(pe, 0); | |
957 | ||
958 | else | |
959 | /* Update the metadata if we are persistent */ | |
493df71c JB |
960 | s->store->type->commit_exception(s->store, &pe->e, |
961 | commit_callback, pe); | |
1da177e4 LT |
962 | } |
963 | ||
964 | /* | |
965 | * Dispatches the copy operation to kcopyd. | |
966 | */ | |
028867ac | 967 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
968 | { |
969 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 970 | struct dm_io_region src, dest; |
1da177e4 LT |
971 | struct block_device *bdev = s->origin->bdev; |
972 | sector_t dev_size; | |
973 | ||
974 | dev_size = get_dev_size(bdev); | |
975 | ||
976 | src.bdev = bdev; | |
71fab00a | 977 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 978 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 979 | |
49beb2b8 | 980 | dest.bdev = s->store->cow->bdev; |
71fab00a | 981 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
982 | dest.count = src.count; |
983 | ||
984 | /* Hand over to kcopyd */ | |
eb69aca5 | 985 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
986 | &src, 1, &dest, 0, copy_callback, pe); |
987 | } | |
988 | ||
2913808e MP |
989 | static struct dm_snap_pending_exception * |
990 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
991 | { | |
3510cb94 | 992 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
993 | |
994 | if (!e) | |
995 | return NULL; | |
996 | ||
997 | return container_of(e, struct dm_snap_pending_exception, e); | |
998 | } | |
999 | ||
1da177e4 LT |
1000 | /* |
1001 | * Looks to see if this snapshot already has a pending exception | |
1002 | * for this chunk, otherwise it allocates a new one and inserts | |
1003 | * it into the pending table. | |
1004 | * | |
1005 | * NOTE: a write lock must be held on snap->lock before calling | |
1006 | * this. | |
1007 | */ | |
028867ac | 1008 | static struct dm_snap_pending_exception * |
c6621392 MP |
1009 | __find_pending_exception(struct dm_snapshot *s, |
1010 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1011 | { |
c6621392 | 1012 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1013 | |
2913808e MP |
1014 | pe2 = __lookup_pending_exception(s, chunk); |
1015 | if (pe2) { | |
76df1c65 | 1016 | free_pending_exception(pe); |
2913808e | 1017 | return pe2; |
1da177e4 LT |
1018 | } |
1019 | ||
76df1c65 AK |
1020 | pe->e.old_chunk = chunk; |
1021 | bio_list_init(&pe->origin_bios); | |
1022 | bio_list_init(&pe->snapshot_bios); | |
1023 | pe->primary_pe = NULL; | |
4b832e8d | 1024 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1025 | pe->started = 0; |
1026 | ||
493df71c | 1027 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1028 | free_pending_exception(pe); |
1029 | return NULL; | |
1030 | } | |
1031 | ||
4b832e8d | 1032 | get_pending_exception(pe); |
3510cb94 | 1033 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1034 | |
1da177e4 LT |
1035 | return pe; |
1036 | } | |
1037 | ||
1d4989c8 | 1038 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1039 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1040 | { |
49beb2b8 | 1041 | bio->bi_bdev = s->store->cow->bdev; |
71fab00a JB |
1042 | bio->bi_sector = chunk_to_sector(s->store, |
1043 | dm_chunk_number(e->new_chunk) + | |
1044 | (chunk - e->old_chunk)) + | |
1045 | (bio->bi_sector & | |
1046 | s->store->chunk_mask); | |
1da177e4 LT |
1047 | } |
1048 | ||
1049 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1050 | union map_info *map_context) | |
1051 | { | |
1d4989c8 | 1052 | struct dm_exception *e; |
028867ac | 1053 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1054 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1055 | chunk_t chunk; |
028867ac | 1056 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1057 | |
494b3ee7 MP |
1058 | if (unlikely(bio_empty_barrier(bio))) { |
1059 | bio->bi_bdev = s->store->cow->bdev; | |
1060 | return DM_MAPIO_REMAPPED; | |
1061 | } | |
1062 | ||
71fab00a | 1063 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1064 | |
1065 | /* Full snapshots are not usable */ | |
76df1c65 | 1066 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1067 | if (!s->valid) |
f6a80ea8 | 1068 | return -EIO; |
1da177e4 | 1069 | |
ba40a2aa AK |
1070 | /* FIXME: should only take write lock if we need |
1071 | * to copy an exception */ | |
1072 | down_write(&s->lock); | |
1073 | ||
1074 | if (!s->valid) { | |
1075 | r = -EIO; | |
1076 | goto out_unlock; | |
1077 | } | |
1078 | ||
1079 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1080 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1081 | if (e) { |
d74f81f8 | 1082 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1083 | goto out_unlock; |
1084 | } | |
1085 | ||
1da177e4 LT |
1086 | /* |
1087 | * Write to snapshot - higher level takes care of RW/RO | |
1088 | * flags so we should only get this if we are | |
1089 | * writeable. | |
1090 | */ | |
1091 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1092 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1093 | if (!pe) { |
c6621392 MP |
1094 | up_write(&s->lock); |
1095 | pe = alloc_pending_exception(s); | |
1096 | down_write(&s->lock); | |
1097 | ||
1098 | if (!s->valid) { | |
1099 | free_pending_exception(pe); | |
1100 | r = -EIO; | |
1101 | goto out_unlock; | |
1102 | } | |
1103 | ||
3510cb94 | 1104 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1105 | if (e) { |
1106 | free_pending_exception(pe); | |
1107 | remap_exception(s, e, bio, chunk); | |
1108 | goto out_unlock; | |
1109 | } | |
1110 | ||
c6621392 | 1111 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1112 | if (!pe) { |
1113 | __invalidate_snapshot(s, -ENOMEM); | |
1114 | r = -EIO; | |
1115 | goto out_unlock; | |
1116 | } | |
1da177e4 LT |
1117 | } |
1118 | ||
d74f81f8 | 1119 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1120 | bio_list_add(&pe->snapshot_bios, bio); |
1121 | ||
d2a7ad29 | 1122 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1123 | |
76df1c65 AK |
1124 | if (!pe->started) { |
1125 | /* this is protected by snap->lock */ | |
1126 | pe->started = 1; | |
ba40a2aa | 1127 | up_write(&s->lock); |
76df1c65 | 1128 | start_copy(pe); |
ba40a2aa AK |
1129 | goto out; |
1130 | } | |
cd45daff | 1131 | } else { |
ba40a2aa | 1132 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1133 | map_context->ptr = track_chunk(s, chunk); |
1134 | } | |
1da177e4 | 1135 | |
ba40a2aa AK |
1136 | out_unlock: |
1137 | up_write(&s->lock); | |
1138 | out: | |
1da177e4 LT |
1139 | return r; |
1140 | } | |
1141 | ||
cd45daff MP |
1142 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1143 | int error, union map_info *map_context) | |
1144 | { | |
1145 | struct dm_snapshot *s = ti->private; | |
1146 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1147 | ||
1148 | if (c) | |
1149 | stop_tracking_chunk(s, c); | |
1150 | ||
1151 | return 0; | |
1152 | } | |
1153 | ||
1da177e4 LT |
1154 | static void snapshot_resume(struct dm_target *ti) |
1155 | { | |
028867ac | 1156 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1157 | |
aa14edeb AK |
1158 | down_write(&s->lock); |
1159 | s->active = 1; | |
1160 | up_write(&s->lock); | |
1da177e4 LT |
1161 | } |
1162 | ||
1163 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1164 | char *result, unsigned int maxlen) | |
1165 | { | |
2e4a31df | 1166 | unsigned sz = 0; |
028867ac | 1167 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1168 | |
1169 | switch (type) { | |
1170 | case STATUSTYPE_INFO: | |
94e76572 MP |
1171 | |
1172 | down_write(&snap->lock); | |
1173 | ||
1da177e4 | 1174 | if (!snap->valid) |
2e4a31df | 1175 | DMEMIT("Invalid"); |
1da177e4 | 1176 | else { |
493df71c | 1177 | if (snap->store->type->fraction_full) { |
1da177e4 | 1178 | sector_t numerator, denominator; |
493df71c JB |
1179 | snap->store->type->fraction_full(snap->store, |
1180 | &numerator, | |
1181 | &denominator); | |
2e4a31df JB |
1182 | DMEMIT("%llu/%llu", |
1183 | (unsigned long long)numerator, | |
1184 | (unsigned long long)denominator); | |
1da177e4 LT |
1185 | } |
1186 | else | |
2e4a31df | 1187 | DMEMIT("Unknown"); |
1da177e4 | 1188 | } |
94e76572 MP |
1189 | |
1190 | up_write(&snap->lock); | |
1191 | ||
1da177e4 LT |
1192 | break; |
1193 | ||
1194 | case STATUSTYPE_TABLE: | |
1195 | /* | |
1196 | * kdevname returns a static pointer so we need | |
1197 | * to make private copies if the output is to | |
1198 | * make sense. | |
1199 | */ | |
2e4a31df | 1200 | DMEMIT("%s", snap->origin->name); |
1e302a92 JB |
1201 | snap->store->type->status(snap->store, type, result + sz, |
1202 | maxlen - sz); | |
1da177e4 LT |
1203 | break; |
1204 | } | |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
8811f46c MS |
1209 | static int snapshot_iterate_devices(struct dm_target *ti, |
1210 | iterate_devices_callout_fn fn, void *data) | |
1211 | { | |
1212 | struct dm_snapshot *snap = ti->private; | |
1213 | ||
1214 | return fn(ti, snap->origin, 0, ti->len, data); | |
1215 | } | |
1216 | ||
1217 | ||
1da177e4 LT |
1218 | /*----------------------------------------------------------------- |
1219 | * Origin methods | |
1220 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1221 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1222 | { | |
d2a7ad29 | 1223 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1224 | struct dm_snapshot *snap; |
1d4989c8 | 1225 | struct dm_exception *e; |
028867ac | 1226 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; |
1da177e4 | 1227 | chunk_t chunk; |
eccf0817 | 1228 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1229 | |
1230 | /* Do all the snapshots on this origin */ | |
1231 | list_for_each_entry (snap, snapshots, list) { | |
1232 | ||
76df1c65 AK |
1233 | down_write(&snap->lock); |
1234 | ||
aa14edeb AK |
1235 | /* Only deal with valid and active snapshots */ |
1236 | if (!snap->valid || !snap->active) | |
76df1c65 | 1237 | goto next_snapshot; |
1da177e4 | 1238 | |
d5e404c1 | 1239 | /* Nothing to do if writing beyond end of snapshot */ |
0cea9c78 | 1240 | if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table)) |
76df1c65 | 1241 | goto next_snapshot; |
1da177e4 LT |
1242 | |
1243 | /* | |
1244 | * Remember, different snapshots can have | |
1245 | * different chunk sizes. | |
1246 | */ | |
71fab00a | 1247 | chunk = sector_to_chunk(snap->store, bio->bi_sector); |
1da177e4 LT |
1248 | |
1249 | /* | |
1250 | * Check exception table to see if block | |
1251 | * is already remapped in this snapshot | |
1252 | * and trigger an exception if not. | |
b4b610f6 | 1253 | * |
4b832e8d | 1254 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1255 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 | 1256 | */ |
3510cb94 | 1257 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1258 | if (e) |
1259 | goto next_snapshot; | |
1260 | ||
2913808e | 1261 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1262 | if (!pe) { |
c6621392 MP |
1263 | up_write(&snap->lock); |
1264 | pe = alloc_pending_exception(snap); | |
1265 | down_write(&snap->lock); | |
1266 | ||
1267 | if (!snap->valid) { | |
1268 | free_pending_exception(pe); | |
1269 | goto next_snapshot; | |
1270 | } | |
1271 | ||
3510cb94 | 1272 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1273 | if (e) { |
1274 | free_pending_exception(pe); | |
1275 | goto next_snapshot; | |
1276 | } | |
1277 | ||
c6621392 | 1278 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1279 | if (!pe) { |
1280 | __invalidate_snapshot(snap, -ENOMEM); | |
1281 | goto next_snapshot; | |
1282 | } | |
76df1c65 AK |
1283 | } |
1284 | ||
1285 | if (!primary_pe) { | |
1286 | /* | |
1287 | * Either every pe here has same | |
1288 | * primary_pe or none has one yet. | |
1289 | */ | |
1290 | if (pe->primary_pe) | |
1291 | primary_pe = pe->primary_pe; | |
1292 | else { | |
1293 | primary_pe = pe; | |
1294 | first = 1; | |
1da177e4 | 1295 | } |
76df1c65 AK |
1296 | |
1297 | bio_list_add(&primary_pe->origin_bios, bio); | |
1298 | ||
d2a7ad29 | 1299 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1300 | } |
1301 | ||
1302 | if (!pe->primary_pe) { | |
76df1c65 | 1303 | pe->primary_pe = primary_pe; |
4b832e8d | 1304 | get_pending_exception(primary_pe); |
76df1c65 AK |
1305 | } |
1306 | ||
1307 | if (!pe->started) { | |
1308 | pe->started = 1; | |
1309 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1310 | } |
1311 | ||
76df1c65 | 1312 | next_snapshot: |
1da177e4 LT |
1313 | up_write(&snap->lock); |
1314 | } | |
1315 | ||
b4b610f6 | 1316 | if (!primary_pe) |
4b832e8d | 1317 | return r; |
b4b610f6 AK |
1318 | |
1319 | /* | |
1320 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1321 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1322 | * got completed while we were in the loop above, so it falls to |
1323 | * us here to remove the primary_pe and submit any origin_bios. | |
1324 | */ | |
1325 | ||
4b832e8d | 1326 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1327 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1328 | free_pending_exception(primary_pe); | |
1329 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1330 | return r; |
b4b610f6 AK |
1331 | } |
1332 | ||
1da177e4 LT |
1333 | /* |
1334 | * Now that we have a complete pe list we can start the copying. | |
1335 | */ | |
eccf0817 AK |
1336 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1337 | start_copy(pe); | |
1da177e4 LT |
1338 | |
1339 | return r; | |
1340 | } | |
1341 | ||
1342 | /* | |
1343 | * Called on a write from the origin driver. | |
1344 | */ | |
1345 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1346 | { | |
1347 | struct origin *o; | |
d2a7ad29 | 1348 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1349 | |
1350 | down_read(&_origins_lock); | |
1351 | o = __lookup_origin(origin->bdev); | |
1352 | if (o) | |
1353 | r = __origin_write(&o->snapshots, bio); | |
1354 | up_read(&_origins_lock); | |
1355 | ||
1356 | return r; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1361 | */ | |
1362 | ||
1363 | /* | |
1364 | * Construct an origin mapping: <dev_path> | |
1365 | * The context for an origin is merely a 'struct dm_dev *' | |
1366 | * pointing to the real device. | |
1367 | */ | |
1368 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1369 | { | |
1370 | int r; | |
1371 | struct dm_dev *dev; | |
1372 | ||
1373 | if (argc != 1) { | |
72d94861 | 1374 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1375 | return -EINVAL; |
1376 | } | |
1377 | ||
1378 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1379 | dm_table_get_mode(ti->table), &dev); | |
1380 | if (r) { | |
1381 | ti->error = "Cannot get target device"; | |
1382 | return r; | |
1383 | } | |
1384 | ||
1385 | ti->private = dev; | |
494b3ee7 MP |
1386 | ti->num_flush_requests = 1; |
1387 | ||
1da177e4 LT |
1388 | return 0; |
1389 | } | |
1390 | ||
1391 | static void origin_dtr(struct dm_target *ti) | |
1392 | { | |
028867ac | 1393 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1394 | dm_put_device(ti, dev); |
1395 | } | |
1396 | ||
1397 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1398 | union map_info *map_context) | |
1399 | { | |
028867ac | 1400 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1401 | bio->bi_bdev = dev->bdev; |
1402 | ||
494b3ee7 MP |
1403 | if (unlikely(bio_empty_barrier(bio))) |
1404 | return DM_MAPIO_REMAPPED; | |
1405 | ||
1da177e4 | 1406 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 1407 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1408 | } |
1409 | ||
1da177e4 LT |
1410 | /* |
1411 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1412 | * chunk sizes. | |
1413 | */ | |
1414 | static void origin_resume(struct dm_target *ti) | |
1415 | { | |
028867ac | 1416 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1417 | |
1418 | down_read(&_origins_lock); | |
1da177e4 | 1419 | |
7e201b35 MP |
1420 | ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev)); |
1421 | ||
1422 | up_read(&_origins_lock); | |
1da177e4 LT |
1423 | } |
1424 | ||
1425 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1426 | unsigned int maxlen) | |
1427 | { | |
028867ac | 1428 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1429 | |
1430 | switch (type) { | |
1431 | case STATUSTYPE_INFO: | |
1432 | result[0] = '\0'; | |
1433 | break; | |
1434 | ||
1435 | case STATUSTYPE_TABLE: | |
1436 | snprintf(result, maxlen, "%s", dev->name); | |
1437 | break; | |
1438 | } | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | ||
8811f46c MS |
1443 | static int origin_iterate_devices(struct dm_target *ti, |
1444 | iterate_devices_callout_fn fn, void *data) | |
1445 | { | |
1446 | struct dm_dev *dev = ti->private; | |
1447 | ||
1448 | return fn(ti, dev, 0, ti->len, data); | |
1449 | } | |
1450 | ||
1da177e4 LT |
1451 | static struct target_type origin_target = { |
1452 | .name = "snapshot-origin", | |
8811f46c | 1453 | .version = {1, 7, 0}, |
1da177e4 LT |
1454 | .module = THIS_MODULE, |
1455 | .ctr = origin_ctr, | |
1456 | .dtr = origin_dtr, | |
1457 | .map = origin_map, | |
1458 | .resume = origin_resume, | |
1459 | .status = origin_status, | |
8811f46c | 1460 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
1461 | }; |
1462 | ||
1463 | static struct target_type snapshot_target = { | |
1464 | .name = "snapshot", | |
8811f46c | 1465 | .version = {1, 7, 0}, |
1da177e4 LT |
1466 | .module = THIS_MODULE, |
1467 | .ctr = snapshot_ctr, | |
1468 | .dtr = snapshot_dtr, | |
1469 | .map = snapshot_map, | |
cd45daff | 1470 | .end_io = snapshot_end_io, |
1da177e4 LT |
1471 | .resume = snapshot_resume, |
1472 | .status = snapshot_status, | |
8811f46c | 1473 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
1474 | }; |
1475 | ||
1476 | static int __init dm_snapshot_init(void) | |
1477 | { | |
1478 | int r; | |
1479 | ||
4db6bfe0 AK |
1480 | r = dm_exception_store_init(); |
1481 | if (r) { | |
1482 | DMERR("Failed to initialize exception stores"); | |
1483 | return r; | |
1484 | } | |
1485 | ||
1da177e4 LT |
1486 | r = dm_register_target(&snapshot_target); |
1487 | if (r) { | |
1488 | DMERR("snapshot target register failed %d", r); | |
034a186d | 1489 | goto bad_register_snapshot_target; |
1da177e4 LT |
1490 | } |
1491 | ||
1492 | r = dm_register_target(&origin_target); | |
1493 | if (r < 0) { | |
72d94861 | 1494 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1495 | goto bad1; |
1496 | } | |
1497 | ||
1498 | r = init_origin_hash(); | |
1499 | if (r) { | |
1500 | DMERR("init_origin_hash failed."); | |
1501 | goto bad2; | |
1502 | } | |
1503 | ||
1d4989c8 | 1504 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
1505 | if (!exception_cache) { |
1506 | DMERR("Couldn't create exception cache."); | |
1507 | r = -ENOMEM; | |
1508 | goto bad3; | |
1509 | } | |
1510 | ||
028867ac | 1511 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1512 | if (!pending_cache) { |
1513 | DMERR("Couldn't create pending cache."); | |
1514 | r = -ENOMEM; | |
1515 | goto bad4; | |
1516 | } | |
1517 | ||
cd45daff MP |
1518 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1519 | if (!tracked_chunk_cache) { | |
1520 | DMERR("Couldn't create cache to track chunks in use."); | |
1521 | r = -ENOMEM; | |
1522 | goto bad5; | |
1523 | } | |
1524 | ||
ca3a931f AK |
1525 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1526 | if (!ksnapd) { | |
1527 | DMERR("Failed to create ksnapd workqueue."); | |
1528 | r = -ENOMEM; | |
92e86812 | 1529 | goto bad_pending_pool; |
ca3a931f AK |
1530 | } |
1531 | ||
1da177e4 LT |
1532 | return 0; |
1533 | ||
4db6bfe0 | 1534 | bad_pending_pool: |
cd45daff | 1535 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1536 | bad5: |
1da177e4 | 1537 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1538 | bad4: |
1da177e4 | 1539 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1540 | bad3: |
1da177e4 | 1541 | exit_origin_hash(); |
4db6bfe0 | 1542 | bad2: |
1da177e4 | 1543 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1544 | bad1: |
1da177e4 | 1545 | dm_unregister_target(&snapshot_target); |
034a186d JB |
1546 | |
1547 | bad_register_snapshot_target: | |
1548 | dm_exception_store_exit(); | |
1da177e4 LT |
1549 | return r; |
1550 | } | |
1551 | ||
1552 | static void __exit dm_snapshot_exit(void) | |
1553 | { | |
ca3a931f AK |
1554 | destroy_workqueue(ksnapd); |
1555 | ||
10d3bd09 MP |
1556 | dm_unregister_target(&snapshot_target); |
1557 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1558 | |
1559 | exit_origin_hash(); | |
1da177e4 LT |
1560 | kmem_cache_destroy(pending_cache); |
1561 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1562 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1563 | |
1564 | dm_exception_store_exit(); | |
1da177e4 LT |
1565 | } |
1566 | ||
1567 | /* Module hooks */ | |
1568 | module_init(dm_snapshot_init); | |
1569 | module_exit(dm_snapshot_exit); | |
1570 | ||
1571 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1572 | MODULE_AUTHOR("Joe Thornber"); | |
1573 | MODULE_LICENSE("GPL"); |