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> | |
90fa1527 | 12 | #include <linux/delay.h> |
1da177e4 LT |
13 | #include <linux/fs.h> |
14 | #include <linux/init.h> | |
15 | #include <linux/kdev_t.h> | |
16 | #include <linux/list.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/vmalloc.h> | |
6f3c3f0a | 21 | #include <linux/log2.h> |
a765e20e | 22 | #include <linux/dm-kcopyd.h> |
ccc45ea8 | 23 | #include <linux/workqueue.h> |
1da177e4 | 24 | |
aea53d92 | 25 | #include "dm-exception-store.h" |
1da177e4 | 26 | #include "dm-bio-list.h" |
1da177e4 | 27 | |
72d94861 AK |
28 | #define DM_MSG_PREFIX "snapshots" |
29 | ||
1da177e4 LT |
30 | /* |
31 | * The percentage increment we will wake up users at | |
32 | */ | |
33 | #define WAKE_UP_PERCENT 5 | |
34 | ||
35 | /* | |
36 | * kcopyd priority of snapshot operations | |
37 | */ | |
38 | #define SNAPSHOT_COPY_PRIORITY 2 | |
39 | ||
40 | /* | |
8ee2767a | 41 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 42 | */ |
8ee2767a | 43 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 44 | |
cd45daff MP |
45 | /* |
46 | * The size of the mempool used to track chunks in use. | |
47 | */ | |
48 | #define MIN_IOS 256 | |
49 | ||
ccc45ea8 JB |
50 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
51 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
52 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
53 | ||
54 | struct exception_table { | |
55 | uint32_t hash_mask; | |
56 | unsigned hash_shift; | |
57 | struct list_head *table; | |
58 | }; | |
59 | ||
60 | struct dm_snapshot { | |
61 | struct rw_semaphore lock; | |
62 | ||
63 | struct dm_dev *origin; | |
64 | ||
65 | /* List of snapshots per Origin */ | |
66 | struct list_head list; | |
67 | ||
68 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
69 | int valid; | |
70 | ||
71 | /* Origin writes don't trigger exceptions until this is set */ | |
72 | int active; | |
73 | ||
74 | /* Used for display of table */ | |
75 | char type; | |
76 | ||
77 | mempool_t *pending_pool; | |
78 | ||
79 | atomic_t pending_exceptions_count; | |
80 | ||
81 | struct exception_table pending; | |
82 | struct exception_table complete; | |
83 | ||
84 | /* | |
85 | * pe_lock protects all pending_exception operations and access | |
86 | * as well as the snapshot_bios list. | |
87 | */ | |
88 | spinlock_t pe_lock; | |
89 | ||
90 | /* The on disk metadata handler */ | |
91 | struct dm_exception_store *store; | |
92 | ||
93 | struct dm_kcopyd_client *kcopyd_client; | |
94 | ||
95 | /* Queue of snapshot writes for ksnapd to flush */ | |
96 | struct bio_list queued_bios; | |
97 | struct work_struct queued_bios_work; | |
98 | ||
99 | /* Chunks with outstanding reads */ | |
100 | mempool_t *tracked_chunk_pool; | |
101 | spinlock_t tracked_chunk_lock; | |
102 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
103 | }; | |
104 | ||
c642f9e0 | 105 | static struct workqueue_struct *ksnapd; |
c4028958 | 106 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 107 | |
ccc45ea8 JB |
108 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
109 | chunk_t chunk) | |
110 | { | |
111 | return chunk << store->chunk_shift; | |
112 | } | |
113 | ||
114 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
115 | { | |
116 | /* | |
117 | * There is only ever one instance of a particular block | |
118 | * device so we can compare pointers safely. | |
119 | */ | |
120 | return lhs == rhs; | |
121 | } | |
122 | ||
028867ac AK |
123 | struct dm_snap_pending_exception { |
124 | struct dm_snap_exception e; | |
1da177e4 LT |
125 | |
126 | /* | |
127 | * Origin buffers waiting for this to complete are held | |
128 | * in a bio list | |
129 | */ | |
130 | struct bio_list origin_bios; | |
131 | struct bio_list snapshot_bios; | |
132 | ||
eccf0817 AK |
133 | /* |
134 | * Short-term queue of pending exceptions prior to submission. | |
135 | */ | |
136 | struct list_head list; | |
137 | ||
1da177e4 | 138 | /* |
b4b610f6 | 139 | * The primary pending_exception is the one that holds |
4b832e8d | 140 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
141 | * group of pending_exceptions. It is always last to get freed. |
142 | * These fields get set up when writing to the origin. | |
1da177e4 | 143 | */ |
028867ac | 144 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
145 | |
146 | /* | |
147 | * Number of pending_exceptions processing this chunk. | |
148 | * When this drops to zero we must complete the origin bios. | |
149 | * If incrementing or decrementing this, hold pe->snap->lock for | |
150 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
151 | * they are the same. | |
152 | */ | |
4b832e8d | 153 | atomic_t ref_count; |
1da177e4 LT |
154 | |
155 | /* Pointer back to snapshot context */ | |
156 | struct dm_snapshot *snap; | |
157 | ||
158 | /* | |
159 | * 1 indicates the exception has already been sent to | |
160 | * kcopyd. | |
161 | */ | |
162 | int started; | |
163 | }; | |
164 | ||
165 | /* | |
166 | * Hash table mapping origin volumes to lists of snapshots and | |
167 | * a lock to protect it | |
168 | */ | |
e18b890b CL |
169 | static struct kmem_cache *exception_cache; |
170 | static struct kmem_cache *pending_cache; | |
1da177e4 | 171 | |
cd45daff MP |
172 | struct dm_snap_tracked_chunk { |
173 | struct hlist_node node; | |
174 | chunk_t chunk; | |
175 | }; | |
176 | ||
177 | static struct kmem_cache *tracked_chunk_cache; | |
178 | ||
179 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
180 | chunk_t chunk) | |
181 | { | |
182 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
183 | GFP_NOIO); | |
184 | unsigned long flags; | |
185 | ||
186 | c->chunk = chunk; | |
187 | ||
188 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
189 | hlist_add_head(&c->node, | |
190 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
191 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
192 | ||
193 | return c; | |
194 | } | |
195 | ||
196 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
197 | struct dm_snap_tracked_chunk *c) | |
198 | { | |
199 | unsigned long flags; | |
200 | ||
201 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
202 | hlist_del(&c->node); | |
203 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
204 | ||
205 | mempool_free(c, s->tracked_chunk_pool); | |
206 | } | |
207 | ||
a8d41b59 MP |
208 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
209 | { | |
210 | struct dm_snap_tracked_chunk *c; | |
211 | struct hlist_node *hn; | |
212 | int found = 0; | |
213 | ||
214 | spin_lock_irq(&s->tracked_chunk_lock); | |
215 | ||
216 | hlist_for_each_entry(c, hn, | |
217 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
218 | if (c->chunk == chunk) { | |
219 | found = 1; | |
220 | break; | |
221 | } | |
222 | } | |
223 | ||
224 | spin_unlock_irq(&s->tracked_chunk_lock); | |
225 | ||
226 | return found; | |
227 | } | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * One of these per registered origin, held in the snapshot_origins hash | |
231 | */ | |
232 | struct origin { | |
233 | /* The origin device */ | |
234 | struct block_device *bdev; | |
235 | ||
236 | struct list_head hash_list; | |
237 | ||
238 | /* List of snapshots for this origin */ | |
239 | struct list_head snapshots; | |
240 | }; | |
241 | ||
242 | /* | |
243 | * Size of the hash table for origin volumes. If we make this | |
244 | * the size of the minors list then it should be nearly perfect | |
245 | */ | |
246 | #define ORIGIN_HASH_SIZE 256 | |
247 | #define ORIGIN_MASK 0xFF | |
248 | static struct list_head *_origins; | |
249 | static struct rw_semaphore _origins_lock; | |
250 | ||
251 | static int init_origin_hash(void) | |
252 | { | |
253 | int i; | |
254 | ||
255 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
256 | GFP_KERNEL); | |
257 | if (!_origins) { | |
72d94861 | 258 | DMERR("unable to allocate memory"); |
1da177e4 LT |
259 | return -ENOMEM; |
260 | } | |
261 | ||
262 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
263 | INIT_LIST_HEAD(_origins + i); | |
264 | init_rwsem(&_origins_lock); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | static void exit_origin_hash(void) | |
270 | { | |
271 | kfree(_origins); | |
272 | } | |
273 | ||
028867ac | 274 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
275 | { |
276 | return bdev->bd_dev & ORIGIN_MASK; | |
277 | } | |
278 | ||
279 | static struct origin *__lookup_origin(struct block_device *origin) | |
280 | { | |
281 | struct list_head *ol; | |
282 | struct origin *o; | |
283 | ||
284 | ol = &_origins[origin_hash(origin)]; | |
285 | list_for_each_entry (o, ol, hash_list) | |
286 | if (bdev_equal(o->bdev, origin)) | |
287 | return o; | |
288 | ||
289 | return NULL; | |
290 | } | |
291 | ||
292 | static void __insert_origin(struct origin *o) | |
293 | { | |
294 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
295 | list_add_tail(&o->hash_list, sl); | |
296 | } | |
297 | ||
298 | /* | |
299 | * Make a note of the snapshot and its origin so we can look it | |
300 | * up when the origin has a write on it. | |
301 | */ | |
302 | static int register_snapshot(struct dm_snapshot *snap) | |
303 | { | |
60c856c8 | 304 | struct origin *o, *new_o; |
1da177e4 LT |
305 | struct block_device *bdev = snap->origin->bdev; |
306 | ||
60c856c8 MP |
307 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
308 | if (!new_o) | |
309 | return -ENOMEM; | |
310 | ||
1da177e4 LT |
311 | down_write(&_origins_lock); |
312 | o = __lookup_origin(bdev); | |
313 | ||
60c856c8 MP |
314 | if (o) |
315 | kfree(new_o); | |
316 | else { | |
1da177e4 | 317 | /* New origin */ |
60c856c8 | 318 | o = new_o; |
1da177e4 LT |
319 | |
320 | /* Initialise the struct */ | |
321 | INIT_LIST_HEAD(&o->snapshots); | |
322 | o->bdev = bdev; | |
323 | ||
324 | __insert_origin(o); | |
325 | } | |
326 | ||
327 | list_add_tail(&snap->list, &o->snapshots); | |
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 | */ |
d74f81f8 MB |
354 | static int init_exception_table(struct exception_table *et, uint32_t size, |
355 | 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 | ||
e18b890b | 371 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
372 | { |
373 | struct list_head *slot; | |
028867ac | 374 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
375 | int i, size; |
376 | ||
377 | size = et->hash_mask + 1; | |
378 | for (i = 0; i < size; i++) { | |
379 | slot = et->table + i; | |
380 | ||
381 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
382 | kmem_cache_free(mem, ex); | |
383 | } | |
384 | ||
385 | vfree(et->table); | |
386 | } | |
387 | ||
028867ac | 388 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 389 | { |
d74f81f8 | 390 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
391 | } |
392 | ||
028867ac AK |
393 | static void insert_exception(struct exception_table *eh, |
394 | struct dm_snap_exception *e) | |
1da177e4 LT |
395 | { |
396 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
397 | list_add(&e->hash_list, l); | |
398 | } | |
399 | ||
028867ac | 400 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
401 | { |
402 | list_del(&e->hash_list); | |
403 | } | |
404 | ||
405 | /* | |
406 | * Return the exception data for a sector, or NULL if not | |
407 | * remapped. | |
408 | */ | |
028867ac AK |
409 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
410 | chunk_t chunk) | |
1da177e4 LT |
411 | { |
412 | struct list_head *slot; | |
028867ac | 413 | struct dm_snap_exception *e; |
1da177e4 LT |
414 | |
415 | slot = &et->table[exception_hash(et, chunk)]; | |
416 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
417 | if (chunk >= e->old_chunk && |
418 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
419 | return e; |
420 | ||
421 | return NULL; | |
422 | } | |
423 | ||
028867ac | 424 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 425 | { |
028867ac | 426 | struct dm_snap_exception *e; |
1da177e4 LT |
427 | |
428 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
429 | if (!e) | |
430 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
431 | ||
432 | return e; | |
433 | } | |
434 | ||
028867ac | 435 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
436 | { |
437 | kmem_cache_free(exception_cache, e); | |
438 | } | |
439 | ||
92e86812 | 440 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 441 | { |
92e86812 MP |
442 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
443 | GFP_NOIO); | |
444 | ||
879129d2 | 445 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
446 | pe->snap = s; |
447 | ||
448 | return pe; | |
1da177e4 LT |
449 | } |
450 | ||
028867ac | 451 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 452 | { |
879129d2 MP |
453 | struct dm_snapshot *s = pe->snap; |
454 | ||
455 | mempool_free(pe, s->pending_pool); | |
456 | smp_mb__before_atomic_dec(); | |
457 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
458 | } |
459 | ||
d74f81f8 MB |
460 | static void insert_completed_exception(struct dm_snapshot *s, |
461 | struct dm_snap_exception *new_e) | |
462 | { | |
463 | struct exception_table *eh = &s->complete; | |
464 | struct list_head *l; | |
465 | struct dm_snap_exception *e = NULL; | |
466 | ||
467 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
468 | ||
469 | /* Add immediately if this table doesn't support consecutive chunks */ | |
470 | if (!eh->hash_shift) | |
471 | goto out; | |
472 | ||
473 | /* List is ordered by old_chunk */ | |
474 | list_for_each_entry_reverse(e, l, hash_list) { | |
475 | /* Insert after an existing chunk? */ | |
476 | if (new_e->old_chunk == (e->old_chunk + | |
477 | dm_consecutive_chunk_count(e) + 1) && | |
478 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
479 | dm_consecutive_chunk_count(e) + 1)) { | |
480 | dm_consecutive_chunk_count_inc(e); | |
481 | free_exception(new_e); | |
482 | return; | |
483 | } | |
484 | ||
485 | /* Insert before an existing chunk? */ | |
486 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
487 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
488 | dm_consecutive_chunk_count_inc(e); | |
489 | e->old_chunk--; | |
490 | e->new_chunk--; | |
491 | free_exception(new_e); | |
492 | return; | |
493 | } | |
494 | ||
495 | if (new_e->old_chunk > e->old_chunk) | |
496 | break; | |
497 | } | |
498 | ||
499 | out: | |
500 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
501 | } | |
502 | ||
a159c1ac JB |
503 | /* |
504 | * Callback used by the exception stores to load exceptions when | |
505 | * initialising. | |
506 | */ | |
507 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 508 | { |
a159c1ac | 509 | struct dm_snapshot *s = context; |
028867ac | 510 | struct dm_snap_exception *e; |
1da177e4 LT |
511 | |
512 | e = alloc_exception(); | |
513 | if (!e) | |
514 | return -ENOMEM; | |
515 | ||
516 | e->old_chunk = old; | |
d74f81f8 MB |
517 | |
518 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 519 | e->new_chunk = new; |
d74f81f8 MB |
520 | |
521 | insert_completed_exception(s, e); | |
522 | ||
1da177e4 LT |
523 | return 0; |
524 | } | |
525 | ||
526 | /* | |
527 | * Hard coded magic. | |
528 | */ | |
529 | static int calc_max_buckets(void) | |
530 | { | |
531 | /* use a fixed size of 2MB */ | |
532 | unsigned long mem = 2 * 1024 * 1024; | |
533 | mem /= sizeof(struct list_head); | |
534 | ||
535 | return mem; | |
536 | } | |
537 | ||
1da177e4 LT |
538 | /* |
539 | * Allocate room for a suitable hash table. | |
540 | */ | |
49beb2b8 JB |
541 | static int init_hash_tables(struct dm_snapshot *s, chunk_t chunk_shift, |
542 | struct dm_dev *cow) | |
1da177e4 LT |
543 | { |
544 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
545 | ||
546 | /* | |
547 | * Calculate based on the size of the original volume or | |
548 | * the COW volume... | |
549 | */ | |
49beb2b8 | 550 | cow_dev_size = get_dev_size(cow->bdev); |
1da177e4 LT |
551 | origin_dev_size = get_dev_size(s->origin->bdev); |
552 | max_buckets = calc_max_buckets(); | |
553 | ||
d0216849 | 554 | hash_size = min(origin_dev_size, cow_dev_size) >> chunk_shift; |
1da177e4 LT |
555 | hash_size = min(hash_size, max_buckets); |
556 | ||
8defd830 | 557 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
558 | if (init_exception_table(&s->complete, hash_size, |
559 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
560 | return -ENOMEM; |
561 | ||
562 | /* | |
563 | * Allocate hash table for in-flight exceptions | |
564 | * Make this smaller than the real hash table | |
565 | */ | |
566 | hash_size >>= 3; | |
567 | if (hash_size < 64) | |
568 | hash_size = 64; | |
569 | ||
d74f81f8 | 570 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
571 | exit_exception_table(&s->complete, exception_cache); |
572 | return -ENOMEM; | |
573 | } | |
574 | ||
575 | return 0; | |
576 | } | |
577 | ||
578 | /* | |
579 | * Round a number up to the nearest 'size' boundary. size must | |
580 | * be a power of 2. | |
581 | */ | |
028867ac | 582 | static ulong round_up(ulong n, ulong size) |
1da177e4 LT |
583 | { |
584 | size--; | |
585 | return (n + size) & ~size; | |
586 | } | |
587 | ||
4c7e3bf4 | 588 | static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg, |
d0216849 | 589 | chunk_t *chunk_size, chunk_t *chunk_mask, |
49beb2b8 JB |
590 | chunk_t *chunk_shift, struct dm_dev *cow, |
591 | char **error) | |
4c7e3bf4 | 592 | { |
d0216849 | 593 | unsigned long chunk_size_ulong; |
4c7e3bf4 MM |
594 | char *value; |
595 | ||
d0216849 | 596 | chunk_size_ulong = simple_strtoul(chunk_size_arg, &value, 10); |
4c7e3bf4 MM |
597 | if (*chunk_size_arg == '\0' || *value != '\0') { |
598 | *error = "Invalid chunk size"; | |
599 | return -EINVAL; | |
600 | } | |
601 | ||
d0216849 JB |
602 | if (!chunk_size_ulong) { |
603 | *chunk_size = *chunk_mask = *chunk_shift = 0; | |
4c7e3bf4 MM |
604 | return 0; |
605 | } | |
606 | ||
607 | /* | |
608 | * Chunk size must be multiple of page size. Silently | |
609 | * round up if it's not. | |
610 | */ | |
d0216849 | 611 | chunk_size_ulong = round_up(chunk_size_ulong, PAGE_SIZE >> 9); |
4c7e3bf4 MM |
612 | |
613 | /* Check chunk_size is a power of 2 */ | |
d0216849 | 614 | if (!is_power_of_2(chunk_size_ulong)) { |
4c7e3bf4 MM |
615 | *error = "Chunk size is not a power of 2"; |
616 | return -EINVAL; | |
617 | } | |
618 | ||
619 | /* Validate the chunk size against the device block size */ | |
49beb2b8 | 620 | if (chunk_size_ulong % (bdev_hardsect_size(cow->bdev) >> 9)) { |
4c7e3bf4 MM |
621 | *error = "Chunk size is not a multiple of device blocksize"; |
622 | return -EINVAL; | |
623 | } | |
624 | ||
d0216849 JB |
625 | *chunk_size = chunk_size_ulong; |
626 | *chunk_mask = chunk_size_ulong - 1; | |
627 | *chunk_shift = ffs(chunk_size_ulong) - 1; | |
4c7e3bf4 MM |
628 | |
629 | return 0; | |
630 | } | |
631 | ||
1da177e4 LT |
632 | /* |
633 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
634 | */ | |
635 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
636 | { | |
637 | struct dm_snapshot *s; | |
cd45daff | 638 | int i; |
1da177e4 LT |
639 | int r = -EINVAL; |
640 | char persistent; | |
641 | char *origin_path; | |
642 | char *cow_path; | |
d0216849 | 643 | chunk_t chunk_size, chunk_mask, chunk_shift; |
49beb2b8 | 644 | struct dm_dev *cow; |
1da177e4 | 645 | |
4c7e3bf4 | 646 | if (argc != 4) { |
72d94861 | 647 | ti->error = "requires exactly 4 arguments"; |
1da177e4 LT |
648 | r = -EINVAL; |
649 | goto bad1; | |
650 | } | |
651 | ||
652 | origin_path = argv[0]; | |
653 | cow_path = argv[1]; | |
654 | persistent = toupper(*argv[2]); | |
655 | ||
656 | if (persistent != 'P' && persistent != 'N') { | |
657 | ti->error = "Persistent flag is not P or N"; | |
658 | r = -EINVAL; | |
659 | goto bad1; | |
660 | } | |
661 | ||
1da177e4 LT |
662 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
663 | if (s == NULL) { | |
664 | ti->error = "Cannot allocate snapshot context private " | |
665 | "structure"; | |
666 | r = -ENOMEM; | |
667 | goto bad1; | |
668 | } | |
669 | ||
670 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
671 | if (r) { | |
672 | ti->error = "Cannot get origin device"; | |
673 | goto bad2; | |
674 | } | |
675 | ||
676 | r = dm_get_device(ti, cow_path, 0, 0, | |
49beb2b8 | 677 | FMODE_READ | FMODE_WRITE, &cow); |
1da177e4 LT |
678 | if (r) { |
679 | dm_put_device(ti, s->origin); | |
680 | ti->error = "Cannot get COW device"; | |
681 | goto bad2; | |
682 | } | |
683 | ||
d0216849 | 684 | r = set_chunk_size(s, argv[3], &chunk_size, &chunk_mask, &chunk_shift, |
49beb2b8 | 685 | cow, &ti->error); |
4c7e3bf4 | 686 | if (r) |
1da177e4 | 687 | goto bad3; |
1da177e4 | 688 | |
1da177e4 | 689 | s->valid = 1; |
aa14edeb | 690 | s->active = 0; |
879129d2 | 691 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 692 | init_rwsem(&s->lock); |
ca3a931f | 693 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
694 | |
695 | /* Allocate hash table for COW data */ | |
49beb2b8 | 696 | if (init_hash_tables(s, chunk_shift, cow)) { |
1da177e4 LT |
697 | ti->error = "Unable to allocate hash table space"; |
698 | r = -ENOMEM; | |
699 | goto bad3; | |
700 | } | |
701 | ||
d0216849 | 702 | r = dm_exception_store_create(argv[2], ti, chunk_size, chunk_mask, |
49beb2b8 | 703 | chunk_shift, cow, &s->store); |
1da177e4 LT |
704 | if (r) { |
705 | ti->error = "Couldn't create exception store"; | |
706 | r = -EINVAL; | |
707 | goto bad4; | |
708 | } | |
709 | ||
eb69aca5 | 710 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
711 | if (r) { |
712 | ti->error = "Could not create kcopyd client"; | |
713 | goto bad5; | |
714 | } | |
715 | ||
92e86812 MP |
716 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
717 | if (!s->pending_pool) { | |
718 | ti->error = "Could not allocate mempool for pending exceptions"; | |
719 | goto bad6; | |
720 | } | |
721 | ||
cd45daff MP |
722 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
723 | tracked_chunk_cache); | |
724 | if (!s->tracked_chunk_pool) { | |
725 | ti->error = "Could not allocate tracked_chunk mempool for " | |
726 | "tracking reads"; | |
92e86812 | 727 | goto bad_tracked_chunk_pool; |
cd45daff MP |
728 | } |
729 | ||
730 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
731 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
732 | ||
733 | spin_lock_init(&s->tracked_chunk_lock); | |
734 | ||
aa14edeb | 735 | /* Metadata must only be loaded into one table at once */ |
493df71c JB |
736 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
737 | (void *)s); | |
0764147b | 738 | if (r < 0) { |
f9cea4f7 | 739 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 740 | goto bad_load_and_register; |
0764147b MB |
741 | } else if (r > 0) { |
742 | s->valid = 0; | |
743 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 744 | } |
aa14edeb | 745 | |
ca3a931f | 746 | bio_list_init(&s->queued_bios); |
c4028958 | 747 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 748 | |
1da177e4 | 749 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 750 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
751 | if (register_snapshot(s)) { |
752 | r = -EINVAL; | |
753 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 754 | goto bad_load_and_register; |
1da177e4 LT |
755 | } |
756 | ||
757 | ti->private = s; | |
d0216849 | 758 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
759 | |
760 | return 0; | |
761 | ||
cd45daff MP |
762 | bad_load_and_register: |
763 | mempool_destroy(s->tracked_chunk_pool); | |
764 | ||
92e86812 MP |
765 | bad_tracked_chunk_pool: |
766 | mempool_destroy(s->pending_pool); | |
767 | ||
1da177e4 | 768 | bad6: |
eb69aca5 | 769 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 LT |
770 | |
771 | bad5: | |
493df71c | 772 | s->store->type->dtr(s->store); |
1da177e4 LT |
773 | |
774 | bad4: | |
775 | exit_exception_table(&s->pending, pending_cache); | |
776 | exit_exception_table(&s->complete, exception_cache); | |
777 | ||
778 | bad3: | |
49beb2b8 | 779 | dm_put_device(ti, cow); |
1da177e4 LT |
780 | dm_put_device(ti, s->origin); |
781 | ||
782 | bad2: | |
783 | kfree(s); | |
784 | ||
785 | bad1: | |
786 | return r; | |
787 | } | |
788 | ||
31c93a0c MB |
789 | static void __free_exceptions(struct dm_snapshot *s) |
790 | { | |
eb69aca5 | 791 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
792 | s->kcopyd_client = NULL; |
793 | ||
794 | exit_exception_table(&s->pending, pending_cache); | |
795 | exit_exception_table(&s->complete, exception_cache); | |
796 | ||
493df71c | 797 | s->store->type->dtr(s->store); |
31c93a0c MB |
798 | } |
799 | ||
1da177e4 LT |
800 | static void snapshot_dtr(struct dm_target *ti) |
801 | { | |
cd45daff MP |
802 | #ifdef CONFIG_DM_DEBUG |
803 | int i; | |
804 | #endif | |
028867ac | 805 | struct dm_snapshot *s = ti->private; |
49beb2b8 | 806 | struct dm_dev *cow = s->store->cow; |
1da177e4 | 807 | |
ca3a931f AK |
808 | flush_workqueue(ksnapd); |
809 | ||
138728dc AK |
810 | /* Prevent further origin writes from using this snapshot. */ |
811 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
812 | unregister_snapshot(s); |
813 | ||
879129d2 | 814 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 815 | msleep(1); |
879129d2 MP |
816 | /* |
817 | * Ensure instructions in mempool_destroy aren't reordered | |
818 | * before atomic_read. | |
819 | */ | |
820 | smp_mb(); | |
821 | ||
cd45daff MP |
822 | #ifdef CONFIG_DM_DEBUG |
823 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
824 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
825 | #endif | |
826 | ||
827 | mempool_destroy(s->tracked_chunk_pool); | |
828 | ||
31c93a0c | 829 | __free_exceptions(s); |
1da177e4 | 830 | |
92e86812 MP |
831 | mempool_destroy(s->pending_pool); |
832 | ||
1da177e4 | 833 | dm_put_device(ti, s->origin); |
49beb2b8 | 834 | dm_put_device(ti, cow); |
138728dc | 835 | |
1da177e4 LT |
836 | kfree(s); |
837 | } | |
838 | ||
839 | /* | |
840 | * Flush a list of buffers. | |
841 | */ | |
842 | static void flush_bios(struct bio *bio) | |
843 | { | |
844 | struct bio *n; | |
845 | ||
846 | while (bio) { | |
847 | n = bio->bi_next; | |
848 | bio->bi_next = NULL; | |
849 | generic_make_request(bio); | |
850 | bio = n; | |
851 | } | |
852 | } | |
853 | ||
c4028958 | 854 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 855 | { |
c4028958 DH |
856 | struct dm_snapshot *s = |
857 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
858 | struct bio *queued_bios; |
859 | unsigned long flags; | |
860 | ||
861 | spin_lock_irqsave(&s->pe_lock, flags); | |
862 | queued_bios = bio_list_get(&s->queued_bios); | |
863 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
864 | ||
865 | flush_bios(queued_bios); | |
866 | } | |
867 | ||
1da177e4 LT |
868 | /* |
869 | * Error a list of buffers. | |
870 | */ | |
871 | static void error_bios(struct bio *bio) | |
872 | { | |
873 | struct bio *n; | |
874 | ||
875 | while (bio) { | |
876 | n = bio->bi_next; | |
877 | bio->bi_next = NULL; | |
6712ecf8 | 878 | bio_io_error(bio); |
1da177e4 LT |
879 | bio = n; |
880 | } | |
881 | } | |
882 | ||
695368ac | 883 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
884 | { |
885 | if (!s->valid) | |
886 | return; | |
887 | ||
888 | if (err == -EIO) | |
889 | DMERR("Invalidating snapshot: Error reading/writing."); | |
890 | else if (err == -ENOMEM) | |
891 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
892 | ||
493df71c JB |
893 | if (s->store->type->drop_snapshot) |
894 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
895 | |
896 | s->valid = 0; | |
897 | ||
0cea9c78 | 898 | dm_table_event(s->store->ti->table); |
76df1c65 AK |
899 | } |
900 | ||
028867ac | 901 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
902 | { |
903 | atomic_inc(&pe->ref_count); | |
904 | } | |
905 | ||
028867ac | 906 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 907 | { |
028867ac | 908 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
909 | struct bio *origin_bios = NULL; |
910 | ||
911 | primary_pe = pe->primary_pe; | |
912 | ||
913 | /* | |
914 | * If this pe is involved in a write to the origin and | |
915 | * it is the last sibling to complete then release | |
916 | * the bios for the original write to the origin. | |
917 | */ | |
918 | if (primary_pe && | |
7c5f78b9 | 919 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 920 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
921 | free_pending_exception(primary_pe); |
922 | } | |
4b832e8d AK |
923 | |
924 | /* | |
925 | * Free the pe if it's not linked to an origin write or if | |
926 | * it's not itself a primary pe. | |
927 | */ | |
928 | if (!primary_pe || primary_pe != pe) | |
929 | free_pending_exception(pe); | |
930 | ||
4b832e8d AK |
931 | return origin_bios; |
932 | } | |
933 | ||
028867ac | 934 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 935 | { |
028867ac | 936 | struct dm_snap_exception *e; |
1da177e4 | 937 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
938 | struct bio *origin_bios = NULL; |
939 | struct bio *snapshot_bios = NULL; | |
940 | int error = 0; | |
1da177e4 | 941 | |
76df1c65 AK |
942 | if (!success) { |
943 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 944 | down_write(&s->lock); |
695368ac | 945 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 946 | error = 1; |
76df1c65 AK |
947 | goto out; |
948 | } | |
949 | ||
950 | e = alloc_exception(); | |
951 | if (!e) { | |
1da177e4 | 952 | down_write(&s->lock); |
695368ac | 953 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 954 | error = 1; |
76df1c65 AK |
955 | goto out; |
956 | } | |
957 | *e = pe->e; | |
1da177e4 | 958 | |
76df1c65 AK |
959 | down_write(&s->lock); |
960 | if (!s->valid) { | |
76df1c65 | 961 | free_exception(e); |
9d493fa8 | 962 | error = 1; |
76df1c65 | 963 | goto out; |
1da177e4 LT |
964 | } |
965 | ||
a8d41b59 MP |
966 | /* |
967 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 968 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
969 | */ |
970 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 971 | msleep(1); |
a8d41b59 | 972 | |
9d493fa8 AK |
973 | /* |
974 | * Add a proper exception, and remove the | |
975 | * in-flight exception from the list. | |
976 | */ | |
d74f81f8 | 977 | insert_completed_exception(s, e); |
76df1c65 | 978 | |
1da177e4 | 979 | out: |
695368ac | 980 | remove_exception(&pe->e); |
9d493fa8 | 981 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 982 | origin_bios = put_pending_exception(pe); |
1da177e4 | 983 | |
9d493fa8 AK |
984 | up_write(&s->lock); |
985 | ||
986 | /* Submit any pending write bios */ | |
987 | if (error) | |
988 | error_bios(snapshot_bios); | |
989 | else | |
990 | flush_bios(snapshot_bios); | |
991 | ||
992 | flush_bios(origin_bios); | |
1da177e4 LT |
993 | } |
994 | ||
995 | static void commit_callback(void *context, int success) | |
996 | { | |
028867ac AK |
997 | struct dm_snap_pending_exception *pe = context; |
998 | ||
1da177e4 LT |
999 | pending_complete(pe, success); |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Called when the copy I/O has finished. kcopyd actually runs | |
1004 | * this code so don't block. | |
1005 | */ | |
4cdc1d1f | 1006 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1007 | { |
028867ac | 1008 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1009 | struct dm_snapshot *s = pe->snap; |
1010 | ||
1011 | if (read_err || write_err) | |
1012 | pending_complete(pe, 0); | |
1013 | ||
1014 | else | |
1015 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1016 | s->store->type->commit_exception(s->store, &pe->e, |
1017 | commit_callback, pe); | |
1da177e4 LT |
1018 | } |
1019 | ||
1020 | /* | |
1021 | * Dispatches the copy operation to kcopyd. | |
1022 | */ | |
028867ac | 1023 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1024 | { |
1025 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1026 | struct dm_io_region src, dest; |
1da177e4 LT |
1027 | struct block_device *bdev = s->origin->bdev; |
1028 | sector_t dev_size; | |
1029 | ||
1030 | dev_size = get_dev_size(bdev); | |
1031 | ||
1032 | src.bdev = bdev; | |
71fab00a | 1033 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
d0216849 | 1034 | src.count = min(s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1035 | |
49beb2b8 | 1036 | dest.bdev = s->store->cow->bdev; |
71fab00a | 1037 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1038 | dest.count = src.count; |
1039 | ||
1040 | /* Hand over to kcopyd */ | |
eb69aca5 | 1041 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1042 | &src, 1, &dest, 0, copy_callback, pe); |
1043 | } | |
1044 | ||
2913808e MP |
1045 | static struct dm_snap_pending_exception * |
1046 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1047 | { | |
1048 | struct dm_snap_exception *e = lookup_exception(&s->pending, chunk); | |
1049 | ||
1050 | if (!e) | |
1051 | return NULL; | |
1052 | ||
1053 | return container_of(e, struct dm_snap_pending_exception, e); | |
1054 | } | |
1055 | ||
1da177e4 LT |
1056 | /* |
1057 | * Looks to see if this snapshot already has a pending exception | |
1058 | * for this chunk, otherwise it allocates a new one and inserts | |
1059 | * it into the pending table. | |
1060 | * | |
1061 | * NOTE: a write lock must be held on snap->lock before calling | |
1062 | * this. | |
1063 | */ | |
028867ac | 1064 | static struct dm_snap_pending_exception * |
c6621392 MP |
1065 | __find_pending_exception(struct dm_snapshot *s, |
1066 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1067 | { |
c6621392 | 1068 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1069 | |
2913808e MP |
1070 | pe2 = __lookup_pending_exception(s, chunk); |
1071 | if (pe2) { | |
76df1c65 | 1072 | free_pending_exception(pe); |
2913808e | 1073 | return pe2; |
1da177e4 LT |
1074 | } |
1075 | ||
76df1c65 AK |
1076 | pe->e.old_chunk = chunk; |
1077 | bio_list_init(&pe->origin_bios); | |
1078 | bio_list_init(&pe->snapshot_bios); | |
1079 | pe->primary_pe = NULL; | |
4b832e8d | 1080 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1081 | pe->started = 0; |
1082 | ||
493df71c | 1083 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1084 | free_pending_exception(pe); |
1085 | return NULL; | |
1086 | } | |
1087 | ||
4b832e8d | 1088 | get_pending_exception(pe); |
76df1c65 AK |
1089 | insert_exception(&s->pending, &pe->e); |
1090 | ||
1da177e4 LT |
1091 | return pe; |
1092 | } | |
1093 | ||
028867ac | 1094 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1095 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1096 | { |
49beb2b8 | 1097 | bio->bi_bdev = s->store->cow->bdev; |
71fab00a JB |
1098 | bio->bi_sector = chunk_to_sector(s->store, |
1099 | dm_chunk_number(e->new_chunk) + | |
1100 | (chunk - e->old_chunk)) + | |
1101 | (bio->bi_sector & | |
1102 | s->store->chunk_mask); | |
1da177e4 LT |
1103 | } |
1104 | ||
1105 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1106 | union map_info *map_context) | |
1107 | { | |
028867ac AK |
1108 | struct dm_snap_exception *e; |
1109 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1110 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1111 | chunk_t chunk; |
028867ac | 1112 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1113 | |
71fab00a | 1114 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1115 | |
1116 | /* Full snapshots are not usable */ | |
76df1c65 | 1117 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1118 | if (!s->valid) |
f6a80ea8 | 1119 | return -EIO; |
1da177e4 | 1120 | |
ba40a2aa AK |
1121 | /* FIXME: should only take write lock if we need |
1122 | * to copy an exception */ | |
1123 | down_write(&s->lock); | |
1124 | ||
1125 | if (!s->valid) { | |
1126 | r = -EIO; | |
1127 | goto out_unlock; | |
1128 | } | |
1129 | ||
1130 | /* If the block is already remapped - use that, else remap it */ | |
1131 | e = lookup_exception(&s->complete, chunk); | |
1132 | if (e) { | |
d74f81f8 | 1133 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1134 | goto out_unlock; |
1135 | } | |
1136 | ||
1da177e4 LT |
1137 | /* |
1138 | * Write to snapshot - higher level takes care of RW/RO | |
1139 | * flags so we should only get this if we are | |
1140 | * writeable. | |
1141 | */ | |
1142 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1143 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1144 | if (!pe) { |
c6621392 MP |
1145 | up_write(&s->lock); |
1146 | pe = alloc_pending_exception(s); | |
1147 | down_write(&s->lock); | |
1148 | ||
1149 | if (!s->valid) { | |
1150 | free_pending_exception(pe); | |
1151 | r = -EIO; | |
1152 | goto out_unlock; | |
1153 | } | |
1154 | ||
35bf659b MP |
1155 | e = lookup_exception(&s->complete, chunk); |
1156 | if (e) { | |
1157 | free_pending_exception(pe); | |
1158 | remap_exception(s, e, bio, chunk); | |
1159 | goto out_unlock; | |
1160 | } | |
1161 | ||
c6621392 | 1162 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1163 | if (!pe) { |
1164 | __invalidate_snapshot(s, -ENOMEM); | |
1165 | r = -EIO; | |
1166 | goto out_unlock; | |
1167 | } | |
1da177e4 LT |
1168 | } |
1169 | ||
d74f81f8 | 1170 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1171 | bio_list_add(&pe->snapshot_bios, bio); |
1172 | ||
d2a7ad29 | 1173 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1174 | |
76df1c65 AK |
1175 | if (!pe->started) { |
1176 | /* this is protected by snap->lock */ | |
1177 | pe->started = 1; | |
ba40a2aa | 1178 | up_write(&s->lock); |
76df1c65 | 1179 | start_copy(pe); |
ba40a2aa AK |
1180 | goto out; |
1181 | } | |
cd45daff | 1182 | } else { |
ba40a2aa | 1183 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1184 | map_context->ptr = track_chunk(s, chunk); |
1185 | } | |
1da177e4 | 1186 | |
ba40a2aa AK |
1187 | out_unlock: |
1188 | up_write(&s->lock); | |
1189 | out: | |
1da177e4 LT |
1190 | return r; |
1191 | } | |
1192 | ||
cd45daff MP |
1193 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1194 | int error, union map_info *map_context) | |
1195 | { | |
1196 | struct dm_snapshot *s = ti->private; | |
1197 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1198 | ||
1199 | if (c) | |
1200 | stop_tracking_chunk(s, c); | |
1201 | ||
1202 | return 0; | |
1203 | } | |
1204 | ||
1da177e4 LT |
1205 | static void snapshot_resume(struct dm_target *ti) |
1206 | { | |
028867ac | 1207 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1208 | |
aa14edeb AK |
1209 | down_write(&s->lock); |
1210 | s->active = 1; | |
1211 | up_write(&s->lock); | |
1da177e4 LT |
1212 | } |
1213 | ||
1214 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1215 | char *result, unsigned int maxlen) | |
1216 | { | |
2e4a31df | 1217 | unsigned sz = 0; |
028867ac | 1218 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1219 | |
1220 | switch (type) { | |
1221 | case STATUSTYPE_INFO: | |
1222 | if (!snap->valid) | |
2e4a31df | 1223 | DMEMIT("Invalid"); |
1da177e4 | 1224 | else { |
493df71c | 1225 | if (snap->store->type->fraction_full) { |
1da177e4 | 1226 | sector_t numerator, denominator; |
493df71c JB |
1227 | snap->store->type->fraction_full(snap->store, |
1228 | &numerator, | |
1229 | &denominator); | |
2e4a31df JB |
1230 | DMEMIT("%llu/%llu", |
1231 | (unsigned long long)numerator, | |
1232 | (unsigned long long)denominator); | |
1da177e4 LT |
1233 | } |
1234 | else | |
2e4a31df | 1235 | DMEMIT("Unknown"); |
1da177e4 LT |
1236 | } |
1237 | break; | |
1238 | ||
1239 | case STATUSTYPE_TABLE: | |
1240 | /* | |
1241 | * kdevname returns a static pointer so we need | |
1242 | * to make private copies if the output is to | |
1243 | * make sense. | |
1244 | */ | |
2e4a31df JB |
1245 | DMEMIT("%s", snap->origin->name); |
1246 | DMEMIT(" %s %s %llu", snap->store->cow->name, | |
1247 | snap->store->type->name, | |
1248 | (unsigned long long)snap->store->chunk_size); | |
1da177e4 LT |
1249 | break; |
1250 | } | |
1251 | ||
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | /*----------------------------------------------------------------- | |
1256 | * Origin methods | |
1257 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1258 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1259 | { | |
d2a7ad29 | 1260 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1261 | struct dm_snapshot *snap; |
028867ac AK |
1262 | struct dm_snap_exception *e; |
1263 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1264 | chunk_t chunk; |
eccf0817 | 1265 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1266 | |
1267 | /* Do all the snapshots on this origin */ | |
1268 | list_for_each_entry (snap, snapshots, list) { | |
1269 | ||
76df1c65 AK |
1270 | down_write(&snap->lock); |
1271 | ||
aa14edeb AK |
1272 | /* Only deal with valid and active snapshots */ |
1273 | if (!snap->valid || !snap->active) | |
76df1c65 | 1274 | goto next_snapshot; |
1da177e4 | 1275 | |
d5e404c1 | 1276 | /* Nothing to do if writing beyond end of snapshot */ |
0cea9c78 | 1277 | if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table)) |
76df1c65 | 1278 | goto next_snapshot; |
1da177e4 LT |
1279 | |
1280 | /* | |
1281 | * Remember, different snapshots can have | |
1282 | * different chunk sizes. | |
1283 | */ | |
71fab00a | 1284 | chunk = sector_to_chunk(snap->store, bio->bi_sector); |
1da177e4 LT |
1285 | |
1286 | /* | |
1287 | * Check exception table to see if block | |
1288 | * is already remapped in this snapshot | |
1289 | * and trigger an exception if not. | |
b4b610f6 | 1290 | * |
4b832e8d | 1291 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1292 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1293 | */ |
1294 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1295 | if (e) |
1296 | goto next_snapshot; | |
1297 | ||
2913808e | 1298 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1299 | if (!pe) { |
c6621392 MP |
1300 | up_write(&snap->lock); |
1301 | pe = alloc_pending_exception(snap); | |
1302 | down_write(&snap->lock); | |
1303 | ||
1304 | if (!snap->valid) { | |
1305 | free_pending_exception(pe); | |
1306 | goto next_snapshot; | |
1307 | } | |
1308 | ||
35bf659b MP |
1309 | e = lookup_exception(&snap->complete, chunk); |
1310 | if (e) { | |
1311 | free_pending_exception(pe); | |
1312 | goto next_snapshot; | |
1313 | } | |
1314 | ||
c6621392 | 1315 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1316 | if (!pe) { |
1317 | __invalidate_snapshot(snap, -ENOMEM); | |
1318 | goto next_snapshot; | |
1319 | } | |
76df1c65 AK |
1320 | } |
1321 | ||
1322 | if (!primary_pe) { | |
1323 | /* | |
1324 | * Either every pe here has same | |
1325 | * primary_pe or none has one yet. | |
1326 | */ | |
1327 | if (pe->primary_pe) | |
1328 | primary_pe = pe->primary_pe; | |
1329 | else { | |
1330 | primary_pe = pe; | |
1331 | first = 1; | |
1da177e4 | 1332 | } |
76df1c65 AK |
1333 | |
1334 | bio_list_add(&primary_pe->origin_bios, bio); | |
1335 | ||
d2a7ad29 | 1336 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1337 | } |
1338 | ||
1339 | if (!pe->primary_pe) { | |
76df1c65 | 1340 | pe->primary_pe = primary_pe; |
4b832e8d | 1341 | get_pending_exception(primary_pe); |
76df1c65 AK |
1342 | } |
1343 | ||
1344 | if (!pe->started) { | |
1345 | pe->started = 1; | |
1346 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1347 | } |
1348 | ||
76df1c65 | 1349 | next_snapshot: |
1da177e4 LT |
1350 | up_write(&snap->lock); |
1351 | } | |
1352 | ||
b4b610f6 | 1353 | if (!primary_pe) |
4b832e8d | 1354 | return r; |
b4b610f6 AK |
1355 | |
1356 | /* | |
1357 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1358 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1359 | * got completed while we were in the loop above, so it falls to |
1360 | * us here to remove the primary_pe and submit any origin_bios. | |
1361 | */ | |
1362 | ||
4b832e8d | 1363 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1364 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1365 | free_pending_exception(primary_pe); | |
1366 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1367 | return r; |
b4b610f6 AK |
1368 | } |
1369 | ||
1da177e4 LT |
1370 | /* |
1371 | * Now that we have a complete pe list we can start the copying. | |
1372 | */ | |
eccf0817 AK |
1373 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1374 | start_copy(pe); | |
1da177e4 LT |
1375 | |
1376 | return r; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Called on a write from the origin driver. | |
1381 | */ | |
1382 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1383 | { | |
1384 | struct origin *o; | |
d2a7ad29 | 1385 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1386 | |
1387 | down_read(&_origins_lock); | |
1388 | o = __lookup_origin(origin->bdev); | |
1389 | if (o) | |
1390 | r = __origin_write(&o->snapshots, bio); | |
1391 | up_read(&_origins_lock); | |
1392 | ||
1393 | return r; | |
1394 | } | |
1395 | ||
1396 | /* | |
1397 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1398 | */ | |
1399 | ||
1400 | /* | |
1401 | * Construct an origin mapping: <dev_path> | |
1402 | * The context for an origin is merely a 'struct dm_dev *' | |
1403 | * pointing to the real device. | |
1404 | */ | |
1405 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1406 | { | |
1407 | int r; | |
1408 | struct dm_dev *dev; | |
1409 | ||
1410 | if (argc != 1) { | |
72d94861 | 1411 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1412 | return -EINVAL; |
1413 | } | |
1414 | ||
1415 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1416 | dm_table_get_mode(ti->table), &dev); | |
1417 | if (r) { | |
1418 | ti->error = "Cannot get target device"; | |
1419 | return r; | |
1420 | } | |
1421 | ||
1422 | ti->private = dev; | |
1423 | return 0; | |
1424 | } | |
1425 | ||
1426 | static void origin_dtr(struct dm_target *ti) | |
1427 | { | |
028867ac | 1428 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1429 | dm_put_device(ti, dev); |
1430 | } | |
1431 | ||
1432 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1433 | union map_info *map_context) | |
1434 | { | |
028867ac | 1435 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1436 | bio->bi_bdev = dev->bdev; |
1437 | ||
1438 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1439 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1440 | } |
1441 | ||
1442 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1443 | ||
1444 | /* | |
1445 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1446 | * chunk sizes. | |
1447 | */ | |
1448 | static void origin_resume(struct dm_target *ti) | |
1449 | { | |
028867ac | 1450 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1451 | struct dm_snapshot *snap; |
1452 | struct origin *o; | |
1453 | chunk_t chunk_size = 0; | |
1454 | ||
1455 | down_read(&_origins_lock); | |
1456 | o = __lookup_origin(dev->bdev); | |
1457 | if (o) | |
1458 | list_for_each_entry (snap, &o->snapshots, list) | |
d0216849 JB |
1459 | chunk_size = min_not_zero(chunk_size, |
1460 | snap->store->chunk_size); | |
1da177e4 LT |
1461 | up_read(&_origins_lock); |
1462 | ||
1463 | ti->split_io = chunk_size; | |
1464 | } | |
1465 | ||
1466 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1467 | unsigned int maxlen) | |
1468 | { | |
028867ac | 1469 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1470 | |
1471 | switch (type) { | |
1472 | case STATUSTYPE_INFO: | |
1473 | result[0] = '\0'; | |
1474 | break; | |
1475 | ||
1476 | case STATUSTYPE_TABLE: | |
1477 | snprintf(result, maxlen, "%s", dev->name); | |
1478 | break; | |
1479 | } | |
1480 | ||
1481 | return 0; | |
1482 | } | |
1483 | ||
1484 | static struct target_type origin_target = { | |
1485 | .name = "snapshot-origin", | |
d74f81f8 | 1486 | .version = {1, 6, 0}, |
1da177e4 LT |
1487 | .module = THIS_MODULE, |
1488 | .ctr = origin_ctr, | |
1489 | .dtr = origin_dtr, | |
1490 | .map = origin_map, | |
1491 | .resume = origin_resume, | |
1492 | .status = origin_status, | |
1493 | }; | |
1494 | ||
1495 | static struct target_type snapshot_target = { | |
1496 | .name = "snapshot", | |
d74f81f8 | 1497 | .version = {1, 6, 0}, |
1da177e4 LT |
1498 | .module = THIS_MODULE, |
1499 | .ctr = snapshot_ctr, | |
1500 | .dtr = snapshot_dtr, | |
1501 | .map = snapshot_map, | |
cd45daff | 1502 | .end_io = snapshot_end_io, |
1da177e4 LT |
1503 | .resume = snapshot_resume, |
1504 | .status = snapshot_status, | |
1505 | }; | |
1506 | ||
1507 | static int __init dm_snapshot_init(void) | |
1508 | { | |
1509 | int r; | |
1510 | ||
4db6bfe0 AK |
1511 | r = dm_exception_store_init(); |
1512 | if (r) { | |
1513 | DMERR("Failed to initialize exception stores"); | |
1514 | return r; | |
1515 | } | |
1516 | ||
1da177e4 LT |
1517 | r = dm_register_target(&snapshot_target); |
1518 | if (r) { | |
1519 | DMERR("snapshot target register failed %d", r); | |
1520 | return r; | |
1521 | } | |
1522 | ||
1523 | r = dm_register_target(&origin_target); | |
1524 | if (r < 0) { | |
72d94861 | 1525 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1526 | goto bad1; |
1527 | } | |
1528 | ||
1529 | r = init_origin_hash(); | |
1530 | if (r) { | |
1531 | DMERR("init_origin_hash failed."); | |
1532 | goto bad2; | |
1533 | } | |
1534 | ||
028867ac | 1535 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1536 | if (!exception_cache) { |
1537 | DMERR("Couldn't create exception cache."); | |
1538 | r = -ENOMEM; | |
1539 | goto bad3; | |
1540 | } | |
1541 | ||
028867ac | 1542 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1543 | if (!pending_cache) { |
1544 | DMERR("Couldn't create pending cache."); | |
1545 | r = -ENOMEM; | |
1546 | goto bad4; | |
1547 | } | |
1548 | ||
cd45daff MP |
1549 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1550 | if (!tracked_chunk_cache) { | |
1551 | DMERR("Couldn't create cache to track chunks in use."); | |
1552 | r = -ENOMEM; | |
1553 | goto bad5; | |
1554 | } | |
1555 | ||
ca3a931f AK |
1556 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1557 | if (!ksnapd) { | |
1558 | DMERR("Failed to create ksnapd workqueue."); | |
1559 | r = -ENOMEM; | |
92e86812 | 1560 | goto bad_pending_pool; |
ca3a931f AK |
1561 | } |
1562 | ||
1da177e4 LT |
1563 | return 0; |
1564 | ||
4db6bfe0 | 1565 | bad_pending_pool: |
cd45daff | 1566 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1567 | bad5: |
1da177e4 | 1568 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1569 | bad4: |
1da177e4 | 1570 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1571 | bad3: |
1da177e4 | 1572 | exit_origin_hash(); |
4db6bfe0 | 1573 | bad2: |
1da177e4 | 1574 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1575 | bad1: |
1da177e4 LT |
1576 | dm_unregister_target(&snapshot_target); |
1577 | return r; | |
1578 | } | |
1579 | ||
1580 | static void __exit dm_snapshot_exit(void) | |
1581 | { | |
ca3a931f AK |
1582 | destroy_workqueue(ksnapd); |
1583 | ||
10d3bd09 MP |
1584 | dm_unregister_target(&snapshot_target); |
1585 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1586 | |
1587 | exit_origin_hash(); | |
1da177e4 LT |
1588 | kmem_cache_destroy(pending_cache); |
1589 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1590 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1591 | |
1592 | dm_exception_store_exit(); | |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | /* Module hooks */ | |
1596 | module_init(dm_snapshot_init); | |
1597 | module_exit(dm_snapshot_exit); | |
1598 | ||
1599 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1600 | MODULE_AUTHOR("Joe Thornber"); | |
1601 | MODULE_LICENSE("GPL"); |