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 | #include "dm-bio-list.h" |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "snapshots" |
28 | ||
1da177e4 LT |
29 | /* |
30 | * The percentage increment we will wake up users at | |
31 | */ | |
32 | #define WAKE_UP_PERCENT 5 | |
33 | ||
34 | /* | |
35 | * kcopyd priority of snapshot operations | |
36 | */ | |
37 | #define SNAPSHOT_COPY_PRIORITY 2 | |
38 | ||
39 | /* | |
8ee2767a | 40 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 41 | */ |
8ee2767a | 42 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 43 | |
cd45daff MP |
44 | /* |
45 | * The size of the mempool used to track chunks in use. | |
46 | */ | |
47 | #define MIN_IOS 256 | |
48 | ||
ccc45ea8 JB |
49 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
50 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
51 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
52 | ||
53 | struct exception_table { | |
54 | uint32_t hash_mask; | |
55 | unsigned hash_shift; | |
56 | struct list_head *table; | |
57 | }; | |
58 | ||
59 | struct dm_snapshot { | |
60 | struct rw_semaphore lock; | |
61 | ||
62 | struct dm_dev *origin; | |
63 | ||
64 | /* List of snapshots per Origin */ | |
65 | struct list_head list; | |
66 | ||
67 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
68 | int valid; | |
69 | ||
70 | /* Origin writes don't trigger exceptions until this is set */ | |
71 | int active; | |
72 | ||
73 | /* Used for display of table */ | |
74 | char type; | |
75 | ||
76 | mempool_t *pending_pool; | |
77 | ||
78 | atomic_t pending_exceptions_count; | |
79 | ||
80 | struct exception_table pending; | |
81 | struct exception_table complete; | |
82 | ||
83 | /* | |
84 | * pe_lock protects all pending_exception operations and access | |
85 | * as well as the snapshot_bios list. | |
86 | */ | |
87 | spinlock_t pe_lock; | |
88 | ||
89 | /* The on disk metadata handler */ | |
90 | struct dm_exception_store *store; | |
91 | ||
92 | struct dm_kcopyd_client *kcopyd_client; | |
93 | ||
94 | /* Queue of snapshot writes for ksnapd to flush */ | |
95 | struct bio_list queued_bios; | |
96 | struct work_struct queued_bios_work; | |
97 | ||
98 | /* Chunks with outstanding reads */ | |
99 | mempool_t *tracked_chunk_pool; | |
100 | spinlock_t tracked_chunk_lock; | |
101 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
102 | }; | |
103 | ||
c642f9e0 | 104 | static struct workqueue_struct *ksnapd; |
c4028958 | 105 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 106 | |
ccc45ea8 JB |
107 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
108 | chunk_t chunk) | |
109 | { | |
110 | return chunk << store->chunk_shift; | |
111 | } | |
112 | ||
113 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
114 | { | |
115 | /* | |
116 | * There is only ever one instance of a particular block | |
117 | * device so we can compare pointers safely. | |
118 | */ | |
119 | return lhs == rhs; | |
120 | } | |
121 | ||
028867ac AK |
122 | struct dm_snap_pending_exception { |
123 | struct dm_snap_exception e; | |
1da177e4 LT |
124 | |
125 | /* | |
126 | * Origin buffers waiting for this to complete are held | |
127 | * in a bio list | |
128 | */ | |
129 | struct bio_list origin_bios; | |
130 | struct bio_list snapshot_bios; | |
131 | ||
eccf0817 AK |
132 | /* |
133 | * Short-term queue of pending exceptions prior to submission. | |
134 | */ | |
135 | struct list_head list; | |
136 | ||
1da177e4 | 137 | /* |
b4b610f6 | 138 | * The primary pending_exception is the one that holds |
4b832e8d | 139 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
140 | * group of pending_exceptions. It is always last to get freed. |
141 | * These fields get set up when writing to the origin. | |
1da177e4 | 142 | */ |
028867ac | 143 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
144 | |
145 | /* | |
146 | * Number of pending_exceptions processing this chunk. | |
147 | * When this drops to zero we must complete the origin bios. | |
148 | * If incrementing or decrementing this, hold pe->snap->lock for | |
149 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
150 | * they are the same. | |
151 | */ | |
4b832e8d | 152 | atomic_t ref_count; |
1da177e4 LT |
153 | |
154 | /* Pointer back to snapshot context */ | |
155 | struct dm_snapshot *snap; | |
156 | ||
157 | /* | |
158 | * 1 indicates the exception has already been sent to | |
159 | * kcopyd. | |
160 | */ | |
161 | int started; | |
162 | }; | |
163 | ||
164 | /* | |
165 | * Hash table mapping origin volumes to lists of snapshots and | |
166 | * a lock to protect it | |
167 | */ | |
e18b890b CL |
168 | static struct kmem_cache *exception_cache; |
169 | static struct kmem_cache *pending_cache; | |
1da177e4 | 170 | |
cd45daff MP |
171 | struct dm_snap_tracked_chunk { |
172 | struct hlist_node node; | |
173 | chunk_t chunk; | |
174 | }; | |
175 | ||
176 | static struct kmem_cache *tracked_chunk_cache; | |
177 | ||
178 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
179 | chunk_t chunk) | |
180 | { | |
181 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
182 | GFP_NOIO); | |
183 | unsigned long flags; | |
184 | ||
185 | c->chunk = chunk; | |
186 | ||
187 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
188 | hlist_add_head(&c->node, | |
189 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
190 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
191 | ||
192 | return c; | |
193 | } | |
194 | ||
195 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
196 | struct dm_snap_tracked_chunk *c) | |
197 | { | |
198 | unsigned long flags; | |
199 | ||
200 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
201 | hlist_del(&c->node); | |
202 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
203 | ||
204 | mempool_free(c, s->tracked_chunk_pool); | |
205 | } | |
206 | ||
a8d41b59 MP |
207 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
208 | { | |
209 | struct dm_snap_tracked_chunk *c; | |
210 | struct hlist_node *hn; | |
211 | int found = 0; | |
212 | ||
213 | spin_lock_irq(&s->tracked_chunk_lock); | |
214 | ||
215 | hlist_for_each_entry(c, hn, | |
216 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
217 | if (c->chunk == chunk) { | |
218 | found = 1; | |
219 | break; | |
220 | } | |
221 | } | |
222 | ||
223 | spin_unlock_irq(&s->tracked_chunk_lock); | |
224 | ||
225 | return found; | |
226 | } | |
227 | ||
1da177e4 LT |
228 | /* |
229 | * One of these per registered origin, held in the snapshot_origins hash | |
230 | */ | |
231 | struct origin { | |
232 | /* The origin device */ | |
233 | struct block_device *bdev; | |
234 | ||
235 | struct list_head hash_list; | |
236 | ||
237 | /* List of snapshots for this origin */ | |
238 | struct list_head snapshots; | |
239 | }; | |
240 | ||
241 | /* | |
242 | * Size of the hash table for origin volumes. If we make this | |
243 | * the size of the minors list then it should be nearly perfect | |
244 | */ | |
245 | #define ORIGIN_HASH_SIZE 256 | |
246 | #define ORIGIN_MASK 0xFF | |
247 | static struct list_head *_origins; | |
248 | static struct rw_semaphore _origins_lock; | |
249 | ||
250 | static int init_origin_hash(void) | |
251 | { | |
252 | int i; | |
253 | ||
254 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
255 | GFP_KERNEL); | |
256 | if (!_origins) { | |
72d94861 | 257 | DMERR("unable to allocate memory"); |
1da177e4 LT |
258 | return -ENOMEM; |
259 | } | |
260 | ||
261 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
262 | INIT_LIST_HEAD(_origins + i); | |
263 | init_rwsem(&_origins_lock); | |
264 | ||
265 | return 0; | |
266 | } | |
267 | ||
268 | static void exit_origin_hash(void) | |
269 | { | |
270 | kfree(_origins); | |
271 | } | |
272 | ||
028867ac | 273 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
274 | { |
275 | return bdev->bd_dev & ORIGIN_MASK; | |
276 | } | |
277 | ||
278 | static struct origin *__lookup_origin(struct block_device *origin) | |
279 | { | |
280 | struct list_head *ol; | |
281 | struct origin *o; | |
282 | ||
283 | ol = &_origins[origin_hash(origin)]; | |
284 | list_for_each_entry (o, ol, hash_list) | |
285 | if (bdev_equal(o->bdev, origin)) | |
286 | return o; | |
287 | ||
288 | return NULL; | |
289 | } | |
290 | ||
291 | static void __insert_origin(struct origin *o) | |
292 | { | |
293 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
294 | list_add_tail(&o->hash_list, sl); | |
295 | } | |
296 | ||
297 | /* | |
298 | * Make a note of the snapshot and its origin so we can look it | |
299 | * up when the origin has a write on it. | |
300 | */ | |
301 | static int register_snapshot(struct dm_snapshot *snap) | |
302 | { | |
60c856c8 | 303 | struct origin *o, *new_o; |
1da177e4 LT |
304 | struct block_device *bdev = snap->origin->bdev; |
305 | ||
60c856c8 MP |
306 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
307 | if (!new_o) | |
308 | return -ENOMEM; | |
309 | ||
1da177e4 LT |
310 | down_write(&_origins_lock); |
311 | o = __lookup_origin(bdev); | |
312 | ||
60c856c8 MP |
313 | if (o) |
314 | kfree(new_o); | |
315 | else { | |
1da177e4 | 316 | /* New origin */ |
60c856c8 | 317 | o = new_o; |
1da177e4 LT |
318 | |
319 | /* Initialise the struct */ | |
320 | INIT_LIST_HEAD(&o->snapshots); | |
321 | o->bdev = bdev; | |
322 | ||
323 | __insert_origin(o); | |
324 | } | |
325 | ||
326 | list_add_tail(&snap->list, &o->snapshots); | |
327 | ||
328 | up_write(&_origins_lock); | |
329 | return 0; | |
330 | } | |
331 | ||
332 | static void unregister_snapshot(struct dm_snapshot *s) | |
333 | { | |
334 | struct origin *o; | |
335 | ||
336 | down_write(&_origins_lock); | |
337 | o = __lookup_origin(s->origin->bdev); | |
338 | ||
339 | list_del(&s->list); | |
340 | if (list_empty(&o->snapshots)) { | |
341 | list_del(&o->hash_list); | |
342 | kfree(o); | |
343 | } | |
344 | ||
345 | up_write(&_origins_lock); | |
346 | } | |
347 | ||
348 | /* | |
349 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
350 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
351 | * some consecutive chunks to be grouped together. | |
1da177e4 | 352 | */ |
d74f81f8 MB |
353 | static int init_exception_table(struct exception_table *et, uint32_t size, |
354 | unsigned hash_shift) | |
1da177e4 LT |
355 | { |
356 | unsigned int i; | |
357 | ||
d74f81f8 | 358 | et->hash_shift = hash_shift; |
1da177e4 LT |
359 | et->hash_mask = size - 1; |
360 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
361 | if (!et->table) | |
362 | return -ENOMEM; | |
363 | ||
364 | for (i = 0; i < size; i++) | |
365 | INIT_LIST_HEAD(et->table + i); | |
366 | ||
367 | return 0; | |
368 | } | |
369 | ||
e18b890b | 370 | static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem) |
1da177e4 LT |
371 | { |
372 | struct list_head *slot; | |
028867ac | 373 | struct dm_snap_exception *ex, *next; |
1da177e4 LT |
374 | int i, size; |
375 | ||
376 | size = et->hash_mask + 1; | |
377 | for (i = 0; i < size; i++) { | |
378 | slot = et->table + i; | |
379 | ||
380 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
381 | kmem_cache_free(mem, ex); | |
382 | } | |
383 | ||
384 | vfree(et->table); | |
385 | } | |
386 | ||
028867ac | 387 | static uint32_t exception_hash(struct exception_table *et, chunk_t chunk) |
1da177e4 | 388 | { |
d74f81f8 | 389 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
390 | } |
391 | ||
028867ac AK |
392 | static void insert_exception(struct exception_table *eh, |
393 | struct dm_snap_exception *e) | |
1da177e4 LT |
394 | { |
395 | struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; | |
396 | list_add(&e->hash_list, l); | |
397 | } | |
398 | ||
028867ac | 399 | static void remove_exception(struct dm_snap_exception *e) |
1da177e4 LT |
400 | { |
401 | list_del(&e->hash_list); | |
402 | } | |
403 | ||
404 | /* | |
405 | * Return the exception data for a sector, or NULL if not | |
406 | * remapped. | |
407 | */ | |
028867ac AK |
408 | static struct dm_snap_exception *lookup_exception(struct exception_table *et, |
409 | chunk_t chunk) | |
1da177e4 LT |
410 | { |
411 | struct list_head *slot; | |
028867ac | 412 | struct dm_snap_exception *e; |
1da177e4 LT |
413 | |
414 | slot = &et->table[exception_hash(et, chunk)]; | |
415 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
416 | if (chunk >= e->old_chunk && |
417 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
418 | return e; |
419 | ||
420 | return NULL; | |
421 | } | |
422 | ||
028867ac | 423 | static struct dm_snap_exception *alloc_exception(void) |
1da177e4 | 424 | { |
028867ac | 425 | struct dm_snap_exception *e; |
1da177e4 LT |
426 | |
427 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
428 | if (!e) | |
429 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
430 | ||
431 | return e; | |
432 | } | |
433 | ||
028867ac | 434 | static void free_exception(struct dm_snap_exception *e) |
1da177e4 LT |
435 | { |
436 | kmem_cache_free(exception_cache, e); | |
437 | } | |
438 | ||
92e86812 | 439 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 440 | { |
92e86812 MP |
441 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
442 | GFP_NOIO); | |
443 | ||
879129d2 | 444 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
445 | pe->snap = s; |
446 | ||
447 | return pe; | |
1da177e4 LT |
448 | } |
449 | ||
028867ac | 450 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 451 | { |
879129d2 MP |
452 | struct dm_snapshot *s = pe->snap; |
453 | ||
454 | mempool_free(pe, s->pending_pool); | |
455 | smp_mb__before_atomic_dec(); | |
456 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
457 | } |
458 | ||
d74f81f8 MB |
459 | static void insert_completed_exception(struct dm_snapshot *s, |
460 | struct dm_snap_exception *new_e) | |
461 | { | |
462 | struct exception_table *eh = &s->complete; | |
463 | struct list_head *l; | |
464 | struct dm_snap_exception *e = NULL; | |
465 | ||
466 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
467 | ||
468 | /* Add immediately if this table doesn't support consecutive chunks */ | |
469 | if (!eh->hash_shift) | |
470 | goto out; | |
471 | ||
472 | /* List is ordered by old_chunk */ | |
473 | list_for_each_entry_reverse(e, l, hash_list) { | |
474 | /* Insert after an existing chunk? */ | |
475 | if (new_e->old_chunk == (e->old_chunk + | |
476 | dm_consecutive_chunk_count(e) + 1) && | |
477 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
478 | dm_consecutive_chunk_count(e) + 1)) { | |
479 | dm_consecutive_chunk_count_inc(e); | |
480 | free_exception(new_e); | |
481 | return; | |
482 | } | |
483 | ||
484 | /* Insert before an existing chunk? */ | |
485 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
486 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
487 | dm_consecutive_chunk_count_inc(e); | |
488 | e->old_chunk--; | |
489 | e->new_chunk--; | |
490 | free_exception(new_e); | |
491 | return; | |
492 | } | |
493 | ||
494 | if (new_e->old_chunk > e->old_chunk) | |
495 | break; | |
496 | } | |
497 | ||
498 | out: | |
499 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
500 | } | |
501 | ||
a159c1ac JB |
502 | /* |
503 | * Callback used by the exception stores to load exceptions when | |
504 | * initialising. | |
505 | */ | |
506 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 507 | { |
a159c1ac | 508 | struct dm_snapshot *s = context; |
028867ac | 509 | struct dm_snap_exception *e; |
1da177e4 LT |
510 | |
511 | e = alloc_exception(); | |
512 | if (!e) | |
513 | return -ENOMEM; | |
514 | ||
515 | e->old_chunk = old; | |
d74f81f8 MB |
516 | |
517 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 518 | e->new_chunk = new; |
d74f81f8 MB |
519 | |
520 | insert_completed_exception(s, e); | |
521 | ||
1da177e4 LT |
522 | return 0; |
523 | } | |
524 | ||
525 | /* | |
526 | * Hard coded magic. | |
527 | */ | |
528 | static int calc_max_buckets(void) | |
529 | { | |
530 | /* use a fixed size of 2MB */ | |
531 | unsigned long mem = 2 * 1024 * 1024; | |
532 | mem /= sizeof(struct list_head); | |
533 | ||
534 | return mem; | |
535 | } | |
536 | ||
1da177e4 LT |
537 | /* |
538 | * Allocate room for a suitable hash table. | |
539 | */ | |
fee1998e | 540 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
541 | { |
542 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
543 | ||
544 | /* | |
545 | * Calculate based on the size of the original volume or | |
546 | * the COW volume... | |
547 | */ | |
fee1998e | 548 | cow_dev_size = get_dev_size(s->store->cow->bdev); |
1da177e4 LT |
549 | origin_dev_size = get_dev_size(s->origin->bdev); |
550 | max_buckets = calc_max_buckets(); | |
551 | ||
fee1998e | 552 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
553 | hash_size = min(hash_size, max_buckets); |
554 | ||
8defd830 | 555 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
556 | if (init_exception_table(&s->complete, hash_size, |
557 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
558 | return -ENOMEM; |
559 | ||
560 | /* | |
561 | * Allocate hash table for in-flight exceptions | |
562 | * Make this smaller than the real hash table | |
563 | */ | |
564 | hash_size >>= 3; | |
565 | if (hash_size < 64) | |
566 | hash_size = 64; | |
567 | ||
d74f81f8 | 568 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
569 | exit_exception_table(&s->complete, exception_cache); |
570 | return -ENOMEM; | |
571 | } | |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
1da177e4 LT |
576 | /* |
577 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
578 | */ | |
579 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
580 | { | |
581 | struct dm_snapshot *s; | |
cd45daff | 582 | int i; |
1da177e4 | 583 | int r = -EINVAL; |
1da177e4 | 584 | char *origin_path; |
fee1998e JB |
585 | struct dm_exception_store *store; |
586 | unsigned args_used; | |
1da177e4 | 587 | |
4c7e3bf4 | 588 | if (argc != 4) { |
72d94861 | 589 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 590 | r = -EINVAL; |
fee1998e | 591 | goto bad_args; |
1da177e4 LT |
592 | } |
593 | ||
594 | origin_path = argv[0]; | |
fee1998e JB |
595 | argv++; |
596 | argc--; | |
1da177e4 | 597 | |
fee1998e JB |
598 | r = dm_exception_store_create(ti, argc, argv, &args_used, &store); |
599 | if (r) { | |
600 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 601 | r = -EINVAL; |
fee1998e | 602 | goto bad_args; |
1da177e4 LT |
603 | } |
604 | ||
fee1998e JB |
605 | argv += args_used; |
606 | argc -= args_used; | |
607 | ||
1da177e4 | 608 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
fee1998e | 609 | if (!s) { |
1da177e4 LT |
610 | ti->error = "Cannot allocate snapshot context private " |
611 | "structure"; | |
612 | r = -ENOMEM; | |
fee1998e | 613 | goto bad_snap; |
1da177e4 LT |
614 | } |
615 | ||
616 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
617 | if (r) { | |
618 | ti->error = "Cannot get origin device"; | |
fee1998e | 619 | goto bad_origin; |
1da177e4 LT |
620 | } |
621 | ||
fee1998e | 622 | s->store = store; |
1da177e4 | 623 | s->valid = 1; |
aa14edeb | 624 | s->active = 0; |
879129d2 | 625 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 626 | init_rwsem(&s->lock); |
ca3a931f | 627 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
628 | |
629 | /* Allocate hash table for COW data */ | |
fee1998e | 630 | if (init_hash_tables(s)) { |
1da177e4 LT |
631 | ti->error = "Unable to allocate hash table space"; |
632 | r = -ENOMEM; | |
fee1998e | 633 | goto bad_hash_tables; |
1da177e4 LT |
634 | } |
635 | ||
eb69aca5 | 636 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
637 | if (r) { |
638 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 639 | goto bad_kcopyd; |
1da177e4 LT |
640 | } |
641 | ||
92e86812 MP |
642 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
643 | if (!s->pending_pool) { | |
644 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 645 | goto bad_pending_pool; |
92e86812 MP |
646 | } |
647 | ||
cd45daff MP |
648 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
649 | tracked_chunk_cache); | |
650 | if (!s->tracked_chunk_pool) { | |
651 | ti->error = "Could not allocate tracked_chunk mempool for " | |
652 | "tracking reads"; | |
92e86812 | 653 | goto bad_tracked_chunk_pool; |
cd45daff MP |
654 | } |
655 | ||
656 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
657 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
658 | ||
659 | spin_lock_init(&s->tracked_chunk_lock); | |
660 | ||
aa14edeb | 661 | /* Metadata must only be loaded into one table at once */ |
493df71c JB |
662 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
663 | (void *)s); | |
0764147b | 664 | if (r < 0) { |
f9cea4f7 | 665 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 666 | goto bad_load_and_register; |
0764147b MB |
667 | } else if (r > 0) { |
668 | s->valid = 0; | |
669 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 670 | } |
aa14edeb | 671 | |
ca3a931f | 672 | bio_list_init(&s->queued_bios); |
c4028958 | 673 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 674 | |
1da177e4 | 675 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 676 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
677 | if (register_snapshot(s)) { |
678 | r = -EINVAL; | |
679 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 680 | goto bad_load_and_register; |
1da177e4 LT |
681 | } |
682 | ||
683 | ti->private = s; | |
d0216849 | 684 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
685 | |
686 | return 0; | |
687 | ||
fee1998e | 688 | bad_load_and_register: |
cd45daff MP |
689 | mempool_destroy(s->tracked_chunk_pool); |
690 | ||
fee1998e | 691 | bad_tracked_chunk_pool: |
92e86812 MP |
692 | mempool_destroy(s->pending_pool); |
693 | ||
fee1998e | 694 | bad_pending_pool: |
eb69aca5 | 695 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 696 | |
fee1998e | 697 | bad_kcopyd: |
1da177e4 LT |
698 | exit_exception_table(&s->pending, pending_cache); |
699 | exit_exception_table(&s->complete, exception_cache); | |
700 | ||
fee1998e | 701 | bad_hash_tables: |
1da177e4 LT |
702 | dm_put_device(ti, s->origin); |
703 | ||
fee1998e | 704 | bad_origin: |
1da177e4 LT |
705 | kfree(s); |
706 | ||
fee1998e JB |
707 | bad_snap: |
708 | dm_exception_store_destroy(store); | |
709 | ||
710 | bad_args: | |
1da177e4 LT |
711 | return r; |
712 | } | |
713 | ||
31c93a0c MB |
714 | static void __free_exceptions(struct dm_snapshot *s) |
715 | { | |
eb69aca5 | 716 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
717 | s->kcopyd_client = NULL; |
718 | ||
719 | exit_exception_table(&s->pending, pending_cache); | |
720 | exit_exception_table(&s->complete, exception_cache); | |
31c93a0c MB |
721 | } |
722 | ||
1da177e4 LT |
723 | static void snapshot_dtr(struct dm_target *ti) |
724 | { | |
cd45daff MP |
725 | #ifdef CONFIG_DM_DEBUG |
726 | int i; | |
727 | #endif | |
028867ac | 728 | struct dm_snapshot *s = ti->private; |
1da177e4 | 729 | |
ca3a931f AK |
730 | flush_workqueue(ksnapd); |
731 | ||
138728dc AK |
732 | /* Prevent further origin writes from using this snapshot. */ |
733 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
734 | unregister_snapshot(s); |
735 | ||
879129d2 | 736 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 737 | msleep(1); |
879129d2 MP |
738 | /* |
739 | * Ensure instructions in mempool_destroy aren't reordered | |
740 | * before atomic_read. | |
741 | */ | |
742 | smp_mb(); | |
743 | ||
cd45daff MP |
744 | #ifdef CONFIG_DM_DEBUG |
745 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
746 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
747 | #endif | |
748 | ||
749 | mempool_destroy(s->tracked_chunk_pool); | |
750 | ||
31c93a0c | 751 | __free_exceptions(s); |
1da177e4 | 752 | |
92e86812 MP |
753 | mempool_destroy(s->pending_pool); |
754 | ||
1da177e4 | 755 | dm_put_device(ti, s->origin); |
fee1998e JB |
756 | |
757 | dm_exception_store_destroy(s->store); | |
138728dc | 758 | |
1da177e4 LT |
759 | kfree(s); |
760 | } | |
761 | ||
762 | /* | |
763 | * Flush a list of buffers. | |
764 | */ | |
765 | static void flush_bios(struct bio *bio) | |
766 | { | |
767 | struct bio *n; | |
768 | ||
769 | while (bio) { | |
770 | n = bio->bi_next; | |
771 | bio->bi_next = NULL; | |
772 | generic_make_request(bio); | |
773 | bio = n; | |
774 | } | |
775 | } | |
776 | ||
c4028958 | 777 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 778 | { |
c4028958 DH |
779 | struct dm_snapshot *s = |
780 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
781 | struct bio *queued_bios; |
782 | unsigned long flags; | |
783 | ||
784 | spin_lock_irqsave(&s->pe_lock, flags); | |
785 | queued_bios = bio_list_get(&s->queued_bios); | |
786 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
787 | ||
788 | flush_bios(queued_bios); | |
789 | } | |
790 | ||
1da177e4 LT |
791 | /* |
792 | * Error a list of buffers. | |
793 | */ | |
794 | static void error_bios(struct bio *bio) | |
795 | { | |
796 | struct bio *n; | |
797 | ||
798 | while (bio) { | |
799 | n = bio->bi_next; | |
800 | bio->bi_next = NULL; | |
6712ecf8 | 801 | bio_io_error(bio); |
1da177e4 LT |
802 | bio = n; |
803 | } | |
804 | } | |
805 | ||
695368ac | 806 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
807 | { |
808 | if (!s->valid) | |
809 | return; | |
810 | ||
811 | if (err == -EIO) | |
812 | DMERR("Invalidating snapshot: Error reading/writing."); | |
813 | else if (err == -ENOMEM) | |
814 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
815 | ||
493df71c JB |
816 | if (s->store->type->drop_snapshot) |
817 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
818 | |
819 | s->valid = 0; | |
820 | ||
0cea9c78 | 821 | dm_table_event(s->store->ti->table); |
76df1c65 AK |
822 | } |
823 | ||
028867ac | 824 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
825 | { |
826 | atomic_inc(&pe->ref_count); | |
827 | } | |
828 | ||
028867ac | 829 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 830 | { |
028867ac | 831 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
832 | struct bio *origin_bios = NULL; |
833 | ||
834 | primary_pe = pe->primary_pe; | |
835 | ||
836 | /* | |
837 | * If this pe is involved in a write to the origin and | |
838 | * it is the last sibling to complete then release | |
839 | * the bios for the original write to the origin. | |
840 | */ | |
841 | if (primary_pe && | |
7c5f78b9 | 842 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 843 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
844 | free_pending_exception(primary_pe); |
845 | } | |
4b832e8d AK |
846 | |
847 | /* | |
848 | * Free the pe if it's not linked to an origin write or if | |
849 | * it's not itself a primary pe. | |
850 | */ | |
851 | if (!primary_pe || primary_pe != pe) | |
852 | free_pending_exception(pe); | |
853 | ||
4b832e8d AK |
854 | return origin_bios; |
855 | } | |
856 | ||
028867ac | 857 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 858 | { |
028867ac | 859 | struct dm_snap_exception *e; |
1da177e4 | 860 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
861 | struct bio *origin_bios = NULL; |
862 | struct bio *snapshot_bios = NULL; | |
863 | int error = 0; | |
1da177e4 | 864 | |
76df1c65 AK |
865 | if (!success) { |
866 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 867 | down_write(&s->lock); |
695368ac | 868 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 869 | error = 1; |
76df1c65 AK |
870 | goto out; |
871 | } | |
872 | ||
873 | e = alloc_exception(); | |
874 | if (!e) { | |
1da177e4 | 875 | down_write(&s->lock); |
695368ac | 876 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 877 | error = 1; |
76df1c65 AK |
878 | goto out; |
879 | } | |
880 | *e = pe->e; | |
1da177e4 | 881 | |
76df1c65 AK |
882 | down_write(&s->lock); |
883 | if (!s->valid) { | |
76df1c65 | 884 | free_exception(e); |
9d493fa8 | 885 | error = 1; |
76df1c65 | 886 | goto out; |
1da177e4 LT |
887 | } |
888 | ||
a8d41b59 MP |
889 | /* |
890 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 891 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
892 | */ |
893 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 894 | msleep(1); |
a8d41b59 | 895 | |
9d493fa8 AK |
896 | /* |
897 | * Add a proper exception, and remove the | |
898 | * in-flight exception from the list. | |
899 | */ | |
d74f81f8 | 900 | insert_completed_exception(s, e); |
76df1c65 | 901 | |
1da177e4 | 902 | out: |
695368ac | 903 | remove_exception(&pe->e); |
9d493fa8 | 904 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 905 | origin_bios = put_pending_exception(pe); |
1da177e4 | 906 | |
9d493fa8 AK |
907 | up_write(&s->lock); |
908 | ||
909 | /* Submit any pending write bios */ | |
910 | if (error) | |
911 | error_bios(snapshot_bios); | |
912 | else | |
913 | flush_bios(snapshot_bios); | |
914 | ||
915 | flush_bios(origin_bios); | |
1da177e4 LT |
916 | } |
917 | ||
918 | static void commit_callback(void *context, int success) | |
919 | { | |
028867ac AK |
920 | struct dm_snap_pending_exception *pe = context; |
921 | ||
1da177e4 LT |
922 | pending_complete(pe, success); |
923 | } | |
924 | ||
925 | /* | |
926 | * Called when the copy I/O has finished. kcopyd actually runs | |
927 | * this code so don't block. | |
928 | */ | |
4cdc1d1f | 929 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 930 | { |
028867ac | 931 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
932 | struct dm_snapshot *s = pe->snap; |
933 | ||
934 | if (read_err || write_err) | |
935 | pending_complete(pe, 0); | |
936 | ||
937 | else | |
938 | /* Update the metadata if we are persistent */ | |
493df71c JB |
939 | s->store->type->commit_exception(s->store, &pe->e, |
940 | commit_callback, pe); | |
1da177e4 LT |
941 | } |
942 | ||
943 | /* | |
944 | * Dispatches the copy operation to kcopyd. | |
945 | */ | |
028867ac | 946 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
947 | { |
948 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 949 | struct dm_io_region src, dest; |
1da177e4 LT |
950 | struct block_device *bdev = s->origin->bdev; |
951 | sector_t dev_size; | |
952 | ||
953 | dev_size = get_dev_size(bdev); | |
954 | ||
955 | src.bdev = bdev; | |
71fab00a | 956 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
d0216849 | 957 | src.count = min(s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 958 | |
49beb2b8 | 959 | dest.bdev = s->store->cow->bdev; |
71fab00a | 960 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
961 | dest.count = src.count; |
962 | ||
963 | /* Hand over to kcopyd */ | |
eb69aca5 | 964 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
965 | &src, 1, &dest, 0, copy_callback, pe); |
966 | } | |
967 | ||
2913808e MP |
968 | static struct dm_snap_pending_exception * |
969 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
970 | { | |
971 | struct dm_snap_exception *e = lookup_exception(&s->pending, chunk); | |
972 | ||
973 | if (!e) | |
974 | return NULL; | |
975 | ||
976 | return container_of(e, struct dm_snap_pending_exception, e); | |
977 | } | |
978 | ||
1da177e4 LT |
979 | /* |
980 | * Looks to see if this snapshot already has a pending exception | |
981 | * for this chunk, otherwise it allocates a new one and inserts | |
982 | * it into the pending table. | |
983 | * | |
984 | * NOTE: a write lock must be held on snap->lock before calling | |
985 | * this. | |
986 | */ | |
028867ac | 987 | static struct dm_snap_pending_exception * |
c6621392 MP |
988 | __find_pending_exception(struct dm_snapshot *s, |
989 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 990 | { |
c6621392 | 991 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 992 | |
2913808e MP |
993 | pe2 = __lookup_pending_exception(s, chunk); |
994 | if (pe2) { | |
76df1c65 | 995 | free_pending_exception(pe); |
2913808e | 996 | return pe2; |
1da177e4 LT |
997 | } |
998 | ||
76df1c65 AK |
999 | pe->e.old_chunk = chunk; |
1000 | bio_list_init(&pe->origin_bios); | |
1001 | bio_list_init(&pe->snapshot_bios); | |
1002 | pe->primary_pe = NULL; | |
4b832e8d | 1003 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1004 | pe->started = 0; |
1005 | ||
493df71c | 1006 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1007 | free_pending_exception(pe); |
1008 | return NULL; | |
1009 | } | |
1010 | ||
4b832e8d | 1011 | get_pending_exception(pe); |
76df1c65 AK |
1012 | insert_exception(&s->pending, &pe->e); |
1013 | ||
1da177e4 LT |
1014 | return pe; |
1015 | } | |
1016 | ||
028867ac | 1017 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1018 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1019 | { |
49beb2b8 | 1020 | bio->bi_bdev = s->store->cow->bdev; |
71fab00a JB |
1021 | bio->bi_sector = chunk_to_sector(s->store, |
1022 | dm_chunk_number(e->new_chunk) + | |
1023 | (chunk - e->old_chunk)) + | |
1024 | (bio->bi_sector & | |
1025 | s->store->chunk_mask); | |
1da177e4 LT |
1026 | } |
1027 | ||
1028 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1029 | union map_info *map_context) | |
1030 | { | |
028867ac AK |
1031 | struct dm_snap_exception *e; |
1032 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1033 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1034 | chunk_t chunk; |
028867ac | 1035 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1036 | |
71fab00a | 1037 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1038 | |
1039 | /* Full snapshots are not usable */ | |
76df1c65 | 1040 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1041 | if (!s->valid) |
f6a80ea8 | 1042 | return -EIO; |
1da177e4 | 1043 | |
ba40a2aa AK |
1044 | /* FIXME: should only take write lock if we need |
1045 | * to copy an exception */ | |
1046 | down_write(&s->lock); | |
1047 | ||
1048 | if (!s->valid) { | |
1049 | r = -EIO; | |
1050 | goto out_unlock; | |
1051 | } | |
1052 | ||
1053 | /* If the block is already remapped - use that, else remap it */ | |
1054 | e = lookup_exception(&s->complete, chunk); | |
1055 | if (e) { | |
d74f81f8 | 1056 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1057 | goto out_unlock; |
1058 | } | |
1059 | ||
1da177e4 LT |
1060 | /* |
1061 | * Write to snapshot - higher level takes care of RW/RO | |
1062 | * flags so we should only get this if we are | |
1063 | * writeable. | |
1064 | */ | |
1065 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1066 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1067 | if (!pe) { |
c6621392 MP |
1068 | up_write(&s->lock); |
1069 | pe = alloc_pending_exception(s); | |
1070 | down_write(&s->lock); | |
1071 | ||
1072 | if (!s->valid) { | |
1073 | free_pending_exception(pe); | |
1074 | r = -EIO; | |
1075 | goto out_unlock; | |
1076 | } | |
1077 | ||
35bf659b MP |
1078 | e = lookup_exception(&s->complete, chunk); |
1079 | if (e) { | |
1080 | free_pending_exception(pe); | |
1081 | remap_exception(s, e, bio, chunk); | |
1082 | goto out_unlock; | |
1083 | } | |
1084 | ||
c6621392 | 1085 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1086 | if (!pe) { |
1087 | __invalidate_snapshot(s, -ENOMEM); | |
1088 | r = -EIO; | |
1089 | goto out_unlock; | |
1090 | } | |
1da177e4 LT |
1091 | } |
1092 | ||
d74f81f8 | 1093 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1094 | bio_list_add(&pe->snapshot_bios, bio); |
1095 | ||
d2a7ad29 | 1096 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1097 | |
76df1c65 AK |
1098 | if (!pe->started) { |
1099 | /* this is protected by snap->lock */ | |
1100 | pe->started = 1; | |
ba40a2aa | 1101 | up_write(&s->lock); |
76df1c65 | 1102 | start_copy(pe); |
ba40a2aa AK |
1103 | goto out; |
1104 | } | |
cd45daff | 1105 | } else { |
ba40a2aa | 1106 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1107 | map_context->ptr = track_chunk(s, chunk); |
1108 | } | |
1da177e4 | 1109 | |
ba40a2aa AK |
1110 | out_unlock: |
1111 | up_write(&s->lock); | |
1112 | out: | |
1da177e4 LT |
1113 | return r; |
1114 | } | |
1115 | ||
cd45daff MP |
1116 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1117 | int error, union map_info *map_context) | |
1118 | { | |
1119 | struct dm_snapshot *s = ti->private; | |
1120 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1121 | ||
1122 | if (c) | |
1123 | stop_tracking_chunk(s, c); | |
1124 | ||
1125 | return 0; | |
1126 | } | |
1127 | ||
1da177e4 LT |
1128 | static void snapshot_resume(struct dm_target *ti) |
1129 | { | |
028867ac | 1130 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1131 | |
aa14edeb AK |
1132 | down_write(&s->lock); |
1133 | s->active = 1; | |
1134 | up_write(&s->lock); | |
1da177e4 LT |
1135 | } |
1136 | ||
1137 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1138 | char *result, unsigned int maxlen) | |
1139 | { | |
2e4a31df | 1140 | unsigned sz = 0; |
028867ac | 1141 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1142 | |
1143 | switch (type) { | |
1144 | case STATUSTYPE_INFO: | |
1145 | if (!snap->valid) | |
2e4a31df | 1146 | DMEMIT("Invalid"); |
1da177e4 | 1147 | else { |
493df71c | 1148 | if (snap->store->type->fraction_full) { |
1da177e4 | 1149 | sector_t numerator, denominator; |
493df71c JB |
1150 | snap->store->type->fraction_full(snap->store, |
1151 | &numerator, | |
1152 | &denominator); | |
2e4a31df JB |
1153 | DMEMIT("%llu/%llu", |
1154 | (unsigned long long)numerator, | |
1155 | (unsigned long long)denominator); | |
1da177e4 LT |
1156 | } |
1157 | else | |
2e4a31df | 1158 | DMEMIT("Unknown"); |
1da177e4 LT |
1159 | } |
1160 | break; | |
1161 | ||
1162 | case STATUSTYPE_TABLE: | |
1163 | /* | |
1164 | * kdevname returns a static pointer so we need | |
1165 | * to make private copies if the output is to | |
1166 | * make sense. | |
1167 | */ | |
2e4a31df JB |
1168 | DMEMIT("%s", snap->origin->name); |
1169 | DMEMIT(" %s %s %llu", snap->store->cow->name, | |
1170 | snap->store->type->name, | |
1171 | (unsigned long long)snap->store->chunk_size); | |
1da177e4 LT |
1172 | break; |
1173 | } | |
1174 | ||
1175 | return 0; | |
1176 | } | |
1177 | ||
1178 | /*----------------------------------------------------------------- | |
1179 | * Origin methods | |
1180 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1181 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1182 | { | |
d2a7ad29 | 1183 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1184 | struct dm_snapshot *snap; |
028867ac AK |
1185 | struct dm_snap_exception *e; |
1186 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1187 | chunk_t chunk; |
eccf0817 | 1188 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1189 | |
1190 | /* Do all the snapshots on this origin */ | |
1191 | list_for_each_entry (snap, snapshots, list) { | |
1192 | ||
76df1c65 AK |
1193 | down_write(&snap->lock); |
1194 | ||
aa14edeb AK |
1195 | /* Only deal with valid and active snapshots */ |
1196 | if (!snap->valid || !snap->active) | |
76df1c65 | 1197 | goto next_snapshot; |
1da177e4 | 1198 | |
d5e404c1 | 1199 | /* Nothing to do if writing beyond end of snapshot */ |
0cea9c78 | 1200 | if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table)) |
76df1c65 | 1201 | goto next_snapshot; |
1da177e4 LT |
1202 | |
1203 | /* | |
1204 | * Remember, different snapshots can have | |
1205 | * different chunk sizes. | |
1206 | */ | |
71fab00a | 1207 | chunk = sector_to_chunk(snap->store, bio->bi_sector); |
1da177e4 LT |
1208 | |
1209 | /* | |
1210 | * Check exception table to see if block | |
1211 | * is already remapped in this snapshot | |
1212 | * and trigger an exception if not. | |
b4b610f6 | 1213 | * |
4b832e8d | 1214 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1215 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1216 | */ |
1217 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1218 | if (e) |
1219 | goto next_snapshot; | |
1220 | ||
2913808e | 1221 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1222 | if (!pe) { |
c6621392 MP |
1223 | up_write(&snap->lock); |
1224 | pe = alloc_pending_exception(snap); | |
1225 | down_write(&snap->lock); | |
1226 | ||
1227 | if (!snap->valid) { | |
1228 | free_pending_exception(pe); | |
1229 | goto next_snapshot; | |
1230 | } | |
1231 | ||
35bf659b MP |
1232 | e = lookup_exception(&snap->complete, chunk); |
1233 | if (e) { | |
1234 | free_pending_exception(pe); | |
1235 | goto next_snapshot; | |
1236 | } | |
1237 | ||
c6621392 | 1238 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1239 | if (!pe) { |
1240 | __invalidate_snapshot(snap, -ENOMEM); | |
1241 | goto next_snapshot; | |
1242 | } | |
76df1c65 AK |
1243 | } |
1244 | ||
1245 | if (!primary_pe) { | |
1246 | /* | |
1247 | * Either every pe here has same | |
1248 | * primary_pe or none has one yet. | |
1249 | */ | |
1250 | if (pe->primary_pe) | |
1251 | primary_pe = pe->primary_pe; | |
1252 | else { | |
1253 | primary_pe = pe; | |
1254 | first = 1; | |
1da177e4 | 1255 | } |
76df1c65 AK |
1256 | |
1257 | bio_list_add(&primary_pe->origin_bios, bio); | |
1258 | ||
d2a7ad29 | 1259 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1260 | } |
1261 | ||
1262 | if (!pe->primary_pe) { | |
76df1c65 | 1263 | pe->primary_pe = primary_pe; |
4b832e8d | 1264 | get_pending_exception(primary_pe); |
76df1c65 AK |
1265 | } |
1266 | ||
1267 | if (!pe->started) { | |
1268 | pe->started = 1; | |
1269 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1270 | } |
1271 | ||
76df1c65 | 1272 | next_snapshot: |
1da177e4 LT |
1273 | up_write(&snap->lock); |
1274 | } | |
1275 | ||
b4b610f6 | 1276 | if (!primary_pe) |
4b832e8d | 1277 | return r; |
b4b610f6 AK |
1278 | |
1279 | /* | |
1280 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1281 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1282 | * got completed while we were in the loop above, so it falls to |
1283 | * us here to remove the primary_pe and submit any origin_bios. | |
1284 | */ | |
1285 | ||
4b832e8d | 1286 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1287 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1288 | free_pending_exception(primary_pe); | |
1289 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1290 | return r; |
b4b610f6 AK |
1291 | } |
1292 | ||
1da177e4 LT |
1293 | /* |
1294 | * Now that we have a complete pe list we can start the copying. | |
1295 | */ | |
eccf0817 AK |
1296 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1297 | start_copy(pe); | |
1da177e4 LT |
1298 | |
1299 | return r; | |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * Called on a write from the origin driver. | |
1304 | */ | |
1305 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1306 | { | |
1307 | struct origin *o; | |
d2a7ad29 | 1308 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1309 | |
1310 | down_read(&_origins_lock); | |
1311 | o = __lookup_origin(origin->bdev); | |
1312 | if (o) | |
1313 | r = __origin_write(&o->snapshots, bio); | |
1314 | up_read(&_origins_lock); | |
1315 | ||
1316 | return r; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1321 | */ | |
1322 | ||
1323 | /* | |
1324 | * Construct an origin mapping: <dev_path> | |
1325 | * The context for an origin is merely a 'struct dm_dev *' | |
1326 | * pointing to the real device. | |
1327 | */ | |
1328 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1329 | { | |
1330 | int r; | |
1331 | struct dm_dev *dev; | |
1332 | ||
1333 | if (argc != 1) { | |
72d94861 | 1334 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1335 | return -EINVAL; |
1336 | } | |
1337 | ||
1338 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1339 | dm_table_get_mode(ti->table), &dev); | |
1340 | if (r) { | |
1341 | ti->error = "Cannot get target device"; | |
1342 | return r; | |
1343 | } | |
1344 | ||
1345 | ti->private = dev; | |
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | static void origin_dtr(struct dm_target *ti) | |
1350 | { | |
028867ac | 1351 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1352 | dm_put_device(ti, dev); |
1353 | } | |
1354 | ||
1355 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1356 | union map_info *map_context) | |
1357 | { | |
028867ac | 1358 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1359 | bio->bi_bdev = dev->bdev; |
1360 | ||
1361 | /* Only tell snapshots if this is a write */ | |
d2a7ad29 | 1362 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1363 | } |
1364 | ||
1365 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
1366 | ||
1367 | /* | |
1368 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1369 | * chunk sizes. | |
1370 | */ | |
1371 | static void origin_resume(struct dm_target *ti) | |
1372 | { | |
028867ac | 1373 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1374 | struct dm_snapshot *snap; |
1375 | struct origin *o; | |
1376 | chunk_t chunk_size = 0; | |
1377 | ||
1378 | down_read(&_origins_lock); | |
1379 | o = __lookup_origin(dev->bdev); | |
1380 | if (o) | |
1381 | list_for_each_entry (snap, &o->snapshots, list) | |
d0216849 JB |
1382 | chunk_size = min_not_zero(chunk_size, |
1383 | snap->store->chunk_size); | |
1da177e4 LT |
1384 | up_read(&_origins_lock); |
1385 | ||
1386 | ti->split_io = chunk_size; | |
1387 | } | |
1388 | ||
1389 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1390 | unsigned int maxlen) | |
1391 | { | |
028867ac | 1392 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1393 | |
1394 | switch (type) { | |
1395 | case STATUSTYPE_INFO: | |
1396 | result[0] = '\0'; | |
1397 | break; | |
1398 | ||
1399 | case STATUSTYPE_TABLE: | |
1400 | snprintf(result, maxlen, "%s", dev->name); | |
1401 | break; | |
1402 | } | |
1403 | ||
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | static struct target_type origin_target = { | |
1408 | .name = "snapshot-origin", | |
d74f81f8 | 1409 | .version = {1, 6, 0}, |
1da177e4 LT |
1410 | .module = THIS_MODULE, |
1411 | .ctr = origin_ctr, | |
1412 | .dtr = origin_dtr, | |
1413 | .map = origin_map, | |
1414 | .resume = origin_resume, | |
1415 | .status = origin_status, | |
1416 | }; | |
1417 | ||
1418 | static struct target_type snapshot_target = { | |
1419 | .name = "snapshot", | |
d74f81f8 | 1420 | .version = {1, 6, 0}, |
1da177e4 LT |
1421 | .module = THIS_MODULE, |
1422 | .ctr = snapshot_ctr, | |
1423 | .dtr = snapshot_dtr, | |
1424 | .map = snapshot_map, | |
cd45daff | 1425 | .end_io = snapshot_end_io, |
1da177e4 LT |
1426 | .resume = snapshot_resume, |
1427 | .status = snapshot_status, | |
1428 | }; | |
1429 | ||
1430 | static int __init dm_snapshot_init(void) | |
1431 | { | |
1432 | int r; | |
1433 | ||
4db6bfe0 AK |
1434 | r = dm_exception_store_init(); |
1435 | if (r) { | |
1436 | DMERR("Failed to initialize exception stores"); | |
1437 | return r; | |
1438 | } | |
1439 | ||
1da177e4 LT |
1440 | r = dm_register_target(&snapshot_target); |
1441 | if (r) { | |
1442 | DMERR("snapshot target register failed %d", r); | |
1443 | return r; | |
1444 | } | |
1445 | ||
1446 | r = dm_register_target(&origin_target); | |
1447 | if (r < 0) { | |
72d94861 | 1448 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1449 | goto bad1; |
1450 | } | |
1451 | ||
1452 | r = init_origin_hash(); | |
1453 | if (r) { | |
1454 | DMERR("init_origin_hash failed."); | |
1455 | goto bad2; | |
1456 | } | |
1457 | ||
028867ac | 1458 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1459 | if (!exception_cache) { |
1460 | DMERR("Couldn't create exception cache."); | |
1461 | r = -ENOMEM; | |
1462 | goto bad3; | |
1463 | } | |
1464 | ||
028867ac | 1465 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1466 | if (!pending_cache) { |
1467 | DMERR("Couldn't create pending cache."); | |
1468 | r = -ENOMEM; | |
1469 | goto bad4; | |
1470 | } | |
1471 | ||
cd45daff MP |
1472 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1473 | if (!tracked_chunk_cache) { | |
1474 | DMERR("Couldn't create cache to track chunks in use."); | |
1475 | r = -ENOMEM; | |
1476 | goto bad5; | |
1477 | } | |
1478 | ||
ca3a931f AK |
1479 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1480 | if (!ksnapd) { | |
1481 | DMERR("Failed to create ksnapd workqueue."); | |
1482 | r = -ENOMEM; | |
92e86812 | 1483 | goto bad_pending_pool; |
ca3a931f AK |
1484 | } |
1485 | ||
1da177e4 LT |
1486 | return 0; |
1487 | ||
4db6bfe0 | 1488 | bad_pending_pool: |
cd45daff | 1489 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1490 | bad5: |
1da177e4 | 1491 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1492 | bad4: |
1da177e4 | 1493 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1494 | bad3: |
1da177e4 | 1495 | exit_origin_hash(); |
4db6bfe0 | 1496 | bad2: |
1da177e4 | 1497 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1498 | bad1: |
1da177e4 LT |
1499 | dm_unregister_target(&snapshot_target); |
1500 | return r; | |
1501 | } | |
1502 | ||
1503 | static void __exit dm_snapshot_exit(void) | |
1504 | { | |
ca3a931f AK |
1505 | destroy_workqueue(ksnapd); |
1506 | ||
10d3bd09 MP |
1507 | dm_unregister_target(&snapshot_target); |
1508 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1509 | |
1510 | exit_origin_hash(); | |
1da177e4 LT |
1511 | kmem_cache_destroy(pending_cache); |
1512 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1513 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1514 | |
1515 | dm_exception_store_exit(); | |
1da177e4 LT |
1516 | } |
1517 | ||
1518 | /* Module hooks */ | |
1519 | module_init(dm_snapshot_init); | |
1520 | module_exit(dm_snapshot_exit); | |
1521 | ||
1522 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1523 | MODULE_AUTHOR("Joe Thornber"); | |
1524 | MODULE_LICENSE("GPL"); |