]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * dm-snapshot.c | |
3 | * | |
4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
5 | * | |
6 | * This file is released under the GPL. | |
7 | */ | |
8 | ||
9 | #include <linux/blkdev.h> | |
1da177e4 | 10 | #include <linux/device-mapper.h> |
90fa1527 | 11 | #include <linux/delay.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/init.h> | |
14 | #include <linux/kdev_t.h> | |
15 | #include <linux/list.h> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
a765e20e | 21 | #include <linux/dm-kcopyd.h> |
ccc45ea8 | 22 | #include <linux/workqueue.h> |
1da177e4 | 23 | |
aea53d92 | 24 | #include "dm-exception-store.h" |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "snapshots" |
27 | ||
1da177e4 LT |
28 | /* |
29 | * The percentage increment we will wake up users at | |
30 | */ | |
31 | #define WAKE_UP_PERCENT 5 | |
32 | ||
33 | /* | |
34 | * kcopyd priority of snapshot operations | |
35 | */ | |
36 | #define SNAPSHOT_COPY_PRIORITY 2 | |
37 | ||
38 | /* | |
8ee2767a | 39 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 40 | */ |
8ee2767a | 41 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 42 | |
cd45daff MP |
43 | /* |
44 | * The size of the mempool used to track chunks in use. | |
45 | */ | |
46 | #define MIN_IOS 256 | |
47 | ||
ccc45ea8 JB |
48 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
49 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
50 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
51 | ||
52 | struct exception_table { | |
53 | uint32_t hash_mask; | |
54 | unsigned hash_shift; | |
55 | struct list_head *table; | |
56 | }; | |
57 | ||
58 | struct dm_snapshot { | |
59 | struct rw_semaphore lock; | |
60 | ||
61 | struct dm_dev *origin; | |
62 | ||
63 | /* List of snapshots per Origin */ | |
64 | struct list_head list; | |
65 | ||
66 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
67 | int valid; | |
68 | ||
69 | /* Origin writes don't trigger exceptions until this is set */ | |
70 | int active; | |
71 | ||
ccc45ea8 JB |
72 | mempool_t *pending_pool; |
73 | ||
74 | atomic_t pending_exceptions_count; | |
75 | ||
76 | struct exception_table pending; | |
77 | struct exception_table complete; | |
78 | ||
79 | /* | |
80 | * pe_lock protects all pending_exception operations and access | |
81 | * as well as the snapshot_bios list. | |
82 | */ | |
83 | spinlock_t pe_lock; | |
84 | ||
85 | /* The on disk metadata handler */ | |
86 | struct dm_exception_store *store; | |
87 | ||
88 | struct dm_kcopyd_client *kcopyd_client; | |
89 | ||
90 | /* Queue of snapshot writes for ksnapd to flush */ | |
91 | struct bio_list queued_bios; | |
92 | struct work_struct queued_bios_work; | |
93 | ||
94 | /* Chunks with outstanding reads */ | |
95 | mempool_t *tracked_chunk_pool; | |
96 | spinlock_t tracked_chunk_lock; | |
97 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
98 | }; | |
99 | ||
c642f9e0 | 100 | static struct workqueue_struct *ksnapd; |
c4028958 | 101 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 102 | |
ccc45ea8 JB |
103 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
104 | chunk_t chunk) | |
105 | { | |
106 | return chunk << store->chunk_shift; | |
107 | } | |
108 | ||
109 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
110 | { | |
111 | /* | |
112 | * There is only ever one instance of a particular block | |
113 | * device so we can compare pointers safely. | |
114 | */ | |
115 | return lhs == rhs; | |
116 | } | |
117 | ||
028867ac AK |
118 | struct dm_snap_pending_exception { |
119 | struct dm_snap_exception e; | |
1da177e4 LT |
120 | |
121 | /* | |
122 | * Origin buffers waiting for this to complete are held | |
123 | * in a bio list | |
124 | */ | |
125 | struct bio_list origin_bios; | |
126 | struct bio_list snapshot_bios; | |
127 | ||
eccf0817 AK |
128 | /* |
129 | * Short-term queue of pending exceptions prior to submission. | |
130 | */ | |
131 | struct list_head list; | |
132 | ||
1da177e4 | 133 | /* |
b4b610f6 | 134 | * The primary pending_exception is the one that holds |
4b832e8d | 135 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
136 | * group of pending_exceptions. It is always last to get freed. |
137 | * These fields get set up when writing to the origin. | |
1da177e4 | 138 | */ |
028867ac | 139 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
140 | |
141 | /* | |
142 | * Number of pending_exceptions processing this chunk. | |
143 | * When this drops to zero we must complete the origin bios. | |
144 | * If incrementing or decrementing this, hold pe->snap->lock for | |
145 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
146 | * they are the same. | |
147 | */ | |
4b832e8d | 148 | atomic_t ref_count; |
1da177e4 LT |
149 | |
150 | /* Pointer back to snapshot context */ | |
151 | struct dm_snapshot *snap; | |
152 | ||
153 | /* | |
154 | * 1 indicates the exception has already been sent to | |
155 | * kcopyd. | |
156 | */ | |
157 | int started; | |
158 | }; | |
159 | ||
160 | /* | |
161 | * Hash table mapping origin volumes to lists of snapshots and | |
162 | * a lock to protect it | |
163 | */ | |
e18b890b CL |
164 | static struct kmem_cache *exception_cache; |
165 | static struct kmem_cache *pending_cache; | |
1da177e4 | 166 | |
cd45daff MP |
167 | struct dm_snap_tracked_chunk { |
168 | struct hlist_node node; | |
169 | chunk_t chunk; | |
170 | }; | |
171 | ||
172 | static struct kmem_cache *tracked_chunk_cache; | |
173 | ||
174 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
175 | chunk_t chunk) | |
176 | { | |
177 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
178 | GFP_NOIO); | |
179 | unsigned long flags; | |
180 | ||
181 | c->chunk = chunk; | |
182 | ||
183 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
184 | hlist_add_head(&c->node, | |
185 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
186 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
187 | ||
188 | return c; | |
189 | } | |
190 | ||
191 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
192 | struct dm_snap_tracked_chunk *c) | |
193 | { | |
194 | unsigned long flags; | |
195 | ||
196 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
197 | hlist_del(&c->node); | |
198 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
199 | ||
200 | mempool_free(c, s->tracked_chunk_pool); | |
201 | } | |
202 | ||
a8d41b59 MP |
203 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
204 | { | |
205 | struct dm_snap_tracked_chunk *c; | |
206 | struct hlist_node *hn; | |
207 | int found = 0; | |
208 | ||
209 | spin_lock_irq(&s->tracked_chunk_lock); | |
210 | ||
211 | hlist_for_each_entry(c, hn, | |
212 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
213 | if (c->chunk == chunk) { | |
214 | found = 1; | |
215 | break; | |
216 | } | |
217 | } | |
218 | ||
219 | spin_unlock_irq(&s->tracked_chunk_lock); | |
220 | ||
221 | return found; | |
222 | } | |
223 | ||
1da177e4 LT |
224 | /* |
225 | * One of these per registered origin, held in the snapshot_origins hash | |
226 | */ | |
227 | struct origin { | |
228 | /* The origin device */ | |
229 | struct block_device *bdev; | |
230 | ||
231 | struct list_head hash_list; | |
232 | ||
233 | /* List of snapshots for this origin */ | |
234 | struct list_head snapshots; | |
235 | }; | |
236 | ||
237 | /* | |
238 | * Size of the hash table for origin volumes. If we make this | |
239 | * the size of the minors list then it should be nearly perfect | |
240 | */ | |
241 | #define ORIGIN_HASH_SIZE 256 | |
242 | #define ORIGIN_MASK 0xFF | |
243 | static struct list_head *_origins; | |
244 | static struct rw_semaphore _origins_lock; | |
245 | ||
246 | static int init_origin_hash(void) | |
247 | { | |
248 | int i; | |
249 | ||
250 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
251 | GFP_KERNEL); | |
252 | if (!_origins) { | |
72d94861 | 253 | DMERR("unable to allocate memory"); |
1da177e4 LT |
254 | return -ENOMEM; |
255 | } | |
256 | ||
257 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
258 | INIT_LIST_HEAD(_origins + i); | |
259 | init_rwsem(&_origins_lock); | |
260 | ||
261 | return 0; | |
262 | } | |
263 | ||
264 | static void exit_origin_hash(void) | |
265 | { | |
266 | kfree(_origins); | |
267 | } | |
268 | ||
028867ac | 269 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
270 | { |
271 | return bdev->bd_dev & ORIGIN_MASK; | |
272 | } | |
273 | ||
274 | static struct origin *__lookup_origin(struct block_device *origin) | |
275 | { | |
276 | struct list_head *ol; | |
277 | struct origin *o; | |
278 | ||
279 | ol = &_origins[origin_hash(origin)]; | |
280 | list_for_each_entry (o, ol, hash_list) | |
281 | if (bdev_equal(o->bdev, origin)) | |
282 | return o; | |
283 | ||
284 | return NULL; | |
285 | } | |
286 | ||
287 | static void __insert_origin(struct origin *o) | |
288 | { | |
289 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
290 | list_add_tail(&o->hash_list, sl); | |
291 | } | |
292 | ||
293 | /* | |
294 | * Make a note of the snapshot and its origin so we can look it | |
295 | * up when the origin has a write on it. | |
296 | */ | |
297 | static int register_snapshot(struct dm_snapshot *snap) | |
298 | { | |
6d45d93e | 299 | struct dm_snapshot *l; |
60c856c8 | 300 | struct origin *o, *new_o; |
1da177e4 LT |
301 | struct block_device *bdev = snap->origin->bdev; |
302 | ||
60c856c8 MP |
303 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
304 | if (!new_o) | |
305 | return -ENOMEM; | |
306 | ||
1da177e4 LT |
307 | down_write(&_origins_lock); |
308 | o = __lookup_origin(bdev); | |
309 | ||
60c856c8 MP |
310 | if (o) |
311 | kfree(new_o); | |
312 | else { | |
1da177e4 | 313 | /* New origin */ |
60c856c8 | 314 | o = new_o; |
1da177e4 LT |
315 | |
316 | /* Initialise the struct */ | |
317 | INIT_LIST_HEAD(&o->snapshots); | |
318 | o->bdev = bdev; | |
319 | ||
320 | __insert_origin(o); | |
321 | } | |
322 | ||
6d45d93e MP |
323 | /* Sort the list according to chunk size, largest-first smallest-last */ |
324 | list_for_each_entry(l, &o->snapshots, list) | |
325 | if (l->store->chunk_size < snap->store->chunk_size) | |
326 | break; | |
327 | list_add_tail(&snap->list, &l->list); | |
1da177e4 LT |
328 | |
329 | up_write(&_origins_lock); | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static void unregister_snapshot(struct dm_snapshot *s) | |
334 | { | |
335 | struct origin *o; | |
336 | ||
337 | down_write(&_origins_lock); | |
338 | o = __lookup_origin(s->origin->bdev); | |
339 | ||
340 | list_del(&s->list); | |
341 | if (list_empty(&o->snapshots)) { | |
342 | list_del(&o->hash_list); | |
343 | kfree(o); | |
344 | } | |
345 | ||
346 | up_write(&_origins_lock); | |
347 | } | |
348 | ||
349 | /* | |
350 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
351 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
352 | * some consecutive chunks to be grouped together. | |
1da177e4 | 353 | */ |
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 | ||
7e201b35 MP |
526 | #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r))) |
527 | ||
528 | /* | |
529 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
530 | * Return zero if the origin has no snapshots. | |
531 | */ | |
532 | static sector_t __minimum_chunk_size(struct origin *o) | |
533 | { | |
534 | struct dm_snapshot *snap; | |
535 | unsigned chunk_size = 0; | |
536 | ||
537 | if (o) | |
538 | list_for_each_entry(snap, &o->snapshots, list) | |
539 | chunk_size = min_not_zero(chunk_size, | |
540 | snap->store->chunk_size); | |
541 | ||
542 | return chunk_size; | |
543 | } | |
544 | ||
1da177e4 LT |
545 | /* |
546 | * Hard coded magic. | |
547 | */ | |
548 | static int calc_max_buckets(void) | |
549 | { | |
550 | /* use a fixed size of 2MB */ | |
551 | unsigned long mem = 2 * 1024 * 1024; | |
552 | mem /= sizeof(struct list_head); | |
553 | ||
554 | return mem; | |
555 | } | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * Allocate room for a suitable hash table. | |
559 | */ | |
fee1998e | 560 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
561 | { |
562 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
563 | ||
564 | /* | |
565 | * Calculate based on the size of the original volume or | |
566 | * the COW volume... | |
567 | */ | |
fee1998e | 568 | cow_dev_size = get_dev_size(s->store->cow->bdev); |
1da177e4 LT |
569 | origin_dev_size = get_dev_size(s->origin->bdev); |
570 | max_buckets = calc_max_buckets(); | |
571 | ||
fee1998e | 572 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
573 | hash_size = min(hash_size, max_buckets); |
574 | ||
8e87b9b8 MP |
575 | if (hash_size < 64) |
576 | hash_size = 64; | |
8defd830 | 577 | hash_size = rounddown_pow_of_two(hash_size); |
d74f81f8 MB |
578 | if (init_exception_table(&s->complete, hash_size, |
579 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
580 | return -ENOMEM; |
581 | ||
582 | /* | |
583 | * Allocate hash table for in-flight exceptions | |
584 | * Make this smaller than the real hash table | |
585 | */ | |
586 | hash_size >>= 3; | |
587 | if (hash_size < 64) | |
588 | hash_size = 64; | |
589 | ||
d74f81f8 | 590 | if (init_exception_table(&s->pending, hash_size, 0)) { |
1da177e4 LT |
591 | exit_exception_table(&s->complete, exception_cache); |
592 | return -ENOMEM; | |
593 | } | |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
1da177e4 LT |
598 | /* |
599 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
600 | */ | |
601 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
602 | { | |
603 | struct dm_snapshot *s; | |
cd45daff | 604 | int i; |
1da177e4 | 605 | int r = -EINVAL; |
1da177e4 | 606 | char *origin_path; |
fee1998e JB |
607 | struct dm_exception_store *store; |
608 | unsigned args_used; | |
1da177e4 | 609 | |
4c7e3bf4 | 610 | if (argc != 4) { |
72d94861 | 611 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 612 | r = -EINVAL; |
fee1998e | 613 | goto bad_args; |
1da177e4 LT |
614 | } |
615 | ||
616 | origin_path = argv[0]; | |
fee1998e JB |
617 | argv++; |
618 | argc--; | |
1da177e4 | 619 | |
fee1998e JB |
620 | r = dm_exception_store_create(ti, argc, argv, &args_used, &store); |
621 | if (r) { | |
622 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 623 | r = -EINVAL; |
fee1998e | 624 | goto bad_args; |
1da177e4 LT |
625 | } |
626 | ||
fee1998e JB |
627 | argv += args_used; |
628 | argc -= args_used; | |
629 | ||
1da177e4 | 630 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
fee1998e | 631 | if (!s) { |
1da177e4 LT |
632 | ti->error = "Cannot allocate snapshot context private " |
633 | "structure"; | |
634 | r = -ENOMEM; | |
fee1998e | 635 | goto bad_snap; |
1da177e4 LT |
636 | } |
637 | ||
638 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); | |
639 | if (r) { | |
640 | ti->error = "Cannot get origin device"; | |
fee1998e | 641 | goto bad_origin; |
1da177e4 LT |
642 | } |
643 | ||
fee1998e | 644 | s->store = store; |
1da177e4 | 645 | s->valid = 1; |
aa14edeb | 646 | s->active = 0; |
879129d2 | 647 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 648 | init_rwsem(&s->lock); |
ca3a931f | 649 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
650 | |
651 | /* Allocate hash table for COW data */ | |
fee1998e | 652 | if (init_hash_tables(s)) { |
1da177e4 LT |
653 | ti->error = "Unable to allocate hash table space"; |
654 | r = -ENOMEM; | |
fee1998e | 655 | goto bad_hash_tables; |
1da177e4 LT |
656 | } |
657 | ||
eb69aca5 | 658 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
659 | if (r) { |
660 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 661 | goto bad_kcopyd; |
1da177e4 LT |
662 | } |
663 | ||
92e86812 MP |
664 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
665 | if (!s->pending_pool) { | |
666 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 667 | goto bad_pending_pool; |
92e86812 MP |
668 | } |
669 | ||
cd45daff MP |
670 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
671 | tracked_chunk_cache); | |
672 | if (!s->tracked_chunk_pool) { | |
673 | ti->error = "Could not allocate tracked_chunk mempool for " | |
674 | "tracking reads"; | |
92e86812 | 675 | goto bad_tracked_chunk_pool; |
cd45daff MP |
676 | } |
677 | ||
678 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
679 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
680 | ||
681 | spin_lock_init(&s->tracked_chunk_lock); | |
682 | ||
aa14edeb | 683 | /* Metadata must only be loaded into one table at once */ |
493df71c JB |
684 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
685 | (void *)s); | |
0764147b | 686 | if (r < 0) { |
f9cea4f7 | 687 | ti->error = "Failed to read snapshot metadata"; |
cd45daff | 688 | goto bad_load_and_register; |
0764147b MB |
689 | } else if (r > 0) { |
690 | s->valid = 0; | |
691 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 692 | } |
aa14edeb | 693 | |
ca3a931f | 694 | bio_list_init(&s->queued_bios); |
c4028958 | 695 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); |
ca3a931f | 696 | |
3f2412dc MP |
697 | if (!s->store->chunk_size) { |
698 | ti->error = "Chunk size not set"; | |
699 | goto bad_load_and_register; | |
700 | } | |
701 | ||
1da177e4 | 702 | /* Add snapshot to the list of snapshots for this origin */ |
aa14edeb | 703 | /* Exceptions aren't triggered till snapshot_resume() is called */ |
1da177e4 LT |
704 | if (register_snapshot(s)) { |
705 | r = -EINVAL; | |
706 | ti->error = "Cannot register snapshot origin"; | |
cd45daff | 707 | goto bad_load_and_register; |
1da177e4 LT |
708 | } |
709 | ||
710 | ti->private = s; | |
d0216849 | 711 | ti->split_io = s->store->chunk_size; |
494b3ee7 | 712 | ti->num_flush_requests = 1; |
1da177e4 LT |
713 | |
714 | return 0; | |
715 | ||
fee1998e | 716 | bad_load_and_register: |
cd45daff MP |
717 | mempool_destroy(s->tracked_chunk_pool); |
718 | ||
fee1998e | 719 | bad_tracked_chunk_pool: |
92e86812 MP |
720 | mempool_destroy(s->pending_pool); |
721 | ||
fee1998e | 722 | bad_pending_pool: |
eb69aca5 | 723 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 724 | |
fee1998e | 725 | bad_kcopyd: |
1da177e4 LT |
726 | exit_exception_table(&s->pending, pending_cache); |
727 | exit_exception_table(&s->complete, exception_cache); | |
728 | ||
fee1998e | 729 | bad_hash_tables: |
1da177e4 LT |
730 | dm_put_device(ti, s->origin); |
731 | ||
fee1998e | 732 | bad_origin: |
1da177e4 LT |
733 | kfree(s); |
734 | ||
fee1998e JB |
735 | bad_snap: |
736 | dm_exception_store_destroy(store); | |
737 | ||
738 | bad_args: | |
1da177e4 LT |
739 | return r; |
740 | } | |
741 | ||
31c93a0c MB |
742 | static void __free_exceptions(struct dm_snapshot *s) |
743 | { | |
eb69aca5 | 744 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
745 | s->kcopyd_client = NULL; |
746 | ||
747 | exit_exception_table(&s->pending, pending_cache); | |
748 | exit_exception_table(&s->complete, exception_cache); | |
31c93a0c MB |
749 | } |
750 | ||
1da177e4 LT |
751 | static void snapshot_dtr(struct dm_target *ti) |
752 | { | |
cd45daff MP |
753 | #ifdef CONFIG_DM_DEBUG |
754 | int i; | |
755 | #endif | |
028867ac | 756 | struct dm_snapshot *s = ti->private; |
1da177e4 | 757 | |
ca3a931f AK |
758 | flush_workqueue(ksnapd); |
759 | ||
138728dc AK |
760 | /* Prevent further origin writes from using this snapshot. */ |
761 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
762 | unregister_snapshot(s); |
763 | ||
879129d2 | 764 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 765 | msleep(1); |
879129d2 MP |
766 | /* |
767 | * Ensure instructions in mempool_destroy aren't reordered | |
768 | * before atomic_read. | |
769 | */ | |
770 | smp_mb(); | |
771 | ||
cd45daff MP |
772 | #ifdef CONFIG_DM_DEBUG |
773 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
774 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
775 | #endif | |
776 | ||
777 | mempool_destroy(s->tracked_chunk_pool); | |
778 | ||
31c93a0c | 779 | __free_exceptions(s); |
1da177e4 | 780 | |
92e86812 MP |
781 | mempool_destroy(s->pending_pool); |
782 | ||
1da177e4 | 783 | dm_put_device(ti, s->origin); |
fee1998e JB |
784 | |
785 | dm_exception_store_destroy(s->store); | |
138728dc | 786 | |
1da177e4 LT |
787 | kfree(s); |
788 | } | |
789 | ||
790 | /* | |
791 | * Flush a list of buffers. | |
792 | */ | |
793 | static void flush_bios(struct bio *bio) | |
794 | { | |
795 | struct bio *n; | |
796 | ||
797 | while (bio) { | |
798 | n = bio->bi_next; | |
799 | bio->bi_next = NULL; | |
800 | generic_make_request(bio); | |
801 | bio = n; | |
802 | } | |
803 | } | |
804 | ||
c4028958 | 805 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 806 | { |
c4028958 DH |
807 | struct dm_snapshot *s = |
808 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
809 | struct bio *queued_bios; |
810 | unsigned long flags; | |
811 | ||
812 | spin_lock_irqsave(&s->pe_lock, flags); | |
813 | queued_bios = bio_list_get(&s->queued_bios); | |
814 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
815 | ||
816 | flush_bios(queued_bios); | |
817 | } | |
818 | ||
1da177e4 LT |
819 | /* |
820 | * Error a list of buffers. | |
821 | */ | |
822 | static void error_bios(struct bio *bio) | |
823 | { | |
824 | struct bio *n; | |
825 | ||
826 | while (bio) { | |
827 | n = bio->bi_next; | |
828 | bio->bi_next = NULL; | |
6712ecf8 | 829 | bio_io_error(bio); |
1da177e4 LT |
830 | bio = n; |
831 | } | |
832 | } | |
833 | ||
695368ac | 834 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
835 | { |
836 | if (!s->valid) | |
837 | return; | |
838 | ||
839 | if (err == -EIO) | |
840 | DMERR("Invalidating snapshot: Error reading/writing."); | |
841 | else if (err == -ENOMEM) | |
842 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
843 | ||
493df71c JB |
844 | if (s->store->type->drop_snapshot) |
845 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
846 | |
847 | s->valid = 0; | |
848 | ||
0cea9c78 | 849 | dm_table_event(s->store->ti->table); |
76df1c65 AK |
850 | } |
851 | ||
028867ac | 852 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
853 | { |
854 | atomic_inc(&pe->ref_count); | |
855 | } | |
856 | ||
028867ac | 857 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 858 | { |
028867ac | 859 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
860 | struct bio *origin_bios = NULL; |
861 | ||
862 | primary_pe = pe->primary_pe; | |
863 | ||
864 | /* | |
865 | * If this pe is involved in a write to the origin and | |
866 | * it is the last sibling to complete then release | |
867 | * the bios for the original write to the origin. | |
868 | */ | |
869 | if (primary_pe && | |
7c5f78b9 | 870 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 871 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
872 | free_pending_exception(primary_pe); |
873 | } | |
4b832e8d AK |
874 | |
875 | /* | |
876 | * Free the pe if it's not linked to an origin write or if | |
877 | * it's not itself a primary pe. | |
878 | */ | |
879 | if (!primary_pe || primary_pe != pe) | |
880 | free_pending_exception(pe); | |
881 | ||
4b832e8d AK |
882 | return origin_bios; |
883 | } | |
884 | ||
028867ac | 885 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 886 | { |
028867ac | 887 | struct dm_snap_exception *e; |
1da177e4 | 888 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
889 | struct bio *origin_bios = NULL; |
890 | struct bio *snapshot_bios = NULL; | |
891 | int error = 0; | |
1da177e4 | 892 | |
76df1c65 AK |
893 | if (!success) { |
894 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 895 | down_write(&s->lock); |
695368ac | 896 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 897 | error = 1; |
76df1c65 AK |
898 | goto out; |
899 | } | |
900 | ||
901 | e = alloc_exception(); | |
902 | if (!e) { | |
1da177e4 | 903 | down_write(&s->lock); |
695368ac | 904 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 905 | error = 1; |
76df1c65 AK |
906 | goto out; |
907 | } | |
908 | *e = pe->e; | |
1da177e4 | 909 | |
76df1c65 AK |
910 | down_write(&s->lock); |
911 | if (!s->valid) { | |
76df1c65 | 912 | free_exception(e); |
9d493fa8 | 913 | error = 1; |
76df1c65 | 914 | goto out; |
1da177e4 LT |
915 | } |
916 | ||
a8d41b59 MP |
917 | /* |
918 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 919 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
920 | */ |
921 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 922 | msleep(1); |
a8d41b59 | 923 | |
9d493fa8 AK |
924 | /* |
925 | * Add a proper exception, and remove the | |
926 | * in-flight exception from the list. | |
927 | */ | |
d74f81f8 | 928 | insert_completed_exception(s, e); |
76df1c65 | 929 | |
1da177e4 | 930 | out: |
695368ac | 931 | remove_exception(&pe->e); |
9d493fa8 | 932 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 933 | origin_bios = put_pending_exception(pe); |
1da177e4 | 934 | |
9d493fa8 AK |
935 | up_write(&s->lock); |
936 | ||
937 | /* Submit any pending write bios */ | |
938 | if (error) | |
939 | error_bios(snapshot_bios); | |
940 | else | |
941 | flush_bios(snapshot_bios); | |
942 | ||
943 | flush_bios(origin_bios); | |
1da177e4 LT |
944 | } |
945 | ||
946 | static void commit_callback(void *context, int success) | |
947 | { | |
028867ac AK |
948 | struct dm_snap_pending_exception *pe = context; |
949 | ||
1da177e4 LT |
950 | pending_complete(pe, success); |
951 | } | |
952 | ||
953 | /* | |
954 | * Called when the copy I/O has finished. kcopyd actually runs | |
955 | * this code so don't block. | |
956 | */ | |
4cdc1d1f | 957 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 958 | { |
028867ac | 959 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
960 | struct dm_snapshot *s = pe->snap; |
961 | ||
962 | if (read_err || write_err) | |
963 | pending_complete(pe, 0); | |
964 | ||
965 | else | |
966 | /* Update the metadata if we are persistent */ | |
493df71c JB |
967 | s->store->type->commit_exception(s->store, &pe->e, |
968 | commit_callback, pe); | |
1da177e4 LT |
969 | } |
970 | ||
971 | /* | |
972 | * Dispatches the copy operation to kcopyd. | |
973 | */ | |
028867ac | 974 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
975 | { |
976 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 977 | struct dm_io_region src, dest; |
1da177e4 LT |
978 | struct block_device *bdev = s->origin->bdev; |
979 | sector_t dev_size; | |
980 | ||
981 | dev_size = get_dev_size(bdev); | |
982 | ||
983 | src.bdev = bdev; | |
71fab00a | 984 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 985 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 986 | |
49beb2b8 | 987 | dest.bdev = s->store->cow->bdev; |
71fab00a | 988 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
989 | dest.count = src.count; |
990 | ||
991 | /* Hand over to kcopyd */ | |
eb69aca5 | 992 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
993 | &src, 1, &dest, 0, copy_callback, pe); |
994 | } | |
995 | ||
2913808e MP |
996 | static struct dm_snap_pending_exception * |
997 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
998 | { | |
999 | struct dm_snap_exception *e = lookup_exception(&s->pending, chunk); | |
1000 | ||
1001 | if (!e) | |
1002 | return NULL; | |
1003 | ||
1004 | return container_of(e, struct dm_snap_pending_exception, e); | |
1005 | } | |
1006 | ||
1da177e4 LT |
1007 | /* |
1008 | * Looks to see if this snapshot already has a pending exception | |
1009 | * for this chunk, otherwise it allocates a new one and inserts | |
1010 | * it into the pending table. | |
1011 | * | |
1012 | * NOTE: a write lock must be held on snap->lock before calling | |
1013 | * this. | |
1014 | */ | |
028867ac | 1015 | static struct dm_snap_pending_exception * |
c6621392 MP |
1016 | __find_pending_exception(struct dm_snapshot *s, |
1017 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1018 | { |
c6621392 | 1019 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1020 | |
2913808e MP |
1021 | pe2 = __lookup_pending_exception(s, chunk); |
1022 | if (pe2) { | |
76df1c65 | 1023 | free_pending_exception(pe); |
2913808e | 1024 | return pe2; |
1da177e4 LT |
1025 | } |
1026 | ||
76df1c65 AK |
1027 | pe->e.old_chunk = chunk; |
1028 | bio_list_init(&pe->origin_bios); | |
1029 | bio_list_init(&pe->snapshot_bios); | |
1030 | pe->primary_pe = NULL; | |
4b832e8d | 1031 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1032 | pe->started = 0; |
1033 | ||
493df71c | 1034 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1035 | free_pending_exception(pe); |
1036 | return NULL; | |
1037 | } | |
1038 | ||
4b832e8d | 1039 | get_pending_exception(pe); |
76df1c65 AK |
1040 | insert_exception(&s->pending, &pe->e); |
1041 | ||
1da177e4 LT |
1042 | return pe; |
1043 | } | |
1044 | ||
028867ac | 1045 | static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e, |
d74f81f8 | 1046 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1047 | { |
49beb2b8 | 1048 | bio->bi_bdev = s->store->cow->bdev; |
71fab00a JB |
1049 | bio->bi_sector = chunk_to_sector(s->store, |
1050 | dm_chunk_number(e->new_chunk) + | |
1051 | (chunk - e->old_chunk)) + | |
1052 | (bio->bi_sector & | |
1053 | s->store->chunk_mask); | |
1da177e4 LT |
1054 | } |
1055 | ||
1056 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1057 | union map_info *map_context) | |
1058 | { | |
028867ac AK |
1059 | struct dm_snap_exception *e; |
1060 | struct dm_snapshot *s = ti->private; | |
d2a7ad29 | 1061 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1062 | chunk_t chunk; |
028867ac | 1063 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1064 | |
494b3ee7 MP |
1065 | if (unlikely(bio_empty_barrier(bio))) { |
1066 | bio->bi_bdev = s->store->cow->bdev; | |
1067 | return DM_MAPIO_REMAPPED; | |
1068 | } | |
1069 | ||
71fab00a | 1070 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1071 | |
1072 | /* Full snapshots are not usable */ | |
76df1c65 | 1073 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1074 | if (!s->valid) |
f6a80ea8 | 1075 | return -EIO; |
1da177e4 | 1076 | |
ba40a2aa AK |
1077 | /* FIXME: should only take write lock if we need |
1078 | * to copy an exception */ | |
1079 | down_write(&s->lock); | |
1080 | ||
1081 | if (!s->valid) { | |
1082 | r = -EIO; | |
1083 | goto out_unlock; | |
1084 | } | |
1085 | ||
1086 | /* If the block is already remapped - use that, else remap it */ | |
1087 | e = lookup_exception(&s->complete, chunk); | |
1088 | if (e) { | |
d74f81f8 | 1089 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1090 | goto out_unlock; |
1091 | } | |
1092 | ||
1da177e4 LT |
1093 | /* |
1094 | * Write to snapshot - higher level takes care of RW/RO | |
1095 | * flags so we should only get this if we are | |
1096 | * writeable. | |
1097 | */ | |
1098 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1099 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1100 | if (!pe) { |
c6621392 MP |
1101 | up_write(&s->lock); |
1102 | pe = alloc_pending_exception(s); | |
1103 | down_write(&s->lock); | |
1104 | ||
1105 | if (!s->valid) { | |
1106 | free_pending_exception(pe); | |
1107 | r = -EIO; | |
1108 | goto out_unlock; | |
1109 | } | |
1110 | ||
35bf659b MP |
1111 | e = lookup_exception(&s->complete, chunk); |
1112 | if (e) { | |
1113 | free_pending_exception(pe); | |
1114 | remap_exception(s, e, bio, chunk); | |
1115 | goto out_unlock; | |
1116 | } | |
1117 | ||
c6621392 | 1118 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1119 | if (!pe) { |
1120 | __invalidate_snapshot(s, -ENOMEM); | |
1121 | r = -EIO; | |
1122 | goto out_unlock; | |
1123 | } | |
1da177e4 LT |
1124 | } |
1125 | ||
d74f81f8 | 1126 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1127 | bio_list_add(&pe->snapshot_bios, bio); |
1128 | ||
d2a7ad29 | 1129 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1130 | |
76df1c65 AK |
1131 | if (!pe->started) { |
1132 | /* this is protected by snap->lock */ | |
1133 | pe->started = 1; | |
ba40a2aa | 1134 | up_write(&s->lock); |
76df1c65 | 1135 | start_copy(pe); |
ba40a2aa AK |
1136 | goto out; |
1137 | } | |
cd45daff | 1138 | } else { |
ba40a2aa | 1139 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1140 | map_context->ptr = track_chunk(s, chunk); |
1141 | } | |
1da177e4 | 1142 | |
ba40a2aa AK |
1143 | out_unlock: |
1144 | up_write(&s->lock); | |
1145 | out: | |
1da177e4 LT |
1146 | return r; |
1147 | } | |
1148 | ||
cd45daff MP |
1149 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1150 | int error, union map_info *map_context) | |
1151 | { | |
1152 | struct dm_snapshot *s = ti->private; | |
1153 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1154 | ||
1155 | if (c) | |
1156 | stop_tracking_chunk(s, c); | |
1157 | ||
1158 | return 0; | |
1159 | } | |
1160 | ||
1da177e4 LT |
1161 | static void snapshot_resume(struct dm_target *ti) |
1162 | { | |
028867ac | 1163 | struct dm_snapshot *s = ti->private; |
1da177e4 | 1164 | |
aa14edeb AK |
1165 | down_write(&s->lock); |
1166 | s->active = 1; | |
1167 | up_write(&s->lock); | |
1da177e4 LT |
1168 | } |
1169 | ||
1170 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1171 | char *result, unsigned int maxlen) | |
1172 | { | |
2e4a31df | 1173 | unsigned sz = 0; |
028867ac | 1174 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1175 | |
1176 | switch (type) { | |
1177 | case STATUSTYPE_INFO: | |
94e76572 MP |
1178 | |
1179 | down_write(&snap->lock); | |
1180 | ||
1da177e4 | 1181 | if (!snap->valid) |
2e4a31df | 1182 | DMEMIT("Invalid"); |
1da177e4 | 1183 | else { |
493df71c | 1184 | if (snap->store->type->fraction_full) { |
1da177e4 | 1185 | sector_t numerator, denominator; |
493df71c JB |
1186 | snap->store->type->fraction_full(snap->store, |
1187 | &numerator, | |
1188 | &denominator); | |
2e4a31df JB |
1189 | DMEMIT("%llu/%llu", |
1190 | (unsigned long long)numerator, | |
1191 | (unsigned long long)denominator); | |
1da177e4 LT |
1192 | } |
1193 | else | |
2e4a31df | 1194 | DMEMIT("Unknown"); |
1da177e4 | 1195 | } |
94e76572 MP |
1196 | |
1197 | up_write(&snap->lock); | |
1198 | ||
1da177e4 LT |
1199 | break; |
1200 | ||
1201 | case STATUSTYPE_TABLE: | |
1202 | /* | |
1203 | * kdevname returns a static pointer so we need | |
1204 | * to make private copies if the output is to | |
1205 | * make sense. | |
1206 | */ | |
2e4a31df | 1207 | DMEMIT("%s", snap->origin->name); |
1e302a92 JB |
1208 | snap->store->type->status(snap->store, type, result + sz, |
1209 | maxlen - sz); | |
1da177e4 LT |
1210 | break; |
1211 | } | |
1212 | ||
1213 | return 0; | |
1214 | } | |
1215 | ||
8811f46c MS |
1216 | static int snapshot_iterate_devices(struct dm_target *ti, |
1217 | iterate_devices_callout_fn fn, void *data) | |
1218 | { | |
1219 | struct dm_snapshot *snap = ti->private; | |
1220 | ||
1221 | return fn(ti, snap->origin, 0, ti->len, data); | |
1222 | } | |
1223 | ||
1224 | ||
1da177e4 LT |
1225 | /*----------------------------------------------------------------- |
1226 | * Origin methods | |
1227 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1228 | static int __origin_write(struct list_head *snapshots, struct bio *bio) |
1229 | { | |
d2a7ad29 | 1230 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1231 | struct dm_snapshot *snap; |
028867ac AK |
1232 | struct dm_snap_exception *e; |
1233 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; | |
1da177e4 | 1234 | chunk_t chunk; |
eccf0817 | 1235 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1236 | |
1237 | /* Do all the snapshots on this origin */ | |
1238 | list_for_each_entry (snap, snapshots, list) { | |
1239 | ||
76df1c65 AK |
1240 | down_write(&snap->lock); |
1241 | ||
aa14edeb AK |
1242 | /* Only deal with valid and active snapshots */ |
1243 | if (!snap->valid || !snap->active) | |
76df1c65 | 1244 | goto next_snapshot; |
1da177e4 | 1245 | |
d5e404c1 | 1246 | /* Nothing to do if writing beyond end of snapshot */ |
0cea9c78 | 1247 | if (bio->bi_sector >= dm_table_get_size(snap->store->ti->table)) |
76df1c65 | 1248 | goto next_snapshot; |
1da177e4 LT |
1249 | |
1250 | /* | |
1251 | * Remember, different snapshots can have | |
1252 | * different chunk sizes. | |
1253 | */ | |
71fab00a | 1254 | chunk = sector_to_chunk(snap->store, bio->bi_sector); |
1da177e4 LT |
1255 | |
1256 | /* | |
1257 | * Check exception table to see if block | |
1258 | * is already remapped in this snapshot | |
1259 | * and trigger an exception if not. | |
b4b610f6 | 1260 | * |
4b832e8d | 1261 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1262 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 LT |
1263 | */ |
1264 | e = lookup_exception(&snap->complete, chunk); | |
76df1c65 AK |
1265 | if (e) |
1266 | goto next_snapshot; | |
1267 | ||
2913808e | 1268 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1269 | if (!pe) { |
c6621392 MP |
1270 | up_write(&snap->lock); |
1271 | pe = alloc_pending_exception(snap); | |
1272 | down_write(&snap->lock); | |
1273 | ||
1274 | if (!snap->valid) { | |
1275 | free_pending_exception(pe); | |
1276 | goto next_snapshot; | |
1277 | } | |
1278 | ||
35bf659b MP |
1279 | e = lookup_exception(&snap->complete, chunk); |
1280 | if (e) { | |
1281 | free_pending_exception(pe); | |
1282 | goto next_snapshot; | |
1283 | } | |
1284 | ||
c6621392 | 1285 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1286 | if (!pe) { |
1287 | __invalidate_snapshot(snap, -ENOMEM); | |
1288 | goto next_snapshot; | |
1289 | } | |
76df1c65 AK |
1290 | } |
1291 | ||
1292 | if (!primary_pe) { | |
1293 | /* | |
1294 | * Either every pe here has same | |
1295 | * primary_pe or none has one yet. | |
1296 | */ | |
1297 | if (pe->primary_pe) | |
1298 | primary_pe = pe->primary_pe; | |
1299 | else { | |
1300 | primary_pe = pe; | |
1301 | first = 1; | |
1da177e4 | 1302 | } |
76df1c65 AK |
1303 | |
1304 | bio_list_add(&primary_pe->origin_bios, bio); | |
1305 | ||
d2a7ad29 | 1306 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1307 | } |
1308 | ||
1309 | if (!pe->primary_pe) { | |
76df1c65 | 1310 | pe->primary_pe = primary_pe; |
4b832e8d | 1311 | get_pending_exception(primary_pe); |
76df1c65 AK |
1312 | } |
1313 | ||
1314 | if (!pe->started) { | |
1315 | pe->started = 1; | |
1316 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1317 | } |
1318 | ||
76df1c65 | 1319 | next_snapshot: |
1da177e4 LT |
1320 | up_write(&snap->lock); |
1321 | } | |
1322 | ||
b4b610f6 | 1323 | if (!primary_pe) |
4b832e8d | 1324 | return r; |
b4b610f6 AK |
1325 | |
1326 | /* | |
1327 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1328 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1329 | * got completed while we were in the loop above, so it falls to |
1330 | * us here to remove the primary_pe and submit any origin_bios. | |
1331 | */ | |
1332 | ||
4b832e8d | 1333 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1334 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1335 | free_pending_exception(primary_pe); | |
1336 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1337 | return r; |
b4b610f6 AK |
1338 | } |
1339 | ||
1da177e4 LT |
1340 | /* |
1341 | * Now that we have a complete pe list we can start the copying. | |
1342 | */ | |
eccf0817 AK |
1343 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1344 | start_copy(pe); | |
1da177e4 LT |
1345 | |
1346 | return r; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Called on a write from the origin driver. | |
1351 | */ | |
1352 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1353 | { | |
1354 | struct origin *o; | |
d2a7ad29 | 1355 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1356 | |
1357 | down_read(&_origins_lock); | |
1358 | o = __lookup_origin(origin->bdev); | |
1359 | if (o) | |
1360 | r = __origin_write(&o->snapshots, bio); | |
1361 | up_read(&_origins_lock); | |
1362 | ||
1363 | return r; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1368 | */ | |
1369 | ||
1370 | /* | |
1371 | * Construct an origin mapping: <dev_path> | |
1372 | * The context for an origin is merely a 'struct dm_dev *' | |
1373 | * pointing to the real device. | |
1374 | */ | |
1375 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1376 | { | |
1377 | int r; | |
1378 | struct dm_dev *dev; | |
1379 | ||
1380 | if (argc != 1) { | |
72d94861 | 1381 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1382 | return -EINVAL; |
1383 | } | |
1384 | ||
1385 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1386 | dm_table_get_mode(ti->table), &dev); | |
1387 | if (r) { | |
1388 | ti->error = "Cannot get target device"; | |
1389 | return r; | |
1390 | } | |
1391 | ||
1392 | ti->private = dev; | |
494b3ee7 MP |
1393 | ti->num_flush_requests = 1; |
1394 | ||
1da177e4 LT |
1395 | return 0; |
1396 | } | |
1397 | ||
1398 | static void origin_dtr(struct dm_target *ti) | |
1399 | { | |
028867ac | 1400 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1401 | dm_put_device(ti, dev); |
1402 | } | |
1403 | ||
1404 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1405 | union map_info *map_context) | |
1406 | { | |
028867ac | 1407 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1408 | bio->bi_bdev = dev->bdev; |
1409 | ||
494b3ee7 MP |
1410 | if (unlikely(bio_empty_barrier(bio))) |
1411 | return DM_MAPIO_REMAPPED; | |
1412 | ||
1da177e4 | 1413 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 1414 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1415 | } |
1416 | ||
1da177e4 LT |
1417 | /* |
1418 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1419 | * chunk sizes. | |
1420 | */ | |
1421 | static void origin_resume(struct dm_target *ti) | |
1422 | { | |
028867ac | 1423 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1424 | |
1425 | down_read(&_origins_lock); | |
1da177e4 | 1426 | |
7e201b35 MP |
1427 | ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev)); |
1428 | ||
1429 | up_read(&_origins_lock); | |
1da177e4 LT |
1430 | } |
1431 | ||
1432 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1433 | unsigned int maxlen) | |
1434 | { | |
028867ac | 1435 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1436 | |
1437 | switch (type) { | |
1438 | case STATUSTYPE_INFO: | |
1439 | result[0] = '\0'; | |
1440 | break; | |
1441 | ||
1442 | case STATUSTYPE_TABLE: | |
1443 | snprintf(result, maxlen, "%s", dev->name); | |
1444 | break; | |
1445 | } | |
1446 | ||
1447 | return 0; | |
1448 | } | |
1449 | ||
8811f46c MS |
1450 | static int origin_iterate_devices(struct dm_target *ti, |
1451 | iterate_devices_callout_fn fn, void *data) | |
1452 | { | |
1453 | struct dm_dev *dev = ti->private; | |
1454 | ||
1455 | return fn(ti, dev, 0, ti->len, data); | |
1456 | } | |
1457 | ||
1da177e4 LT |
1458 | static struct target_type origin_target = { |
1459 | .name = "snapshot-origin", | |
8811f46c | 1460 | .version = {1, 7, 0}, |
1da177e4 LT |
1461 | .module = THIS_MODULE, |
1462 | .ctr = origin_ctr, | |
1463 | .dtr = origin_dtr, | |
1464 | .map = origin_map, | |
1465 | .resume = origin_resume, | |
1466 | .status = origin_status, | |
8811f46c | 1467 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
1468 | }; |
1469 | ||
1470 | static struct target_type snapshot_target = { | |
1471 | .name = "snapshot", | |
8811f46c | 1472 | .version = {1, 7, 0}, |
1da177e4 LT |
1473 | .module = THIS_MODULE, |
1474 | .ctr = snapshot_ctr, | |
1475 | .dtr = snapshot_dtr, | |
1476 | .map = snapshot_map, | |
cd45daff | 1477 | .end_io = snapshot_end_io, |
1da177e4 LT |
1478 | .resume = snapshot_resume, |
1479 | .status = snapshot_status, | |
8811f46c | 1480 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
1481 | }; |
1482 | ||
1483 | static int __init dm_snapshot_init(void) | |
1484 | { | |
1485 | int r; | |
1486 | ||
4db6bfe0 AK |
1487 | r = dm_exception_store_init(); |
1488 | if (r) { | |
1489 | DMERR("Failed to initialize exception stores"); | |
1490 | return r; | |
1491 | } | |
1492 | ||
1da177e4 LT |
1493 | r = dm_register_target(&snapshot_target); |
1494 | if (r) { | |
1495 | DMERR("snapshot target register failed %d", r); | |
034a186d | 1496 | goto bad_register_snapshot_target; |
1da177e4 LT |
1497 | } |
1498 | ||
1499 | r = dm_register_target(&origin_target); | |
1500 | if (r < 0) { | |
72d94861 | 1501 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1502 | goto bad1; |
1503 | } | |
1504 | ||
1505 | r = init_origin_hash(); | |
1506 | if (r) { | |
1507 | DMERR("init_origin_hash failed."); | |
1508 | goto bad2; | |
1509 | } | |
1510 | ||
028867ac | 1511 | exception_cache = KMEM_CACHE(dm_snap_exception, 0); |
1da177e4 LT |
1512 | if (!exception_cache) { |
1513 | DMERR("Couldn't create exception cache."); | |
1514 | r = -ENOMEM; | |
1515 | goto bad3; | |
1516 | } | |
1517 | ||
028867ac | 1518 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1519 | if (!pending_cache) { |
1520 | DMERR("Couldn't create pending cache."); | |
1521 | r = -ENOMEM; | |
1522 | goto bad4; | |
1523 | } | |
1524 | ||
cd45daff MP |
1525 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1526 | if (!tracked_chunk_cache) { | |
1527 | DMERR("Couldn't create cache to track chunks in use."); | |
1528 | r = -ENOMEM; | |
1529 | goto bad5; | |
1530 | } | |
1531 | ||
ca3a931f AK |
1532 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1533 | if (!ksnapd) { | |
1534 | DMERR("Failed to create ksnapd workqueue."); | |
1535 | r = -ENOMEM; | |
92e86812 | 1536 | goto bad_pending_pool; |
ca3a931f AK |
1537 | } |
1538 | ||
1da177e4 LT |
1539 | return 0; |
1540 | ||
4db6bfe0 | 1541 | bad_pending_pool: |
cd45daff | 1542 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1543 | bad5: |
1da177e4 | 1544 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1545 | bad4: |
1da177e4 | 1546 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1547 | bad3: |
1da177e4 | 1548 | exit_origin_hash(); |
4db6bfe0 | 1549 | bad2: |
1da177e4 | 1550 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1551 | bad1: |
1da177e4 | 1552 | dm_unregister_target(&snapshot_target); |
034a186d JB |
1553 | |
1554 | bad_register_snapshot_target: | |
1555 | dm_exception_store_exit(); | |
1da177e4 LT |
1556 | return r; |
1557 | } | |
1558 | ||
1559 | static void __exit dm_snapshot_exit(void) | |
1560 | { | |
ca3a931f AK |
1561 | destroy_workqueue(ksnapd); |
1562 | ||
10d3bd09 MP |
1563 | dm_unregister_target(&snapshot_target); |
1564 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1565 | |
1566 | exit_origin_hash(); | |
1da177e4 LT |
1567 | kmem_cache_destroy(pending_cache); |
1568 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1569 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1570 | |
1571 | dm_exception_store_exit(); | |
1da177e4 LT |
1572 | } |
1573 | ||
1574 | /* Module hooks */ | |
1575 | module_init(dm_snapshot_init); | |
1576 | module_exit(dm_snapshot_exit); | |
1577 | ||
1578 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1579 | MODULE_AUTHOR("Joe Thornber"); | |
1580 | MODULE_LICENSE("GPL"); |