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 | ||
d698aa45 MP |
28 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
29 | ||
30 | #define dm_target_is_snapshot_merge(ti) \ | |
31 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
32 | ||
1da177e4 LT |
33 | /* |
34 | * The percentage increment we will wake up users at | |
35 | */ | |
36 | #define WAKE_UP_PERCENT 5 | |
37 | ||
38 | /* | |
39 | * kcopyd priority of snapshot operations | |
40 | */ | |
41 | #define SNAPSHOT_COPY_PRIORITY 2 | |
42 | ||
43 | /* | |
8ee2767a | 44 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 45 | */ |
8ee2767a | 46 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 47 | |
cd45daff MP |
48 | /* |
49 | * The size of the mempool used to track chunks in use. | |
50 | */ | |
51 | #define MIN_IOS 256 | |
52 | ||
ccc45ea8 JB |
53 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
54 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
55 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
56 | ||
191437a5 | 57 | struct dm_exception_table { |
ccc45ea8 JB |
58 | uint32_t hash_mask; |
59 | unsigned hash_shift; | |
60 | struct list_head *table; | |
61 | }; | |
62 | ||
63 | struct dm_snapshot { | |
64 | struct rw_semaphore lock; | |
65 | ||
66 | struct dm_dev *origin; | |
fc56f6fb MS |
67 | struct dm_dev *cow; |
68 | ||
69 | struct dm_target *ti; | |
ccc45ea8 JB |
70 | |
71 | /* List of snapshots per Origin */ | |
72 | struct list_head list; | |
73 | ||
74 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
75 | int valid; | |
76 | ||
77 | /* Origin writes don't trigger exceptions until this is set */ | |
78 | int active; | |
79 | ||
c26655ca MS |
80 | /* Whether or not owning mapped_device is suspended */ |
81 | int suspended; | |
82 | ||
ccc45ea8 JB |
83 | mempool_t *pending_pool; |
84 | ||
85 | atomic_t pending_exceptions_count; | |
86 | ||
191437a5 JB |
87 | struct dm_exception_table pending; |
88 | struct dm_exception_table complete; | |
ccc45ea8 JB |
89 | |
90 | /* | |
91 | * pe_lock protects all pending_exception operations and access | |
92 | * as well as the snapshot_bios list. | |
93 | */ | |
94 | spinlock_t pe_lock; | |
95 | ||
96 | /* The on disk metadata handler */ | |
97 | struct dm_exception_store *store; | |
98 | ||
99 | struct dm_kcopyd_client *kcopyd_client; | |
100 | ||
101 | /* Queue of snapshot writes for ksnapd to flush */ | |
102 | struct bio_list queued_bios; | |
103 | struct work_struct queued_bios_work; | |
104 | ||
105 | /* Chunks with outstanding reads */ | |
106 | mempool_t *tracked_chunk_pool; | |
107 | spinlock_t tracked_chunk_lock; | |
108 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
109 | }; | |
110 | ||
fc56f6fb MS |
111 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
112 | { | |
113 | return s->cow; | |
114 | } | |
115 | EXPORT_SYMBOL(dm_snap_cow); | |
116 | ||
c642f9e0 | 117 | static struct workqueue_struct *ksnapd; |
c4028958 | 118 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 119 | |
ccc45ea8 JB |
120 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
121 | chunk_t chunk) | |
122 | { | |
123 | return chunk << store->chunk_shift; | |
124 | } | |
125 | ||
126 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
127 | { | |
128 | /* | |
129 | * There is only ever one instance of a particular block | |
130 | * device so we can compare pointers safely. | |
131 | */ | |
132 | return lhs == rhs; | |
133 | } | |
134 | ||
028867ac | 135 | struct dm_snap_pending_exception { |
1d4989c8 | 136 | struct dm_exception e; |
1da177e4 LT |
137 | |
138 | /* | |
139 | * Origin buffers waiting for this to complete are held | |
140 | * in a bio list | |
141 | */ | |
142 | struct bio_list origin_bios; | |
143 | struct bio_list snapshot_bios; | |
144 | ||
1da177e4 LT |
145 | /* Pointer back to snapshot context */ |
146 | struct dm_snapshot *snap; | |
147 | ||
148 | /* | |
149 | * 1 indicates the exception has already been sent to | |
150 | * kcopyd. | |
151 | */ | |
152 | int started; | |
153 | }; | |
154 | ||
155 | /* | |
156 | * Hash table mapping origin volumes to lists of snapshots and | |
157 | * a lock to protect it | |
158 | */ | |
e18b890b CL |
159 | static struct kmem_cache *exception_cache; |
160 | static struct kmem_cache *pending_cache; | |
1da177e4 | 161 | |
cd45daff MP |
162 | struct dm_snap_tracked_chunk { |
163 | struct hlist_node node; | |
164 | chunk_t chunk; | |
165 | }; | |
166 | ||
167 | static struct kmem_cache *tracked_chunk_cache; | |
168 | ||
169 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
170 | chunk_t chunk) | |
171 | { | |
172 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
173 | GFP_NOIO); | |
174 | unsigned long flags; | |
175 | ||
176 | c->chunk = chunk; | |
177 | ||
178 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
179 | hlist_add_head(&c->node, | |
180 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
181 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
182 | ||
183 | return c; | |
184 | } | |
185 | ||
186 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
187 | struct dm_snap_tracked_chunk *c) | |
188 | { | |
189 | unsigned long flags; | |
190 | ||
191 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
192 | hlist_del(&c->node); | |
193 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
194 | ||
195 | mempool_free(c, s->tracked_chunk_pool); | |
196 | } | |
197 | ||
a8d41b59 MP |
198 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
199 | { | |
200 | struct dm_snap_tracked_chunk *c; | |
201 | struct hlist_node *hn; | |
202 | int found = 0; | |
203 | ||
204 | spin_lock_irq(&s->tracked_chunk_lock); | |
205 | ||
206 | hlist_for_each_entry(c, hn, | |
207 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
208 | if (c->chunk == chunk) { | |
209 | found = 1; | |
210 | break; | |
211 | } | |
212 | } | |
213 | ||
214 | spin_unlock_irq(&s->tracked_chunk_lock); | |
215 | ||
216 | return found; | |
217 | } | |
218 | ||
615d1eb9 MS |
219 | /* |
220 | * This conflicting I/O is extremely improbable in the caller, | |
221 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
222 | */ | |
223 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
224 | { | |
225 | while (__chunk_is_tracked(s, chunk)) | |
226 | msleep(1); | |
227 | } | |
228 | ||
1da177e4 LT |
229 | /* |
230 | * One of these per registered origin, held in the snapshot_origins hash | |
231 | */ | |
232 | struct origin { | |
233 | /* The origin device */ | |
234 | struct block_device *bdev; | |
235 | ||
236 | struct list_head hash_list; | |
237 | ||
238 | /* List of snapshots for this origin */ | |
239 | struct list_head snapshots; | |
240 | }; | |
241 | ||
242 | /* | |
243 | * Size of the hash table for origin volumes. If we make this | |
244 | * the size of the minors list then it should be nearly perfect | |
245 | */ | |
246 | #define ORIGIN_HASH_SIZE 256 | |
247 | #define ORIGIN_MASK 0xFF | |
248 | static struct list_head *_origins; | |
249 | static struct rw_semaphore _origins_lock; | |
250 | ||
251 | static int init_origin_hash(void) | |
252 | { | |
253 | int i; | |
254 | ||
255 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
256 | GFP_KERNEL); | |
257 | if (!_origins) { | |
72d94861 | 258 | DMERR("unable to allocate memory"); |
1da177e4 LT |
259 | return -ENOMEM; |
260 | } | |
261 | ||
262 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
263 | INIT_LIST_HEAD(_origins + i); | |
264 | init_rwsem(&_origins_lock); | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | static void exit_origin_hash(void) | |
270 | { | |
271 | kfree(_origins); | |
272 | } | |
273 | ||
028867ac | 274 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
275 | { |
276 | return bdev->bd_dev & ORIGIN_MASK; | |
277 | } | |
278 | ||
279 | static struct origin *__lookup_origin(struct block_device *origin) | |
280 | { | |
281 | struct list_head *ol; | |
282 | struct origin *o; | |
283 | ||
284 | ol = &_origins[origin_hash(origin)]; | |
285 | list_for_each_entry (o, ol, hash_list) | |
286 | if (bdev_equal(o->bdev, origin)) | |
287 | return o; | |
288 | ||
289 | return NULL; | |
290 | } | |
291 | ||
292 | static void __insert_origin(struct origin *o) | |
293 | { | |
294 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
295 | list_add_tail(&o->hash_list, sl); | |
296 | } | |
297 | ||
c1f0c183 MS |
298 | /* |
299 | * _origins_lock must be held when calling this function. | |
300 | * Returns number of snapshots registered using the supplied cow device, plus: | |
301 | * snap_src - a snapshot suitable for use as a source of exception handover | |
302 | * snap_dest - a snapshot capable of receiving exception handover. | |
303 | * | |
304 | * Possible return values and states: | |
305 | * 0: NULL, NULL - first new snapshot | |
306 | * 1: snap_src, NULL - normal snapshot | |
307 | * 2: snap_src, snap_dest - waiting for handover | |
308 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
309 | * 1: NULL, snap_dest - source got destroyed without handover | |
310 | */ | |
311 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
312 | struct dm_snapshot **snap_src, | |
313 | struct dm_snapshot **snap_dest) | |
314 | { | |
315 | struct dm_snapshot *s; | |
316 | struct origin *o; | |
317 | int count = 0; | |
318 | int active; | |
319 | ||
320 | o = __lookup_origin(snap->origin->bdev); | |
321 | if (!o) | |
322 | goto out; | |
323 | ||
324 | list_for_each_entry(s, &o->snapshots, list) { | |
325 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) | |
326 | continue; | |
327 | ||
328 | down_read(&s->lock); | |
329 | active = s->active; | |
330 | up_read(&s->lock); | |
331 | ||
332 | if (active) { | |
333 | if (snap_src) | |
334 | *snap_src = s; | |
335 | } else if (snap_dest) | |
336 | *snap_dest = s; | |
337 | ||
338 | count++; | |
339 | } | |
340 | ||
341 | out: | |
342 | return count; | |
343 | } | |
344 | ||
345 | /* | |
346 | * On success, returns 1 if this snapshot is a handover destination, | |
347 | * otherwise returns 0. | |
348 | */ | |
349 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
350 | { | |
351 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
352 | ||
353 | /* Does snapshot need exceptions handed over to it? */ | |
354 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) || | |
355 | snap_dest) { | |
356 | snap->ti->error = "Snapshot cow pairing for exception " | |
357 | "table handover failed"; | |
358 | return -EINVAL; | |
359 | } | |
360 | ||
361 | /* | |
362 | * If no snap_src was found, snap cannot become a handover | |
363 | * destination. | |
364 | */ | |
365 | if (!snap_src) | |
366 | return 0; | |
367 | ||
368 | return 1; | |
369 | } | |
370 | ||
371 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
372 | { | |
373 | struct dm_snapshot *l; | |
374 | ||
375 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
376 | list_for_each_entry(l, &o->snapshots, list) | |
377 | if (l->store->chunk_size < s->store->chunk_size) | |
378 | break; | |
379 | list_add_tail(&s->list, &l->list); | |
380 | } | |
381 | ||
1da177e4 LT |
382 | /* |
383 | * Make a note of the snapshot and its origin so we can look it | |
384 | * up when the origin has a write on it. | |
c1f0c183 MS |
385 | * |
386 | * Also validate snapshot exception store handovers. | |
387 | * On success, returns 1 if this registration is a handover destination, | |
388 | * otherwise returns 0. | |
1da177e4 LT |
389 | */ |
390 | static int register_snapshot(struct dm_snapshot *snap) | |
391 | { | |
c1f0c183 | 392 | struct origin *o, *new_o = NULL; |
1da177e4 | 393 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 394 | int r = 0; |
1da177e4 | 395 | |
60c856c8 MP |
396 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
397 | if (!new_o) | |
398 | return -ENOMEM; | |
399 | ||
1da177e4 | 400 | down_write(&_origins_lock); |
1da177e4 | 401 | |
c1f0c183 MS |
402 | r = __validate_exception_handover(snap); |
403 | if (r < 0) { | |
404 | kfree(new_o); | |
405 | goto out; | |
406 | } | |
407 | ||
408 | o = __lookup_origin(bdev); | |
60c856c8 MP |
409 | if (o) |
410 | kfree(new_o); | |
411 | else { | |
1da177e4 | 412 | /* New origin */ |
60c856c8 | 413 | o = new_o; |
1da177e4 LT |
414 | |
415 | /* Initialise the struct */ | |
416 | INIT_LIST_HEAD(&o->snapshots); | |
417 | o->bdev = bdev; | |
418 | ||
419 | __insert_origin(o); | |
420 | } | |
421 | ||
c1f0c183 MS |
422 | __insert_snapshot(o, snap); |
423 | ||
424 | out: | |
425 | up_write(&_origins_lock); | |
426 | ||
427 | return r; | |
428 | } | |
429 | ||
430 | /* | |
431 | * Move snapshot to correct place in list according to chunk size. | |
432 | */ | |
433 | static void reregister_snapshot(struct dm_snapshot *s) | |
434 | { | |
435 | struct block_device *bdev = s->origin->bdev; | |
436 | ||
437 | down_write(&_origins_lock); | |
438 | ||
439 | list_del(&s->list); | |
440 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
441 | |
442 | up_write(&_origins_lock); | |
1da177e4 LT |
443 | } |
444 | ||
445 | static void unregister_snapshot(struct dm_snapshot *s) | |
446 | { | |
447 | struct origin *o; | |
448 | ||
449 | down_write(&_origins_lock); | |
450 | o = __lookup_origin(s->origin->bdev); | |
451 | ||
452 | list_del(&s->list); | |
c1f0c183 | 453 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
454 | list_del(&o->hash_list); |
455 | kfree(o); | |
456 | } | |
457 | ||
458 | up_write(&_origins_lock); | |
459 | } | |
460 | ||
461 | /* | |
462 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
463 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
464 | * some consecutive chunks to be grouped together. | |
1da177e4 | 465 | */ |
3510cb94 JB |
466 | static int dm_exception_table_init(struct dm_exception_table *et, |
467 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
468 | { |
469 | unsigned int i; | |
470 | ||
d74f81f8 | 471 | et->hash_shift = hash_shift; |
1da177e4 LT |
472 | et->hash_mask = size - 1; |
473 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
474 | if (!et->table) | |
475 | return -ENOMEM; | |
476 | ||
477 | for (i = 0; i < size; i++) | |
478 | INIT_LIST_HEAD(et->table + i); | |
479 | ||
480 | return 0; | |
481 | } | |
482 | ||
3510cb94 JB |
483 | static void dm_exception_table_exit(struct dm_exception_table *et, |
484 | struct kmem_cache *mem) | |
1da177e4 LT |
485 | { |
486 | struct list_head *slot; | |
1d4989c8 | 487 | struct dm_exception *ex, *next; |
1da177e4 LT |
488 | int i, size; |
489 | ||
490 | size = et->hash_mask + 1; | |
491 | for (i = 0; i < size; i++) { | |
492 | slot = et->table + i; | |
493 | ||
494 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
495 | kmem_cache_free(mem, ex); | |
496 | } | |
497 | ||
498 | vfree(et->table); | |
499 | } | |
500 | ||
191437a5 | 501 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 502 | { |
d74f81f8 | 503 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
504 | } |
505 | ||
3510cb94 | 506 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
507 | { |
508 | list_del(&e->hash_list); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Return the exception data for a sector, or NULL if not | |
513 | * remapped. | |
514 | */ | |
3510cb94 JB |
515 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
516 | chunk_t chunk) | |
1da177e4 LT |
517 | { |
518 | struct list_head *slot; | |
1d4989c8 | 519 | struct dm_exception *e; |
1da177e4 LT |
520 | |
521 | slot = &et->table[exception_hash(et, chunk)]; | |
522 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
523 | if (chunk >= e->old_chunk && |
524 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
525 | return e; |
526 | ||
527 | return NULL; | |
528 | } | |
529 | ||
3510cb94 | 530 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 531 | { |
1d4989c8 | 532 | struct dm_exception *e; |
1da177e4 LT |
533 | |
534 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
535 | if (!e) | |
536 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
537 | ||
538 | return e; | |
539 | } | |
540 | ||
3510cb94 | 541 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
542 | { |
543 | kmem_cache_free(exception_cache, e); | |
544 | } | |
545 | ||
92e86812 | 546 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 547 | { |
92e86812 MP |
548 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
549 | GFP_NOIO); | |
550 | ||
879129d2 | 551 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
552 | pe->snap = s; |
553 | ||
554 | return pe; | |
1da177e4 LT |
555 | } |
556 | ||
028867ac | 557 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 558 | { |
879129d2 MP |
559 | struct dm_snapshot *s = pe->snap; |
560 | ||
561 | mempool_free(pe, s->pending_pool); | |
562 | smp_mb__before_atomic_dec(); | |
563 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
564 | } |
565 | ||
3510cb94 JB |
566 | static void dm_insert_exception(struct dm_exception_table *eh, |
567 | struct dm_exception *new_e) | |
d74f81f8 | 568 | { |
d74f81f8 | 569 | struct list_head *l; |
1d4989c8 | 570 | struct dm_exception *e = NULL; |
d74f81f8 MB |
571 | |
572 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
573 | ||
574 | /* Add immediately if this table doesn't support consecutive chunks */ | |
575 | if (!eh->hash_shift) | |
576 | goto out; | |
577 | ||
578 | /* List is ordered by old_chunk */ | |
579 | list_for_each_entry_reverse(e, l, hash_list) { | |
580 | /* Insert after an existing chunk? */ | |
581 | if (new_e->old_chunk == (e->old_chunk + | |
582 | dm_consecutive_chunk_count(e) + 1) && | |
583 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
584 | dm_consecutive_chunk_count(e) + 1)) { | |
585 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 586 | free_completed_exception(new_e); |
d74f81f8 MB |
587 | return; |
588 | } | |
589 | ||
590 | /* Insert before an existing chunk? */ | |
591 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
592 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
593 | dm_consecutive_chunk_count_inc(e); | |
594 | e->old_chunk--; | |
595 | e->new_chunk--; | |
3510cb94 | 596 | free_completed_exception(new_e); |
d74f81f8 MB |
597 | return; |
598 | } | |
599 | ||
600 | if (new_e->old_chunk > e->old_chunk) | |
601 | break; | |
602 | } | |
603 | ||
604 | out: | |
605 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
606 | } | |
607 | ||
a159c1ac JB |
608 | /* |
609 | * Callback used by the exception stores to load exceptions when | |
610 | * initialising. | |
611 | */ | |
612 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 613 | { |
a159c1ac | 614 | struct dm_snapshot *s = context; |
1d4989c8 | 615 | struct dm_exception *e; |
1da177e4 | 616 | |
3510cb94 | 617 | e = alloc_completed_exception(); |
1da177e4 LT |
618 | if (!e) |
619 | return -ENOMEM; | |
620 | ||
621 | e->old_chunk = old; | |
d74f81f8 MB |
622 | |
623 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 624 | e->new_chunk = new; |
d74f81f8 | 625 | |
3510cb94 | 626 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 627 | |
1da177e4 LT |
628 | return 0; |
629 | } | |
630 | ||
7e201b35 MP |
631 | #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r))) |
632 | ||
633 | /* | |
634 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
635 | * Return zero if the origin has no snapshots. | |
636 | */ | |
637 | static sector_t __minimum_chunk_size(struct origin *o) | |
638 | { | |
639 | struct dm_snapshot *snap; | |
640 | unsigned chunk_size = 0; | |
641 | ||
642 | if (o) | |
643 | list_for_each_entry(snap, &o->snapshots, list) | |
644 | chunk_size = min_not_zero(chunk_size, | |
645 | snap->store->chunk_size); | |
646 | ||
647 | return chunk_size; | |
648 | } | |
649 | ||
1da177e4 LT |
650 | /* |
651 | * Hard coded magic. | |
652 | */ | |
653 | static int calc_max_buckets(void) | |
654 | { | |
655 | /* use a fixed size of 2MB */ | |
656 | unsigned long mem = 2 * 1024 * 1024; | |
657 | mem /= sizeof(struct list_head); | |
658 | ||
659 | return mem; | |
660 | } | |
661 | ||
1da177e4 LT |
662 | /* |
663 | * Allocate room for a suitable hash table. | |
664 | */ | |
fee1998e | 665 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
666 | { |
667 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
668 | ||
669 | /* | |
670 | * Calculate based on the size of the original volume or | |
671 | * the COW volume... | |
672 | */ | |
fc56f6fb | 673 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
674 | origin_dev_size = get_dev_size(s->origin->bdev); |
675 | max_buckets = calc_max_buckets(); | |
676 | ||
fee1998e | 677 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
678 | hash_size = min(hash_size, max_buckets); |
679 | ||
8e87b9b8 MP |
680 | if (hash_size < 64) |
681 | hash_size = 64; | |
8defd830 | 682 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
683 | if (dm_exception_table_init(&s->complete, hash_size, |
684 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
685 | return -ENOMEM; |
686 | ||
687 | /* | |
688 | * Allocate hash table for in-flight exceptions | |
689 | * Make this smaller than the real hash table | |
690 | */ | |
691 | hash_size >>= 3; | |
692 | if (hash_size < 64) | |
693 | hash_size = 64; | |
694 | ||
3510cb94 JB |
695 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
696 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
697 | return -ENOMEM; |
698 | } | |
699 | ||
700 | return 0; | |
701 | } | |
702 | ||
1da177e4 LT |
703 | /* |
704 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
705 | */ | |
706 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
707 | { | |
708 | struct dm_snapshot *s; | |
cd45daff | 709 | int i; |
1da177e4 | 710 | int r = -EINVAL; |
fc56f6fb | 711 | char *origin_path, *cow_path; |
10b8106a MS |
712 | unsigned args_used, num_flush_requests = 1; |
713 | fmode_t origin_mode = FMODE_READ; | |
1da177e4 | 714 | |
4c7e3bf4 | 715 | if (argc != 4) { |
72d94861 | 716 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 717 | r = -EINVAL; |
fc56f6fb | 718 | goto bad; |
1da177e4 LT |
719 | } |
720 | ||
10b8106a MS |
721 | if (dm_target_is_snapshot_merge(ti)) { |
722 | num_flush_requests = 2; | |
723 | origin_mode = FMODE_WRITE; | |
724 | } | |
725 | ||
1da177e4 | 726 | origin_path = argv[0]; |
fee1998e JB |
727 | argv++; |
728 | argc--; | |
1da177e4 | 729 | |
fc56f6fb MS |
730 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
731 | if (!s) { | |
732 | ti->error = "Cannot allocate snapshot context private " | |
733 | "structure"; | |
734 | r = -ENOMEM; | |
735 | goto bad; | |
736 | } | |
737 | ||
738 | cow_path = argv[0]; | |
739 | argv++; | |
740 | argc--; | |
741 | ||
742 | r = dm_get_device(ti, cow_path, 0, 0, | |
743 | FMODE_READ | FMODE_WRITE, &s->cow); | |
744 | if (r) { | |
745 | ti->error = "Cannot get COW device"; | |
746 | goto bad_cow; | |
747 | } | |
748 | ||
749 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
750 | if (r) { |
751 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 752 | r = -EINVAL; |
fc56f6fb | 753 | goto bad_store; |
1da177e4 LT |
754 | } |
755 | ||
fee1998e JB |
756 | argv += args_used; |
757 | argc -= args_used; | |
758 | ||
10b8106a | 759 | r = dm_get_device(ti, origin_path, 0, ti->len, origin_mode, &s->origin); |
1da177e4 LT |
760 | if (r) { |
761 | ti->error = "Cannot get origin device"; | |
fee1998e | 762 | goto bad_origin; |
1da177e4 LT |
763 | } |
764 | ||
fc56f6fb | 765 | s->ti = ti; |
1da177e4 | 766 | s->valid = 1; |
aa14edeb | 767 | s->active = 0; |
c26655ca | 768 | s->suspended = 0; |
879129d2 | 769 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 770 | init_rwsem(&s->lock); |
c1f0c183 | 771 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 772 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
773 | |
774 | /* Allocate hash table for COW data */ | |
fee1998e | 775 | if (init_hash_tables(s)) { |
1da177e4 LT |
776 | ti->error = "Unable to allocate hash table space"; |
777 | r = -ENOMEM; | |
fee1998e | 778 | goto bad_hash_tables; |
1da177e4 LT |
779 | } |
780 | ||
eb69aca5 | 781 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
782 | if (r) { |
783 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 784 | goto bad_kcopyd; |
1da177e4 LT |
785 | } |
786 | ||
92e86812 MP |
787 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
788 | if (!s->pending_pool) { | |
789 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 790 | goto bad_pending_pool; |
92e86812 MP |
791 | } |
792 | ||
cd45daff MP |
793 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
794 | tracked_chunk_cache); | |
795 | if (!s->tracked_chunk_pool) { | |
796 | ti->error = "Could not allocate tracked_chunk mempool for " | |
797 | "tracking reads"; | |
92e86812 | 798 | goto bad_tracked_chunk_pool; |
cd45daff MP |
799 | } |
800 | ||
801 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
802 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
803 | ||
804 | spin_lock_init(&s->tracked_chunk_lock); | |
805 | ||
c1f0c183 MS |
806 | bio_list_init(&s->queued_bios); |
807 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); | |
808 | ||
809 | ti->private = s; | |
10b8106a | 810 | ti->num_flush_requests = num_flush_requests; |
c1f0c183 MS |
811 | |
812 | /* Add snapshot to the list of snapshots for this origin */ | |
813 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
814 | r = register_snapshot(s); | |
815 | if (r == -ENOMEM) { | |
816 | ti->error = "Snapshot origin struct allocation failed"; | |
817 | goto bad_load_and_register; | |
818 | } else if (r < 0) { | |
819 | /* invalid handover, register_snapshot has set ti->error */ | |
820 | goto bad_load_and_register; | |
821 | } | |
822 | ||
823 | /* | |
824 | * Metadata must only be loaded into one table at once, so skip this | |
825 | * if metadata will be handed over during resume. | |
826 | * Chunk size will be set during the handover - set it to zero to | |
827 | * ensure it's ignored. | |
828 | */ | |
829 | if (r > 0) { | |
830 | s->store->chunk_size = 0; | |
831 | return 0; | |
832 | } | |
833 | ||
493df71c JB |
834 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
835 | (void *)s); | |
0764147b | 836 | if (r < 0) { |
f9cea4f7 | 837 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 838 | goto bad_read_metadata; |
0764147b MB |
839 | } else if (r > 0) { |
840 | s->valid = 0; | |
841 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 842 | } |
aa14edeb | 843 | |
3f2412dc MP |
844 | if (!s->store->chunk_size) { |
845 | ti->error = "Chunk size not set"; | |
c1f0c183 | 846 | goto bad_read_metadata; |
1da177e4 | 847 | } |
d0216849 | 848 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
849 | |
850 | return 0; | |
851 | ||
c1f0c183 MS |
852 | bad_read_metadata: |
853 | unregister_snapshot(s); | |
854 | ||
fee1998e | 855 | bad_load_and_register: |
cd45daff MP |
856 | mempool_destroy(s->tracked_chunk_pool); |
857 | ||
fee1998e | 858 | bad_tracked_chunk_pool: |
92e86812 MP |
859 | mempool_destroy(s->pending_pool); |
860 | ||
fee1998e | 861 | bad_pending_pool: |
eb69aca5 | 862 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 863 | |
fee1998e | 864 | bad_kcopyd: |
3510cb94 JB |
865 | dm_exception_table_exit(&s->pending, pending_cache); |
866 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 867 | |
fee1998e | 868 | bad_hash_tables: |
1da177e4 LT |
869 | dm_put_device(ti, s->origin); |
870 | ||
fee1998e | 871 | bad_origin: |
fc56f6fb | 872 | dm_exception_store_destroy(s->store); |
1da177e4 | 873 | |
fc56f6fb MS |
874 | bad_store: |
875 | dm_put_device(ti, s->cow); | |
fee1998e | 876 | |
fc56f6fb MS |
877 | bad_cow: |
878 | kfree(s); | |
879 | ||
880 | bad: | |
1da177e4 LT |
881 | return r; |
882 | } | |
883 | ||
31c93a0c MB |
884 | static void __free_exceptions(struct dm_snapshot *s) |
885 | { | |
eb69aca5 | 886 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
887 | s->kcopyd_client = NULL; |
888 | ||
3510cb94 JB |
889 | dm_exception_table_exit(&s->pending, pending_cache); |
890 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
891 | } |
892 | ||
c1f0c183 MS |
893 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
894 | struct dm_snapshot *snap_dest) | |
895 | { | |
896 | union { | |
897 | struct dm_exception_table table_swap; | |
898 | struct dm_exception_store *store_swap; | |
899 | } u; | |
900 | ||
901 | /* | |
902 | * Swap all snapshot context information between the two instances. | |
903 | */ | |
904 | u.table_swap = snap_dest->complete; | |
905 | snap_dest->complete = snap_src->complete; | |
906 | snap_src->complete = u.table_swap; | |
907 | ||
908 | u.store_swap = snap_dest->store; | |
909 | snap_dest->store = snap_src->store; | |
910 | snap_src->store = u.store_swap; | |
911 | ||
912 | snap_dest->store->snap = snap_dest; | |
913 | snap_src->store->snap = snap_src; | |
914 | ||
915 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
916 | snap_dest->valid = snap_src->valid; | |
917 | ||
918 | /* | |
919 | * Set source invalid to ensure it receives no further I/O. | |
920 | */ | |
921 | snap_src->valid = 0; | |
922 | } | |
923 | ||
1da177e4 LT |
924 | static void snapshot_dtr(struct dm_target *ti) |
925 | { | |
cd45daff MP |
926 | #ifdef CONFIG_DM_DEBUG |
927 | int i; | |
928 | #endif | |
028867ac | 929 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 930 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 931 | |
ca3a931f AK |
932 | flush_workqueue(ksnapd); |
933 | ||
c1f0c183 MS |
934 | down_read(&_origins_lock); |
935 | /* Check whether exception handover must be cancelled */ | |
936 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
937 | if (snap_src && snap_dest && (s == snap_src)) { | |
938 | down_write(&snap_dest->lock); | |
939 | snap_dest->valid = 0; | |
940 | up_write(&snap_dest->lock); | |
941 | DMERR("Cancelling snapshot handover."); | |
942 | } | |
943 | up_read(&_origins_lock); | |
944 | ||
138728dc AK |
945 | /* Prevent further origin writes from using this snapshot. */ |
946 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
947 | unregister_snapshot(s); |
948 | ||
879129d2 | 949 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 950 | msleep(1); |
879129d2 MP |
951 | /* |
952 | * Ensure instructions in mempool_destroy aren't reordered | |
953 | * before atomic_read. | |
954 | */ | |
955 | smp_mb(); | |
956 | ||
cd45daff MP |
957 | #ifdef CONFIG_DM_DEBUG |
958 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
959 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
960 | #endif | |
961 | ||
962 | mempool_destroy(s->tracked_chunk_pool); | |
963 | ||
31c93a0c | 964 | __free_exceptions(s); |
1da177e4 | 965 | |
92e86812 MP |
966 | mempool_destroy(s->pending_pool); |
967 | ||
1da177e4 | 968 | dm_put_device(ti, s->origin); |
fee1998e JB |
969 | |
970 | dm_exception_store_destroy(s->store); | |
138728dc | 971 | |
fc56f6fb MS |
972 | dm_put_device(ti, s->cow); |
973 | ||
1da177e4 LT |
974 | kfree(s); |
975 | } | |
976 | ||
977 | /* | |
978 | * Flush a list of buffers. | |
979 | */ | |
980 | static void flush_bios(struct bio *bio) | |
981 | { | |
982 | struct bio *n; | |
983 | ||
984 | while (bio) { | |
985 | n = bio->bi_next; | |
986 | bio->bi_next = NULL; | |
987 | generic_make_request(bio); | |
988 | bio = n; | |
989 | } | |
990 | } | |
991 | ||
c4028958 | 992 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 993 | { |
c4028958 DH |
994 | struct dm_snapshot *s = |
995 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
996 | struct bio *queued_bios; |
997 | unsigned long flags; | |
998 | ||
999 | spin_lock_irqsave(&s->pe_lock, flags); | |
1000 | queued_bios = bio_list_get(&s->queued_bios); | |
1001 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
1002 | ||
1003 | flush_bios(queued_bios); | |
1004 | } | |
1005 | ||
515ad66c MP |
1006 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1007 | ||
1008 | /* | |
1009 | * Flush a list of buffers. | |
1010 | */ | |
1011 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1012 | { | |
1013 | struct bio *n; | |
1014 | int r; | |
1015 | ||
1016 | while (bio) { | |
1017 | n = bio->bi_next; | |
1018 | bio->bi_next = NULL; | |
1019 | r = do_origin(s->origin, bio); | |
1020 | if (r == DM_MAPIO_REMAPPED) | |
1021 | generic_make_request(bio); | |
1022 | bio = n; | |
1023 | } | |
1024 | } | |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * Error a list of buffers. | |
1028 | */ | |
1029 | static void error_bios(struct bio *bio) | |
1030 | { | |
1031 | struct bio *n; | |
1032 | ||
1033 | while (bio) { | |
1034 | n = bio->bi_next; | |
1035 | bio->bi_next = NULL; | |
6712ecf8 | 1036 | bio_io_error(bio); |
1da177e4 LT |
1037 | bio = n; |
1038 | } | |
1039 | } | |
1040 | ||
695368ac | 1041 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1042 | { |
1043 | if (!s->valid) | |
1044 | return; | |
1045 | ||
1046 | if (err == -EIO) | |
1047 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1048 | else if (err == -ENOMEM) | |
1049 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1050 | ||
493df71c JB |
1051 | if (s->store->type->drop_snapshot) |
1052 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1053 | |
1054 | s->valid = 0; | |
1055 | ||
fc56f6fb | 1056 | dm_table_event(s->ti->table); |
76df1c65 AK |
1057 | } |
1058 | ||
028867ac | 1059 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1060 | { |
1d4989c8 | 1061 | struct dm_exception *e; |
1da177e4 | 1062 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1063 | struct bio *origin_bios = NULL; |
1064 | struct bio *snapshot_bios = NULL; | |
1065 | int error = 0; | |
1da177e4 | 1066 | |
76df1c65 AK |
1067 | if (!success) { |
1068 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1069 | down_write(&s->lock); |
695368ac | 1070 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1071 | error = 1; |
76df1c65 AK |
1072 | goto out; |
1073 | } | |
1074 | ||
3510cb94 | 1075 | e = alloc_completed_exception(); |
76df1c65 | 1076 | if (!e) { |
1da177e4 | 1077 | down_write(&s->lock); |
695368ac | 1078 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1079 | error = 1; |
76df1c65 AK |
1080 | goto out; |
1081 | } | |
1082 | *e = pe->e; | |
1da177e4 | 1083 | |
76df1c65 AK |
1084 | down_write(&s->lock); |
1085 | if (!s->valid) { | |
3510cb94 | 1086 | free_completed_exception(e); |
9d493fa8 | 1087 | error = 1; |
76df1c65 | 1088 | goto out; |
1da177e4 LT |
1089 | } |
1090 | ||
615d1eb9 MS |
1091 | /* Check for conflicting reads */ |
1092 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1093 | |
9d493fa8 AK |
1094 | /* |
1095 | * Add a proper exception, and remove the | |
1096 | * in-flight exception from the list. | |
1097 | */ | |
3510cb94 | 1098 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1099 | |
1da177e4 | 1100 | out: |
3510cb94 | 1101 | dm_remove_exception(&pe->e); |
9d493fa8 | 1102 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c MP |
1103 | origin_bios = bio_list_get(&pe->origin_bios); |
1104 | free_pending_exception(pe); | |
1da177e4 | 1105 | |
9d493fa8 AK |
1106 | up_write(&s->lock); |
1107 | ||
1108 | /* Submit any pending write bios */ | |
1109 | if (error) | |
1110 | error_bios(snapshot_bios); | |
1111 | else | |
1112 | flush_bios(snapshot_bios); | |
1113 | ||
515ad66c | 1114 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1115 | } |
1116 | ||
1117 | static void commit_callback(void *context, int success) | |
1118 | { | |
028867ac AK |
1119 | struct dm_snap_pending_exception *pe = context; |
1120 | ||
1da177e4 LT |
1121 | pending_complete(pe, success); |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Called when the copy I/O has finished. kcopyd actually runs | |
1126 | * this code so don't block. | |
1127 | */ | |
4cdc1d1f | 1128 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1129 | { |
028867ac | 1130 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1131 | struct dm_snapshot *s = pe->snap; |
1132 | ||
1133 | if (read_err || write_err) | |
1134 | pending_complete(pe, 0); | |
1135 | ||
1136 | else | |
1137 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1138 | s->store->type->commit_exception(s->store, &pe->e, |
1139 | commit_callback, pe); | |
1da177e4 LT |
1140 | } |
1141 | ||
1142 | /* | |
1143 | * Dispatches the copy operation to kcopyd. | |
1144 | */ | |
028867ac | 1145 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1146 | { |
1147 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1148 | struct dm_io_region src, dest; |
1da177e4 LT |
1149 | struct block_device *bdev = s->origin->bdev; |
1150 | sector_t dev_size; | |
1151 | ||
1152 | dev_size = get_dev_size(bdev); | |
1153 | ||
1154 | src.bdev = bdev; | |
71fab00a | 1155 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1156 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1157 | |
fc56f6fb | 1158 | dest.bdev = s->cow->bdev; |
71fab00a | 1159 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1160 | dest.count = src.count; |
1161 | ||
1162 | /* Hand over to kcopyd */ | |
eb69aca5 | 1163 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1164 | &src, 1, &dest, 0, copy_callback, pe); |
1165 | } | |
1166 | ||
2913808e MP |
1167 | static struct dm_snap_pending_exception * |
1168 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1169 | { | |
3510cb94 | 1170 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1171 | |
1172 | if (!e) | |
1173 | return NULL; | |
1174 | ||
1175 | return container_of(e, struct dm_snap_pending_exception, e); | |
1176 | } | |
1177 | ||
1da177e4 LT |
1178 | /* |
1179 | * Looks to see if this snapshot already has a pending exception | |
1180 | * for this chunk, otherwise it allocates a new one and inserts | |
1181 | * it into the pending table. | |
1182 | * | |
1183 | * NOTE: a write lock must be held on snap->lock before calling | |
1184 | * this. | |
1185 | */ | |
028867ac | 1186 | static struct dm_snap_pending_exception * |
c6621392 MP |
1187 | __find_pending_exception(struct dm_snapshot *s, |
1188 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1189 | { |
c6621392 | 1190 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1191 | |
2913808e MP |
1192 | pe2 = __lookup_pending_exception(s, chunk); |
1193 | if (pe2) { | |
76df1c65 | 1194 | free_pending_exception(pe); |
2913808e | 1195 | return pe2; |
1da177e4 LT |
1196 | } |
1197 | ||
76df1c65 AK |
1198 | pe->e.old_chunk = chunk; |
1199 | bio_list_init(&pe->origin_bios); | |
1200 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 AK |
1201 | pe->started = 0; |
1202 | ||
493df71c | 1203 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1204 | free_pending_exception(pe); |
1205 | return NULL; | |
1206 | } | |
1207 | ||
3510cb94 | 1208 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1209 | |
1da177e4 LT |
1210 | return pe; |
1211 | } | |
1212 | ||
1d4989c8 | 1213 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1214 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1215 | { |
fc56f6fb | 1216 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1217 | bio->bi_sector = chunk_to_sector(s->store, |
1218 | dm_chunk_number(e->new_chunk) + | |
1219 | (chunk - e->old_chunk)) + | |
1220 | (bio->bi_sector & | |
1221 | s->store->chunk_mask); | |
1da177e4 LT |
1222 | } |
1223 | ||
1224 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1225 | union map_info *map_context) | |
1226 | { | |
1d4989c8 | 1227 | struct dm_exception *e; |
028867ac | 1228 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1229 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1230 | chunk_t chunk; |
028867ac | 1231 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1232 | |
494b3ee7 | 1233 | if (unlikely(bio_empty_barrier(bio))) { |
fc56f6fb | 1234 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1235 | return DM_MAPIO_REMAPPED; |
1236 | } | |
1237 | ||
71fab00a | 1238 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1239 | |
1240 | /* Full snapshots are not usable */ | |
76df1c65 | 1241 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1242 | if (!s->valid) |
f6a80ea8 | 1243 | return -EIO; |
1da177e4 | 1244 | |
ba40a2aa AK |
1245 | /* FIXME: should only take write lock if we need |
1246 | * to copy an exception */ | |
1247 | down_write(&s->lock); | |
1248 | ||
1249 | if (!s->valid) { | |
1250 | r = -EIO; | |
1251 | goto out_unlock; | |
1252 | } | |
1253 | ||
1254 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1255 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1256 | if (e) { |
d74f81f8 | 1257 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1258 | goto out_unlock; |
1259 | } | |
1260 | ||
1da177e4 LT |
1261 | /* |
1262 | * Write to snapshot - higher level takes care of RW/RO | |
1263 | * flags so we should only get this if we are | |
1264 | * writeable. | |
1265 | */ | |
1266 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1267 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1268 | if (!pe) { |
c6621392 MP |
1269 | up_write(&s->lock); |
1270 | pe = alloc_pending_exception(s); | |
1271 | down_write(&s->lock); | |
1272 | ||
1273 | if (!s->valid) { | |
1274 | free_pending_exception(pe); | |
1275 | r = -EIO; | |
1276 | goto out_unlock; | |
1277 | } | |
1278 | ||
3510cb94 | 1279 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1280 | if (e) { |
1281 | free_pending_exception(pe); | |
1282 | remap_exception(s, e, bio, chunk); | |
1283 | goto out_unlock; | |
1284 | } | |
1285 | ||
c6621392 | 1286 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1287 | if (!pe) { |
1288 | __invalidate_snapshot(s, -ENOMEM); | |
1289 | r = -EIO; | |
1290 | goto out_unlock; | |
1291 | } | |
1da177e4 LT |
1292 | } |
1293 | ||
d74f81f8 | 1294 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1295 | bio_list_add(&pe->snapshot_bios, bio); |
1296 | ||
d2a7ad29 | 1297 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1298 | |
76df1c65 AK |
1299 | if (!pe->started) { |
1300 | /* this is protected by snap->lock */ | |
1301 | pe->started = 1; | |
ba40a2aa | 1302 | up_write(&s->lock); |
76df1c65 | 1303 | start_copy(pe); |
ba40a2aa AK |
1304 | goto out; |
1305 | } | |
cd45daff | 1306 | } else { |
ba40a2aa | 1307 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1308 | map_context->ptr = track_chunk(s, chunk); |
1309 | } | |
1da177e4 | 1310 | |
ba40a2aa AK |
1311 | out_unlock: |
1312 | up_write(&s->lock); | |
1313 | out: | |
1da177e4 LT |
1314 | return r; |
1315 | } | |
1316 | ||
3452c2a1 MP |
1317 | /* |
1318 | * A snapshot-merge target behaves like a combination of a snapshot | |
1319 | * target and a snapshot-origin target. It only generates new | |
1320 | * exceptions in other snapshots and not in the one that is being | |
1321 | * merged. | |
1322 | * | |
1323 | * For each chunk, if there is an existing exception, it is used to | |
1324 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1325 | * which in turn might generate exceptions in other snapshots. | |
1326 | */ | |
1327 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio, | |
1328 | union map_info *map_context) | |
1329 | { | |
1330 | struct dm_exception *e; | |
1331 | struct dm_snapshot *s = ti->private; | |
1332 | int r = DM_MAPIO_REMAPPED; | |
1333 | chunk_t chunk; | |
1334 | ||
10b8106a MS |
1335 | if (unlikely(bio_empty_barrier(bio))) { |
1336 | if (!map_context->flush_request) | |
1337 | bio->bi_bdev = s->origin->bdev; | |
1338 | else | |
1339 | bio->bi_bdev = s->cow->bdev; | |
1340 | map_context->ptr = NULL; | |
1341 | return DM_MAPIO_REMAPPED; | |
1342 | } | |
1343 | ||
3452c2a1 MP |
1344 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1345 | ||
1346 | down_read(&s->lock); | |
1347 | ||
1348 | /* Full snapshots are not usable */ | |
1349 | if (!s->valid) { | |
1350 | r = -EIO; | |
1351 | goto out_unlock; | |
1352 | } | |
1353 | ||
1354 | /* If the block is already remapped - use that */ | |
1355 | e = dm_lookup_exception(&s->complete, chunk); | |
1356 | if (e) { | |
1357 | remap_exception(s, e, bio, chunk); | |
1358 | goto out_unlock; | |
1359 | } | |
1360 | ||
1361 | bio->bi_bdev = s->origin->bdev; | |
1362 | ||
1363 | if (bio_rw(bio) == WRITE) { | |
1364 | up_read(&s->lock); | |
1365 | return do_origin(s->origin, bio); | |
1366 | } | |
1367 | ||
1368 | out_unlock: | |
1369 | up_read(&s->lock); | |
1370 | ||
1371 | return r; | |
1372 | } | |
1373 | ||
cd45daff MP |
1374 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1375 | int error, union map_info *map_context) | |
1376 | { | |
1377 | struct dm_snapshot *s = ti->private; | |
1378 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1379 | ||
1380 | if (c) | |
1381 | stop_tracking_chunk(s, c); | |
1382 | ||
1383 | return 0; | |
1384 | } | |
1385 | ||
c26655ca MS |
1386 | static void snapshot_postsuspend(struct dm_target *ti) |
1387 | { | |
1388 | struct dm_snapshot *s = ti->private; | |
1389 | ||
1390 | down_write(&s->lock); | |
1391 | s->suspended = 1; | |
1392 | up_write(&s->lock); | |
1393 | } | |
1394 | ||
c1f0c183 MS |
1395 | static int snapshot_preresume(struct dm_target *ti) |
1396 | { | |
1397 | int r = 0; | |
1398 | struct dm_snapshot *s = ti->private; | |
1399 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1400 | ||
1401 | down_read(&_origins_lock); | |
1402 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1403 | if (snap_src && snap_dest) { | |
1404 | down_read(&snap_src->lock); | |
1405 | if (s == snap_src) { | |
1406 | DMERR("Unable to resume snapshot source until " | |
1407 | "handover completes."); | |
1408 | r = -EINVAL; | |
1409 | } else if (!snap_src->suspended) { | |
1410 | DMERR("Unable to perform snapshot handover until " | |
1411 | "source is suspended."); | |
1412 | r = -EINVAL; | |
1413 | } | |
1414 | up_read(&snap_src->lock); | |
1415 | } | |
1416 | up_read(&_origins_lock); | |
1417 | ||
1418 | return r; | |
1419 | } | |
1420 | ||
1da177e4 LT |
1421 | static void snapshot_resume(struct dm_target *ti) |
1422 | { | |
028867ac | 1423 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1424 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1425 | ||
1426 | down_read(&_origins_lock); | |
1427 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1428 | if (snap_src && snap_dest) { | |
1429 | down_write(&snap_src->lock); | |
1430 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1431 | __handover_exceptions(snap_src, snap_dest); | |
1432 | up_write(&snap_dest->lock); | |
1433 | up_write(&snap_src->lock); | |
1434 | } | |
1435 | up_read(&_origins_lock); | |
1436 | ||
1437 | /* Now we have correct chunk size, reregister */ | |
1438 | reregister_snapshot(s); | |
1da177e4 | 1439 | |
aa14edeb AK |
1440 | down_write(&s->lock); |
1441 | s->active = 1; | |
c26655ca | 1442 | s->suspended = 0; |
aa14edeb | 1443 | up_write(&s->lock); |
1da177e4 LT |
1444 | } |
1445 | ||
1446 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1447 | char *result, unsigned int maxlen) | |
1448 | { | |
2e4a31df | 1449 | unsigned sz = 0; |
028867ac | 1450 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1451 | |
1452 | switch (type) { | |
1453 | case STATUSTYPE_INFO: | |
94e76572 MP |
1454 | |
1455 | down_write(&snap->lock); | |
1456 | ||
1da177e4 | 1457 | if (!snap->valid) |
2e4a31df | 1458 | DMEMIT("Invalid"); |
1da177e4 | 1459 | else { |
985903bb MS |
1460 | if (snap->store->type->usage) { |
1461 | sector_t total_sectors, sectors_allocated, | |
1462 | metadata_sectors; | |
1463 | snap->store->type->usage(snap->store, | |
1464 | &total_sectors, | |
1465 | §ors_allocated, | |
1466 | &metadata_sectors); | |
1467 | DMEMIT("%llu/%llu %llu", | |
1468 | (unsigned long long)sectors_allocated, | |
1469 | (unsigned long long)total_sectors, | |
1470 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1471 | } |
1472 | else | |
2e4a31df | 1473 | DMEMIT("Unknown"); |
1da177e4 | 1474 | } |
94e76572 MP |
1475 | |
1476 | up_write(&snap->lock); | |
1477 | ||
1da177e4 LT |
1478 | break; |
1479 | ||
1480 | case STATUSTYPE_TABLE: | |
1481 | /* | |
1482 | * kdevname returns a static pointer so we need | |
1483 | * to make private copies if the output is to | |
1484 | * make sense. | |
1485 | */ | |
fc56f6fb | 1486 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1487 | snap->store->type->status(snap->store, type, result + sz, |
1488 | maxlen - sz); | |
1da177e4 LT |
1489 | break; |
1490 | } | |
1491 | ||
1492 | return 0; | |
1493 | } | |
1494 | ||
8811f46c MS |
1495 | static int snapshot_iterate_devices(struct dm_target *ti, |
1496 | iterate_devices_callout_fn fn, void *data) | |
1497 | { | |
1498 | struct dm_snapshot *snap = ti->private; | |
1499 | ||
1500 | return fn(ti, snap->origin, 0, ti->len, data); | |
1501 | } | |
1502 | ||
1503 | ||
1da177e4 LT |
1504 | /*----------------------------------------------------------------- |
1505 | * Origin methods | |
1506 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1507 | |
1508 | /* | |
1509 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1510 | * supplied bio was ignored. The caller may submit it immediately. | |
1511 | * (No remapping actually occurs as the origin is always a direct linear | |
1512 | * map.) | |
1513 | * | |
1514 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1515 | * and any supplied bio is added to a list to be submitted once all | |
1516 | * the necessary exceptions exist. | |
1517 | */ | |
1518 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1519 | struct bio *bio) | |
1da177e4 | 1520 | { |
515ad66c | 1521 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1522 | struct dm_snapshot *snap; |
1d4989c8 | 1523 | struct dm_exception *e; |
515ad66c MP |
1524 | struct dm_snap_pending_exception *pe; |
1525 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1526 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1527 | chunk_t chunk; |
1528 | ||
1529 | /* Do all the snapshots on this origin */ | |
1530 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
1531 | /* |
1532 | * Don't make new exceptions in a merging snapshot | |
1533 | * because it has effectively been deleted | |
1534 | */ | |
1535 | if (dm_target_is_snapshot_merge(snap->ti)) | |
1536 | continue; | |
1537 | ||
76df1c65 AK |
1538 | down_write(&snap->lock); |
1539 | ||
aa14edeb AK |
1540 | /* Only deal with valid and active snapshots */ |
1541 | if (!snap->valid || !snap->active) | |
76df1c65 | 1542 | goto next_snapshot; |
1da177e4 | 1543 | |
d5e404c1 | 1544 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1545 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1546 | goto next_snapshot; |
1da177e4 LT |
1547 | |
1548 | /* | |
1549 | * Remember, different snapshots can have | |
1550 | * different chunk sizes. | |
1551 | */ | |
9eaae8ff | 1552 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1553 | |
1554 | /* | |
1555 | * Check exception table to see if block | |
1556 | * is already remapped in this snapshot | |
1557 | * and trigger an exception if not. | |
1558 | */ | |
3510cb94 | 1559 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1560 | if (e) |
1561 | goto next_snapshot; | |
1562 | ||
2913808e | 1563 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1564 | if (!pe) { |
c6621392 MP |
1565 | up_write(&snap->lock); |
1566 | pe = alloc_pending_exception(snap); | |
1567 | down_write(&snap->lock); | |
1568 | ||
1569 | if (!snap->valid) { | |
1570 | free_pending_exception(pe); | |
1571 | goto next_snapshot; | |
1572 | } | |
1573 | ||
3510cb94 | 1574 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1575 | if (e) { |
1576 | free_pending_exception(pe); | |
1577 | goto next_snapshot; | |
1578 | } | |
1579 | ||
c6621392 | 1580 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1581 | if (!pe) { |
1582 | __invalidate_snapshot(snap, -ENOMEM); | |
1583 | goto next_snapshot; | |
1584 | } | |
76df1c65 AK |
1585 | } |
1586 | ||
515ad66c | 1587 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 1588 | |
515ad66c MP |
1589 | /* |
1590 | * If an origin bio was supplied, queue it to wait for the | |
1591 | * completion of this exception, and start this one last, | |
1592 | * at the end of the function. | |
1593 | */ | |
1594 | if (bio) { | |
1595 | bio_list_add(&pe->origin_bios, bio); | |
1596 | bio = NULL; | |
76df1c65 | 1597 | |
515ad66c MP |
1598 | if (!pe->started) { |
1599 | pe->started = 1; | |
1600 | pe_to_start_last = pe; | |
1601 | } | |
76df1c65 AK |
1602 | } |
1603 | ||
1604 | if (!pe->started) { | |
1605 | pe->started = 1; | |
515ad66c | 1606 | pe_to_start_now = pe; |
1da177e4 LT |
1607 | } |
1608 | ||
76df1c65 | 1609 | next_snapshot: |
1da177e4 | 1610 | up_write(&snap->lock); |
1da177e4 | 1611 | |
515ad66c MP |
1612 | if (pe_to_start_now) { |
1613 | start_copy(pe_to_start_now); | |
1614 | pe_to_start_now = NULL; | |
1615 | } | |
b4b610f6 AK |
1616 | } |
1617 | ||
1da177e4 | 1618 | /* |
515ad66c MP |
1619 | * Submit the exception against which the bio is queued last, |
1620 | * to give the other exceptions a head start. | |
1da177e4 | 1621 | */ |
515ad66c MP |
1622 | if (pe_to_start_last) |
1623 | start_copy(pe_to_start_last); | |
1da177e4 LT |
1624 | |
1625 | return r; | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * Called on a write from the origin driver. | |
1630 | */ | |
1631 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1632 | { | |
1633 | struct origin *o; | |
d2a7ad29 | 1634 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1635 | |
1636 | down_read(&_origins_lock); | |
1637 | o = __lookup_origin(origin->bdev); | |
1638 | if (o) | |
9eaae8ff | 1639 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
1640 | up_read(&_origins_lock); |
1641 | ||
1642 | return r; | |
1643 | } | |
1644 | ||
1645 | /* | |
1646 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1647 | */ | |
1648 | ||
1649 | /* | |
1650 | * Construct an origin mapping: <dev_path> | |
1651 | * The context for an origin is merely a 'struct dm_dev *' | |
1652 | * pointing to the real device. | |
1653 | */ | |
1654 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1655 | { | |
1656 | int r; | |
1657 | struct dm_dev *dev; | |
1658 | ||
1659 | if (argc != 1) { | |
72d94861 | 1660 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1661 | return -EINVAL; |
1662 | } | |
1663 | ||
1664 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1665 | dm_table_get_mode(ti->table), &dev); | |
1666 | if (r) { | |
1667 | ti->error = "Cannot get target device"; | |
1668 | return r; | |
1669 | } | |
1670 | ||
1671 | ti->private = dev; | |
494b3ee7 MP |
1672 | ti->num_flush_requests = 1; |
1673 | ||
1da177e4 LT |
1674 | return 0; |
1675 | } | |
1676 | ||
1677 | static void origin_dtr(struct dm_target *ti) | |
1678 | { | |
028867ac | 1679 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1680 | dm_put_device(ti, dev); |
1681 | } | |
1682 | ||
1683 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1684 | union map_info *map_context) | |
1685 | { | |
028867ac | 1686 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1687 | bio->bi_bdev = dev->bdev; |
1688 | ||
494b3ee7 MP |
1689 | if (unlikely(bio_empty_barrier(bio))) |
1690 | return DM_MAPIO_REMAPPED; | |
1691 | ||
1da177e4 | 1692 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 1693 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1694 | } |
1695 | ||
1da177e4 LT |
1696 | /* |
1697 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1698 | * chunk sizes. | |
1699 | */ | |
1700 | static void origin_resume(struct dm_target *ti) | |
1701 | { | |
028867ac | 1702 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1703 | |
1704 | down_read(&_origins_lock); | |
1da177e4 | 1705 | |
7e201b35 MP |
1706 | ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev)); |
1707 | ||
1708 | up_read(&_origins_lock); | |
1da177e4 LT |
1709 | } |
1710 | ||
1711 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1712 | unsigned int maxlen) | |
1713 | { | |
028867ac | 1714 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1715 | |
1716 | switch (type) { | |
1717 | case STATUSTYPE_INFO: | |
1718 | result[0] = '\0'; | |
1719 | break; | |
1720 | ||
1721 | case STATUSTYPE_TABLE: | |
1722 | snprintf(result, maxlen, "%s", dev->name); | |
1723 | break; | |
1724 | } | |
1725 | ||
1726 | return 0; | |
1727 | } | |
1728 | ||
8811f46c MS |
1729 | static int origin_iterate_devices(struct dm_target *ti, |
1730 | iterate_devices_callout_fn fn, void *data) | |
1731 | { | |
1732 | struct dm_dev *dev = ti->private; | |
1733 | ||
1734 | return fn(ti, dev, 0, ti->len, data); | |
1735 | } | |
1736 | ||
1da177e4 LT |
1737 | static struct target_type origin_target = { |
1738 | .name = "snapshot-origin", | |
8811f46c | 1739 | .version = {1, 7, 0}, |
1da177e4 LT |
1740 | .module = THIS_MODULE, |
1741 | .ctr = origin_ctr, | |
1742 | .dtr = origin_dtr, | |
1743 | .map = origin_map, | |
1744 | .resume = origin_resume, | |
1745 | .status = origin_status, | |
8811f46c | 1746 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
1747 | }; |
1748 | ||
1749 | static struct target_type snapshot_target = { | |
1750 | .name = "snapshot", | |
c26655ca | 1751 | .version = {1, 9, 0}, |
1da177e4 LT |
1752 | .module = THIS_MODULE, |
1753 | .ctr = snapshot_ctr, | |
1754 | .dtr = snapshot_dtr, | |
1755 | .map = snapshot_map, | |
cd45daff | 1756 | .end_io = snapshot_end_io, |
c26655ca | 1757 | .postsuspend = snapshot_postsuspend, |
c1f0c183 | 1758 | .preresume = snapshot_preresume, |
1da177e4 LT |
1759 | .resume = snapshot_resume, |
1760 | .status = snapshot_status, | |
8811f46c | 1761 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
1762 | }; |
1763 | ||
d698aa45 MP |
1764 | static struct target_type merge_target = { |
1765 | .name = dm_snapshot_merge_target_name, | |
1766 | .version = {1, 0, 0}, | |
1767 | .module = THIS_MODULE, | |
1768 | .ctr = snapshot_ctr, | |
1769 | .dtr = snapshot_dtr, | |
3452c2a1 | 1770 | .map = snapshot_merge_map, |
d698aa45 MP |
1771 | .end_io = snapshot_end_io, |
1772 | .postsuspend = snapshot_postsuspend, | |
1773 | .preresume = snapshot_preresume, | |
1774 | .resume = snapshot_resume, | |
1775 | .status = snapshot_status, | |
1776 | .iterate_devices = snapshot_iterate_devices, | |
1777 | }; | |
1778 | ||
1da177e4 LT |
1779 | static int __init dm_snapshot_init(void) |
1780 | { | |
1781 | int r; | |
1782 | ||
4db6bfe0 AK |
1783 | r = dm_exception_store_init(); |
1784 | if (r) { | |
1785 | DMERR("Failed to initialize exception stores"); | |
1786 | return r; | |
1787 | } | |
1788 | ||
1da177e4 | 1789 | r = dm_register_target(&snapshot_target); |
d698aa45 | 1790 | if (r < 0) { |
1da177e4 | 1791 | DMERR("snapshot target register failed %d", r); |
034a186d | 1792 | goto bad_register_snapshot_target; |
1da177e4 LT |
1793 | } |
1794 | ||
1795 | r = dm_register_target(&origin_target); | |
1796 | if (r < 0) { | |
72d94861 | 1797 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
1798 | goto bad_register_origin_target; |
1799 | } | |
1800 | ||
1801 | r = dm_register_target(&merge_target); | |
1802 | if (r < 0) { | |
1803 | DMERR("Merge target register failed %d", r); | |
1804 | goto bad_register_merge_target; | |
1da177e4 LT |
1805 | } |
1806 | ||
1807 | r = init_origin_hash(); | |
1808 | if (r) { | |
1809 | DMERR("init_origin_hash failed."); | |
d698aa45 | 1810 | goto bad_origin_hash; |
1da177e4 LT |
1811 | } |
1812 | ||
1d4989c8 | 1813 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
1814 | if (!exception_cache) { |
1815 | DMERR("Couldn't create exception cache."); | |
1816 | r = -ENOMEM; | |
d698aa45 | 1817 | goto bad_exception_cache; |
1da177e4 LT |
1818 | } |
1819 | ||
028867ac | 1820 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1821 | if (!pending_cache) { |
1822 | DMERR("Couldn't create pending cache."); | |
1823 | r = -ENOMEM; | |
d698aa45 | 1824 | goto bad_pending_cache; |
1da177e4 LT |
1825 | } |
1826 | ||
cd45daff MP |
1827 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1828 | if (!tracked_chunk_cache) { | |
1829 | DMERR("Couldn't create cache to track chunks in use."); | |
1830 | r = -ENOMEM; | |
d698aa45 | 1831 | goto bad_tracked_chunk_cache; |
cd45daff MP |
1832 | } |
1833 | ||
ca3a931f AK |
1834 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1835 | if (!ksnapd) { | |
1836 | DMERR("Failed to create ksnapd workqueue."); | |
1837 | r = -ENOMEM; | |
92e86812 | 1838 | goto bad_pending_pool; |
ca3a931f AK |
1839 | } |
1840 | ||
1da177e4 LT |
1841 | return 0; |
1842 | ||
4db6bfe0 | 1843 | bad_pending_pool: |
cd45daff | 1844 | kmem_cache_destroy(tracked_chunk_cache); |
d698aa45 | 1845 | bad_tracked_chunk_cache: |
1da177e4 | 1846 | kmem_cache_destroy(pending_cache); |
d698aa45 | 1847 | bad_pending_cache: |
1da177e4 | 1848 | kmem_cache_destroy(exception_cache); |
d698aa45 | 1849 | bad_exception_cache: |
1da177e4 | 1850 | exit_origin_hash(); |
d698aa45 MP |
1851 | bad_origin_hash: |
1852 | dm_unregister_target(&merge_target); | |
1853 | bad_register_merge_target: | |
1da177e4 | 1854 | dm_unregister_target(&origin_target); |
d698aa45 | 1855 | bad_register_origin_target: |
1da177e4 | 1856 | dm_unregister_target(&snapshot_target); |
034a186d JB |
1857 | bad_register_snapshot_target: |
1858 | dm_exception_store_exit(); | |
d698aa45 | 1859 | |
1da177e4 LT |
1860 | return r; |
1861 | } | |
1862 | ||
1863 | static void __exit dm_snapshot_exit(void) | |
1864 | { | |
ca3a931f AK |
1865 | destroy_workqueue(ksnapd); |
1866 | ||
10d3bd09 MP |
1867 | dm_unregister_target(&snapshot_target); |
1868 | dm_unregister_target(&origin_target); | |
d698aa45 | 1869 | dm_unregister_target(&merge_target); |
1da177e4 LT |
1870 | |
1871 | exit_origin_hash(); | |
1da177e4 LT |
1872 | kmem_cache_destroy(pending_cache); |
1873 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1874 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1875 | |
1876 | dm_exception_store_exit(); | |
1da177e4 LT |
1877 | } |
1878 | ||
1879 | /* Module hooks */ | |
1880 | module_init(dm_snapshot_init); | |
1881 | module_exit(dm_snapshot_exit); | |
1882 | ||
1883 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1884 | MODULE_AUTHOR("Joe Thornber"); | |
1885 | MODULE_LICENSE("GPL"); |