]>
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> |
1da177e4 | 22 | |
b735fede MP |
23 | #include "dm.h" |
24 | ||
aea53d92 | 25 | #include "dm-exception-store.h" |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "snapshots" |
28 | ||
d698aa45 MP |
29 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
30 | ||
31 | #define dm_target_is_snapshot_merge(ti) \ | |
32 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
33 | ||
cd45daff MP |
34 | /* |
35 | * The size of the mempool used to track chunks in use. | |
36 | */ | |
37 | #define MIN_IOS 256 | |
38 | ||
ccc45ea8 JB |
39 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
40 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
41 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
42 | ||
191437a5 | 43 | struct dm_exception_table { |
ccc45ea8 JB |
44 | uint32_t hash_mask; |
45 | unsigned hash_shift; | |
46 | struct list_head *table; | |
47 | }; | |
48 | ||
49 | struct dm_snapshot { | |
50 | struct rw_semaphore lock; | |
51 | ||
52 | struct dm_dev *origin; | |
fc56f6fb MS |
53 | struct dm_dev *cow; |
54 | ||
55 | struct dm_target *ti; | |
ccc45ea8 JB |
56 | |
57 | /* List of snapshots per Origin */ | |
58 | struct list_head list; | |
59 | ||
d8ddb1cf MS |
60 | /* |
61 | * You can't use a snapshot if this is 0 (e.g. if full). | |
62 | * A snapshot-merge target never clears this. | |
63 | */ | |
ccc45ea8 JB |
64 | int valid; |
65 | ||
66 | /* Origin writes don't trigger exceptions until this is set */ | |
67 | int active; | |
68 | ||
ccc45ea8 JB |
69 | atomic_t pending_exceptions_count; |
70 | ||
230c83af MP |
71 | /* Protected by "lock" */ |
72 | sector_t exception_start_sequence; | |
73 | ||
74 | /* Protected by kcopyd single-threaded callback */ | |
75 | sector_t exception_complete_sequence; | |
76 | ||
77 | /* | |
78 | * A list of pending exceptions that completed out of order. | |
79 | * Protected by kcopyd single-threaded callback. | |
80 | */ | |
81 | struct list_head out_of_order_list; | |
82 | ||
924e600d MS |
83 | mempool_t *pending_pool; |
84 | ||
191437a5 JB |
85 | struct dm_exception_table pending; |
86 | struct dm_exception_table complete; | |
ccc45ea8 JB |
87 | |
88 | /* | |
89 | * pe_lock protects all pending_exception operations and access | |
90 | * as well as the snapshot_bios list. | |
91 | */ | |
92 | spinlock_t pe_lock; | |
93 | ||
924e600d MS |
94 | /* Chunks with outstanding reads */ |
95 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
96 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
97 | ||
ccc45ea8 JB |
98 | /* The on disk metadata handler */ |
99 | struct dm_exception_store *store; | |
100 | ||
101 | struct dm_kcopyd_client *kcopyd_client; | |
102 | ||
924e600d MS |
103 | /* Wait for events based on state_bits */ |
104 | unsigned long state_bits; | |
105 | ||
106 | /* Range of chunks currently being merged. */ | |
107 | chunk_t first_merging_chunk; | |
108 | int num_merging_chunks; | |
1e03f97e | 109 | |
d8ddb1cf MS |
110 | /* |
111 | * The merge operation failed if this flag is set. | |
112 | * Failure modes are handled as follows: | |
113 | * - I/O error reading the header | |
114 | * => don't load the target; abort. | |
115 | * - Header does not have "valid" flag set | |
116 | * => use the origin; forget about the snapshot. | |
117 | * - I/O error when reading exceptions | |
118 | * => don't load the target; abort. | |
119 | * (We can't use the intermediate origin state.) | |
120 | * - I/O error while merging | |
121 | * => stop merging; set merge_failed; process I/O normally. | |
122 | */ | |
123 | int merge_failed; | |
124 | ||
9fe86254 MP |
125 | /* |
126 | * Incoming bios that overlap with chunks being merged must wait | |
127 | * for them to be committed. | |
128 | */ | |
129 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
130 | }; |
131 | ||
1e03f97e MP |
132 | /* |
133 | * state_bits: | |
134 | * RUNNING_MERGE - Merge operation is in progress. | |
135 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
136 | * cleared afterwards. | |
137 | */ | |
138 | #define RUNNING_MERGE 0 | |
139 | #define SHUTDOWN_MERGE 1 | |
140 | ||
df5d2e90 MP |
141 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
142 | "A percentage of time allocated for copy on write"); | |
143 | ||
c2411045 MP |
144 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
145 | { | |
146 | return s->origin; | |
147 | } | |
148 | EXPORT_SYMBOL(dm_snap_origin); | |
149 | ||
fc56f6fb MS |
150 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
151 | { | |
152 | return s->cow; | |
153 | } | |
154 | EXPORT_SYMBOL(dm_snap_cow); | |
155 | ||
ccc45ea8 JB |
156 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
157 | chunk_t chunk) | |
158 | { | |
159 | return chunk << store->chunk_shift; | |
160 | } | |
161 | ||
162 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
163 | { | |
164 | /* | |
165 | * There is only ever one instance of a particular block | |
166 | * device so we can compare pointers safely. | |
167 | */ | |
168 | return lhs == rhs; | |
169 | } | |
170 | ||
028867ac | 171 | struct dm_snap_pending_exception { |
1d4989c8 | 172 | struct dm_exception e; |
1da177e4 LT |
173 | |
174 | /* | |
175 | * Origin buffers waiting for this to complete are held | |
176 | * in a bio list | |
177 | */ | |
178 | struct bio_list origin_bios; | |
179 | struct bio_list snapshot_bios; | |
180 | ||
1da177e4 LT |
181 | /* Pointer back to snapshot context */ |
182 | struct dm_snapshot *snap; | |
183 | ||
184 | /* | |
185 | * 1 indicates the exception has already been sent to | |
186 | * kcopyd. | |
187 | */ | |
188 | int started; | |
a6e50b40 | 189 | |
230c83af MP |
190 | /* There was copying error. */ |
191 | int copy_error; | |
192 | ||
193 | /* A sequence number, it is used for in-order completion. */ | |
194 | sector_t exception_sequence; | |
195 | ||
196 | struct list_head out_of_order_entry; | |
197 | ||
a6e50b40 MP |
198 | /* |
199 | * For writing a complete chunk, bypassing the copy. | |
200 | */ | |
201 | struct bio *full_bio; | |
202 | bio_end_io_t *full_bio_end_io; | |
203 | void *full_bio_private; | |
1da177e4 LT |
204 | }; |
205 | ||
206 | /* | |
207 | * Hash table mapping origin volumes to lists of snapshots and | |
208 | * a lock to protect it | |
209 | */ | |
e18b890b CL |
210 | static struct kmem_cache *exception_cache; |
211 | static struct kmem_cache *pending_cache; | |
1da177e4 | 212 | |
cd45daff MP |
213 | struct dm_snap_tracked_chunk { |
214 | struct hlist_node node; | |
215 | chunk_t chunk; | |
216 | }; | |
217 | ||
ee18026a MP |
218 | static void init_tracked_chunk(struct bio *bio) |
219 | { | |
220 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
221 | INIT_HLIST_NODE(&c->node); | |
222 | } | |
223 | ||
224 | static bool is_bio_tracked(struct bio *bio) | |
225 | { | |
226 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
227 | return !hlist_unhashed(&c->node); | |
228 | } | |
229 | ||
230 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 231 | { |
42bc954f | 232 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
233 | |
234 | c->chunk = chunk; | |
235 | ||
9aa0c0e6 | 236 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
237 | hlist_add_head(&c->node, |
238 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 239 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
240 | } |
241 | ||
ee18026a | 242 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 243 | { |
ee18026a | 244 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
245 | unsigned long flags; |
246 | ||
247 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
248 | hlist_del(&c->node); | |
249 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
250 | } |
251 | ||
a8d41b59 MP |
252 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
253 | { | |
254 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
255 | int found = 0; |
256 | ||
257 | spin_lock_irq(&s->tracked_chunk_lock); | |
258 | ||
b67bfe0d | 259 | hlist_for_each_entry(c, |
a8d41b59 MP |
260 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
261 | if (c->chunk == chunk) { | |
262 | found = 1; | |
263 | break; | |
264 | } | |
265 | } | |
266 | ||
267 | spin_unlock_irq(&s->tracked_chunk_lock); | |
268 | ||
269 | return found; | |
270 | } | |
271 | ||
615d1eb9 MS |
272 | /* |
273 | * This conflicting I/O is extremely improbable in the caller, | |
274 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
275 | */ | |
276 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
277 | { | |
278 | while (__chunk_is_tracked(s, chunk)) | |
279 | msleep(1); | |
280 | } | |
281 | ||
1da177e4 LT |
282 | /* |
283 | * One of these per registered origin, held in the snapshot_origins hash | |
284 | */ | |
285 | struct origin { | |
286 | /* The origin device */ | |
287 | struct block_device *bdev; | |
288 | ||
289 | struct list_head hash_list; | |
290 | ||
291 | /* List of snapshots for this origin */ | |
292 | struct list_head snapshots; | |
293 | }; | |
294 | ||
b735fede MP |
295 | /* |
296 | * This structure is allocated for each origin target | |
297 | */ | |
298 | struct dm_origin { | |
299 | struct dm_dev *dev; | |
300 | struct dm_target *ti; | |
301 | unsigned split_boundary; | |
302 | struct list_head hash_list; | |
303 | }; | |
304 | ||
1da177e4 LT |
305 | /* |
306 | * Size of the hash table for origin volumes. If we make this | |
307 | * the size of the minors list then it should be nearly perfect | |
308 | */ | |
309 | #define ORIGIN_HASH_SIZE 256 | |
310 | #define ORIGIN_MASK 0xFF | |
311 | static struct list_head *_origins; | |
b735fede | 312 | static struct list_head *_dm_origins; |
1da177e4 LT |
313 | static struct rw_semaphore _origins_lock; |
314 | ||
73dfd078 MP |
315 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
316 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
317 | static uint64_t _pending_exceptions_done_count; | |
318 | ||
1da177e4 LT |
319 | static int init_origin_hash(void) |
320 | { | |
321 | int i; | |
322 | ||
323 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
324 | GFP_KERNEL); | |
325 | if (!_origins) { | |
b735fede | 326 | DMERR("unable to allocate memory for _origins"); |
1da177e4 LT |
327 | return -ENOMEM; |
328 | } | |
1da177e4 LT |
329 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
330 | INIT_LIST_HEAD(_origins + i); | |
b735fede MP |
331 | |
332 | _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
333 | GFP_KERNEL); | |
334 | if (!_dm_origins) { | |
335 | DMERR("unable to allocate memory for _dm_origins"); | |
336 | kfree(_origins); | |
337 | return -ENOMEM; | |
338 | } | |
339 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
340 | INIT_LIST_HEAD(_dm_origins + i); | |
341 | ||
1da177e4 LT |
342 | init_rwsem(&_origins_lock); |
343 | ||
344 | return 0; | |
345 | } | |
346 | ||
347 | static void exit_origin_hash(void) | |
348 | { | |
349 | kfree(_origins); | |
b735fede | 350 | kfree(_dm_origins); |
1da177e4 LT |
351 | } |
352 | ||
028867ac | 353 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
354 | { |
355 | return bdev->bd_dev & ORIGIN_MASK; | |
356 | } | |
357 | ||
358 | static struct origin *__lookup_origin(struct block_device *origin) | |
359 | { | |
360 | struct list_head *ol; | |
361 | struct origin *o; | |
362 | ||
363 | ol = &_origins[origin_hash(origin)]; | |
364 | list_for_each_entry (o, ol, hash_list) | |
365 | if (bdev_equal(o->bdev, origin)) | |
366 | return o; | |
367 | ||
368 | return NULL; | |
369 | } | |
370 | ||
371 | static void __insert_origin(struct origin *o) | |
372 | { | |
373 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
374 | list_add_tail(&o->hash_list, sl); | |
375 | } | |
376 | ||
b735fede MP |
377 | static struct dm_origin *__lookup_dm_origin(struct block_device *origin) |
378 | { | |
379 | struct list_head *ol; | |
380 | struct dm_origin *o; | |
381 | ||
382 | ol = &_dm_origins[origin_hash(origin)]; | |
383 | list_for_each_entry (o, ol, hash_list) | |
384 | if (bdev_equal(o->dev->bdev, origin)) | |
385 | return o; | |
386 | ||
387 | return NULL; | |
388 | } | |
389 | ||
390 | static void __insert_dm_origin(struct dm_origin *o) | |
391 | { | |
392 | struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)]; | |
393 | list_add_tail(&o->hash_list, sl); | |
394 | } | |
395 | ||
396 | static void __remove_dm_origin(struct dm_origin *o) | |
397 | { | |
398 | list_del(&o->hash_list); | |
399 | } | |
400 | ||
c1f0c183 MS |
401 | /* |
402 | * _origins_lock must be held when calling this function. | |
403 | * Returns number of snapshots registered using the supplied cow device, plus: | |
404 | * snap_src - a snapshot suitable for use as a source of exception handover | |
405 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
406 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
407 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 408 | * |
9d3b15c4 | 409 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
410 | * 0: NULL, NULL - first new snapshot |
411 | * 1: snap_src, NULL - normal snapshot | |
412 | * 2: snap_src, snap_dest - waiting for handover | |
413 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
414 | * 1: NULL, snap_dest - source got destroyed without handover | |
415 | */ | |
416 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
417 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
418 | struct dm_snapshot **snap_dest, |
419 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
420 | { |
421 | struct dm_snapshot *s; | |
422 | struct origin *o; | |
423 | int count = 0; | |
424 | int active; | |
425 | ||
426 | o = __lookup_origin(snap->origin->bdev); | |
427 | if (!o) | |
428 | goto out; | |
429 | ||
430 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
431 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
432 | *snap_merge = s; | |
c1f0c183 MS |
433 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
434 | continue; | |
435 | ||
436 | down_read(&s->lock); | |
437 | active = s->active; | |
438 | up_read(&s->lock); | |
439 | ||
440 | if (active) { | |
441 | if (snap_src) | |
442 | *snap_src = s; | |
443 | } else if (snap_dest) | |
444 | *snap_dest = s; | |
445 | ||
446 | count++; | |
447 | } | |
448 | ||
449 | out: | |
450 | return count; | |
451 | } | |
452 | ||
453 | /* | |
454 | * On success, returns 1 if this snapshot is a handover destination, | |
455 | * otherwise returns 0. | |
456 | */ | |
457 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
458 | { | |
459 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 460 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
461 | |
462 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
463 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
464 | &snap_merge) == 2) || | |
c1f0c183 MS |
465 | snap_dest) { |
466 | snap->ti->error = "Snapshot cow pairing for exception " | |
467 | "table handover failed"; | |
468 | return -EINVAL; | |
469 | } | |
470 | ||
471 | /* | |
472 | * If no snap_src was found, snap cannot become a handover | |
473 | * destination. | |
474 | */ | |
475 | if (!snap_src) | |
476 | return 0; | |
477 | ||
9d3b15c4 MP |
478 | /* |
479 | * Non-snapshot-merge handover? | |
480 | */ | |
481 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
482 | return 1; | |
483 | ||
484 | /* | |
485 | * Do not allow more than one merging snapshot. | |
486 | */ | |
487 | if (snap_merge) { | |
488 | snap->ti->error = "A snapshot is already merging."; | |
489 | return -EINVAL; | |
490 | } | |
491 | ||
1e03f97e MP |
492 | if (!snap_src->store->type->prepare_merge || |
493 | !snap_src->store->type->commit_merge) { | |
494 | snap->ti->error = "Snapshot exception store does not " | |
495 | "support snapshot-merge."; | |
496 | return -EINVAL; | |
497 | } | |
498 | ||
c1f0c183 MS |
499 | return 1; |
500 | } | |
501 | ||
502 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
503 | { | |
504 | struct dm_snapshot *l; | |
505 | ||
506 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
507 | list_for_each_entry(l, &o->snapshots, list) | |
508 | if (l->store->chunk_size < s->store->chunk_size) | |
509 | break; | |
510 | list_add_tail(&s->list, &l->list); | |
511 | } | |
512 | ||
1da177e4 LT |
513 | /* |
514 | * Make a note of the snapshot and its origin so we can look it | |
515 | * up when the origin has a write on it. | |
c1f0c183 MS |
516 | * |
517 | * Also validate snapshot exception store handovers. | |
518 | * On success, returns 1 if this registration is a handover destination, | |
519 | * otherwise returns 0. | |
1da177e4 LT |
520 | */ |
521 | static int register_snapshot(struct dm_snapshot *snap) | |
522 | { | |
c1f0c183 | 523 | struct origin *o, *new_o = NULL; |
1da177e4 | 524 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 525 | int r = 0; |
1da177e4 | 526 | |
60c856c8 MP |
527 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
528 | if (!new_o) | |
529 | return -ENOMEM; | |
530 | ||
1da177e4 | 531 | down_write(&_origins_lock); |
1da177e4 | 532 | |
c1f0c183 MS |
533 | r = __validate_exception_handover(snap); |
534 | if (r < 0) { | |
535 | kfree(new_o); | |
536 | goto out; | |
537 | } | |
538 | ||
539 | o = __lookup_origin(bdev); | |
60c856c8 MP |
540 | if (o) |
541 | kfree(new_o); | |
542 | else { | |
1da177e4 | 543 | /* New origin */ |
60c856c8 | 544 | o = new_o; |
1da177e4 LT |
545 | |
546 | /* Initialise the struct */ | |
547 | INIT_LIST_HEAD(&o->snapshots); | |
548 | o->bdev = bdev; | |
549 | ||
550 | __insert_origin(o); | |
551 | } | |
552 | ||
c1f0c183 MS |
553 | __insert_snapshot(o, snap); |
554 | ||
555 | out: | |
556 | up_write(&_origins_lock); | |
557 | ||
558 | return r; | |
559 | } | |
560 | ||
561 | /* | |
562 | * Move snapshot to correct place in list according to chunk size. | |
563 | */ | |
564 | static void reregister_snapshot(struct dm_snapshot *s) | |
565 | { | |
566 | struct block_device *bdev = s->origin->bdev; | |
567 | ||
568 | down_write(&_origins_lock); | |
569 | ||
570 | list_del(&s->list); | |
571 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
572 | |
573 | up_write(&_origins_lock); | |
1da177e4 LT |
574 | } |
575 | ||
576 | static void unregister_snapshot(struct dm_snapshot *s) | |
577 | { | |
578 | struct origin *o; | |
579 | ||
580 | down_write(&_origins_lock); | |
581 | o = __lookup_origin(s->origin->bdev); | |
582 | ||
583 | list_del(&s->list); | |
c1f0c183 | 584 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
585 | list_del(&o->hash_list); |
586 | kfree(o); | |
587 | } | |
588 | ||
589 | up_write(&_origins_lock); | |
590 | } | |
591 | ||
592 | /* | |
593 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
594 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
595 | * some consecutive chunks to be grouped together. | |
1da177e4 | 596 | */ |
3510cb94 JB |
597 | static int dm_exception_table_init(struct dm_exception_table *et, |
598 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
599 | { |
600 | unsigned int i; | |
601 | ||
d74f81f8 | 602 | et->hash_shift = hash_shift; |
1da177e4 LT |
603 | et->hash_mask = size - 1; |
604 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
605 | if (!et->table) | |
606 | return -ENOMEM; | |
607 | ||
608 | for (i = 0; i < size; i++) | |
609 | INIT_LIST_HEAD(et->table + i); | |
610 | ||
611 | return 0; | |
612 | } | |
613 | ||
3510cb94 JB |
614 | static void dm_exception_table_exit(struct dm_exception_table *et, |
615 | struct kmem_cache *mem) | |
1da177e4 LT |
616 | { |
617 | struct list_head *slot; | |
1d4989c8 | 618 | struct dm_exception *ex, *next; |
1da177e4 LT |
619 | int i, size; |
620 | ||
621 | size = et->hash_mask + 1; | |
622 | for (i = 0; i < size; i++) { | |
623 | slot = et->table + i; | |
624 | ||
625 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
626 | kmem_cache_free(mem, ex); | |
627 | } | |
628 | ||
629 | vfree(et->table); | |
630 | } | |
631 | ||
191437a5 | 632 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 633 | { |
d74f81f8 | 634 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
635 | } |
636 | ||
3510cb94 | 637 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
638 | { |
639 | list_del(&e->hash_list); | |
640 | } | |
641 | ||
642 | /* | |
643 | * Return the exception data for a sector, or NULL if not | |
644 | * remapped. | |
645 | */ | |
3510cb94 JB |
646 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
647 | chunk_t chunk) | |
1da177e4 LT |
648 | { |
649 | struct list_head *slot; | |
1d4989c8 | 650 | struct dm_exception *e; |
1da177e4 LT |
651 | |
652 | slot = &et->table[exception_hash(et, chunk)]; | |
653 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
654 | if (chunk >= e->old_chunk && |
655 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
656 | return e; |
657 | ||
658 | return NULL; | |
659 | } | |
660 | ||
119bc547 | 661 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 662 | { |
1d4989c8 | 663 | struct dm_exception *e; |
1da177e4 | 664 | |
119bc547 MP |
665 | e = kmem_cache_alloc(exception_cache, gfp); |
666 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
667 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
668 | ||
669 | return e; | |
670 | } | |
671 | ||
3510cb94 | 672 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
673 | { |
674 | kmem_cache_free(exception_cache, e); | |
675 | } | |
676 | ||
92e86812 | 677 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 678 | { |
92e86812 MP |
679 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
680 | GFP_NOIO); | |
681 | ||
879129d2 | 682 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
683 | pe->snap = s; |
684 | ||
685 | return pe; | |
1da177e4 LT |
686 | } |
687 | ||
028867ac | 688 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 689 | { |
879129d2 MP |
690 | struct dm_snapshot *s = pe->snap; |
691 | ||
692 | mempool_free(pe, s->pending_pool); | |
4e857c58 | 693 | smp_mb__before_atomic(); |
879129d2 | 694 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
695 | } |
696 | ||
3510cb94 JB |
697 | static void dm_insert_exception(struct dm_exception_table *eh, |
698 | struct dm_exception *new_e) | |
d74f81f8 | 699 | { |
d74f81f8 | 700 | struct list_head *l; |
1d4989c8 | 701 | struct dm_exception *e = NULL; |
d74f81f8 MB |
702 | |
703 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
704 | ||
705 | /* Add immediately if this table doesn't support consecutive chunks */ | |
706 | if (!eh->hash_shift) | |
707 | goto out; | |
708 | ||
709 | /* List is ordered by old_chunk */ | |
710 | list_for_each_entry_reverse(e, l, hash_list) { | |
711 | /* Insert after an existing chunk? */ | |
712 | if (new_e->old_chunk == (e->old_chunk + | |
713 | dm_consecutive_chunk_count(e) + 1) && | |
714 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
715 | dm_consecutive_chunk_count(e) + 1)) { | |
716 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 717 | free_completed_exception(new_e); |
d74f81f8 MB |
718 | return; |
719 | } | |
720 | ||
721 | /* Insert before an existing chunk? */ | |
722 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
723 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
724 | dm_consecutive_chunk_count_inc(e); | |
725 | e->old_chunk--; | |
726 | e->new_chunk--; | |
3510cb94 | 727 | free_completed_exception(new_e); |
d74f81f8 MB |
728 | return; |
729 | } | |
730 | ||
731 | if (new_e->old_chunk > e->old_chunk) | |
732 | break; | |
733 | } | |
734 | ||
735 | out: | |
736 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
737 | } | |
738 | ||
a159c1ac JB |
739 | /* |
740 | * Callback used by the exception stores to load exceptions when | |
741 | * initialising. | |
742 | */ | |
743 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 744 | { |
a159c1ac | 745 | struct dm_snapshot *s = context; |
1d4989c8 | 746 | struct dm_exception *e; |
1da177e4 | 747 | |
119bc547 | 748 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
749 | if (!e) |
750 | return -ENOMEM; | |
751 | ||
752 | e->old_chunk = old; | |
d74f81f8 MB |
753 | |
754 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 755 | e->new_chunk = new; |
d74f81f8 | 756 | |
3510cb94 | 757 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 758 | |
1da177e4 LT |
759 | return 0; |
760 | } | |
761 | ||
7e201b35 MP |
762 | /* |
763 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
764 | * Return zero if the origin has no snapshots. | |
765 | */ | |
542f9038 | 766 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
767 | { |
768 | struct dm_snapshot *snap; | |
769 | unsigned chunk_size = 0; | |
770 | ||
771 | if (o) | |
772 | list_for_each_entry(snap, &o->snapshots, list) | |
773 | chunk_size = min_not_zero(chunk_size, | |
774 | snap->store->chunk_size); | |
775 | ||
542f9038 | 776 | return (uint32_t) chunk_size; |
7e201b35 MP |
777 | } |
778 | ||
1da177e4 LT |
779 | /* |
780 | * Hard coded magic. | |
781 | */ | |
782 | static int calc_max_buckets(void) | |
783 | { | |
784 | /* use a fixed size of 2MB */ | |
785 | unsigned long mem = 2 * 1024 * 1024; | |
786 | mem /= sizeof(struct list_head); | |
787 | ||
788 | return mem; | |
789 | } | |
790 | ||
1da177e4 LT |
791 | /* |
792 | * Allocate room for a suitable hash table. | |
793 | */ | |
fee1998e | 794 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 795 | { |
60e356f3 | 796 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
797 | |
798 | /* | |
799 | * Calculate based on the size of the original volume or | |
800 | * the COW volume... | |
801 | */ | |
fc56f6fb | 802 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
803 | max_buckets = calc_max_buckets(); |
804 | ||
60e356f3 | 805 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
806 | hash_size = min(hash_size, max_buckets); |
807 | ||
8e87b9b8 MP |
808 | if (hash_size < 64) |
809 | hash_size = 64; | |
8defd830 | 810 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
811 | if (dm_exception_table_init(&s->complete, hash_size, |
812 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
813 | return -ENOMEM; |
814 | ||
815 | /* | |
816 | * Allocate hash table for in-flight exceptions | |
817 | * Make this smaller than the real hash table | |
818 | */ | |
819 | hash_size >>= 3; | |
820 | if (hash_size < 64) | |
821 | hash_size = 64; | |
822 | ||
3510cb94 JB |
823 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
824 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
825 | return -ENOMEM; |
826 | } | |
827 | ||
828 | return 0; | |
829 | } | |
830 | ||
1e03f97e MP |
831 | static void merge_shutdown(struct dm_snapshot *s) |
832 | { | |
833 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 834 | smp_mb__after_atomic(); |
1e03f97e MP |
835 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
836 | } | |
837 | ||
9fe86254 MP |
838 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
839 | { | |
840 | s->first_merging_chunk = 0; | |
841 | s->num_merging_chunks = 0; | |
842 | ||
843 | return bio_list_get(&s->bios_queued_during_merge); | |
844 | } | |
845 | ||
1e03f97e MP |
846 | /* |
847 | * Remove one chunk from the index of completed exceptions. | |
848 | */ | |
849 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
850 | chunk_t old_chunk) | |
851 | { | |
852 | struct dm_exception *e; | |
853 | ||
1e03f97e MP |
854 | e = dm_lookup_exception(&s->complete, old_chunk); |
855 | if (!e) { | |
856 | DMERR("Corruption detected: exception for block %llu is " | |
857 | "on disk but not in memory", | |
858 | (unsigned long long)old_chunk); | |
859 | return -EINVAL; | |
860 | } | |
861 | ||
862 | /* | |
863 | * If this is the only chunk using this exception, remove exception. | |
864 | */ | |
865 | if (!dm_consecutive_chunk_count(e)) { | |
866 | dm_remove_exception(e); | |
867 | free_completed_exception(e); | |
868 | return 0; | |
869 | } | |
870 | ||
871 | /* | |
872 | * The chunk may be either at the beginning or the end of a | |
873 | * group of consecutive chunks - never in the middle. We are | |
874 | * removing chunks in the opposite order to that in which they | |
875 | * were added, so this should always be true. | |
876 | * Decrement the consecutive chunk counter and adjust the | |
877 | * starting point if necessary. | |
878 | */ | |
879 | if (old_chunk == e->old_chunk) { | |
880 | e->old_chunk++; | |
881 | e->new_chunk++; | |
882 | } else if (old_chunk != e->old_chunk + | |
883 | dm_consecutive_chunk_count(e)) { | |
884 | DMERR("Attempt to merge block %llu from the " | |
885 | "middle of a chunk range [%llu - %llu]", | |
886 | (unsigned long long)old_chunk, | |
887 | (unsigned long long)e->old_chunk, | |
888 | (unsigned long long) | |
889 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
890 | return -EINVAL; | |
891 | } | |
892 | ||
893 | dm_consecutive_chunk_count_dec(e); | |
894 | ||
895 | return 0; | |
896 | } | |
897 | ||
9fe86254 MP |
898 | static void flush_bios(struct bio *bio); |
899 | ||
900 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 901 | { |
9fe86254 MP |
902 | struct bio *b = NULL; |
903 | int r; | |
904 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
905 | |
906 | down_write(&s->lock); | |
9fe86254 MP |
907 | |
908 | /* | |
909 | * Process chunks (and associated exceptions) in reverse order | |
910 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
911 | */ | |
912 | do { | |
913 | r = __remove_single_exception_chunk(s, old_chunk); | |
914 | if (r) | |
915 | goto out; | |
916 | } while (old_chunk-- > s->first_merging_chunk); | |
917 | ||
918 | b = __release_queued_bios_after_merge(s); | |
919 | ||
920 | out: | |
1e03f97e | 921 | up_write(&s->lock); |
9fe86254 MP |
922 | if (b) |
923 | flush_bios(b); | |
1e03f97e MP |
924 | |
925 | return r; | |
926 | } | |
927 | ||
73dfd078 MP |
928 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
929 | sector_t sector, unsigned chunk_size); | |
930 | ||
1e03f97e MP |
931 | static void merge_callback(int read_err, unsigned long write_err, |
932 | void *context); | |
933 | ||
73dfd078 MP |
934 | static uint64_t read_pending_exceptions_done_count(void) |
935 | { | |
936 | uint64_t pending_exceptions_done; | |
937 | ||
938 | spin_lock(&_pending_exceptions_done_spinlock); | |
939 | pending_exceptions_done = _pending_exceptions_done_count; | |
940 | spin_unlock(&_pending_exceptions_done_spinlock); | |
941 | ||
942 | return pending_exceptions_done; | |
943 | } | |
944 | ||
945 | static void increment_pending_exceptions_done_count(void) | |
946 | { | |
947 | spin_lock(&_pending_exceptions_done_spinlock); | |
948 | _pending_exceptions_done_count++; | |
949 | spin_unlock(&_pending_exceptions_done_spinlock); | |
950 | ||
951 | wake_up_all(&_pending_exceptions_done); | |
952 | } | |
953 | ||
1e03f97e MP |
954 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
955 | { | |
8a2d5286 | 956 | int i, linear_chunks; |
1e03f97e MP |
957 | chunk_t old_chunk, new_chunk; |
958 | struct dm_io_region src, dest; | |
8a2d5286 | 959 | sector_t io_size; |
73dfd078 | 960 | uint64_t previous_count; |
1e03f97e MP |
961 | |
962 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
963 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
964 | goto shut; | |
965 | ||
966 | /* | |
967 | * valid flag never changes during merge, so no lock required. | |
968 | */ | |
969 | if (!s->valid) { | |
970 | DMERR("Snapshot is invalid: can't merge"); | |
971 | goto shut; | |
972 | } | |
973 | ||
8a2d5286 MS |
974 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
975 | &new_chunk); | |
976 | if (linear_chunks <= 0) { | |
d8ddb1cf | 977 | if (linear_chunks < 0) { |
1e03f97e MP |
978 | DMERR("Read error in exception store: " |
979 | "shutting down merge"); | |
d8ddb1cf MS |
980 | down_write(&s->lock); |
981 | s->merge_failed = 1; | |
982 | up_write(&s->lock); | |
983 | } | |
1e03f97e MP |
984 | goto shut; |
985 | } | |
986 | ||
8a2d5286 MS |
987 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
988 | old_chunk = old_chunk + 1 - linear_chunks; | |
989 | new_chunk = new_chunk + 1 - linear_chunks; | |
990 | ||
991 | /* | |
992 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
993 | * from the exception store to the origin | |
994 | */ | |
995 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 996 | |
1e03f97e MP |
997 | dest.bdev = s->origin->bdev; |
998 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 999 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
1000 | |
1001 | src.bdev = s->cow->bdev; | |
1002 | src.sector = chunk_to_sector(s->store, new_chunk); | |
1003 | src.count = dest.count; | |
1004 | ||
73dfd078 MP |
1005 | /* |
1006 | * Reallocate any exceptions needed in other snapshots then | |
1007 | * wait for the pending exceptions to complete. | |
1008 | * Each time any pending exception (globally on the system) | |
1009 | * completes we are woken and repeat the process to find out | |
1010 | * if we can proceed. While this may not seem a particularly | |
1011 | * efficient algorithm, it is not expected to have any | |
1012 | * significant impact on performance. | |
1013 | */ | |
1014 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 1015 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
1016 | wait_event(_pending_exceptions_done, |
1017 | (read_pending_exceptions_done_count() != | |
1018 | previous_count)); | |
1019 | /* Retry after the wait, until all exceptions are done. */ | |
1020 | previous_count = read_pending_exceptions_done_count(); | |
1021 | } | |
1022 | ||
9fe86254 MP |
1023 | down_write(&s->lock); |
1024 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 1025 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
1026 | up_write(&s->lock); |
1027 | ||
8a2d5286 MS |
1028 | /* Wait until writes to all 'linear_chunks' drain */ |
1029 | for (i = 0; i < linear_chunks; i++) | |
1030 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 1031 | |
1e03f97e MP |
1032 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
1033 | return; | |
1034 | ||
1035 | shut: | |
1036 | merge_shutdown(s); | |
1037 | } | |
1038 | ||
9fe86254 MP |
1039 | static void error_bios(struct bio *bio); |
1040 | ||
1e03f97e MP |
1041 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
1042 | { | |
1043 | struct dm_snapshot *s = context; | |
9fe86254 | 1044 | struct bio *b = NULL; |
1e03f97e MP |
1045 | |
1046 | if (read_err || write_err) { | |
1047 | if (read_err) | |
1048 | DMERR("Read error: shutting down merge."); | |
1049 | else | |
1050 | DMERR("Write error: shutting down merge."); | |
1051 | goto shut; | |
1052 | } | |
1053 | ||
9fe86254 MP |
1054 | if (s->store->type->commit_merge(s->store, |
1055 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1056 | DMERR("Write error in exception store: shutting down merge"); |
1057 | goto shut; | |
1058 | } | |
1059 | ||
9fe86254 MP |
1060 | if (remove_single_exception_chunk(s) < 0) |
1061 | goto shut; | |
1062 | ||
1e03f97e MP |
1063 | snapshot_merge_next_chunks(s); |
1064 | ||
1065 | return; | |
1066 | ||
1067 | shut: | |
9fe86254 | 1068 | down_write(&s->lock); |
d8ddb1cf | 1069 | s->merge_failed = 1; |
9fe86254 MP |
1070 | b = __release_queued_bios_after_merge(s); |
1071 | up_write(&s->lock); | |
1072 | error_bios(b); | |
1073 | ||
1e03f97e MP |
1074 | merge_shutdown(s); |
1075 | } | |
1076 | ||
1077 | static void start_merge(struct dm_snapshot *s) | |
1078 | { | |
1079 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1080 | snapshot_merge_next_chunks(s); | |
1081 | } | |
1082 | ||
1e03f97e MP |
1083 | /* |
1084 | * Stop the merging process and wait until it finishes. | |
1085 | */ | |
1086 | static void stop_merge(struct dm_snapshot *s) | |
1087 | { | |
1088 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1089 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1090 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1091 | } | |
1092 | ||
1da177e4 LT |
1093 | /* |
1094 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
1095 | */ | |
1096 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1097 | { | |
1098 | struct dm_snapshot *s; | |
cd45daff | 1099 | int i; |
1da177e4 | 1100 | int r = -EINVAL; |
fc56f6fb | 1101 | char *origin_path, *cow_path; |
55a62eef | 1102 | unsigned args_used, num_flush_bios = 1; |
10b8106a | 1103 | fmode_t origin_mode = FMODE_READ; |
1da177e4 | 1104 | |
4c7e3bf4 | 1105 | if (argc != 4) { |
72d94861 | 1106 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1107 | r = -EINVAL; |
fc56f6fb | 1108 | goto bad; |
1da177e4 LT |
1109 | } |
1110 | ||
10b8106a | 1111 | if (dm_target_is_snapshot_merge(ti)) { |
55a62eef | 1112 | num_flush_bios = 2; |
10b8106a MS |
1113 | origin_mode = FMODE_WRITE; |
1114 | } | |
1115 | ||
fc56f6fb MS |
1116 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1117 | if (!s) { | |
a2d2b034 | 1118 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1119 | r = -ENOMEM; |
1120 | goto bad; | |
1121 | } | |
1122 | ||
c2411045 MP |
1123 | origin_path = argv[0]; |
1124 | argv++; | |
1125 | argc--; | |
1126 | ||
1127 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1128 | if (r) { | |
1129 | ti->error = "Cannot get origin device"; | |
1130 | goto bad_origin; | |
1131 | } | |
1132 | ||
fc56f6fb MS |
1133 | cow_path = argv[0]; |
1134 | argv++; | |
1135 | argc--; | |
1136 | ||
024d37e9 | 1137 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1138 | if (r) { |
1139 | ti->error = "Cannot get COW device"; | |
1140 | goto bad_cow; | |
1141 | } | |
1142 | ||
1143 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1144 | if (r) { |
1145 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1146 | r = -EINVAL; |
fc56f6fb | 1147 | goto bad_store; |
1da177e4 LT |
1148 | } |
1149 | ||
fee1998e JB |
1150 | argv += args_used; |
1151 | argc -= args_used; | |
1152 | ||
fc56f6fb | 1153 | s->ti = ti; |
1da177e4 | 1154 | s->valid = 1; |
aa14edeb | 1155 | s->active = 0; |
879129d2 | 1156 | atomic_set(&s->pending_exceptions_count, 0); |
230c83af MP |
1157 | s->exception_start_sequence = 0; |
1158 | s->exception_complete_sequence = 0; | |
1159 | INIT_LIST_HEAD(&s->out_of_order_list); | |
1da177e4 | 1160 | init_rwsem(&s->lock); |
c1f0c183 | 1161 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1162 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1163 | s->state_bits = 0; |
d8ddb1cf | 1164 | s->merge_failed = 0; |
9fe86254 MP |
1165 | s->first_merging_chunk = 0; |
1166 | s->num_merging_chunks = 0; | |
1167 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1168 | |
1169 | /* Allocate hash table for COW data */ | |
fee1998e | 1170 | if (init_hash_tables(s)) { |
1da177e4 LT |
1171 | ti->error = "Unable to allocate hash table space"; |
1172 | r = -ENOMEM; | |
fee1998e | 1173 | goto bad_hash_tables; |
1da177e4 LT |
1174 | } |
1175 | ||
df5d2e90 | 1176 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1177 | if (IS_ERR(s->kcopyd_client)) { |
1178 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1179 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1180 | goto bad_kcopyd; |
1da177e4 LT |
1181 | } |
1182 | ||
92e86812 MP |
1183 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1184 | if (!s->pending_pool) { | |
1185 | ti->error = "Could not allocate mempool for pending exceptions"; | |
09e8b813 | 1186 | r = -ENOMEM; |
fee1998e | 1187 | goto bad_pending_pool; |
92e86812 MP |
1188 | } |
1189 | ||
cd45daff MP |
1190 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1191 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1192 | ||
1193 | spin_lock_init(&s->tracked_chunk_lock); | |
1194 | ||
c1f0c183 | 1195 | ti->private = s; |
55a62eef | 1196 | ti->num_flush_bios = num_flush_bios; |
42bc954f | 1197 | ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1198 | |
1199 | /* Add snapshot to the list of snapshots for this origin */ | |
1200 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1201 | r = register_snapshot(s); | |
1202 | if (r == -ENOMEM) { | |
1203 | ti->error = "Snapshot origin struct allocation failed"; | |
1204 | goto bad_load_and_register; | |
1205 | } else if (r < 0) { | |
1206 | /* invalid handover, register_snapshot has set ti->error */ | |
1207 | goto bad_load_and_register; | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Metadata must only be loaded into one table at once, so skip this | |
1212 | * if metadata will be handed over during resume. | |
1213 | * Chunk size will be set during the handover - set it to zero to | |
1214 | * ensure it's ignored. | |
1215 | */ | |
1216 | if (r > 0) { | |
1217 | s->store->chunk_size = 0; | |
1218 | return 0; | |
1219 | } | |
1220 | ||
493df71c JB |
1221 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1222 | (void *)s); | |
0764147b | 1223 | if (r < 0) { |
f9cea4f7 | 1224 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1225 | goto bad_read_metadata; |
0764147b MB |
1226 | } else if (r > 0) { |
1227 | s->valid = 0; | |
1228 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1229 | } |
aa14edeb | 1230 | |
3f2412dc MP |
1231 | if (!s->store->chunk_size) { |
1232 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1233 | goto bad_read_metadata; |
1da177e4 | 1234 | } |
542f9038 MS |
1235 | |
1236 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1237 | if (r) | |
1238 | goto bad_read_metadata; | |
1da177e4 LT |
1239 | |
1240 | return 0; | |
1241 | ||
c1f0c183 MS |
1242 | bad_read_metadata: |
1243 | unregister_snapshot(s); | |
1244 | ||
fee1998e | 1245 | bad_load_and_register: |
92e86812 MP |
1246 | mempool_destroy(s->pending_pool); |
1247 | ||
fee1998e | 1248 | bad_pending_pool: |
eb69aca5 | 1249 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1250 | |
fee1998e | 1251 | bad_kcopyd: |
3510cb94 JB |
1252 | dm_exception_table_exit(&s->pending, pending_cache); |
1253 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1254 | |
fee1998e | 1255 | bad_hash_tables: |
fc56f6fb | 1256 | dm_exception_store_destroy(s->store); |
1da177e4 | 1257 | |
fc56f6fb MS |
1258 | bad_store: |
1259 | dm_put_device(ti, s->cow); | |
fee1998e | 1260 | |
fc56f6fb | 1261 | bad_cow: |
c2411045 MP |
1262 | dm_put_device(ti, s->origin); |
1263 | ||
1264 | bad_origin: | |
fc56f6fb MS |
1265 | kfree(s); |
1266 | ||
1267 | bad: | |
1da177e4 LT |
1268 | return r; |
1269 | } | |
1270 | ||
31c93a0c MB |
1271 | static void __free_exceptions(struct dm_snapshot *s) |
1272 | { | |
eb69aca5 | 1273 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1274 | s->kcopyd_client = NULL; |
1275 | ||
3510cb94 JB |
1276 | dm_exception_table_exit(&s->pending, pending_cache); |
1277 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1278 | } |
1279 | ||
c1f0c183 MS |
1280 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1281 | struct dm_snapshot *snap_dest) | |
1282 | { | |
1283 | union { | |
1284 | struct dm_exception_table table_swap; | |
1285 | struct dm_exception_store *store_swap; | |
1286 | } u; | |
1287 | ||
1288 | /* | |
1289 | * Swap all snapshot context information between the two instances. | |
1290 | */ | |
1291 | u.table_swap = snap_dest->complete; | |
1292 | snap_dest->complete = snap_src->complete; | |
1293 | snap_src->complete = u.table_swap; | |
1294 | ||
1295 | u.store_swap = snap_dest->store; | |
1296 | snap_dest->store = snap_src->store; | |
1297 | snap_src->store = u.store_swap; | |
1298 | ||
1299 | snap_dest->store->snap = snap_dest; | |
1300 | snap_src->store->snap = snap_src; | |
1301 | ||
542f9038 | 1302 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 MS |
1303 | snap_dest->valid = snap_src->valid; |
1304 | ||
1305 | /* | |
1306 | * Set source invalid to ensure it receives no further I/O. | |
1307 | */ | |
1308 | snap_src->valid = 0; | |
1309 | } | |
1310 | ||
1da177e4 LT |
1311 | static void snapshot_dtr(struct dm_target *ti) |
1312 | { | |
cd45daff MP |
1313 | #ifdef CONFIG_DM_DEBUG |
1314 | int i; | |
1315 | #endif | |
028867ac | 1316 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1317 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1318 | |
c1f0c183 MS |
1319 | down_read(&_origins_lock); |
1320 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1321 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1322 | if (snap_src && snap_dest && (s == snap_src)) { |
1323 | down_write(&snap_dest->lock); | |
1324 | snap_dest->valid = 0; | |
1325 | up_write(&snap_dest->lock); | |
1326 | DMERR("Cancelling snapshot handover."); | |
1327 | } | |
1328 | up_read(&_origins_lock); | |
1329 | ||
1e03f97e MP |
1330 | if (dm_target_is_snapshot_merge(ti)) |
1331 | stop_merge(s); | |
1332 | ||
138728dc AK |
1333 | /* Prevent further origin writes from using this snapshot. */ |
1334 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1335 | unregister_snapshot(s); |
1336 | ||
879129d2 | 1337 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1338 | msleep(1); |
879129d2 MP |
1339 | /* |
1340 | * Ensure instructions in mempool_destroy aren't reordered | |
1341 | * before atomic_read. | |
1342 | */ | |
1343 | smp_mb(); | |
1344 | ||
cd45daff MP |
1345 | #ifdef CONFIG_DM_DEBUG |
1346 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1347 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1348 | #endif | |
1349 | ||
31c93a0c | 1350 | __free_exceptions(s); |
1da177e4 | 1351 | |
92e86812 MP |
1352 | mempool_destroy(s->pending_pool); |
1353 | ||
fee1998e | 1354 | dm_exception_store_destroy(s->store); |
138728dc | 1355 | |
fc56f6fb MS |
1356 | dm_put_device(ti, s->cow); |
1357 | ||
c2411045 MP |
1358 | dm_put_device(ti, s->origin); |
1359 | ||
1da177e4 LT |
1360 | kfree(s); |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * Flush a list of buffers. | |
1365 | */ | |
1366 | static void flush_bios(struct bio *bio) | |
1367 | { | |
1368 | struct bio *n; | |
1369 | ||
1370 | while (bio) { | |
1371 | n = bio->bi_next; | |
1372 | bio->bi_next = NULL; | |
1373 | generic_make_request(bio); | |
1374 | bio = n; | |
1375 | } | |
1376 | } | |
1377 | ||
515ad66c MP |
1378 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1379 | ||
1380 | /* | |
1381 | * Flush a list of buffers. | |
1382 | */ | |
1383 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1384 | { | |
1385 | struct bio *n; | |
1386 | int r; | |
1387 | ||
1388 | while (bio) { | |
1389 | n = bio->bi_next; | |
1390 | bio->bi_next = NULL; | |
1391 | r = do_origin(s->origin, bio); | |
1392 | if (r == DM_MAPIO_REMAPPED) | |
1393 | generic_make_request(bio); | |
1394 | bio = n; | |
1395 | } | |
1396 | } | |
1397 | ||
1da177e4 LT |
1398 | /* |
1399 | * Error a list of buffers. | |
1400 | */ | |
1401 | static void error_bios(struct bio *bio) | |
1402 | { | |
1403 | struct bio *n; | |
1404 | ||
1405 | while (bio) { | |
1406 | n = bio->bi_next; | |
1407 | bio->bi_next = NULL; | |
6712ecf8 | 1408 | bio_io_error(bio); |
1da177e4 LT |
1409 | bio = n; |
1410 | } | |
1411 | } | |
1412 | ||
695368ac | 1413 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1414 | { |
1415 | if (!s->valid) | |
1416 | return; | |
1417 | ||
1418 | if (err == -EIO) | |
1419 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1420 | else if (err == -ENOMEM) | |
1421 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1422 | ||
493df71c JB |
1423 | if (s->store->type->drop_snapshot) |
1424 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1425 | |
1426 | s->valid = 0; | |
1427 | ||
fc56f6fb | 1428 | dm_table_event(s->ti->table); |
76df1c65 AK |
1429 | } |
1430 | ||
028867ac | 1431 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1432 | { |
1d4989c8 | 1433 | struct dm_exception *e; |
1da177e4 | 1434 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1435 | struct bio *origin_bios = NULL; |
1436 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1437 | struct bio *full_bio = NULL; |
9d493fa8 | 1438 | int error = 0; |
1da177e4 | 1439 | |
76df1c65 AK |
1440 | if (!success) { |
1441 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1442 | down_write(&s->lock); |
695368ac | 1443 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1444 | error = 1; |
76df1c65 AK |
1445 | goto out; |
1446 | } | |
1447 | ||
119bc547 | 1448 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1449 | if (!e) { |
1da177e4 | 1450 | down_write(&s->lock); |
695368ac | 1451 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1452 | error = 1; |
76df1c65 AK |
1453 | goto out; |
1454 | } | |
1455 | *e = pe->e; | |
1da177e4 | 1456 | |
76df1c65 AK |
1457 | down_write(&s->lock); |
1458 | if (!s->valid) { | |
3510cb94 | 1459 | free_completed_exception(e); |
9d493fa8 | 1460 | error = 1; |
76df1c65 | 1461 | goto out; |
1da177e4 LT |
1462 | } |
1463 | ||
615d1eb9 MS |
1464 | /* Check for conflicting reads */ |
1465 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1466 | |
9d493fa8 AK |
1467 | /* |
1468 | * Add a proper exception, and remove the | |
1469 | * in-flight exception from the list. | |
1470 | */ | |
3510cb94 | 1471 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1472 | |
a2d2b034 | 1473 | out: |
3510cb94 | 1474 | dm_remove_exception(&pe->e); |
9d493fa8 | 1475 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1476 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 MP |
1477 | full_bio = pe->full_bio; |
1478 | if (full_bio) { | |
1479 | full_bio->bi_end_io = pe->full_bio_end_io; | |
1480 | full_bio->bi_private = pe->full_bio_private; | |
196d38bc | 1481 | atomic_inc(&full_bio->bi_remaining); |
a6e50b40 | 1482 | } |
73dfd078 MP |
1483 | increment_pending_exceptions_done_count(); |
1484 | ||
9d493fa8 AK |
1485 | up_write(&s->lock); |
1486 | ||
1487 | /* Submit any pending write bios */ | |
a6e50b40 MP |
1488 | if (error) { |
1489 | if (full_bio) | |
1490 | bio_io_error(full_bio); | |
9d493fa8 | 1491 | error_bios(snapshot_bios); |
a6e50b40 MP |
1492 | } else { |
1493 | if (full_bio) | |
1494 | bio_endio(full_bio, 0); | |
9d493fa8 | 1495 | flush_bios(snapshot_bios); |
a6e50b40 | 1496 | } |
9d493fa8 | 1497 | |
515ad66c | 1498 | retry_origin_bios(s, origin_bios); |
22aa66a3 MP |
1499 | |
1500 | free_pending_exception(pe); | |
1da177e4 LT |
1501 | } |
1502 | ||
1503 | static void commit_callback(void *context, int success) | |
1504 | { | |
028867ac AK |
1505 | struct dm_snap_pending_exception *pe = context; |
1506 | ||
1da177e4 LT |
1507 | pending_complete(pe, success); |
1508 | } | |
1509 | ||
230c83af MP |
1510 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1511 | { | |
1512 | struct dm_snapshot *s = pe->snap; | |
1513 | ||
1514 | if (unlikely(pe->copy_error)) | |
1515 | pending_complete(pe, 0); | |
1516 | ||
1517 | else | |
1518 | /* Update the metadata if we are persistent */ | |
1519 | s->store->type->commit_exception(s->store, &pe->e, | |
1520 | commit_callback, pe); | |
1521 | } | |
1522 | ||
1da177e4 LT |
1523 | /* |
1524 | * Called when the copy I/O has finished. kcopyd actually runs | |
1525 | * this code so don't block. | |
1526 | */ | |
4cdc1d1f | 1527 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1528 | { |
028867ac | 1529 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1530 | struct dm_snapshot *s = pe->snap; |
1531 | ||
230c83af | 1532 | pe->copy_error = read_err || write_err; |
1da177e4 | 1533 | |
230c83af MP |
1534 | if (pe->exception_sequence == s->exception_complete_sequence) { |
1535 | s->exception_complete_sequence++; | |
1536 | complete_exception(pe); | |
1537 | ||
1538 | while (!list_empty(&s->out_of_order_list)) { | |
1539 | pe = list_entry(s->out_of_order_list.next, | |
1540 | struct dm_snap_pending_exception, out_of_order_entry); | |
1541 | if (pe->exception_sequence != s->exception_complete_sequence) | |
1542 | break; | |
1543 | s->exception_complete_sequence++; | |
1544 | list_del(&pe->out_of_order_entry); | |
1545 | complete_exception(pe); | |
1546 | } | |
1547 | } else { | |
1548 | struct list_head *lh; | |
1549 | struct dm_snap_pending_exception *pe2; | |
1550 | ||
1551 | list_for_each_prev(lh, &s->out_of_order_list) { | |
1552 | pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry); | |
1553 | if (pe2->exception_sequence < pe->exception_sequence) | |
1554 | break; | |
1555 | } | |
1556 | list_add(&pe->out_of_order_entry, lh); | |
1557 | } | |
1da177e4 LT |
1558 | } |
1559 | ||
1560 | /* | |
1561 | * Dispatches the copy operation to kcopyd. | |
1562 | */ | |
028867ac | 1563 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1564 | { |
1565 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1566 | struct dm_io_region src, dest; |
1da177e4 LT |
1567 | struct block_device *bdev = s->origin->bdev; |
1568 | sector_t dev_size; | |
1569 | ||
1570 | dev_size = get_dev_size(bdev); | |
1571 | ||
1572 | src.bdev = bdev; | |
71fab00a | 1573 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1574 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1575 | |
fc56f6fb | 1576 | dest.bdev = s->cow->bdev; |
71fab00a | 1577 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1578 | dest.count = src.count; |
1579 | ||
1580 | /* Hand over to kcopyd */ | |
a2d2b034 | 1581 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1582 | } |
1583 | ||
a6e50b40 MP |
1584 | static void full_bio_end_io(struct bio *bio, int error) |
1585 | { | |
1586 | void *callback_data = bio->bi_private; | |
1587 | ||
1588 | dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0); | |
1589 | } | |
1590 | ||
1591 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1592 | struct bio *bio) | |
1593 | { | |
1594 | struct dm_snapshot *s = pe->snap; | |
1595 | void *callback_data; | |
1596 | ||
1597 | pe->full_bio = bio; | |
1598 | pe->full_bio_end_io = bio->bi_end_io; | |
1599 | pe->full_bio_private = bio->bi_private; | |
1600 | ||
1601 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, | |
1602 | copy_callback, pe); | |
1603 | ||
1604 | bio->bi_end_io = full_bio_end_io; | |
1605 | bio->bi_private = callback_data; | |
1606 | ||
1607 | generic_make_request(bio); | |
1608 | } | |
1609 | ||
2913808e MP |
1610 | static struct dm_snap_pending_exception * |
1611 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1612 | { | |
3510cb94 | 1613 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1614 | |
1615 | if (!e) | |
1616 | return NULL; | |
1617 | ||
1618 | return container_of(e, struct dm_snap_pending_exception, e); | |
1619 | } | |
1620 | ||
1da177e4 LT |
1621 | /* |
1622 | * Looks to see if this snapshot already has a pending exception | |
1623 | * for this chunk, otherwise it allocates a new one and inserts | |
1624 | * it into the pending table. | |
1625 | * | |
1626 | * NOTE: a write lock must be held on snap->lock before calling | |
1627 | * this. | |
1628 | */ | |
028867ac | 1629 | static struct dm_snap_pending_exception * |
c6621392 MP |
1630 | __find_pending_exception(struct dm_snapshot *s, |
1631 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1632 | { |
c6621392 | 1633 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1634 | |
2913808e MP |
1635 | pe2 = __lookup_pending_exception(s, chunk); |
1636 | if (pe2) { | |
76df1c65 | 1637 | free_pending_exception(pe); |
2913808e | 1638 | return pe2; |
1da177e4 LT |
1639 | } |
1640 | ||
76df1c65 AK |
1641 | pe->e.old_chunk = chunk; |
1642 | bio_list_init(&pe->origin_bios); | |
1643 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1644 | pe->started = 0; |
a6e50b40 | 1645 | pe->full_bio = NULL; |
76df1c65 | 1646 | |
493df71c | 1647 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1648 | free_pending_exception(pe); |
1649 | return NULL; | |
1650 | } | |
1651 | ||
230c83af MP |
1652 | pe->exception_sequence = s->exception_start_sequence++; |
1653 | ||
3510cb94 | 1654 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1655 | |
1da177e4 LT |
1656 | return pe; |
1657 | } | |
1658 | ||
1d4989c8 | 1659 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1660 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1661 | { |
fc56f6fb | 1662 | bio->bi_bdev = s->cow->bdev; |
4f024f37 KO |
1663 | bio->bi_iter.bi_sector = |
1664 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1665 | (chunk - e->old_chunk)) + | |
1666 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1667 | } |
1668 | ||
7de3ee57 | 1669 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1670 | { |
1d4989c8 | 1671 | struct dm_exception *e; |
028867ac | 1672 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1673 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1674 | chunk_t chunk; |
028867ac | 1675 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1676 | |
ee18026a MP |
1677 | init_tracked_chunk(bio); |
1678 | ||
d87f4c14 | 1679 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1680 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1681 | return DM_MAPIO_REMAPPED; |
1682 | } | |
1683 | ||
4f024f37 | 1684 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
1da177e4 LT |
1685 | |
1686 | /* Full snapshots are not usable */ | |
76df1c65 | 1687 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1688 | if (!s->valid) |
f6a80ea8 | 1689 | return -EIO; |
1da177e4 | 1690 | |
ba40a2aa AK |
1691 | /* FIXME: should only take write lock if we need |
1692 | * to copy an exception */ | |
1693 | down_write(&s->lock); | |
1694 | ||
1695 | if (!s->valid) { | |
1696 | r = -EIO; | |
1697 | goto out_unlock; | |
1698 | } | |
1699 | ||
1700 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1701 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1702 | if (e) { |
d74f81f8 | 1703 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1704 | goto out_unlock; |
1705 | } | |
1706 | ||
1da177e4 LT |
1707 | /* |
1708 | * Write to snapshot - higher level takes care of RW/RO | |
1709 | * flags so we should only get this if we are | |
1710 | * writeable. | |
1711 | */ | |
1712 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1713 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1714 | if (!pe) { |
c6621392 MP |
1715 | up_write(&s->lock); |
1716 | pe = alloc_pending_exception(s); | |
1717 | down_write(&s->lock); | |
1718 | ||
1719 | if (!s->valid) { | |
1720 | free_pending_exception(pe); | |
1721 | r = -EIO; | |
1722 | goto out_unlock; | |
1723 | } | |
1724 | ||
3510cb94 | 1725 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1726 | if (e) { |
1727 | free_pending_exception(pe); | |
1728 | remap_exception(s, e, bio, chunk); | |
1729 | goto out_unlock; | |
1730 | } | |
1731 | ||
c6621392 | 1732 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1733 | if (!pe) { |
1734 | __invalidate_snapshot(s, -ENOMEM); | |
1735 | r = -EIO; | |
1736 | goto out_unlock; | |
1737 | } | |
1da177e4 LT |
1738 | } |
1739 | ||
d74f81f8 | 1740 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 1741 | |
d2a7ad29 | 1742 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1743 | |
a6e50b40 | 1744 | if (!pe->started && |
4f024f37 KO |
1745 | bio->bi_iter.bi_size == |
1746 | (s->store->chunk_size << SECTOR_SHIFT)) { | |
a6e50b40 MP |
1747 | pe->started = 1; |
1748 | up_write(&s->lock); | |
1749 | start_full_bio(pe, bio); | |
1750 | goto out; | |
1751 | } | |
1752 | ||
1753 | bio_list_add(&pe->snapshot_bios, bio); | |
1754 | ||
76df1c65 AK |
1755 | if (!pe->started) { |
1756 | /* this is protected by snap->lock */ | |
1757 | pe->started = 1; | |
ba40a2aa | 1758 | up_write(&s->lock); |
76df1c65 | 1759 | start_copy(pe); |
ba40a2aa AK |
1760 | goto out; |
1761 | } | |
cd45daff | 1762 | } else { |
ba40a2aa | 1763 | bio->bi_bdev = s->origin->bdev; |
ee18026a | 1764 | track_chunk(s, bio, chunk); |
cd45daff | 1765 | } |
1da177e4 | 1766 | |
a2d2b034 | 1767 | out_unlock: |
ba40a2aa | 1768 | up_write(&s->lock); |
a2d2b034 | 1769 | out: |
1da177e4 LT |
1770 | return r; |
1771 | } | |
1772 | ||
3452c2a1 MP |
1773 | /* |
1774 | * A snapshot-merge target behaves like a combination of a snapshot | |
1775 | * target and a snapshot-origin target. It only generates new | |
1776 | * exceptions in other snapshots and not in the one that is being | |
1777 | * merged. | |
1778 | * | |
1779 | * For each chunk, if there is an existing exception, it is used to | |
1780 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1781 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1782 | * If merging is currently taking place on the chunk in question, the |
1783 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 1784 | */ |
7de3ee57 | 1785 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
1786 | { |
1787 | struct dm_exception *e; | |
1788 | struct dm_snapshot *s = ti->private; | |
1789 | int r = DM_MAPIO_REMAPPED; | |
1790 | chunk_t chunk; | |
1791 | ||
ee18026a MP |
1792 | init_tracked_chunk(bio); |
1793 | ||
d87f4c14 | 1794 | if (bio->bi_rw & REQ_FLUSH) { |
55a62eef | 1795 | if (!dm_bio_get_target_bio_nr(bio)) |
10b8106a MS |
1796 | bio->bi_bdev = s->origin->bdev; |
1797 | else | |
1798 | bio->bi_bdev = s->cow->bdev; | |
10b8106a MS |
1799 | return DM_MAPIO_REMAPPED; |
1800 | } | |
1801 | ||
4f024f37 | 1802 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 1803 | |
9fe86254 | 1804 | down_write(&s->lock); |
3452c2a1 | 1805 | |
d2fdb776 MP |
1806 | /* Full merging snapshots are redirected to the origin */ |
1807 | if (!s->valid) | |
1808 | goto redirect_to_origin; | |
3452c2a1 MP |
1809 | |
1810 | /* If the block is already remapped - use that */ | |
1811 | e = dm_lookup_exception(&s->complete, chunk); | |
1812 | if (e) { | |
9fe86254 MP |
1813 | /* Queue writes overlapping with chunks being merged */ |
1814 | if (bio_rw(bio) == WRITE && | |
1815 | chunk >= s->first_merging_chunk && | |
1816 | chunk < (s->first_merging_chunk + | |
1817 | s->num_merging_chunks)) { | |
1818 | bio->bi_bdev = s->origin->bdev; | |
1819 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1820 | r = DM_MAPIO_SUBMITTED; | |
1821 | goto out_unlock; | |
1822 | } | |
17aa0332 | 1823 | |
3452c2a1 | 1824 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1825 | |
1826 | if (bio_rw(bio) == WRITE) | |
ee18026a | 1827 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
1828 | goto out_unlock; |
1829 | } | |
1830 | ||
d2fdb776 | 1831 | redirect_to_origin: |
3452c2a1 MP |
1832 | bio->bi_bdev = s->origin->bdev; |
1833 | ||
1834 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1835 | up_write(&s->lock); |
3452c2a1 MP |
1836 | return do_origin(s->origin, bio); |
1837 | } | |
1838 | ||
1839 | out_unlock: | |
9fe86254 | 1840 | up_write(&s->lock); |
3452c2a1 MP |
1841 | |
1842 | return r; | |
1843 | } | |
1844 | ||
7de3ee57 | 1845 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error) |
cd45daff MP |
1846 | { |
1847 | struct dm_snapshot *s = ti->private; | |
cd45daff | 1848 | |
ee18026a MP |
1849 | if (is_bio_tracked(bio)) |
1850 | stop_tracking_chunk(s, bio); | |
cd45daff MP |
1851 | |
1852 | return 0; | |
1853 | } | |
1854 | ||
1e03f97e MP |
1855 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1856 | { | |
1857 | struct dm_snapshot *s = ti->private; | |
1858 | ||
1859 | stop_merge(s); | |
1860 | } | |
1861 | ||
c1f0c183 MS |
1862 | static int snapshot_preresume(struct dm_target *ti) |
1863 | { | |
1864 | int r = 0; | |
1865 | struct dm_snapshot *s = ti->private; | |
1866 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1867 | ||
1868 | down_read(&_origins_lock); | |
9d3b15c4 | 1869 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1870 | if (snap_src && snap_dest) { |
1871 | down_read(&snap_src->lock); | |
1872 | if (s == snap_src) { | |
1873 | DMERR("Unable to resume snapshot source until " | |
1874 | "handover completes."); | |
1875 | r = -EINVAL; | |
b83b2f29 | 1876 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
1877 | DMERR("Unable to perform snapshot handover until " |
1878 | "source is suspended."); | |
1879 | r = -EINVAL; | |
1880 | } | |
1881 | up_read(&snap_src->lock); | |
1882 | } | |
1883 | up_read(&_origins_lock); | |
1884 | ||
1885 | return r; | |
1886 | } | |
1887 | ||
1da177e4 LT |
1888 | static void snapshot_resume(struct dm_target *ti) |
1889 | { | |
028867ac | 1890 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1891 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
b735fede MP |
1892 | struct dm_origin *o; |
1893 | struct mapped_device *origin_md = NULL; | |
c1f0c183 MS |
1894 | |
1895 | down_read(&_origins_lock); | |
b735fede MP |
1896 | |
1897 | o = __lookup_dm_origin(s->origin->bdev); | |
1898 | if (o) | |
1899 | origin_md = dm_table_get_md(o->ti->table); | |
1900 | if (origin_md == dm_table_get_md(ti->table)) | |
1901 | origin_md = NULL; | |
1902 | ||
1903 | if (origin_md) | |
1904 | dm_internal_suspend_fast(origin_md); | |
1905 | ||
9d3b15c4 | 1906 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1907 | if (snap_src && snap_dest) { |
1908 | down_write(&snap_src->lock); | |
1909 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1910 | __handover_exceptions(snap_src, snap_dest); | |
1911 | up_write(&snap_dest->lock); | |
1912 | up_write(&snap_src->lock); | |
1913 | } | |
b735fede MP |
1914 | |
1915 | if (origin_md) | |
1916 | dm_internal_resume_fast(origin_md); | |
1917 | ||
c1f0c183 MS |
1918 | up_read(&_origins_lock); |
1919 | ||
1920 | /* Now we have correct chunk size, reregister */ | |
1921 | reregister_snapshot(s); | |
1da177e4 | 1922 | |
aa14edeb AK |
1923 | down_write(&s->lock); |
1924 | s->active = 1; | |
1925 | up_write(&s->lock); | |
1da177e4 LT |
1926 | } |
1927 | ||
542f9038 | 1928 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 1929 | { |
542f9038 | 1930 | uint32_t min_chunksize; |
1e03f97e MP |
1931 | |
1932 | down_read(&_origins_lock); | |
1933 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1934 | up_read(&_origins_lock); | |
1935 | ||
1936 | return min_chunksize; | |
1937 | } | |
1938 | ||
1939 | static void snapshot_merge_resume(struct dm_target *ti) | |
1940 | { | |
1941 | struct dm_snapshot *s = ti->private; | |
1942 | ||
1943 | /* | |
1944 | * Handover exceptions from existing snapshot. | |
1945 | */ | |
1946 | snapshot_resume(ti); | |
1947 | ||
1948 | /* | |
542f9038 | 1949 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 1950 | */ |
542f9038 | 1951 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
1952 | |
1953 | start_merge(s); | |
1954 | } | |
1955 | ||
fd7c092e MP |
1956 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
1957 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 1958 | { |
2e4a31df | 1959 | unsigned sz = 0; |
028867ac | 1960 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1961 | |
1962 | switch (type) { | |
1963 | case STATUSTYPE_INFO: | |
94e76572 MP |
1964 | |
1965 | down_write(&snap->lock); | |
1966 | ||
1da177e4 | 1967 | if (!snap->valid) |
2e4a31df | 1968 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1969 | else if (snap->merge_failed) |
1970 | DMEMIT("Merge failed"); | |
1da177e4 | 1971 | else { |
985903bb MS |
1972 | if (snap->store->type->usage) { |
1973 | sector_t total_sectors, sectors_allocated, | |
1974 | metadata_sectors; | |
1975 | snap->store->type->usage(snap->store, | |
1976 | &total_sectors, | |
1977 | §ors_allocated, | |
1978 | &metadata_sectors); | |
1979 | DMEMIT("%llu/%llu %llu", | |
1980 | (unsigned long long)sectors_allocated, | |
1981 | (unsigned long long)total_sectors, | |
1982 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1983 | } |
1984 | else | |
2e4a31df | 1985 | DMEMIT("Unknown"); |
1da177e4 | 1986 | } |
94e76572 MP |
1987 | |
1988 | up_write(&snap->lock); | |
1989 | ||
1da177e4 LT |
1990 | break; |
1991 | ||
1992 | case STATUSTYPE_TABLE: | |
1993 | /* | |
1994 | * kdevname returns a static pointer so we need | |
1995 | * to make private copies if the output is to | |
1996 | * make sense. | |
1997 | */ | |
fc56f6fb | 1998 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1999 | snap->store->type->status(snap->store, type, result + sz, |
2000 | maxlen - sz); | |
1da177e4 LT |
2001 | break; |
2002 | } | |
1da177e4 LT |
2003 | } |
2004 | ||
8811f46c MS |
2005 | static int snapshot_iterate_devices(struct dm_target *ti, |
2006 | iterate_devices_callout_fn fn, void *data) | |
2007 | { | |
2008 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
2009 | int r; |
2010 | ||
2011 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 2012 | |
1e5554c8 MP |
2013 | if (!r) |
2014 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
2015 | ||
2016 | return r; | |
8811f46c MS |
2017 | } |
2018 | ||
2019 | ||
1da177e4 LT |
2020 | /*----------------------------------------------------------------- |
2021 | * Origin methods | |
2022 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
2023 | |
2024 | /* | |
2025 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
2026 | * supplied bio was ignored. The caller may submit it immediately. | |
2027 | * (No remapping actually occurs as the origin is always a direct linear | |
2028 | * map.) | |
2029 | * | |
2030 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
2031 | * and any supplied bio is added to a list to be submitted once all | |
2032 | * the necessary exceptions exist. | |
2033 | */ | |
2034 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
2035 | struct bio *bio) | |
1da177e4 | 2036 | { |
515ad66c | 2037 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2038 | struct dm_snapshot *snap; |
1d4989c8 | 2039 | struct dm_exception *e; |
515ad66c MP |
2040 | struct dm_snap_pending_exception *pe; |
2041 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
2042 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
2043 | chunk_t chunk; |
2044 | ||
2045 | /* Do all the snapshots on this origin */ | |
2046 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
2047 | /* |
2048 | * Don't make new exceptions in a merging snapshot | |
2049 | * because it has effectively been deleted | |
2050 | */ | |
2051 | if (dm_target_is_snapshot_merge(snap->ti)) | |
2052 | continue; | |
2053 | ||
76df1c65 AK |
2054 | down_write(&snap->lock); |
2055 | ||
aa14edeb AK |
2056 | /* Only deal with valid and active snapshots */ |
2057 | if (!snap->valid || !snap->active) | |
76df1c65 | 2058 | goto next_snapshot; |
1da177e4 | 2059 | |
d5e404c1 | 2060 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 2061 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 2062 | goto next_snapshot; |
1da177e4 LT |
2063 | |
2064 | /* | |
2065 | * Remember, different snapshots can have | |
2066 | * different chunk sizes. | |
2067 | */ | |
9eaae8ff | 2068 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
2069 | |
2070 | /* | |
2071 | * Check exception table to see if block | |
2072 | * is already remapped in this snapshot | |
2073 | * and trigger an exception if not. | |
2074 | */ | |
3510cb94 | 2075 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
2076 | if (e) |
2077 | goto next_snapshot; | |
2078 | ||
2913808e | 2079 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2080 | if (!pe) { |
c6621392 MP |
2081 | up_write(&snap->lock); |
2082 | pe = alloc_pending_exception(snap); | |
2083 | down_write(&snap->lock); | |
2084 | ||
2085 | if (!snap->valid) { | |
2086 | free_pending_exception(pe); | |
2087 | goto next_snapshot; | |
2088 | } | |
2089 | ||
3510cb94 | 2090 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
2091 | if (e) { |
2092 | free_pending_exception(pe); | |
2093 | goto next_snapshot; | |
2094 | } | |
2095 | ||
c6621392 | 2096 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
2097 | if (!pe) { |
2098 | __invalidate_snapshot(snap, -ENOMEM); | |
2099 | goto next_snapshot; | |
2100 | } | |
76df1c65 AK |
2101 | } |
2102 | ||
515ad66c | 2103 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2104 | |
515ad66c MP |
2105 | /* |
2106 | * If an origin bio was supplied, queue it to wait for the | |
2107 | * completion of this exception, and start this one last, | |
2108 | * at the end of the function. | |
2109 | */ | |
2110 | if (bio) { | |
2111 | bio_list_add(&pe->origin_bios, bio); | |
2112 | bio = NULL; | |
76df1c65 | 2113 | |
515ad66c MP |
2114 | if (!pe->started) { |
2115 | pe->started = 1; | |
2116 | pe_to_start_last = pe; | |
2117 | } | |
76df1c65 AK |
2118 | } |
2119 | ||
2120 | if (!pe->started) { | |
2121 | pe->started = 1; | |
515ad66c | 2122 | pe_to_start_now = pe; |
1da177e4 LT |
2123 | } |
2124 | ||
a2d2b034 | 2125 | next_snapshot: |
1da177e4 | 2126 | up_write(&snap->lock); |
1da177e4 | 2127 | |
515ad66c MP |
2128 | if (pe_to_start_now) { |
2129 | start_copy(pe_to_start_now); | |
2130 | pe_to_start_now = NULL; | |
2131 | } | |
b4b610f6 AK |
2132 | } |
2133 | ||
1da177e4 | 2134 | /* |
515ad66c MP |
2135 | * Submit the exception against which the bio is queued last, |
2136 | * to give the other exceptions a head start. | |
1da177e4 | 2137 | */ |
515ad66c MP |
2138 | if (pe_to_start_last) |
2139 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2140 | |
2141 | return r; | |
2142 | } | |
2143 | ||
2144 | /* | |
2145 | * Called on a write from the origin driver. | |
2146 | */ | |
2147 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2148 | { | |
2149 | struct origin *o; | |
d2a7ad29 | 2150 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2151 | |
2152 | down_read(&_origins_lock); | |
2153 | o = __lookup_origin(origin->bdev); | |
2154 | if (o) | |
4f024f37 | 2155 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
1da177e4 LT |
2156 | up_read(&_origins_lock); |
2157 | ||
2158 | return r; | |
2159 | } | |
2160 | ||
73dfd078 MP |
2161 | /* |
2162 | * Trigger exceptions in all non-merging snapshots. | |
2163 | * | |
2164 | * The chunk size of the merging snapshot may be larger than the chunk | |
2165 | * size of some other snapshot so we may need to reallocate multiple | |
2166 | * chunks in other snapshots. | |
2167 | * | |
2168 | * We scan all the overlapping exceptions in the other snapshots. | |
2169 | * Returns 1 if anything was reallocated and must be waited for, | |
2170 | * otherwise returns 0. | |
2171 | * | |
2172 | * size must be a multiple of merging_snap's chunk_size. | |
2173 | */ | |
2174 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2175 | sector_t sector, unsigned size) | |
2176 | { | |
2177 | int must_wait = 0; | |
2178 | sector_t n; | |
2179 | struct origin *o; | |
2180 | ||
2181 | /* | |
542f9038 | 2182 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2183 | * by snapshot_merge_resume(). |
2184 | */ | |
2185 | down_read(&_origins_lock); | |
2186 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2187 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2188 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2189 | DM_MAPIO_SUBMITTED) | |
2190 | must_wait = 1; | |
2191 | up_read(&_origins_lock); | |
2192 | ||
2193 | return must_wait; | |
2194 | } | |
2195 | ||
1da177e4 LT |
2196 | /* |
2197 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2198 | */ | |
2199 | ||
2200 | /* | |
2201 | * Construct an origin mapping: <dev_path> | |
2202 | * The context for an origin is merely a 'struct dm_dev *' | |
2203 | * pointing to the real device. | |
2204 | */ | |
2205 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2206 | { | |
2207 | int r; | |
599cdf3b | 2208 | struct dm_origin *o; |
1da177e4 LT |
2209 | |
2210 | if (argc != 1) { | |
72d94861 | 2211 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2212 | return -EINVAL; |
2213 | } | |
2214 | ||
599cdf3b MP |
2215 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2216 | if (!o) { | |
2217 | ti->error = "Cannot allocate private origin structure"; | |
2218 | r = -ENOMEM; | |
2219 | goto bad_alloc; | |
2220 | } | |
2221 | ||
2222 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2223 | if (r) { |
2224 | ti->error = "Cannot get target device"; | |
599cdf3b | 2225 | goto bad_open; |
1da177e4 LT |
2226 | } |
2227 | ||
b735fede | 2228 | o->ti = ti; |
599cdf3b | 2229 | ti->private = o; |
55a62eef | 2230 | ti->num_flush_bios = 1; |
494b3ee7 | 2231 | |
1da177e4 | 2232 | return 0; |
599cdf3b MP |
2233 | |
2234 | bad_open: | |
2235 | kfree(o); | |
2236 | bad_alloc: | |
2237 | return r; | |
1da177e4 LT |
2238 | } |
2239 | ||
2240 | static void origin_dtr(struct dm_target *ti) | |
2241 | { | |
599cdf3b | 2242 | struct dm_origin *o = ti->private; |
b735fede | 2243 | |
599cdf3b MP |
2244 | dm_put_device(ti, o->dev); |
2245 | kfree(o); | |
1da177e4 LT |
2246 | } |
2247 | ||
7de3ee57 | 2248 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2249 | { |
599cdf3b | 2250 | struct dm_origin *o = ti->private; |
298eaa89 | 2251 | unsigned available_sectors; |
1da177e4 | 2252 | |
599cdf3b | 2253 | bio->bi_bdev = o->dev->bdev; |
1da177e4 | 2254 | |
298eaa89 | 2255 | if (unlikely(bio->bi_rw & REQ_FLUSH)) |
494b3ee7 MP |
2256 | return DM_MAPIO_REMAPPED; |
2257 | ||
298eaa89 | 2258 | if (bio_rw(bio) != WRITE) |
494b3ee7 MP |
2259 | return DM_MAPIO_REMAPPED; |
2260 | ||
298eaa89 MP |
2261 | available_sectors = o->split_boundary - |
2262 | ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1)); | |
2263 | ||
2264 | if (bio_sectors(bio) > available_sectors) | |
2265 | dm_accept_partial_bio(bio, available_sectors); | |
2266 | ||
1da177e4 | 2267 | /* Only tell snapshots if this is a write */ |
298eaa89 | 2268 | return do_origin(o->dev, bio); |
1da177e4 LT |
2269 | } |
2270 | ||
1da177e4 | 2271 | /* |
542f9038 | 2272 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2273 | * chunk sizes. |
2274 | */ | |
2275 | static void origin_resume(struct dm_target *ti) | |
2276 | { | |
599cdf3b | 2277 | struct dm_origin *o = ti->private; |
1da177e4 | 2278 | |
298eaa89 | 2279 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
b735fede MP |
2280 | |
2281 | down_write(&_origins_lock); | |
2282 | __insert_dm_origin(o); | |
2283 | up_write(&_origins_lock); | |
2284 | } | |
2285 | ||
2286 | static void origin_postsuspend(struct dm_target *ti) | |
2287 | { | |
2288 | struct dm_origin *o = ti->private; | |
2289 | ||
2290 | down_write(&_origins_lock); | |
2291 | __remove_dm_origin(o); | |
2292 | up_write(&_origins_lock); | |
1da177e4 LT |
2293 | } |
2294 | ||
fd7c092e MP |
2295 | static void origin_status(struct dm_target *ti, status_type_t type, |
2296 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2297 | { |
599cdf3b | 2298 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2299 | |
2300 | switch (type) { | |
2301 | case STATUSTYPE_INFO: | |
2302 | result[0] = '\0'; | |
2303 | break; | |
2304 | ||
2305 | case STATUSTYPE_TABLE: | |
599cdf3b | 2306 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 LT |
2307 | break; |
2308 | } | |
1da177e4 LT |
2309 | } |
2310 | ||
b1d55528 MP |
2311 | static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
2312 | struct bio_vec *biovec, int max_size) | |
2313 | { | |
599cdf3b MP |
2314 | struct dm_origin *o = ti->private; |
2315 | struct request_queue *q = bdev_get_queue(o->dev->bdev); | |
b1d55528 MP |
2316 | |
2317 | if (!q->merge_bvec_fn) | |
2318 | return max_size; | |
2319 | ||
599cdf3b | 2320 | bvm->bi_bdev = o->dev->bdev; |
b1d55528 MP |
2321 | |
2322 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2323 | } | |
2324 | ||
8811f46c MS |
2325 | static int origin_iterate_devices(struct dm_target *ti, |
2326 | iterate_devices_callout_fn fn, void *data) | |
2327 | { | |
599cdf3b | 2328 | struct dm_origin *o = ti->private; |
8811f46c | 2329 | |
599cdf3b | 2330 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2331 | } |
2332 | ||
1da177e4 LT |
2333 | static struct target_type origin_target = { |
2334 | .name = "snapshot-origin", | |
b735fede | 2335 | .version = {1, 9, 0}, |
1da177e4 LT |
2336 | .module = THIS_MODULE, |
2337 | .ctr = origin_ctr, | |
2338 | .dtr = origin_dtr, | |
2339 | .map = origin_map, | |
2340 | .resume = origin_resume, | |
b735fede | 2341 | .postsuspend = origin_postsuspend, |
1da177e4 | 2342 | .status = origin_status, |
b1d55528 | 2343 | .merge = origin_merge, |
8811f46c | 2344 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2345 | }; |
2346 | ||
2347 | static struct target_type snapshot_target = { | |
2348 | .name = "snapshot", | |
b735fede | 2349 | .version = {1, 13, 0}, |
1da177e4 LT |
2350 | .module = THIS_MODULE, |
2351 | .ctr = snapshot_ctr, | |
2352 | .dtr = snapshot_dtr, | |
2353 | .map = snapshot_map, | |
cd45daff | 2354 | .end_io = snapshot_end_io, |
c1f0c183 | 2355 | .preresume = snapshot_preresume, |
1da177e4 LT |
2356 | .resume = snapshot_resume, |
2357 | .status = snapshot_status, | |
8811f46c | 2358 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2359 | }; |
2360 | ||
d698aa45 MP |
2361 | static struct target_type merge_target = { |
2362 | .name = dm_snapshot_merge_target_name, | |
42bc954f | 2363 | .version = {1, 2, 0}, |
d698aa45 MP |
2364 | .module = THIS_MODULE, |
2365 | .ctr = snapshot_ctr, | |
2366 | .dtr = snapshot_dtr, | |
3452c2a1 | 2367 | .map = snapshot_merge_map, |
d698aa45 | 2368 | .end_io = snapshot_end_io, |
1e03f97e | 2369 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2370 | .preresume = snapshot_preresume, |
1e03f97e | 2371 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2372 | .status = snapshot_status, |
2373 | .iterate_devices = snapshot_iterate_devices, | |
2374 | }; | |
2375 | ||
1da177e4 LT |
2376 | static int __init dm_snapshot_init(void) |
2377 | { | |
2378 | int r; | |
2379 | ||
4db6bfe0 AK |
2380 | r = dm_exception_store_init(); |
2381 | if (r) { | |
2382 | DMERR("Failed to initialize exception stores"); | |
2383 | return r; | |
2384 | } | |
2385 | ||
1da177e4 | 2386 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2387 | if (r < 0) { |
1da177e4 | 2388 | DMERR("snapshot target register failed %d", r); |
034a186d | 2389 | goto bad_register_snapshot_target; |
1da177e4 LT |
2390 | } |
2391 | ||
2392 | r = dm_register_target(&origin_target); | |
2393 | if (r < 0) { | |
72d94861 | 2394 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2395 | goto bad_register_origin_target; |
2396 | } | |
2397 | ||
2398 | r = dm_register_target(&merge_target); | |
2399 | if (r < 0) { | |
2400 | DMERR("Merge target register failed %d", r); | |
2401 | goto bad_register_merge_target; | |
1da177e4 LT |
2402 | } |
2403 | ||
2404 | r = init_origin_hash(); | |
2405 | if (r) { | |
2406 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2407 | goto bad_origin_hash; |
1da177e4 LT |
2408 | } |
2409 | ||
1d4989c8 | 2410 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2411 | if (!exception_cache) { |
2412 | DMERR("Couldn't create exception cache."); | |
2413 | r = -ENOMEM; | |
d698aa45 | 2414 | goto bad_exception_cache; |
1da177e4 LT |
2415 | } |
2416 | ||
028867ac | 2417 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2418 | if (!pending_cache) { |
2419 | DMERR("Couldn't create pending cache."); | |
2420 | r = -ENOMEM; | |
d698aa45 | 2421 | goto bad_pending_cache; |
1da177e4 LT |
2422 | } |
2423 | ||
1da177e4 LT |
2424 | return 0; |
2425 | ||
d698aa45 | 2426 | bad_pending_cache: |
1da177e4 | 2427 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2428 | bad_exception_cache: |
1da177e4 | 2429 | exit_origin_hash(); |
d698aa45 MP |
2430 | bad_origin_hash: |
2431 | dm_unregister_target(&merge_target); | |
2432 | bad_register_merge_target: | |
1da177e4 | 2433 | dm_unregister_target(&origin_target); |
d698aa45 | 2434 | bad_register_origin_target: |
1da177e4 | 2435 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2436 | bad_register_snapshot_target: |
2437 | dm_exception_store_exit(); | |
d698aa45 | 2438 | |
1da177e4 LT |
2439 | return r; |
2440 | } | |
2441 | ||
2442 | static void __exit dm_snapshot_exit(void) | |
2443 | { | |
10d3bd09 MP |
2444 | dm_unregister_target(&snapshot_target); |
2445 | dm_unregister_target(&origin_target); | |
d698aa45 | 2446 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2447 | |
2448 | exit_origin_hash(); | |
1da177e4 LT |
2449 | kmem_cache_destroy(pending_cache); |
2450 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2451 | |
2452 | dm_exception_store_exit(); | |
1da177e4 LT |
2453 | } |
2454 | ||
2455 | /* Module hooks */ | |
2456 | module_init(dm_snapshot_init); | |
2457 | module_exit(dm_snapshot_exit); | |
2458 | ||
2459 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2460 | MODULE_AUTHOR("Joe Thornber"); | |
2461 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2462 | MODULE_ALIAS("dm-snapshot-origin"); |
2463 | MODULE_ALIAS("dm-snapshot-merge"); |