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 | ||
76c44f6d MP |
66 | /* |
67 | * The snapshot overflowed because of a write to the snapshot device. | |
68 | * We don't have to invalidate the snapshot in this case, but we need | |
69 | * to prevent further writes. | |
70 | */ | |
71 | int snapshot_overflowed; | |
72 | ||
ccc45ea8 JB |
73 | /* Origin writes don't trigger exceptions until this is set */ |
74 | int active; | |
75 | ||
ccc45ea8 JB |
76 | atomic_t pending_exceptions_count; |
77 | ||
230c83af MP |
78 | /* Protected by "lock" */ |
79 | sector_t exception_start_sequence; | |
80 | ||
81 | /* Protected by kcopyd single-threaded callback */ | |
82 | sector_t exception_complete_sequence; | |
83 | ||
84 | /* | |
85 | * A list of pending exceptions that completed out of order. | |
86 | * Protected by kcopyd single-threaded callback. | |
87 | */ | |
88 | struct list_head out_of_order_list; | |
89 | ||
924e600d MS |
90 | mempool_t *pending_pool; |
91 | ||
191437a5 JB |
92 | struct dm_exception_table pending; |
93 | struct dm_exception_table complete; | |
ccc45ea8 JB |
94 | |
95 | /* | |
96 | * pe_lock protects all pending_exception operations and access | |
97 | * as well as the snapshot_bios list. | |
98 | */ | |
99 | spinlock_t pe_lock; | |
100 | ||
924e600d MS |
101 | /* Chunks with outstanding reads */ |
102 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
103 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
104 | ||
ccc45ea8 JB |
105 | /* The on disk metadata handler */ |
106 | struct dm_exception_store *store; | |
107 | ||
108 | struct dm_kcopyd_client *kcopyd_client; | |
109 | ||
924e600d MS |
110 | /* Wait for events based on state_bits */ |
111 | unsigned long state_bits; | |
112 | ||
113 | /* Range of chunks currently being merged. */ | |
114 | chunk_t first_merging_chunk; | |
115 | int num_merging_chunks; | |
1e03f97e | 116 | |
d8ddb1cf MS |
117 | /* |
118 | * The merge operation failed if this flag is set. | |
119 | * Failure modes are handled as follows: | |
120 | * - I/O error reading the header | |
121 | * => don't load the target; abort. | |
122 | * - Header does not have "valid" flag set | |
123 | * => use the origin; forget about the snapshot. | |
124 | * - I/O error when reading exceptions | |
125 | * => don't load the target; abort. | |
126 | * (We can't use the intermediate origin state.) | |
127 | * - I/O error while merging | |
128 | * => stop merging; set merge_failed; process I/O normally. | |
129 | */ | |
130 | int merge_failed; | |
131 | ||
9fe86254 MP |
132 | /* |
133 | * Incoming bios that overlap with chunks being merged must wait | |
134 | * for them to be committed. | |
135 | */ | |
136 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
137 | }; |
138 | ||
1e03f97e MP |
139 | /* |
140 | * state_bits: | |
141 | * RUNNING_MERGE - Merge operation is in progress. | |
142 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
143 | * cleared afterwards. | |
144 | */ | |
145 | #define RUNNING_MERGE 0 | |
146 | #define SHUTDOWN_MERGE 1 | |
147 | ||
df5d2e90 MP |
148 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
149 | "A percentage of time allocated for copy on write"); | |
150 | ||
c2411045 MP |
151 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
152 | { | |
153 | return s->origin; | |
154 | } | |
155 | EXPORT_SYMBOL(dm_snap_origin); | |
156 | ||
fc56f6fb MS |
157 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
158 | { | |
159 | return s->cow; | |
160 | } | |
161 | EXPORT_SYMBOL(dm_snap_cow); | |
162 | ||
ccc45ea8 JB |
163 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
164 | chunk_t chunk) | |
165 | { | |
166 | return chunk << store->chunk_shift; | |
167 | } | |
168 | ||
169 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
170 | { | |
171 | /* | |
172 | * There is only ever one instance of a particular block | |
173 | * device so we can compare pointers safely. | |
174 | */ | |
175 | return lhs == rhs; | |
176 | } | |
177 | ||
028867ac | 178 | struct dm_snap_pending_exception { |
1d4989c8 | 179 | struct dm_exception e; |
1da177e4 LT |
180 | |
181 | /* | |
182 | * Origin buffers waiting for this to complete are held | |
183 | * in a bio list | |
184 | */ | |
185 | struct bio_list origin_bios; | |
186 | struct bio_list snapshot_bios; | |
187 | ||
1da177e4 LT |
188 | /* Pointer back to snapshot context */ |
189 | struct dm_snapshot *snap; | |
190 | ||
191 | /* | |
192 | * 1 indicates the exception has already been sent to | |
193 | * kcopyd. | |
194 | */ | |
195 | int started; | |
a6e50b40 | 196 | |
230c83af MP |
197 | /* There was copying error. */ |
198 | int copy_error; | |
199 | ||
200 | /* A sequence number, it is used for in-order completion. */ | |
201 | sector_t exception_sequence; | |
202 | ||
203 | struct list_head out_of_order_entry; | |
204 | ||
a6e50b40 MP |
205 | /* |
206 | * For writing a complete chunk, bypassing the copy. | |
207 | */ | |
208 | struct bio *full_bio; | |
209 | bio_end_io_t *full_bio_end_io; | |
1da177e4 LT |
210 | }; |
211 | ||
212 | /* | |
213 | * Hash table mapping origin volumes to lists of snapshots and | |
214 | * a lock to protect it | |
215 | */ | |
e18b890b CL |
216 | static struct kmem_cache *exception_cache; |
217 | static struct kmem_cache *pending_cache; | |
1da177e4 | 218 | |
cd45daff MP |
219 | struct dm_snap_tracked_chunk { |
220 | struct hlist_node node; | |
221 | chunk_t chunk; | |
222 | }; | |
223 | ||
ee18026a MP |
224 | static void init_tracked_chunk(struct bio *bio) |
225 | { | |
226 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
227 | INIT_HLIST_NODE(&c->node); | |
228 | } | |
229 | ||
230 | static bool is_bio_tracked(struct bio *bio) | |
231 | { | |
232 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
233 | return !hlist_unhashed(&c->node); | |
234 | } | |
235 | ||
236 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 237 | { |
42bc954f | 238 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
239 | |
240 | c->chunk = chunk; | |
241 | ||
9aa0c0e6 | 242 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
243 | hlist_add_head(&c->node, |
244 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 245 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
246 | } |
247 | ||
ee18026a | 248 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 249 | { |
ee18026a | 250 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
251 | unsigned long flags; |
252 | ||
253 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
254 | hlist_del(&c->node); | |
255 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
256 | } |
257 | ||
a8d41b59 MP |
258 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
259 | { | |
260 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
261 | int found = 0; |
262 | ||
263 | spin_lock_irq(&s->tracked_chunk_lock); | |
264 | ||
b67bfe0d | 265 | hlist_for_each_entry(c, |
a8d41b59 MP |
266 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
267 | if (c->chunk == chunk) { | |
268 | found = 1; | |
269 | break; | |
270 | } | |
271 | } | |
272 | ||
273 | spin_unlock_irq(&s->tracked_chunk_lock); | |
274 | ||
275 | return found; | |
276 | } | |
277 | ||
615d1eb9 MS |
278 | /* |
279 | * This conflicting I/O is extremely improbable in the caller, | |
280 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
281 | */ | |
282 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
283 | { | |
284 | while (__chunk_is_tracked(s, chunk)) | |
285 | msleep(1); | |
286 | } | |
287 | ||
1da177e4 LT |
288 | /* |
289 | * One of these per registered origin, held in the snapshot_origins hash | |
290 | */ | |
291 | struct origin { | |
292 | /* The origin device */ | |
293 | struct block_device *bdev; | |
294 | ||
295 | struct list_head hash_list; | |
296 | ||
297 | /* List of snapshots for this origin */ | |
298 | struct list_head snapshots; | |
299 | }; | |
300 | ||
b735fede MP |
301 | /* |
302 | * This structure is allocated for each origin target | |
303 | */ | |
304 | struct dm_origin { | |
305 | struct dm_dev *dev; | |
306 | struct dm_target *ti; | |
307 | unsigned split_boundary; | |
308 | struct list_head hash_list; | |
309 | }; | |
310 | ||
1da177e4 LT |
311 | /* |
312 | * Size of the hash table for origin volumes. If we make this | |
313 | * the size of the minors list then it should be nearly perfect | |
314 | */ | |
315 | #define ORIGIN_HASH_SIZE 256 | |
316 | #define ORIGIN_MASK 0xFF | |
317 | static struct list_head *_origins; | |
b735fede | 318 | static struct list_head *_dm_origins; |
1da177e4 LT |
319 | static struct rw_semaphore _origins_lock; |
320 | ||
73dfd078 MP |
321 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
322 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
323 | static uint64_t _pending_exceptions_done_count; | |
324 | ||
1da177e4 LT |
325 | static int init_origin_hash(void) |
326 | { | |
327 | int i; | |
328 | ||
329 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
330 | GFP_KERNEL); | |
331 | if (!_origins) { | |
b735fede | 332 | DMERR("unable to allocate memory for _origins"); |
1da177e4 LT |
333 | return -ENOMEM; |
334 | } | |
1da177e4 LT |
335 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
336 | INIT_LIST_HEAD(_origins + i); | |
b735fede MP |
337 | |
338 | _dm_origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
339 | GFP_KERNEL); | |
340 | if (!_dm_origins) { | |
341 | DMERR("unable to allocate memory for _dm_origins"); | |
342 | kfree(_origins); | |
343 | return -ENOMEM; | |
344 | } | |
345 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
346 | INIT_LIST_HEAD(_dm_origins + i); | |
347 | ||
1da177e4 LT |
348 | init_rwsem(&_origins_lock); |
349 | ||
350 | return 0; | |
351 | } | |
352 | ||
353 | static void exit_origin_hash(void) | |
354 | { | |
355 | kfree(_origins); | |
b735fede | 356 | kfree(_dm_origins); |
1da177e4 LT |
357 | } |
358 | ||
028867ac | 359 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
360 | { |
361 | return bdev->bd_dev & ORIGIN_MASK; | |
362 | } | |
363 | ||
364 | static struct origin *__lookup_origin(struct block_device *origin) | |
365 | { | |
366 | struct list_head *ol; | |
367 | struct origin *o; | |
368 | ||
369 | ol = &_origins[origin_hash(origin)]; | |
370 | list_for_each_entry (o, ol, hash_list) | |
371 | if (bdev_equal(o->bdev, origin)) | |
372 | return o; | |
373 | ||
374 | return NULL; | |
375 | } | |
376 | ||
377 | static void __insert_origin(struct origin *o) | |
378 | { | |
379 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
380 | list_add_tail(&o->hash_list, sl); | |
381 | } | |
382 | ||
b735fede MP |
383 | static struct dm_origin *__lookup_dm_origin(struct block_device *origin) |
384 | { | |
385 | struct list_head *ol; | |
386 | struct dm_origin *o; | |
387 | ||
388 | ol = &_dm_origins[origin_hash(origin)]; | |
389 | list_for_each_entry (o, ol, hash_list) | |
390 | if (bdev_equal(o->dev->bdev, origin)) | |
391 | return o; | |
392 | ||
393 | return NULL; | |
394 | } | |
395 | ||
396 | static void __insert_dm_origin(struct dm_origin *o) | |
397 | { | |
398 | struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)]; | |
399 | list_add_tail(&o->hash_list, sl); | |
400 | } | |
401 | ||
402 | static void __remove_dm_origin(struct dm_origin *o) | |
403 | { | |
404 | list_del(&o->hash_list); | |
405 | } | |
406 | ||
c1f0c183 MS |
407 | /* |
408 | * _origins_lock must be held when calling this function. | |
409 | * Returns number of snapshots registered using the supplied cow device, plus: | |
410 | * snap_src - a snapshot suitable for use as a source of exception handover | |
411 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
412 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
413 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 414 | * |
9d3b15c4 | 415 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
416 | * 0: NULL, NULL - first new snapshot |
417 | * 1: snap_src, NULL - normal snapshot | |
418 | * 2: snap_src, snap_dest - waiting for handover | |
419 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
420 | * 1: NULL, snap_dest - source got destroyed without handover | |
421 | */ | |
422 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
423 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
424 | struct dm_snapshot **snap_dest, |
425 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
426 | { |
427 | struct dm_snapshot *s; | |
428 | struct origin *o; | |
429 | int count = 0; | |
430 | int active; | |
431 | ||
432 | o = __lookup_origin(snap->origin->bdev); | |
433 | if (!o) | |
434 | goto out; | |
435 | ||
436 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
437 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
438 | *snap_merge = s; | |
c1f0c183 MS |
439 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
440 | continue; | |
441 | ||
442 | down_read(&s->lock); | |
443 | active = s->active; | |
444 | up_read(&s->lock); | |
445 | ||
446 | if (active) { | |
447 | if (snap_src) | |
448 | *snap_src = s; | |
449 | } else if (snap_dest) | |
450 | *snap_dest = s; | |
451 | ||
452 | count++; | |
453 | } | |
454 | ||
455 | out: | |
456 | return count; | |
457 | } | |
458 | ||
459 | /* | |
460 | * On success, returns 1 if this snapshot is a handover destination, | |
461 | * otherwise returns 0. | |
462 | */ | |
463 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
464 | { | |
465 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 466 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
467 | |
468 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
469 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
470 | &snap_merge) == 2) || | |
c1f0c183 MS |
471 | snap_dest) { |
472 | snap->ti->error = "Snapshot cow pairing for exception " | |
473 | "table handover failed"; | |
474 | return -EINVAL; | |
475 | } | |
476 | ||
477 | /* | |
478 | * If no snap_src was found, snap cannot become a handover | |
479 | * destination. | |
480 | */ | |
481 | if (!snap_src) | |
482 | return 0; | |
483 | ||
9d3b15c4 MP |
484 | /* |
485 | * Non-snapshot-merge handover? | |
486 | */ | |
487 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
488 | return 1; | |
489 | ||
490 | /* | |
491 | * Do not allow more than one merging snapshot. | |
492 | */ | |
493 | if (snap_merge) { | |
494 | snap->ti->error = "A snapshot is already merging."; | |
495 | return -EINVAL; | |
496 | } | |
497 | ||
1e03f97e MP |
498 | if (!snap_src->store->type->prepare_merge || |
499 | !snap_src->store->type->commit_merge) { | |
500 | snap->ti->error = "Snapshot exception store does not " | |
501 | "support snapshot-merge."; | |
502 | return -EINVAL; | |
503 | } | |
504 | ||
c1f0c183 MS |
505 | return 1; |
506 | } | |
507 | ||
508 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
509 | { | |
510 | struct dm_snapshot *l; | |
511 | ||
512 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
513 | list_for_each_entry(l, &o->snapshots, list) | |
514 | if (l->store->chunk_size < s->store->chunk_size) | |
515 | break; | |
516 | list_add_tail(&s->list, &l->list); | |
517 | } | |
518 | ||
1da177e4 LT |
519 | /* |
520 | * Make a note of the snapshot and its origin so we can look it | |
521 | * up when the origin has a write on it. | |
c1f0c183 MS |
522 | * |
523 | * Also validate snapshot exception store handovers. | |
524 | * On success, returns 1 if this registration is a handover destination, | |
525 | * otherwise returns 0. | |
1da177e4 LT |
526 | */ |
527 | static int register_snapshot(struct dm_snapshot *snap) | |
528 | { | |
c1f0c183 | 529 | struct origin *o, *new_o = NULL; |
1da177e4 | 530 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 531 | int r = 0; |
1da177e4 | 532 | |
60c856c8 MP |
533 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
534 | if (!new_o) | |
535 | return -ENOMEM; | |
536 | ||
1da177e4 | 537 | down_write(&_origins_lock); |
1da177e4 | 538 | |
c1f0c183 MS |
539 | r = __validate_exception_handover(snap); |
540 | if (r < 0) { | |
541 | kfree(new_o); | |
542 | goto out; | |
543 | } | |
544 | ||
545 | o = __lookup_origin(bdev); | |
60c856c8 MP |
546 | if (o) |
547 | kfree(new_o); | |
548 | else { | |
1da177e4 | 549 | /* New origin */ |
60c856c8 | 550 | o = new_o; |
1da177e4 LT |
551 | |
552 | /* Initialise the struct */ | |
553 | INIT_LIST_HEAD(&o->snapshots); | |
554 | o->bdev = bdev; | |
555 | ||
556 | __insert_origin(o); | |
557 | } | |
558 | ||
c1f0c183 MS |
559 | __insert_snapshot(o, snap); |
560 | ||
561 | out: | |
562 | up_write(&_origins_lock); | |
563 | ||
564 | return r; | |
565 | } | |
566 | ||
567 | /* | |
568 | * Move snapshot to correct place in list according to chunk size. | |
569 | */ | |
570 | static void reregister_snapshot(struct dm_snapshot *s) | |
571 | { | |
572 | struct block_device *bdev = s->origin->bdev; | |
573 | ||
574 | down_write(&_origins_lock); | |
575 | ||
576 | list_del(&s->list); | |
577 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
578 | |
579 | up_write(&_origins_lock); | |
1da177e4 LT |
580 | } |
581 | ||
582 | static void unregister_snapshot(struct dm_snapshot *s) | |
583 | { | |
584 | struct origin *o; | |
585 | ||
586 | down_write(&_origins_lock); | |
587 | o = __lookup_origin(s->origin->bdev); | |
588 | ||
589 | list_del(&s->list); | |
c1f0c183 | 590 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
591 | list_del(&o->hash_list); |
592 | kfree(o); | |
593 | } | |
594 | ||
595 | up_write(&_origins_lock); | |
596 | } | |
597 | ||
598 | /* | |
599 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
600 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
601 | * some consecutive chunks to be grouped together. | |
1da177e4 | 602 | */ |
3510cb94 JB |
603 | static int dm_exception_table_init(struct dm_exception_table *et, |
604 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
605 | { |
606 | unsigned int i; | |
607 | ||
d74f81f8 | 608 | et->hash_shift = hash_shift; |
1da177e4 LT |
609 | et->hash_mask = size - 1; |
610 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
611 | if (!et->table) | |
612 | return -ENOMEM; | |
613 | ||
614 | for (i = 0; i < size; i++) | |
615 | INIT_LIST_HEAD(et->table + i); | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
3510cb94 JB |
620 | static void dm_exception_table_exit(struct dm_exception_table *et, |
621 | struct kmem_cache *mem) | |
1da177e4 LT |
622 | { |
623 | struct list_head *slot; | |
1d4989c8 | 624 | struct dm_exception *ex, *next; |
1da177e4 LT |
625 | int i, size; |
626 | ||
627 | size = et->hash_mask + 1; | |
628 | for (i = 0; i < size; i++) { | |
629 | slot = et->table + i; | |
630 | ||
631 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
632 | kmem_cache_free(mem, ex); | |
633 | } | |
634 | ||
635 | vfree(et->table); | |
636 | } | |
637 | ||
191437a5 | 638 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 639 | { |
d74f81f8 | 640 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
641 | } |
642 | ||
3510cb94 | 643 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
644 | { |
645 | list_del(&e->hash_list); | |
646 | } | |
647 | ||
648 | /* | |
649 | * Return the exception data for a sector, or NULL if not | |
650 | * remapped. | |
651 | */ | |
3510cb94 JB |
652 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
653 | chunk_t chunk) | |
1da177e4 LT |
654 | { |
655 | struct list_head *slot; | |
1d4989c8 | 656 | struct dm_exception *e; |
1da177e4 LT |
657 | |
658 | slot = &et->table[exception_hash(et, chunk)]; | |
659 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
660 | if (chunk >= e->old_chunk && |
661 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
662 | return e; |
663 | ||
664 | return NULL; | |
665 | } | |
666 | ||
119bc547 | 667 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 668 | { |
1d4989c8 | 669 | struct dm_exception *e; |
1da177e4 | 670 | |
119bc547 MP |
671 | e = kmem_cache_alloc(exception_cache, gfp); |
672 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
673 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
674 | ||
675 | return e; | |
676 | } | |
677 | ||
3510cb94 | 678 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
679 | { |
680 | kmem_cache_free(exception_cache, e); | |
681 | } | |
682 | ||
92e86812 | 683 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 684 | { |
92e86812 MP |
685 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
686 | GFP_NOIO); | |
687 | ||
879129d2 | 688 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
689 | pe->snap = s; |
690 | ||
691 | return pe; | |
1da177e4 LT |
692 | } |
693 | ||
028867ac | 694 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 695 | { |
879129d2 MP |
696 | struct dm_snapshot *s = pe->snap; |
697 | ||
698 | mempool_free(pe, s->pending_pool); | |
4e857c58 | 699 | smp_mb__before_atomic(); |
879129d2 | 700 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
701 | } |
702 | ||
3510cb94 JB |
703 | static void dm_insert_exception(struct dm_exception_table *eh, |
704 | struct dm_exception *new_e) | |
d74f81f8 | 705 | { |
d74f81f8 | 706 | struct list_head *l; |
1d4989c8 | 707 | struct dm_exception *e = NULL; |
d74f81f8 MB |
708 | |
709 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
710 | ||
711 | /* Add immediately if this table doesn't support consecutive chunks */ | |
712 | if (!eh->hash_shift) | |
713 | goto out; | |
714 | ||
715 | /* List is ordered by old_chunk */ | |
716 | list_for_each_entry_reverse(e, l, hash_list) { | |
717 | /* Insert after an existing chunk? */ | |
718 | if (new_e->old_chunk == (e->old_chunk + | |
719 | dm_consecutive_chunk_count(e) + 1) && | |
720 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
721 | dm_consecutive_chunk_count(e) + 1)) { | |
722 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 723 | free_completed_exception(new_e); |
d74f81f8 MB |
724 | return; |
725 | } | |
726 | ||
727 | /* Insert before an existing chunk? */ | |
728 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
729 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
730 | dm_consecutive_chunk_count_inc(e); | |
731 | e->old_chunk--; | |
732 | e->new_chunk--; | |
3510cb94 | 733 | free_completed_exception(new_e); |
d74f81f8 MB |
734 | return; |
735 | } | |
736 | ||
737 | if (new_e->old_chunk > e->old_chunk) | |
738 | break; | |
739 | } | |
740 | ||
741 | out: | |
742 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
743 | } | |
744 | ||
a159c1ac JB |
745 | /* |
746 | * Callback used by the exception stores to load exceptions when | |
747 | * initialising. | |
748 | */ | |
749 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 750 | { |
a159c1ac | 751 | struct dm_snapshot *s = context; |
1d4989c8 | 752 | struct dm_exception *e; |
1da177e4 | 753 | |
119bc547 | 754 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
755 | if (!e) |
756 | return -ENOMEM; | |
757 | ||
758 | e->old_chunk = old; | |
d74f81f8 MB |
759 | |
760 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 761 | e->new_chunk = new; |
d74f81f8 | 762 | |
3510cb94 | 763 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 764 | |
1da177e4 LT |
765 | return 0; |
766 | } | |
767 | ||
7e201b35 MP |
768 | /* |
769 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
770 | * Return zero if the origin has no snapshots. | |
771 | */ | |
542f9038 | 772 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
773 | { |
774 | struct dm_snapshot *snap; | |
775 | unsigned chunk_size = 0; | |
776 | ||
777 | if (o) | |
778 | list_for_each_entry(snap, &o->snapshots, list) | |
779 | chunk_size = min_not_zero(chunk_size, | |
780 | snap->store->chunk_size); | |
781 | ||
542f9038 | 782 | return (uint32_t) chunk_size; |
7e201b35 MP |
783 | } |
784 | ||
1da177e4 LT |
785 | /* |
786 | * Hard coded magic. | |
787 | */ | |
788 | static int calc_max_buckets(void) | |
789 | { | |
790 | /* use a fixed size of 2MB */ | |
791 | unsigned long mem = 2 * 1024 * 1024; | |
792 | mem /= sizeof(struct list_head); | |
793 | ||
794 | return mem; | |
795 | } | |
796 | ||
1da177e4 LT |
797 | /* |
798 | * Allocate room for a suitable hash table. | |
799 | */ | |
fee1998e | 800 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 801 | { |
60e356f3 | 802 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
803 | |
804 | /* | |
805 | * Calculate based on the size of the original volume or | |
806 | * the COW volume... | |
807 | */ | |
fc56f6fb | 808 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
809 | max_buckets = calc_max_buckets(); |
810 | ||
60e356f3 | 811 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
812 | hash_size = min(hash_size, max_buckets); |
813 | ||
8e87b9b8 MP |
814 | if (hash_size < 64) |
815 | hash_size = 64; | |
8defd830 | 816 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
817 | if (dm_exception_table_init(&s->complete, hash_size, |
818 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
819 | return -ENOMEM; |
820 | ||
821 | /* | |
822 | * Allocate hash table for in-flight exceptions | |
823 | * Make this smaller than the real hash table | |
824 | */ | |
825 | hash_size >>= 3; | |
826 | if (hash_size < 64) | |
827 | hash_size = 64; | |
828 | ||
3510cb94 JB |
829 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
830 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
831 | return -ENOMEM; |
832 | } | |
833 | ||
834 | return 0; | |
835 | } | |
836 | ||
1e03f97e MP |
837 | static void merge_shutdown(struct dm_snapshot *s) |
838 | { | |
839 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 840 | smp_mb__after_atomic(); |
1e03f97e MP |
841 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
842 | } | |
843 | ||
9fe86254 MP |
844 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
845 | { | |
846 | s->first_merging_chunk = 0; | |
847 | s->num_merging_chunks = 0; | |
848 | ||
849 | return bio_list_get(&s->bios_queued_during_merge); | |
850 | } | |
851 | ||
1e03f97e MP |
852 | /* |
853 | * Remove one chunk from the index of completed exceptions. | |
854 | */ | |
855 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
856 | chunk_t old_chunk) | |
857 | { | |
858 | struct dm_exception *e; | |
859 | ||
1e03f97e MP |
860 | e = dm_lookup_exception(&s->complete, old_chunk); |
861 | if (!e) { | |
862 | DMERR("Corruption detected: exception for block %llu is " | |
863 | "on disk but not in memory", | |
864 | (unsigned long long)old_chunk); | |
865 | return -EINVAL; | |
866 | } | |
867 | ||
868 | /* | |
869 | * If this is the only chunk using this exception, remove exception. | |
870 | */ | |
871 | if (!dm_consecutive_chunk_count(e)) { | |
872 | dm_remove_exception(e); | |
873 | free_completed_exception(e); | |
874 | return 0; | |
875 | } | |
876 | ||
877 | /* | |
878 | * The chunk may be either at the beginning or the end of a | |
879 | * group of consecutive chunks - never in the middle. We are | |
880 | * removing chunks in the opposite order to that in which they | |
881 | * were added, so this should always be true. | |
882 | * Decrement the consecutive chunk counter and adjust the | |
883 | * starting point if necessary. | |
884 | */ | |
885 | if (old_chunk == e->old_chunk) { | |
886 | e->old_chunk++; | |
887 | e->new_chunk++; | |
888 | } else if (old_chunk != e->old_chunk + | |
889 | dm_consecutive_chunk_count(e)) { | |
890 | DMERR("Attempt to merge block %llu from the " | |
891 | "middle of a chunk range [%llu - %llu]", | |
892 | (unsigned long long)old_chunk, | |
893 | (unsigned long long)e->old_chunk, | |
894 | (unsigned long long) | |
895 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
896 | return -EINVAL; | |
897 | } | |
898 | ||
899 | dm_consecutive_chunk_count_dec(e); | |
900 | ||
901 | return 0; | |
902 | } | |
903 | ||
9fe86254 MP |
904 | static void flush_bios(struct bio *bio); |
905 | ||
906 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 907 | { |
9fe86254 MP |
908 | struct bio *b = NULL; |
909 | int r; | |
910 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
911 | |
912 | down_write(&s->lock); | |
9fe86254 MP |
913 | |
914 | /* | |
915 | * Process chunks (and associated exceptions) in reverse order | |
916 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
917 | */ | |
918 | do { | |
919 | r = __remove_single_exception_chunk(s, old_chunk); | |
920 | if (r) | |
921 | goto out; | |
922 | } while (old_chunk-- > s->first_merging_chunk); | |
923 | ||
924 | b = __release_queued_bios_after_merge(s); | |
925 | ||
926 | out: | |
1e03f97e | 927 | up_write(&s->lock); |
9fe86254 MP |
928 | if (b) |
929 | flush_bios(b); | |
1e03f97e MP |
930 | |
931 | return r; | |
932 | } | |
933 | ||
73dfd078 MP |
934 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
935 | sector_t sector, unsigned chunk_size); | |
936 | ||
1e03f97e MP |
937 | static void merge_callback(int read_err, unsigned long write_err, |
938 | void *context); | |
939 | ||
73dfd078 MP |
940 | static uint64_t read_pending_exceptions_done_count(void) |
941 | { | |
942 | uint64_t pending_exceptions_done; | |
943 | ||
944 | spin_lock(&_pending_exceptions_done_spinlock); | |
945 | pending_exceptions_done = _pending_exceptions_done_count; | |
946 | spin_unlock(&_pending_exceptions_done_spinlock); | |
947 | ||
948 | return pending_exceptions_done; | |
949 | } | |
950 | ||
951 | static void increment_pending_exceptions_done_count(void) | |
952 | { | |
953 | spin_lock(&_pending_exceptions_done_spinlock); | |
954 | _pending_exceptions_done_count++; | |
955 | spin_unlock(&_pending_exceptions_done_spinlock); | |
956 | ||
957 | wake_up_all(&_pending_exceptions_done); | |
958 | } | |
959 | ||
1e03f97e MP |
960 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
961 | { | |
8a2d5286 | 962 | int i, linear_chunks; |
1e03f97e MP |
963 | chunk_t old_chunk, new_chunk; |
964 | struct dm_io_region src, dest; | |
8a2d5286 | 965 | sector_t io_size; |
73dfd078 | 966 | uint64_t previous_count; |
1e03f97e MP |
967 | |
968 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
969 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
970 | goto shut; | |
971 | ||
972 | /* | |
973 | * valid flag never changes during merge, so no lock required. | |
974 | */ | |
975 | if (!s->valid) { | |
976 | DMERR("Snapshot is invalid: can't merge"); | |
977 | goto shut; | |
978 | } | |
979 | ||
8a2d5286 MS |
980 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
981 | &new_chunk); | |
982 | if (linear_chunks <= 0) { | |
d8ddb1cf | 983 | if (linear_chunks < 0) { |
1e03f97e MP |
984 | DMERR("Read error in exception store: " |
985 | "shutting down merge"); | |
d8ddb1cf MS |
986 | down_write(&s->lock); |
987 | s->merge_failed = 1; | |
988 | up_write(&s->lock); | |
989 | } | |
1e03f97e MP |
990 | goto shut; |
991 | } | |
992 | ||
8a2d5286 MS |
993 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
994 | old_chunk = old_chunk + 1 - linear_chunks; | |
995 | new_chunk = new_chunk + 1 - linear_chunks; | |
996 | ||
997 | /* | |
998 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
999 | * from the exception store to the origin | |
1000 | */ | |
1001 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 1002 | |
1e03f97e MP |
1003 | dest.bdev = s->origin->bdev; |
1004 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 1005 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
1006 | |
1007 | src.bdev = s->cow->bdev; | |
1008 | src.sector = chunk_to_sector(s->store, new_chunk); | |
1009 | src.count = dest.count; | |
1010 | ||
73dfd078 MP |
1011 | /* |
1012 | * Reallocate any exceptions needed in other snapshots then | |
1013 | * wait for the pending exceptions to complete. | |
1014 | * Each time any pending exception (globally on the system) | |
1015 | * completes we are woken and repeat the process to find out | |
1016 | * if we can proceed. While this may not seem a particularly | |
1017 | * efficient algorithm, it is not expected to have any | |
1018 | * significant impact on performance. | |
1019 | */ | |
1020 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 1021 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
1022 | wait_event(_pending_exceptions_done, |
1023 | (read_pending_exceptions_done_count() != | |
1024 | previous_count)); | |
1025 | /* Retry after the wait, until all exceptions are done. */ | |
1026 | previous_count = read_pending_exceptions_done_count(); | |
1027 | } | |
1028 | ||
9fe86254 MP |
1029 | down_write(&s->lock); |
1030 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 1031 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
1032 | up_write(&s->lock); |
1033 | ||
8a2d5286 MS |
1034 | /* Wait until writes to all 'linear_chunks' drain */ |
1035 | for (i = 0; i < linear_chunks; i++) | |
1036 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 1037 | |
1e03f97e MP |
1038 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
1039 | return; | |
1040 | ||
1041 | shut: | |
1042 | merge_shutdown(s); | |
1043 | } | |
1044 | ||
9fe86254 MP |
1045 | static void error_bios(struct bio *bio); |
1046 | ||
1e03f97e MP |
1047 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
1048 | { | |
1049 | struct dm_snapshot *s = context; | |
9fe86254 | 1050 | struct bio *b = NULL; |
1e03f97e MP |
1051 | |
1052 | if (read_err || write_err) { | |
1053 | if (read_err) | |
1054 | DMERR("Read error: shutting down merge."); | |
1055 | else | |
1056 | DMERR("Write error: shutting down merge."); | |
1057 | goto shut; | |
1058 | } | |
1059 | ||
9fe86254 MP |
1060 | if (s->store->type->commit_merge(s->store, |
1061 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1062 | DMERR("Write error in exception store: shutting down merge"); |
1063 | goto shut; | |
1064 | } | |
1065 | ||
9fe86254 MP |
1066 | if (remove_single_exception_chunk(s) < 0) |
1067 | goto shut; | |
1068 | ||
1e03f97e MP |
1069 | snapshot_merge_next_chunks(s); |
1070 | ||
1071 | return; | |
1072 | ||
1073 | shut: | |
9fe86254 | 1074 | down_write(&s->lock); |
d8ddb1cf | 1075 | s->merge_failed = 1; |
9fe86254 MP |
1076 | b = __release_queued_bios_after_merge(s); |
1077 | up_write(&s->lock); | |
1078 | error_bios(b); | |
1079 | ||
1e03f97e MP |
1080 | merge_shutdown(s); |
1081 | } | |
1082 | ||
1083 | static void start_merge(struct dm_snapshot *s) | |
1084 | { | |
1085 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1086 | snapshot_merge_next_chunks(s); | |
1087 | } | |
1088 | ||
1e03f97e MP |
1089 | /* |
1090 | * Stop the merging process and wait until it finishes. | |
1091 | */ | |
1092 | static void stop_merge(struct dm_snapshot *s) | |
1093 | { | |
1094 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1095 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1096 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1097 | } | |
1098 | ||
1da177e4 | 1099 | /* |
b0d3cc01 | 1100 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p|po|n> <chunk-size> |
1da177e4 LT |
1101 | */ |
1102 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1103 | { | |
1104 | struct dm_snapshot *s; | |
cd45daff | 1105 | int i; |
1da177e4 | 1106 | int r = -EINVAL; |
fc56f6fb | 1107 | char *origin_path, *cow_path; |
55a62eef | 1108 | unsigned args_used, num_flush_bios = 1; |
10b8106a | 1109 | fmode_t origin_mode = FMODE_READ; |
1da177e4 | 1110 | |
4c7e3bf4 | 1111 | if (argc != 4) { |
72d94861 | 1112 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1113 | r = -EINVAL; |
fc56f6fb | 1114 | goto bad; |
1da177e4 LT |
1115 | } |
1116 | ||
10b8106a | 1117 | if (dm_target_is_snapshot_merge(ti)) { |
55a62eef | 1118 | num_flush_bios = 2; |
10b8106a MS |
1119 | origin_mode = FMODE_WRITE; |
1120 | } | |
1121 | ||
fc56f6fb MS |
1122 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1123 | if (!s) { | |
a2d2b034 | 1124 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1125 | r = -ENOMEM; |
1126 | goto bad; | |
1127 | } | |
1128 | ||
c2411045 MP |
1129 | origin_path = argv[0]; |
1130 | argv++; | |
1131 | argc--; | |
1132 | ||
1133 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1134 | if (r) { | |
1135 | ti->error = "Cannot get origin device"; | |
1136 | goto bad_origin; | |
1137 | } | |
1138 | ||
fc56f6fb MS |
1139 | cow_path = argv[0]; |
1140 | argv++; | |
1141 | argc--; | |
1142 | ||
024d37e9 | 1143 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1144 | if (r) { |
1145 | ti->error = "Cannot get COW device"; | |
1146 | goto bad_cow; | |
1147 | } | |
1148 | ||
1149 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1150 | if (r) { |
1151 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1152 | r = -EINVAL; |
fc56f6fb | 1153 | goto bad_store; |
1da177e4 LT |
1154 | } |
1155 | ||
fee1998e JB |
1156 | argv += args_used; |
1157 | argc -= args_used; | |
1158 | ||
fc56f6fb | 1159 | s->ti = ti; |
1da177e4 | 1160 | s->valid = 1; |
76c44f6d | 1161 | s->snapshot_overflowed = 0; |
aa14edeb | 1162 | s->active = 0; |
879129d2 | 1163 | atomic_set(&s->pending_exceptions_count, 0); |
230c83af MP |
1164 | s->exception_start_sequence = 0; |
1165 | s->exception_complete_sequence = 0; | |
1166 | INIT_LIST_HEAD(&s->out_of_order_list); | |
1da177e4 | 1167 | init_rwsem(&s->lock); |
c1f0c183 | 1168 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1169 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1170 | s->state_bits = 0; |
d8ddb1cf | 1171 | s->merge_failed = 0; |
9fe86254 MP |
1172 | s->first_merging_chunk = 0; |
1173 | s->num_merging_chunks = 0; | |
1174 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1175 | |
1176 | /* Allocate hash table for COW data */ | |
fee1998e | 1177 | if (init_hash_tables(s)) { |
1da177e4 LT |
1178 | ti->error = "Unable to allocate hash table space"; |
1179 | r = -ENOMEM; | |
fee1998e | 1180 | goto bad_hash_tables; |
1da177e4 LT |
1181 | } |
1182 | ||
df5d2e90 | 1183 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1184 | if (IS_ERR(s->kcopyd_client)) { |
1185 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1186 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1187 | goto bad_kcopyd; |
1da177e4 LT |
1188 | } |
1189 | ||
92e86812 MP |
1190 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1191 | if (!s->pending_pool) { | |
1192 | ti->error = "Could not allocate mempool for pending exceptions"; | |
09e8b813 | 1193 | r = -ENOMEM; |
fee1998e | 1194 | goto bad_pending_pool; |
92e86812 MP |
1195 | } |
1196 | ||
cd45daff MP |
1197 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1198 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1199 | ||
1200 | spin_lock_init(&s->tracked_chunk_lock); | |
1201 | ||
c1f0c183 | 1202 | ti->private = s; |
55a62eef | 1203 | ti->num_flush_bios = num_flush_bios; |
42bc954f | 1204 | ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1205 | |
1206 | /* Add snapshot to the list of snapshots for this origin */ | |
1207 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1208 | r = register_snapshot(s); | |
1209 | if (r == -ENOMEM) { | |
1210 | ti->error = "Snapshot origin struct allocation failed"; | |
1211 | goto bad_load_and_register; | |
1212 | } else if (r < 0) { | |
1213 | /* invalid handover, register_snapshot has set ti->error */ | |
1214 | goto bad_load_and_register; | |
1215 | } | |
1216 | ||
1217 | /* | |
1218 | * Metadata must only be loaded into one table at once, so skip this | |
1219 | * if metadata will be handed over during resume. | |
1220 | * Chunk size will be set during the handover - set it to zero to | |
1221 | * ensure it's ignored. | |
1222 | */ | |
1223 | if (r > 0) { | |
1224 | s->store->chunk_size = 0; | |
1225 | return 0; | |
1226 | } | |
1227 | ||
493df71c JB |
1228 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1229 | (void *)s); | |
0764147b | 1230 | if (r < 0) { |
f9cea4f7 | 1231 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1232 | goto bad_read_metadata; |
0764147b MB |
1233 | } else if (r > 0) { |
1234 | s->valid = 0; | |
1235 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1236 | } |
aa14edeb | 1237 | |
3f2412dc MP |
1238 | if (!s->store->chunk_size) { |
1239 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1240 | goto bad_read_metadata; |
1da177e4 | 1241 | } |
542f9038 MS |
1242 | |
1243 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1244 | if (r) | |
1245 | goto bad_read_metadata; | |
1da177e4 LT |
1246 | |
1247 | return 0; | |
1248 | ||
c1f0c183 MS |
1249 | bad_read_metadata: |
1250 | unregister_snapshot(s); | |
1251 | ||
fee1998e | 1252 | bad_load_and_register: |
92e86812 MP |
1253 | mempool_destroy(s->pending_pool); |
1254 | ||
fee1998e | 1255 | bad_pending_pool: |
eb69aca5 | 1256 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1257 | |
fee1998e | 1258 | bad_kcopyd: |
3510cb94 JB |
1259 | dm_exception_table_exit(&s->pending, pending_cache); |
1260 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1261 | |
fee1998e | 1262 | bad_hash_tables: |
fc56f6fb | 1263 | dm_exception_store_destroy(s->store); |
1da177e4 | 1264 | |
fc56f6fb MS |
1265 | bad_store: |
1266 | dm_put_device(ti, s->cow); | |
fee1998e | 1267 | |
fc56f6fb | 1268 | bad_cow: |
c2411045 MP |
1269 | dm_put_device(ti, s->origin); |
1270 | ||
1271 | bad_origin: | |
fc56f6fb MS |
1272 | kfree(s); |
1273 | ||
1274 | bad: | |
1da177e4 LT |
1275 | return r; |
1276 | } | |
1277 | ||
31c93a0c MB |
1278 | static void __free_exceptions(struct dm_snapshot *s) |
1279 | { | |
eb69aca5 | 1280 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1281 | s->kcopyd_client = NULL; |
1282 | ||
3510cb94 JB |
1283 | dm_exception_table_exit(&s->pending, pending_cache); |
1284 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1285 | } |
1286 | ||
c1f0c183 MS |
1287 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1288 | struct dm_snapshot *snap_dest) | |
1289 | { | |
1290 | union { | |
1291 | struct dm_exception_table table_swap; | |
1292 | struct dm_exception_store *store_swap; | |
1293 | } u; | |
1294 | ||
1295 | /* | |
1296 | * Swap all snapshot context information between the two instances. | |
1297 | */ | |
1298 | u.table_swap = snap_dest->complete; | |
1299 | snap_dest->complete = snap_src->complete; | |
1300 | snap_src->complete = u.table_swap; | |
1301 | ||
1302 | u.store_swap = snap_dest->store; | |
1303 | snap_dest->store = snap_src->store; | |
b0d3cc01 | 1304 | snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow; |
c1f0c183 MS |
1305 | snap_src->store = u.store_swap; |
1306 | ||
1307 | snap_dest->store->snap = snap_dest; | |
1308 | snap_src->store->snap = snap_src; | |
1309 | ||
542f9038 | 1310 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 | 1311 | snap_dest->valid = snap_src->valid; |
76c44f6d | 1312 | snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed; |
c1f0c183 MS |
1313 | |
1314 | /* | |
1315 | * Set source invalid to ensure it receives no further I/O. | |
1316 | */ | |
1317 | snap_src->valid = 0; | |
1318 | } | |
1319 | ||
1da177e4 LT |
1320 | static void snapshot_dtr(struct dm_target *ti) |
1321 | { | |
cd45daff MP |
1322 | #ifdef CONFIG_DM_DEBUG |
1323 | int i; | |
1324 | #endif | |
028867ac | 1325 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1326 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1327 | |
c1f0c183 MS |
1328 | down_read(&_origins_lock); |
1329 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1330 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1331 | if (snap_src && snap_dest && (s == snap_src)) { |
1332 | down_write(&snap_dest->lock); | |
1333 | snap_dest->valid = 0; | |
1334 | up_write(&snap_dest->lock); | |
1335 | DMERR("Cancelling snapshot handover."); | |
1336 | } | |
1337 | up_read(&_origins_lock); | |
1338 | ||
1e03f97e MP |
1339 | if (dm_target_is_snapshot_merge(ti)) |
1340 | stop_merge(s); | |
1341 | ||
138728dc AK |
1342 | /* Prevent further origin writes from using this snapshot. */ |
1343 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1344 | unregister_snapshot(s); |
1345 | ||
879129d2 | 1346 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1347 | msleep(1); |
879129d2 MP |
1348 | /* |
1349 | * Ensure instructions in mempool_destroy aren't reordered | |
1350 | * before atomic_read. | |
1351 | */ | |
1352 | smp_mb(); | |
1353 | ||
cd45daff MP |
1354 | #ifdef CONFIG_DM_DEBUG |
1355 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1356 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1357 | #endif | |
1358 | ||
31c93a0c | 1359 | __free_exceptions(s); |
1da177e4 | 1360 | |
92e86812 MP |
1361 | mempool_destroy(s->pending_pool); |
1362 | ||
fee1998e | 1363 | dm_exception_store_destroy(s->store); |
138728dc | 1364 | |
fc56f6fb MS |
1365 | dm_put_device(ti, s->cow); |
1366 | ||
c2411045 MP |
1367 | dm_put_device(ti, s->origin); |
1368 | ||
1da177e4 LT |
1369 | kfree(s); |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * Flush a list of buffers. | |
1374 | */ | |
1375 | static void flush_bios(struct bio *bio) | |
1376 | { | |
1377 | struct bio *n; | |
1378 | ||
1379 | while (bio) { | |
1380 | n = bio->bi_next; | |
1381 | bio->bi_next = NULL; | |
1382 | generic_make_request(bio); | |
1383 | bio = n; | |
1384 | } | |
1385 | } | |
1386 | ||
515ad66c MP |
1387 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1388 | ||
1389 | /* | |
1390 | * Flush a list of buffers. | |
1391 | */ | |
1392 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1393 | { | |
1394 | struct bio *n; | |
1395 | int r; | |
1396 | ||
1397 | while (bio) { | |
1398 | n = bio->bi_next; | |
1399 | bio->bi_next = NULL; | |
1400 | r = do_origin(s->origin, bio); | |
1401 | if (r == DM_MAPIO_REMAPPED) | |
1402 | generic_make_request(bio); | |
1403 | bio = n; | |
1404 | } | |
1405 | } | |
1406 | ||
1da177e4 LT |
1407 | /* |
1408 | * Error a list of buffers. | |
1409 | */ | |
1410 | static void error_bios(struct bio *bio) | |
1411 | { | |
1412 | struct bio *n; | |
1413 | ||
1414 | while (bio) { | |
1415 | n = bio->bi_next; | |
1416 | bio->bi_next = NULL; | |
6712ecf8 | 1417 | bio_io_error(bio); |
1da177e4 LT |
1418 | bio = n; |
1419 | } | |
1420 | } | |
1421 | ||
695368ac | 1422 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1423 | { |
1424 | if (!s->valid) | |
1425 | return; | |
1426 | ||
1427 | if (err == -EIO) | |
1428 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1429 | else if (err == -ENOMEM) | |
1430 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1431 | ||
493df71c JB |
1432 | if (s->store->type->drop_snapshot) |
1433 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1434 | |
1435 | s->valid = 0; | |
1436 | ||
fc56f6fb | 1437 | dm_table_event(s->ti->table); |
76df1c65 AK |
1438 | } |
1439 | ||
385277bf | 1440 | static void pending_complete(void *context, int success) |
1da177e4 | 1441 | { |
385277bf | 1442 | struct dm_snap_pending_exception *pe = context; |
1d4989c8 | 1443 | struct dm_exception *e; |
1da177e4 | 1444 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1445 | struct bio *origin_bios = NULL; |
1446 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1447 | struct bio *full_bio = NULL; |
9d493fa8 | 1448 | int error = 0; |
1da177e4 | 1449 | |
76df1c65 AK |
1450 | if (!success) { |
1451 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1452 | down_write(&s->lock); |
695368ac | 1453 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1454 | error = 1; |
76df1c65 AK |
1455 | goto out; |
1456 | } | |
1457 | ||
119bc547 | 1458 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1459 | if (!e) { |
1da177e4 | 1460 | down_write(&s->lock); |
695368ac | 1461 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1462 | error = 1; |
76df1c65 AK |
1463 | goto out; |
1464 | } | |
1465 | *e = pe->e; | |
1da177e4 | 1466 | |
76df1c65 AK |
1467 | down_write(&s->lock); |
1468 | if (!s->valid) { | |
3510cb94 | 1469 | free_completed_exception(e); |
9d493fa8 | 1470 | error = 1; |
76df1c65 | 1471 | goto out; |
1da177e4 LT |
1472 | } |
1473 | ||
615d1eb9 MS |
1474 | /* Check for conflicting reads */ |
1475 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1476 | |
9d493fa8 AK |
1477 | /* |
1478 | * Add a proper exception, and remove the | |
1479 | * in-flight exception from the list. | |
1480 | */ | |
3510cb94 | 1481 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1482 | |
a2d2b034 | 1483 | out: |
3510cb94 | 1484 | dm_remove_exception(&pe->e); |
9d493fa8 | 1485 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1486 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 | 1487 | full_bio = pe->full_bio; |
fe3265b1 | 1488 | if (full_bio) |
a6e50b40 | 1489 | full_bio->bi_end_io = pe->full_bio_end_io; |
73dfd078 MP |
1490 | increment_pending_exceptions_done_count(); |
1491 | ||
9d493fa8 AK |
1492 | up_write(&s->lock); |
1493 | ||
1494 | /* Submit any pending write bios */ | |
a6e50b40 MP |
1495 | if (error) { |
1496 | if (full_bio) | |
1497 | bio_io_error(full_bio); | |
9d493fa8 | 1498 | error_bios(snapshot_bios); |
a6e50b40 MP |
1499 | } else { |
1500 | if (full_bio) | |
4246a0b6 | 1501 | bio_endio(full_bio); |
9d493fa8 | 1502 | flush_bios(snapshot_bios); |
a6e50b40 | 1503 | } |
9d493fa8 | 1504 | |
515ad66c | 1505 | retry_origin_bios(s, origin_bios); |
22aa66a3 MP |
1506 | |
1507 | free_pending_exception(pe); | |
1da177e4 LT |
1508 | } |
1509 | ||
230c83af MP |
1510 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1511 | { | |
1512 | struct dm_snapshot *s = pe->snap; | |
1513 | ||
385277bf MP |
1514 | /* Update the metadata if we are persistent */ |
1515 | s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error, | |
1516 | pending_complete, pe); | |
230c83af MP |
1517 | } |
1518 | ||
1da177e4 LT |
1519 | /* |
1520 | * Called when the copy I/O has finished. kcopyd actually runs | |
1521 | * this code so don't block. | |
1522 | */ | |
4cdc1d1f | 1523 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1524 | { |
028867ac | 1525 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1526 | struct dm_snapshot *s = pe->snap; |
1527 | ||
230c83af | 1528 | pe->copy_error = read_err || write_err; |
1da177e4 | 1529 | |
230c83af MP |
1530 | if (pe->exception_sequence == s->exception_complete_sequence) { |
1531 | s->exception_complete_sequence++; | |
1532 | complete_exception(pe); | |
1533 | ||
1534 | while (!list_empty(&s->out_of_order_list)) { | |
1535 | pe = list_entry(s->out_of_order_list.next, | |
1536 | struct dm_snap_pending_exception, out_of_order_entry); | |
1537 | if (pe->exception_sequence != s->exception_complete_sequence) | |
1538 | break; | |
1539 | s->exception_complete_sequence++; | |
1540 | list_del(&pe->out_of_order_entry); | |
1541 | complete_exception(pe); | |
1542 | } | |
1543 | } else { | |
1544 | struct list_head *lh; | |
1545 | struct dm_snap_pending_exception *pe2; | |
1546 | ||
1547 | list_for_each_prev(lh, &s->out_of_order_list) { | |
1548 | pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry); | |
1549 | if (pe2->exception_sequence < pe->exception_sequence) | |
1550 | break; | |
1551 | } | |
1552 | list_add(&pe->out_of_order_entry, lh); | |
1553 | } | |
1da177e4 LT |
1554 | } |
1555 | ||
1556 | /* | |
1557 | * Dispatches the copy operation to kcopyd. | |
1558 | */ | |
028867ac | 1559 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1560 | { |
1561 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1562 | struct dm_io_region src, dest; |
1da177e4 LT |
1563 | struct block_device *bdev = s->origin->bdev; |
1564 | sector_t dev_size; | |
1565 | ||
1566 | dev_size = get_dev_size(bdev); | |
1567 | ||
1568 | src.bdev = bdev; | |
71fab00a | 1569 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1570 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1571 | |
fc56f6fb | 1572 | dest.bdev = s->cow->bdev; |
71fab00a | 1573 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1574 | dest.count = src.count; |
1575 | ||
1576 | /* Hand over to kcopyd */ | |
a2d2b034 | 1577 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1578 | } |
1579 | ||
4246a0b6 | 1580 | static void full_bio_end_io(struct bio *bio) |
a6e50b40 MP |
1581 | { |
1582 | void *callback_data = bio->bi_private; | |
1583 | ||
4246a0b6 | 1584 | dm_kcopyd_do_callback(callback_data, 0, bio->bi_error ? 1 : 0); |
a6e50b40 MP |
1585 | } |
1586 | ||
1587 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1588 | struct bio *bio) | |
1589 | { | |
1590 | struct dm_snapshot *s = pe->snap; | |
1591 | void *callback_data; | |
1592 | ||
1593 | pe->full_bio = bio; | |
1594 | pe->full_bio_end_io = bio->bi_end_io; | |
a6e50b40 MP |
1595 | |
1596 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, | |
1597 | copy_callback, pe); | |
1598 | ||
1599 | bio->bi_end_io = full_bio_end_io; | |
1600 | bio->bi_private = callback_data; | |
1601 | ||
1602 | generic_make_request(bio); | |
1603 | } | |
1604 | ||
2913808e MP |
1605 | static struct dm_snap_pending_exception * |
1606 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1607 | { | |
3510cb94 | 1608 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1609 | |
1610 | if (!e) | |
1611 | return NULL; | |
1612 | ||
1613 | return container_of(e, struct dm_snap_pending_exception, e); | |
1614 | } | |
1615 | ||
1da177e4 LT |
1616 | /* |
1617 | * Looks to see if this snapshot already has a pending exception | |
1618 | * for this chunk, otherwise it allocates a new one and inserts | |
1619 | * it into the pending table. | |
1620 | * | |
1621 | * NOTE: a write lock must be held on snap->lock before calling | |
1622 | * this. | |
1623 | */ | |
028867ac | 1624 | static struct dm_snap_pending_exception * |
c6621392 MP |
1625 | __find_pending_exception(struct dm_snapshot *s, |
1626 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1627 | { |
c6621392 | 1628 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1629 | |
2913808e MP |
1630 | pe2 = __lookup_pending_exception(s, chunk); |
1631 | if (pe2) { | |
76df1c65 | 1632 | free_pending_exception(pe); |
2913808e | 1633 | return pe2; |
1da177e4 LT |
1634 | } |
1635 | ||
76df1c65 AK |
1636 | pe->e.old_chunk = chunk; |
1637 | bio_list_init(&pe->origin_bios); | |
1638 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1639 | pe->started = 0; |
a6e50b40 | 1640 | pe->full_bio = NULL; |
76df1c65 | 1641 | |
493df71c | 1642 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1643 | free_pending_exception(pe); |
1644 | return NULL; | |
1645 | } | |
1646 | ||
230c83af MP |
1647 | pe->exception_sequence = s->exception_start_sequence++; |
1648 | ||
3510cb94 | 1649 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1650 | |
1da177e4 LT |
1651 | return pe; |
1652 | } | |
1653 | ||
1d4989c8 | 1654 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1655 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1656 | { |
fc56f6fb | 1657 | bio->bi_bdev = s->cow->bdev; |
4f024f37 KO |
1658 | bio->bi_iter.bi_sector = |
1659 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1660 | (chunk - e->old_chunk)) + | |
1661 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1662 | } |
1663 | ||
7de3ee57 | 1664 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1665 | { |
1d4989c8 | 1666 | struct dm_exception *e; |
028867ac | 1667 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1668 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1669 | chunk_t chunk; |
028867ac | 1670 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1671 | |
ee18026a MP |
1672 | init_tracked_chunk(bio); |
1673 | ||
d87f4c14 | 1674 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1675 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1676 | return DM_MAPIO_REMAPPED; |
1677 | } | |
1678 | ||
4f024f37 | 1679 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
1da177e4 LT |
1680 | |
1681 | /* Full snapshots are not usable */ | |
76df1c65 | 1682 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1683 | if (!s->valid) |
f6a80ea8 | 1684 | return -EIO; |
1da177e4 | 1685 | |
ba40a2aa AK |
1686 | /* FIXME: should only take write lock if we need |
1687 | * to copy an exception */ | |
1688 | down_write(&s->lock); | |
1689 | ||
76c44f6d | 1690 | if (!s->valid || (unlikely(s->snapshot_overflowed) && bio_rw(bio) == WRITE)) { |
ba40a2aa AK |
1691 | r = -EIO; |
1692 | goto out_unlock; | |
1693 | } | |
1694 | ||
1695 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1696 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1697 | if (e) { |
d74f81f8 | 1698 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1699 | goto out_unlock; |
1700 | } | |
1701 | ||
1da177e4 LT |
1702 | /* |
1703 | * Write to snapshot - higher level takes care of RW/RO | |
1704 | * flags so we should only get this if we are | |
1705 | * writeable. | |
1706 | */ | |
1707 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1708 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1709 | if (!pe) { |
c6621392 MP |
1710 | up_write(&s->lock); |
1711 | pe = alloc_pending_exception(s); | |
1712 | down_write(&s->lock); | |
1713 | ||
76c44f6d | 1714 | if (!s->valid || s->snapshot_overflowed) { |
c6621392 MP |
1715 | free_pending_exception(pe); |
1716 | r = -EIO; | |
1717 | goto out_unlock; | |
1718 | } | |
1719 | ||
3510cb94 | 1720 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1721 | if (e) { |
1722 | free_pending_exception(pe); | |
1723 | remap_exception(s, e, bio, chunk); | |
1724 | goto out_unlock; | |
1725 | } | |
1726 | ||
c6621392 | 1727 | pe = __find_pending_exception(s, pe, chunk); |
2913808e | 1728 | if (!pe) { |
b0d3cc01 MS |
1729 | if (s->store->userspace_supports_overflow) { |
1730 | s->snapshot_overflowed = 1; | |
1731 | DMERR("Snapshot overflowed: Unable to allocate exception."); | |
1732 | } else | |
1733 | __invalidate_snapshot(s, -ENOMEM); | |
2913808e MP |
1734 | r = -EIO; |
1735 | goto out_unlock; | |
1736 | } | |
1da177e4 LT |
1737 | } |
1738 | ||
d74f81f8 | 1739 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 1740 | |
d2a7ad29 | 1741 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1742 | |
a6e50b40 | 1743 | if (!pe->started && |
4f024f37 KO |
1744 | bio->bi_iter.bi_size == |
1745 | (s->store->chunk_size << SECTOR_SHIFT)) { | |
a6e50b40 MP |
1746 | pe->started = 1; |
1747 | up_write(&s->lock); | |
1748 | start_full_bio(pe, bio); | |
1749 | goto out; | |
1750 | } | |
1751 | ||
1752 | bio_list_add(&pe->snapshot_bios, bio); | |
1753 | ||
76df1c65 AK |
1754 | if (!pe->started) { |
1755 | /* this is protected by snap->lock */ | |
1756 | pe->started = 1; | |
ba40a2aa | 1757 | up_write(&s->lock); |
76df1c65 | 1758 | start_copy(pe); |
ba40a2aa AK |
1759 | goto out; |
1760 | } | |
cd45daff | 1761 | } else { |
ba40a2aa | 1762 | bio->bi_bdev = s->origin->bdev; |
ee18026a | 1763 | track_chunk(s, bio, chunk); |
cd45daff | 1764 | } |
1da177e4 | 1765 | |
a2d2b034 | 1766 | out_unlock: |
ba40a2aa | 1767 | up_write(&s->lock); |
a2d2b034 | 1768 | out: |
1da177e4 LT |
1769 | return r; |
1770 | } | |
1771 | ||
3452c2a1 MP |
1772 | /* |
1773 | * A snapshot-merge target behaves like a combination of a snapshot | |
1774 | * target and a snapshot-origin target. It only generates new | |
1775 | * exceptions in other snapshots and not in the one that is being | |
1776 | * merged. | |
1777 | * | |
1778 | * For each chunk, if there is an existing exception, it is used to | |
1779 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1780 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1781 | * If merging is currently taking place on the chunk in question, the |
1782 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 1783 | */ |
7de3ee57 | 1784 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
1785 | { |
1786 | struct dm_exception *e; | |
1787 | struct dm_snapshot *s = ti->private; | |
1788 | int r = DM_MAPIO_REMAPPED; | |
1789 | chunk_t chunk; | |
1790 | ||
ee18026a MP |
1791 | init_tracked_chunk(bio); |
1792 | ||
d87f4c14 | 1793 | if (bio->bi_rw & REQ_FLUSH) { |
55a62eef | 1794 | if (!dm_bio_get_target_bio_nr(bio)) |
10b8106a MS |
1795 | bio->bi_bdev = s->origin->bdev; |
1796 | else | |
1797 | bio->bi_bdev = s->cow->bdev; | |
10b8106a MS |
1798 | return DM_MAPIO_REMAPPED; |
1799 | } | |
1800 | ||
4f024f37 | 1801 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 1802 | |
9fe86254 | 1803 | down_write(&s->lock); |
3452c2a1 | 1804 | |
d2fdb776 MP |
1805 | /* Full merging snapshots are redirected to the origin */ |
1806 | if (!s->valid) | |
1807 | goto redirect_to_origin; | |
3452c2a1 MP |
1808 | |
1809 | /* If the block is already remapped - use that */ | |
1810 | e = dm_lookup_exception(&s->complete, chunk); | |
1811 | if (e) { | |
9fe86254 MP |
1812 | /* Queue writes overlapping with chunks being merged */ |
1813 | if (bio_rw(bio) == WRITE && | |
1814 | chunk >= s->first_merging_chunk && | |
1815 | chunk < (s->first_merging_chunk + | |
1816 | s->num_merging_chunks)) { | |
1817 | bio->bi_bdev = s->origin->bdev; | |
1818 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1819 | r = DM_MAPIO_SUBMITTED; | |
1820 | goto out_unlock; | |
1821 | } | |
17aa0332 | 1822 | |
3452c2a1 | 1823 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1824 | |
1825 | if (bio_rw(bio) == WRITE) | |
ee18026a | 1826 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
1827 | goto out_unlock; |
1828 | } | |
1829 | ||
d2fdb776 | 1830 | redirect_to_origin: |
3452c2a1 MP |
1831 | bio->bi_bdev = s->origin->bdev; |
1832 | ||
1833 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1834 | up_write(&s->lock); |
3452c2a1 MP |
1835 | return do_origin(s->origin, bio); |
1836 | } | |
1837 | ||
1838 | out_unlock: | |
9fe86254 | 1839 | up_write(&s->lock); |
3452c2a1 MP |
1840 | |
1841 | return r; | |
1842 | } | |
1843 | ||
7de3ee57 | 1844 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error) |
cd45daff MP |
1845 | { |
1846 | struct dm_snapshot *s = ti->private; | |
cd45daff | 1847 | |
ee18026a MP |
1848 | if (is_bio_tracked(bio)) |
1849 | stop_tracking_chunk(s, bio); | |
cd45daff MP |
1850 | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1e03f97e MP |
1854 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1855 | { | |
1856 | struct dm_snapshot *s = ti->private; | |
1857 | ||
1858 | stop_merge(s); | |
1859 | } | |
1860 | ||
c1f0c183 MS |
1861 | static int snapshot_preresume(struct dm_target *ti) |
1862 | { | |
1863 | int r = 0; | |
1864 | struct dm_snapshot *s = ti->private; | |
1865 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1866 | ||
1867 | down_read(&_origins_lock); | |
9d3b15c4 | 1868 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1869 | if (snap_src && snap_dest) { |
1870 | down_read(&snap_src->lock); | |
1871 | if (s == snap_src) { | |
1872 | DMERR("Unable to resume snapshot source until " | |
1873 | "handover completes."); | |
1874 | r = -EINVAL; | |
b83b2f29 | 1875 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
1876 | DMERR("Unable to perform snapshot handover until " |
1877 | "source is suspended."); | |
1878 | r = -EINVAL; | |
1879 | } | |
1880 | up_read(&snap_src->lock); | |
1881 | } | |
1882 | up_read(&_origins_lock); | |
1883 | ||
1884 | return r; | |
1885 | } | |
1886 | ||
1da177e4 LT |
1887 | static void snapshot_resume(struct dm_target *ti) |
1888 | { | |
028867ac | 1889 | struct dm_snapshot *s = ti->private; |
09ee96b2 | 1890 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL; |
b735fede MP |
1891 | struct dm_origin *o; |
1892 | struct mapped_device *origin_md = NULL; | |
09ee96b2 | 1893 | bool must_restart_merging = false; |
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); | |
09ee96b2 MP |
1900 | if (!origin_md) { |
1901 | (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging); | |
1902 | if (snap_merging) | |
1903 | origin_md = dm_table_get_md(snap_merging->ti->table); | |
1904 | } | |
b735fede MP |
1905 | if (origin_md == dm_table_get_md(ti->table)) |
1906 | origin_md = NULL; | |
09ee96b2 MP |
1907 | if (origin_md) { |
1908 | if (dm_hold(origin_md)) | |
1909 | origin_md = NULL; | |
1910 | } | |
b735fede | 1911 | |
09ee96b2 MP |
1912 | up_read(&_origins_lock); |
1913 | ||
1914 | if (origin_md) { | |
b735fede | 1915 | dm_internal_suspend_fast(origin_md); |
09ee96b2 MP |
1916 | if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) { |
1917 | must_restart_merging = true; | |
1918 | stop_merge(snap_merging); | |
1919 | } | |
1920 | } | |
1921 | ||
1922 | down_read(&_origins_lock); | |
b735fede | 1923 | |
9d3b15c4 | 1924 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1925 | if (snap_src && snap_dest) { |
1926 | down_write(&snap_src->lock); | |
1927 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1928 | __handover_exceptions(snap_src, snap_dest); | |
1929 | up_write(&snap_dest->lock); | |
1930 | up_write(&snap_src->lock); | |
1931 | } | |
b735fede | 1932 | |
c1f0c183 MS |
1933 | up_read(&_origins_lock); |
1934 | ||
09ee96b2 MP |
1935 | if (origin_md) { |
1936 | if (must_restart_merging) | |
1937 | start_merge(snap_merging); | |
1938 | dm_internal_resume_fast(origin_md); | |
1939 | dm_put(origin_md); | |
1940 | } | |
1941 | ||
c1f0c183 MS |
1942 | /* Now we have correct chunk size, reregister */ |
1943 | reregister_snapshot(s); | |
1da177e4 | 1944 | |
aa14edeb AK |
1945 | down_write(&s->lock); |
1946 | s->active = 1; | |
1947 | up_write(&s->lock); | |
1da177e4 LT |
1948 | } |
1949 | ||
542f9038 | 1950 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 1951 | { |
542f9038 | 1952 | uint32_t min_chunksize; |
1e03f97e MP |
1953 | |
1954 | down_read(&_origins_lock); | |
1955 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1956 | up_read(&_origins_lock); | |
1957 | ||
1958 | return min_chunksize; | |
1959 | } | |
1960 | ||
1961 | static void snapshot_merge_resume(struct dm_target *ti) | |
1962 | { | |
1963 | struct dm_snapshot *s = ti->private; | |
1964 | ||
1965 | /* | |
1966 | * Handover exceptions from existing snapshot. | |
1967 | */ | |
1968 | snapshot_resume(ti); | |
1969 | ||
1970 | /* | |
542f9038 | 1971 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 1972 | */ |
542f9038 | 1973 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
1974 | |
1975 | start_merge(s); | |
1976 | } | |
1977 | ||
fd7c092e MP |
1978 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
1979 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 1980 | { |
2e4a31df | 1981 | unsigned sz = 0; |
028867ac | 1982 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1983 | |
1984 | switch (type) { | |
1985 | case STATUSTYPE_INFO: | |
94e76572 MP |
1986 | |
1987 | down_write(&snap->lock); | |
1988 | ||
1da177e4 | 1989 | if (!snap->valid) |
2e4a31df | 1990 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1991 | else if (snap->merge_failed) |
1992 | DMEMIT("Merge failed"); | |
76c44f6d MP |
1993 | else if (snap->snapshot_overflowed) |
1994 | DMEMIT("Overflow"); | |
1da177e4 | 1995 | else { |
985903bb MS |
1996 | if (snap->store->type->usage) { |
1997 | sector_t total_sectors, sectors_allocated, | |
1998 | metadata_sectors; | |
1999 | snap->store->type->usage(snap->store, | |
2000 | &total_sectors, | |
2001 | §ors_allocated, | |
2002 | &metadata_sectors); | |
2003 | DMEMIT("%llu/%llu %llu", | |
2004 | (unsigned long long)sectors_allocated, | |
2005 | (unsigned long long)total_sectors, | |
2006 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
2007 | } |
2008 | else | |
2e4a31df | 2009 | DMEMIT("Unknown"); |
1da177e4 | 2010 | } |
94e76572 MP |
2011 | |
2012 | up_write(&snap->lock); | |
2013 | ||
1da177e4 LT |
2014 | break; |
2015 | ||
2016 | case STATUSTYPE_TABLE: | |
2017 | /* | |
2018 | * kdevname returns a static pointer so we need | |
2019 | * to make private copies if the output is to | |
2020 | * make sense. | |
2021 | */ | |
fc56f6fb | 2022 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
2023 | snap->store->type->status(snap->store, type, result + sz, |
2024 | maxlen - sz); | |
1da177e4 LT |
2025 | break; |
2026 | } | |
1da177e4 LT |
2027 | } |
2028 | ||
8811f46c MS |
2029 | static int snapshot_iterate_devices(struct dm_target *ti, |
2030 | iterate_devices_callout_fn fn, void *data) | |
2031 | { | |
2032 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
2033 | int r; |
2034 | ||
2035 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 2036 | |
1e5554c8 MP |
2037 | if (!r) |
2038 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
2039 | ||
2040 | return r; | |
8811f46c MS |
2041 | } |
2042 | ||
2043 | ||
1da177e4 LT |
2044 | /*----------------------------------------------------------------- |
2045 | * Origin methods | |
2046 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
2047 | |
2048 | /* | |
2049 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
2050 | * supplied bio was ignored. The caller may submit it immediately. | |
2051 | * (No remapping actually occurs as the origin is always a direct linear | |
2052 | * map.) | |
2053 | * | |
2054 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
2055 | * and any supplied bio is added to a list to be submitted once all | |
2056 | * the necessary exceptions exist. | |
2057 | */ | |
2058 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
2059 | struct bio *bio) | |
1da177e4 | 2060 | { |
515ad66c | 2061 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2062 | struct dm_snapshot *snap; |
1d4989c8 | 2063 | struct dm_exception *e; |
515ad66c MP |
2064 | struct dm_snap_pending_exception *pe; |
2065 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
2066 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
2067 | chunk_t chunk; |
2068 | ||
2069 | /* Do all the snapshots on this origin */ | |
2070 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
2071 | /* |
2072 | * Don't make new exceptions in a merging snapshot | |
2073 | * because it has effectively been deleted | |
2074 | */ | |
2075 | if (dm_target_is_snapshot_merge(snap->ti)) | |
2076 | continue; | |
2077 | ||
76df1c65 AK |
2078 | down_write(&snap->lock); |
2079 | ||
aa14edeb AK |
2080 | /* Only deal with valid and active snapshots */ |
2081 | if (!snap->valid || !snap->active) | |
76df1c65 | 2082 | goto next_snapshot; |
1da177e4 | 2083 | |
d5e404c1 | 2084 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 2085 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 2086 | goto next_snapshot; |
1da177e4 LT |
2087 | |
2088 | /* | |
2089 | * Remember, different snapshots can have | |
2090 | * different chunk sizes. | |
2091 | */ | |
9eaae8ff | 2092 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
2093 | |
2094 | /* | |
2095 | * Check exception table to see if block | |
2096 | * is already remapped in this snapshot | |
2097 | * and trigger an exception if not. | |
2098 | */ | |
3510cb94 | 2099 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
2100 | if (e) |
2101 | goto next_snapshot; | |
2102 | ||
2913808e | 2103 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2104 | if (!pe) { |
c6621392 MP |
2105 | up_write(&snap->lock); |
2106 | pe = alloc_pending_exception(snap); | |
2107 | down_write(&snap->lock); | |
2108 | ||
2109 | if (!snap->valid) { | |
2110 | free_pending_exception(pe); | |
2111 | goto next_snapshot; | |
2112 | } | |
2113 | ||
3510cb94 | 2114 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
2115 | if (e) { |
2116 | free_pending_exception(pe); | |
2117 | goto next_snapshot; | |
2118 | } | |
2119 | ||
c6621392 | 2120 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
2121 | if (!pe) { |
2122 | __invalidate_snapshot(snap, -ENOMEM); | |
2123 | goto next_snapshot; | |
2124 | } | |
76df1c65 AK |
2125 | } |
2126 | ||
515ad66c | 2127 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2128 | |
515ad66c MP |
2129 | /* |
2130 | * If an origin bio was supplied, queue it to wait for the | |
2131 | * completion of this exception, and start this one last, | |
2132 | * at the end of the function. | |
2133 | */ | |
2134 | if (bio) { | |
2135 | bio_list_add(&pe->origin_bios, bio); | |
2136 | bio = NULL; | |
76df1c65 | 2137 | |
515ad66c MP |
2138 | if (!pe->started) { |
2139 | pe->started = 1; | |
2140 | pe_to_start_last = pe; | |
2141 | } | |
76df1c65 AK |
2142 | } |
2143 | ||
2144 | if (!pe->started) { | |
2145 | pe->started = 1; | |
515ad66c | 2146 | pe_to_start_now = pe; |
1da177e4 LT |
2147 | } |
2148 | ||
a2d2b034 | 2149 | next_snapshot: |
1da177e4 | 2150 | up_write(&snap->lock); |
1da177e4 | 2151 | |
515ad66c MP |
2152 | if (pe_to_start_now) { |
2153 | start_copy(pe_to_start_now); | |
2154 | pe_to_start_now = NULL; | |
2155 | } | |
b4b610f6 AK |
2156 | } |
2157 | ||
1da177e4 | 2158 | /* |
515ad66c MP |
2159 | * Submit the exception against which the bio is queued last, |
2160 | * to give the other exceptions a head start. | |
1da177e4 | 2161 | */ |
515ad66c MP |
2162 | if (pe_to_start_last) |
2163 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2164 | |
2165 | return r; | |
2166 | } | |
2167 | ||
2168 | /* | |
2169 | * Called on a write from the origin driver. | |
2170 | */ | |
2171 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2172 | { | |
2173 | struct origin *o; | |
d2a7ad29 | 2174 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2175 | |
2176 | down_read(&_origins_lock); | |
2177 | o = __lookup_origin(origin->bdev); | |
2178 | if (o) | |
4f024f37 | 2179 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
1da177e4 LT |
2180 | up_read(&_origins_lock); |
2181 | ||
2182 | return r; | |
2183 | } | |
2184 | ||
73dfd078 MP |
2185 | /* |
2186 | * Trigger exceptions in all non-merging snapshots. | |
2187 | * | |
2188 | * The chunk size of the merging snapshot may be larger than the chunk | |
2189 | * size of some other snapshot so we may need to reallocate multiple | |
2190 | * chunks in other snapshots. | |
2191 | * | |
2192 | * We scan all the overlapping exceptions in the other snapshots. | |
2193 | * Returns 1 if anything was reallocated and must be waited for, | |
2194 | * otherwise returns 0. | |
2195 | * | |
2196 | * size must be a multiple of merging_snap's chunk_size. | |
2197 | */ | |
2198 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2199 | sector_t sector, unsigned size) | |
2200 | { | |
2201 | int must_wait = 0; | |
2202 | sector_t n; | |
2203 | struct origin *o; | |
2204 | ||
2205 | /* | |
542f9038 | 2206 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2207 | * by snapshot_merge_resume(). |
2208 | */ | |
2209 | down_read(&_origins_lock); | |
2210 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2211 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2212 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2213 | DM_MAPIO_SUBMITTED) | |
2214 | must_wait = 1; | |
2215 | up_read(&_origins_lock); | |
2216 | ||
2217 | return must_wait; | |
2218 | } | |
2219 | ||
1da177e4 LT |
2220 | /* |
2221 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2222 | */ | |
2223 | ||
2224 | /* | |
2225 | * Construct an origin mapping: <dev_path> | |
2226 | * The context for an origin is merely a 'struct dm_dev *' | |
2227 | * pointing to the real device. | |
2228 | */ | |
2229 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2230 | { | |
2231 | int r; | |
599cdf3b | 2232 | struct dm_origin *o; |
1da177e4 LT |
2233 | |
2234 | if (argc != 1) { | |
72d94861 | 2235 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2236 | return -EINVAL; |
2237 | } | |
2238 | ||
599cdf3b MP |
2239 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2240 | if (!o) { | |
2241 | ti->error = "Cannot allocate private origin structure"; | |
2242 | r = -ENOMEM; | |
2243 | goto bad_alloc; | |
2244 | } | |
2245 | ||
2246 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2247 | if (r) { |
2248 | ti->error = "Cannot get target device"; | |
599cdf3b | 2249 | goto bad_open; |
1da177e4 LT |
2250 | } |
2251 | ||
b735fede | 2252 | o->ti = ti; |
599cdf3b | 2253 | ti->private = o; |
55a62eef | 2254 | ti->num_flush_bios = 1; |
494b3ee7 | 2255 | |
1da177e4 | 2256 | return 0; |
599cdf3b MP |
2257 | |
2258 | bad_open: | |
2259 | kfree(o); | |
2260 | bad_alloc: | |
2261 | return r; | |
1da177e4 LT |
2262 | } |
2263 | ||
2264 | static void origin_dtr(struct dm_target *ti) | |
2265 | { | |
599cdf3b | 2266 | struct dm_origin *o = ti->private; |
b735fede | 2267 | |
599cdf3b MP |
2268 | dm_put_device(ti, o->dev); |
2269 | kfree(o); | |
1da177e4 LT |
2270 | } |
2271 | ||
7de3ee57 | 2272 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2273 | { |
599cdf3b | 2274 | struct dm_origin *o = ti->private; |
298eaa89 | 2275 | unsigned available_sectors; |
1da177e4 | 2276 | |
599cdf3b | 2277 | bio->bi_bdev = o->dev->bdev; |
1da177e4 | 2278 | |
298eaa89 | 2279 | if (unlikely(bio->bi_rw & REQ_FLUSH)) |
494b3ee7 MP |
2280 | return DM_MAPIO_REMAPPED; |
2281 | ||
298eaa89 | 2282 | if (bio_rw(bio) != WRITE) |
494b3ee7 MP |
2283 | return DM_MAPIO_REMAPPED; |
2284 | ||
298eaa89 MP |
2285 | available_sectors = o->split_boundary - |
2286 | ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1)); | |
2287 | ||
2288 | if (bio_sectors(bio) > available_sectors) | |
2289 | dm_accept_partial_bio(bio, available_sectors); | |
2290 | ||
1da177e4 | 2291 | /* Only tell snapshots if this is a write */ |
298eaa89 | 2292 | return do_origin(o->dev, bio); |
1da177e4 LT |
2293 | } |
2294 | ||
1da177e4 | 2295 | /* |
542f9038 | 2296 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2297 | * chunk sizes. |
2298 | */ | |
2299 | static void origin_resume(struct dm_target *ti) | |
2300 | { | |
599cdf3b | 2301 | struct dm_origin *o = ti->private; |
1da177e4 | 2302 | |
298eaa89 | 2303 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
b735fede MP |
2304 | |
2305 | down_write(&_origins_lock); | |
2306 | __insert_dm_origin(o); | |
2307 | up_write(&_origins_lock); | |
2308 | } | |
2309 | ||
2310 | static void origin_postsuspend(struct dm_target *ti) | |
2311 | { | |
2312 | struct dm_origin *o = ti->private; | |
2313 | ||
2314 | down_write(&_origins_lock); | |
2315 | __remove_dm_origin(o); | |
2316 | up_write(&_origins_lock); | |
1da177e4 LT |
2317 | } |
2318 | ||
fd7c092e MP |
2319 | static void origin_status(struct dm_target *ti, status_type_t type, |
2320 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2321 | { |
599cdf3b | 2322 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2323 | |
2324 | switch (type) { | |
2325 | case STATUSTYPE_INFO: | |
2326 | result[0] = '\0'; | |
2327 | break; | |
2328 | ||
2329 | case STATUSTYPE_TABLE: | |
599cdf3b | 2330 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 LT |
2331 | break; |
2332 | } | |
1da177e4 LT |
2333 | } |
2334 | ||
8811f46c MS |
2335 | static int origin_iterate_devices(struct dm_target *ti, |
2336 | iterate_devices_callout_fn fn, void *data) | |
2337 | { | |
599cdf3b | 2338 | struct dm_origin *o = ti->private; |
8811f46c | 2339 | |
599cdf3b | 2340 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2341 | } |
2342 | ||
1da177e4 LT |
2343 | static struct target_type origin_target = { |
2344 | .name = "snapshot-origin", | |
b735fede | 2345 | .version = {1, 9, 0}, |
1da177e4 LT |
2346 | .module = THIS_MODULE, |
2347 | .ctr = origin_ctr, | |
2348 | .dtr = origin_dtr, | |
2349 | .map = origin_map, | |
2350 | .resume = origin_resume, | |
b735fede | 2351 | .postsuspend = origin_postsuspend, |
1da177e4 | 2352 | .status = origin_status, |
8811f46c | 2353 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2354 | }; |
2355 | ||
2356 | static struct target_type snapshot_target = { | |
2357 | .name = "snapshot", | |
b0d3cc01 | 2358 | .version = {1, 15, 0}, |
1da177e4 LT |
2359 | .module = THIS_MODULE, |
2360 | .ctr = snapshot_ctr, | |
2361 | .dtr = snapshot_dtr, | |
2362 | .map = snapshot_map, | |
cd45daff | 2363 | .end_io = snapshot_end_io, |
c1f0c183 | 2364 | .preresume = snapshot_preresume, |
1da177e4 LT |
2365 | .resume = snapshot_resume, |
2366 | .status = snapshot_status, | |
8811f46c | 2367 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2368 | }; |
2369 | ||
d698aa45 MP |
2370 | static struct target_type merge_target = { |
2371 | .name = dm_snapshot_merge_target_name, | |
b0d3cc01 | 2372 | .version = {1, 4, 0}, |
d698aa45 MP |
2373 | .module = THIS_MODULE, |
2374 | .ctr = snapshot_ctr, | |
2375 | .dtr = snapshot_dtr, | |
3452c2a1 | 2376 | .map = snapshot_merge_map, |
d698aa45 | 2377 | .end_io = snapshot_end_io, |
1e03f97e | 2378 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2379 | .preresume = snapshot_preresume, |
1e03f97e | 2380 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2381 | .status = snapshot_status, |
2382 | .iterate_devices = snapshot_iterate_devices, | |
2383 | }; | |
2384 | ||
1da177e4 LT |
2385 | static int __init dm_snapshot_init(void) |
2386 | { | |
2387 | int r; | |
2388 | ||
4db6bfe0 AK |
2389 | r = dm_exception_store_init(); |
2390 | if (r) { | |
2391 | DMERR("Failed to initialize exception stores"); | |
2392 | return r; | |
2393 | } | |
2394 | ||
1da177e4 | 2395 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2396 | if (r < 0) { |
1da177e4 | 2397 | DMERR("snapshot target register failed %d", r); |
034a186d | 2398 | goto bad_register_snapshot_target; |
1da177e4 LT |
2399 | } |
2400 | ||
2401 | r = dm_register_target(&origin_target); | |
2402 | if (r < 0) { | |
72d94861 | 2403 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2404 | goto bad_register_origin_target; |
2405 | } | |
2406 | ||
2407 | r = dm_register_target(&merge_target); | |
2408 | if (r < 0) { | |
2409 | DMERR("Merge target register failed %d", r); | |
2410 | goto bad_register_merge_target; | |
1da177e4 LT |
2411 | } |
2412 | ||
2413 | r = init_origin_hash(); | |
2414 | if (r) { | |
2415 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2416 | goto bad_origin_hash; |
1da177e4 LT |
2417 | } |
2418 | ||
1d4989c8 | 2419 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2420 | if (!exception_cache) { |
2421 | DMERR("Couldn't create exception cache."); | |
2422 | r = -ENOMEM; | |
d698aa45 | 2423 | goto bad_exception_cache; |
1da177e4 LT |
2424 | } |
2425 | ||
028867ac | 2426 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2427 | if (!pending_cache) { |
2428 | DMERR("Couldn't create pending cache."); | |
2429 | r = -ENOMEM; | |
d698aa45 | 2430 | goto bad_pending_cache; |
1da177e4 LT |
2431 | } |
2432 | ||
1da177e4 LT |
2433 | return 0; |
2434 | ||
d698aa45 | 2435 | bad_pending_cache: |
1da177e4 | 2436 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2437 | bad_exception_cache: |
1da177e4 | 2438 | exit_origin_hash(); |
d698aa45 MP |
2439 | bad_origin_hash: |
2440 | dm_unregister_target(&merge_target); | |
2441 | bad_register_merge_target: | |
1da177e4 | 2442 | dm_unregister_target(&origin_target); |
d698aa45 | 2443 | bad_register_origin_target: |
1da177e4 | 2444 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2445 | bad_register_snapshot_target: |
2446 | dm_exception_store_exit(); | |
d698aa45 | 2447 | |
1da177e4 LT |
2448 | return r; |
2449 | } | |
2450 | ||
2451 | static void __exit dm_snapshot_exit(void) | |
2452 | { | |
10d3bd09 MP |
2453 | dm_unregister_target(&snapshot_target); |
2454 | dm_unregister_target(&origin_target); | |
d698aa45 | 2455 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2456 | |
2457 | exit_origin_hash(); | |
1da177e4 LT |
2458 | kmem_cache_destroy(pending_cache); |
2459 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2460 | |
2461 | dm_exception_store_exit(); | |
1da177e4 LT |
2462 | } |
2463 | ||
2464 | /* Module hooks */ | |
2465 | module_init(dm_snapshot_init); | |
2466 | module_exit(dm_snapshot_exit); | |
2467 | ||
2468 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2469 | MODULE_AUTHOR("Joe Thornber"); | |
2470 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2471 | MODULE_ALIAS("dm-snapshot-origin"); |
2472 | MODULE_ALIAS("dm-snapshot-merge"); |