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 | |
aea53d92 | 23 | #include "dm-exception-store.h" |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "snapshots" |
26 | ||
d698aa45 MP |
27 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
28 | ||
29 | #define dm_target_is_snapshot_merge(ti) \ | |
30 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
31 | ||
cd45daff MP |
32 | /* |
33 | * The size of the mempool used to track chunks in use. | |
34 | */ | |
35 | #define MIN_IOS 256 | |
36 | ||
ccc45ea8 JB |
37 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
38 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
39 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
40 | ||
191437a5 | 41 | struct dm_exception_table { |
ccc45ea8 JB |
42 | uint32_t hash_mask; |
43 | unsigned hash_shift; | |
44 | struct list_head *table; | |
45 | }; | |
46 | ||
47 | struct dm_snapshot { | |
48 | struct rw_semaphore lock; | |
49 | ||
50 | struct dm_dev *origin; | |
fc56f6fb MS |
51 | struct dm_dev *cow; |
52 | ||
53 | struct dm_target *ti; | |
ccc45ea8 JB |
54 | |
55 | /* List of snapshots per Origin */ | |
56 | struct list_head list; | |
57 | ||
d8ddb1cf MS |
58 | /* |
59 | * You can't use a snapshot if this is 0 (e.g. if full). | |
60 | * A snapshot-merge target never clears this. | |
61 | */ | |
ccc45ea8 JB |
62 | int valid; |
63 | ||
64 | /* Origin writes don't trigger exceptions until this is set */ | |
65 | int active; | |
66 | ||
ccc45ea8 JB |
67 | atomic_t pending_exceptions_count; |
68 | ||
230c83af MP |
69 | /* Protected by "lock" */ |
70 | sector_t exception_start_sequence; | |
71 | ||
72 | /* Protected by kcopyd single-threaded callback */ | |
73 | sector_t exception_complete_sequence; | |
74 | ||
75 | /* | |
76 | * A list of pending exceptions that completed out of order. | |
77 | * Protected by kcopyd single-threaded callback. | |
78 | */ | |
79 | struct list_head out_of_order_list; | |
80 | ||
924e600d MS |
81 | mempool_t *pending_pool; |
82 | ||
191437a5 JB |
83 | struct dm_exception_table pending; |
84 | struct dm_exception_table complete; | |
ccc45ea8 JB |
85 | |
86 | /* | |
87 | * pe_lock protects all pending_exception operations and access | |
88 | * as well as the snapshot_bios list. | |
89 | */ | |
90 | spinlock_t pe_lock; | |
91 | ||
924e600d MS |
92 | /* Chunks with outstanding reads */ |
93 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
94 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
95 | ||
ccc45ea8 JB |
96 | /* The on disk metadata handler */ |
97 | struct dm_exception_store *store; | |
98 | ||
99 | struct dm_kcopyd_client *kcopyd_client; | |
100 | ||
924e600d MS |
101 | /* Wait for events based on state_bits */ |
102 | unsigned long state_bits; | |
103 | ||
104 | /* Range of chunks currently being merged. */ | |
105 | chunk_t first_merging_chunk; | |
106 | int num_merging_chunks; | |
1e03f97e | 107 | |
d8ddb1cf MS |
108 | /* |
109 | * The merge operation failed if this flag is set. | |
110 | * Failure modes are handled as follows: | |
111 | * - I/O error reading the header | |
112 | * => don't load the target; abort. | |
113 | * - Header does not have "valid" flag set | |
114 | * => use the origin; forget about the snapshot. | |
115 | * - I/O error when reading exceptions | |
116 | * => don't load the target; abort. | |
117 | * (We can't use the intermediate origin state.) | |
118 | * - I/O error while merging | |
119 | * => stop merging; set merge_failed; process I/O normally. | |
120 | */ | |
121 | int merge_failed; | |
122 | ||
9fe86254 MP |
123 | /* |
124 | * Incoming bios that overlap with chunks being merged must wait | |
125 | * for them to be committed. | |
126 | */ | |
127 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
128 | }; |
129 | ||
1e03f97e MP |
130 | /* |
131 | * state_bits: | |
132 | * RUNNING_MERGE - Merge operation is in progress. | |
133 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
134 | * cleared afterwards. | |
135 | */ | |
136 | #define RUNNING_MERGE 0 | |
137 | #define SHUTDOWN_MERGE 1 | |
138 | ||
df5d2e90 MP |
139 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
140 | "A percentage of time allocated for copy on write"); | |
141 | ||
c2411045 MP |
142 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
143 | { | |
144 | return s->origin; | |
145 | } | |
146 | EXPORT_SYMBOL(dm_snap_origin); | |
147 | ||
fc56f6fb MS |
148 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
149 | { | |
150 | return s->cow; | |
151 | } | |
152 | EXPORT_SYMBOL(dm_snap_cow); | |
153 | ||
ccc45ea8 JB |
154 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
155 | chunk_t chunk) | |
156 | { | |
157 | return chunk << store->chunk_shift; | |
158 | } | |
159 | ||
160 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
161 | { | |
162 | /* | |
163 | * There is only ever one instance of a particular block | |
164 | * device so we can compare pointers safely. | |
165 | */ | |
166 | return lhs == rhs; | |
167 | } | |
168 | ||
028867ac | 169 | struct dm_snap_pending_exception { |
1d4989c8 | 170 | struct dm_exception e; |
1da177e4 LT |
171 | |
172 | /* | |
173 | * Origin buffers waiting for this to complete are held | |
174 | * in a bio list | |
175 | */ | |
176 | struct bio_list origin_bios; | |
177 | struct bio_list snapshot_bios; | |
178 | ||
1da177e4 LT |
179 | /* Pointer back to snapshot context */ |
180 | struct dm_snapshot *snap; | |
181 | ||
182 | /* | |
183 | * 1 indicates the exception has already been sent to | |
184 | * kcopyd. | |
185 | */ | |
186 | int started; | |
a6e50b40 | 187 | |
230c83af MP |
188 | /* There was copying error. */ |
189 | int copy_error; | |
190 | ||
191 | /* A sequence number, it is used for in-order completion. */ | |
192 | sector_t exception_sequence; | |
193 | ||
194 | struct list_head out_of_order_entry; | |
195 | ||
a6e50b40 MP |
196 | /* |
197 | * For writing a complete chunk, bypassing the copy. | |
198 | */ | |
199 | struct bio *full_bio; | |
200 | bio_end_io_t *full_bio_end_io; | |
201 | void *full_bio_private; | |
1da177e4 LT |
202 | }; |
203 | ||
204 | /* | |
205 | * Hash table mapping origin volumes to lists of snapshots and | |
206 | * a lock to protect it | |
207 | */ | |
e18b890b CL |
208 | static struct kmem_cache *exception_cache; |
209 | static struct kmem_cache *pending_cache; | |
1da177e4 | 210 | |
cd45daff MP |
211 | struct dm_snap_tracked_chunk { |
212 | struct hlist_node node; | |
213 | chunk_t chunk; | |
214 | }; | |
215 | ||
ee18026a MP |
216 | static void init_tracked_chunk(struct bio *bio) |
217 | { | |
218 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
219 | INIT_HLIST_NODE(&c->node); | |
220 | } | |
221 | ||
222 | static bool is_bio_tracked(struct bio *bio) | |
223 | { | |
224 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
225 | return !hlist_unhashed(&c->node); | |
226 | } | |
227 | ||
228 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 229 | { |
42bc954f | 230 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
231 | |
232 | c->chunk = chunk; | |
233 | ||
9aa0c0e6 | 234 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
235 | hlist_add_head(&c->node, |
236 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 237 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
238 | } |
239 | ||
ee18026a | 240 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 241 | { |
ee18026a | 242 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
243 | unsigned long flags; |
244 | ||
245 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
246 | hlist_del(&c->node); | |
247 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
248 | } |
249 | ||
a8d41b59 MP |
250 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
251 | { | |
252 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
253 | int found = 0; |
254 | ||
255 | spin_lock_irq(&s->tracked_chunk_lock); | |
256 | ||
b67bfe0d | 257 | hlist_for_each_entry(c, |
a8d41b59 MP |
258 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
259 | if (c->chunk == chunk) { | |
260 | found = 1; | |
261 | break; | |
262 | } | |
263 | } | |
264 | ||
265 | spin_unlock_irq(&s->tracked_chunk_lock); | |
266 | ||
267 | return found; | |
268 | } | |
269 | ||
615d1eb9 MS |
270 | /* |
271 | * This conflicting I/O is extremely improbable in the caller, | |
272 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
273 | */ | |
274 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
275 | { | |
276 | while (__chunk_is_tracked(s, chunk)) | |
277 | msleep(1); | |
278 | } | |
279 | ||
1da177e4 LT |
280 | /* |
281 | * One of these per registered origin, held in the snapshot_origins hash | |
282 | */ | |
283 | struct origin { | |
284 | /* The origin device */ | |
285 | struct block_device *bdev; | |
286 | ||
287 | struct list_head hash_list; | |
288 | ||
289 | /* List of snapshots for this origin */ | |
290 | struct list_head snapshots; | |
291 | }; | |
292 | ||
293 | /* | |
294 | * Size of the hash table for origin volumes. If we make this | |
295 | * the size of the minors list then it should be nearly perfect | |
296 | */ | |
297 | #define ORIGIN_HASH_SIZE 256 | |
298 | #define ORIGIN_MASK 0xFF | |
299 | static struct list_head *_origins; | |
300 | static struct rw_semaphore _origins_lock; | |
301 | ||
73dfd078 MP |
302 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
303 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
304 | static uint64_t _pending_exceptions_done_count; | |
305 | ||
1da177e4 LT |
306 | static int init_origin_hash(void) |
307 | { | |
308 | int i; | |
309 | ||
310 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
311 | GFP_KERNEL); | |
312 | if (!_origins) { | |
72d94861 | 313 | DMERR("unable to allocate memory"); |
1da177e4 LT |
314 | return -ENOMEM; |
315 | } | |
316 | ||
317 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
318 | INIT_LIST_HEAD(_origins + i); | |
319 | init_rwsem(&_origins_lock); | |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
324 | static void exit_origin_hash(void) | |
325 | { | |
326 | kfree(_origins); | |
327 | } | |
328 | ||
028867ac | 329 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
330 | { |
331 | return bdev->bd_dev & ORIGIN_MASK; | |
332 | } | |
333 | ||
334 | static struct origin *__lookup_origin(struct block_device *origin) | |
335 | { | |
336 | struct list_head *ol; | |
337 | struct origin *o; | |
338 | ||
339 | ol = &_origins[origin_hash(origin)]; | |
340 | list_for_each_entry (o, ol, hash_list) | |
341 | if (bdev_equal(o->bdev, origin)) | |
342 | return o; | |
343 | ||
344 | return NULL; | |
345 | } | |
346 | ||
347 | static void __insert_origin(struct origin *o) | |
348 | { | |
349 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
350 | list_add_tail(&o->hash_list, sl); | |
351 | } | |
352 | ||
c1f0c183 MS |
353 | /* |
354 | * _origins_lock must be held when calling this function. | |
355 | * Returns number of snapshots registered using the supplied cow device, plus: | |
356 | * snap_src - a snapshot suitable for use as a source of exception handover | |
357 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
358 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
359 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 360 | * |
9d3b15c4 | 361 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
362 | * 0: NULL, NULL - first new snapshot |
363 | * 1: snap_src, NULL - normal snapshot | |
364 | * 2: snap_src, snap_dest - waiting for handover | |
365 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
366 | * 1: NULL, snap_dest - source got destroyed without handover | |
367 | */ | |
368 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
369 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
370 | struct dm_snapshot **snap_dest, |
371 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
372 | { |
373 | struct dm_snapshot *s; | |
374 | struct origin *o; | |
375 | int count = 0; | |
376 | int active; | |
377 | ||
378 | o = __lookup_origin(snap->origin->bdev); | |
379 | if (!o) | |
380 | goto out; | |
381 | ||
382 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
383 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
384 | *snap_merge = s; | |
c1f0c183 MS |
385 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
386 | continue; | |
387 | ||
388 | down_read(&s->lock); | |
389 | active = s->active; | |
390 | up_read(&s->lock); | |
391 | ||
392 | if (active) { | |
393 | if (snap_src) | |
394 | *snap_src = s; | |
395 | } else if (snap_dest) | |
396 | *snap_dest = s; | |
397 | ||
398 | count++; | |
399 | } | |
400 | ||
401 | out: | |
402 | return count; | |
403 | } | |
404 | ||
405 | /* | |
406 | * On success, returns 1 if this snapshot is a handover destination, | |
407 | * otherwise returns 0. | |
408 | */ | |
409 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
410 | { | |
411 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 412 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
413 | |
414 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
415 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
416 | &snap_merge) == 2) || | |
c1f0c183 MS |
417 | snap_dest) { |
418 | snap->ti->error = "Snapshot cow pairing for exception " | |
419 | "table handover failed"; | |
420 | return -EINVAL; | |
421 | } | |
422 | ||
423 | /* | |
424 | * If no snap_src was found, snap cannot become a handover | |
425 | * destination. | |
426 | */ | |
427 | if (!snap_src) | |
428 | return 0; | |
429 | ||
9d3b15c4 MP |
430 | /* |
431 | * Non-snapshot-merge handover? | |
432 | */ | |
433 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
434 | return 1; | |
435 | ||
436 | /* | |
437 | * Do not allow more than one merging snapshot. | |
438 | */ | |
439 | if (snap_merge) { | |
440 | snap->ti->error = "A snapshot is already merging."; | |
441 | return -EINVAL; | |
442 | } | |
443 | ||
1e03f97e MP |
444 | if (!snap_src->store->type->prepare_merge || |
445 | !snap_src->store->type->commit_merge) { | |
446 | snap->ti->error = "Snapshot exception store does not " | |
447 | "support snapshot-merge."; | |
448 | return -EINVAL; | |
449 | } | |
450 | ||
c1f0c183 MS |
451 | return 1; |
452 | } | |
453 | ||
454 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
455 | { | |
456 | struct dm_snapshot *l; | |
457 | ||
458 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
459 | list_for_each_entry(l, &o->snapshots, list) | |
460 | if (l->store->chunk_size < s->store->chunk_size) | |
461 | break; | |
462 | list_add_tail(&s->list, &l->list); | |
463 | } | |
464 | ||
1da177e4 LT |
465 | /* |
466 | * Make a note of the snapshot and its origin so we can look it | |
467 | * up when the origin has a write on it. | |
c1f0c183 MS |
468 | * |
469 | * Also validate snapshot exception store handovers. | |
470 | * On success, returns 1 if this registration is a handover destination, | |
471 | * otherwise returns 0. | |
1da177e4 LT |
472 | */ |
473 | static int register_snapshot(struct dm_snapshot *snap) | |
474 | { | |
c1f0c183 | 475 | struct origin *o, *new_o = NULL; |
1da177e4 | 476 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 477 | int r = 0; |
1da177e4 | 478 | |
60c856c8 MP |
479 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
480 | if (!new_o) | |
481 | return -ENOMEM; | |
482 | ||
1da177e4 | 483 | down_write(&_origins_lock); |
1da177e4 | 484 | |
c1f0c183 MS |
485 | r = __validate_exception_handover(snap); |
486 | if (r < 0) { | |
487 | kfree(new_o); | |
488 | goto out; | |
489 | } | |
490 | ||
491 | o = __lookup_origin(bdev); | |
60c856c8 MP |
492 | if (o) |
493 | kfree(new_o); | |
494 | else { | |
1da177e4 | 495 | /* New origin */ |
60c856c8 | 496 | o = new_o; |
1da177e4 LT |
497 | |
498 | /* Initialise the struct */ | |
499 | INIT_LIST_HEAD(&o->snapshots); | |
500 | o->bdev = bdev; | |
501 | ||
502 | __insert_origin(o); | |
503 | } | |
504 | ||
c1f0c183 MS |
505 | __insert_snapshot(o, snap); |
506 | ||
507 | out: | |
508 | up_write(&_origins_lock); | |
509 | ||
510 | return r; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Move snapshot to correct place in list according to chunk size. | |
515 | */ | |
516 | static void reregister_snapshot(struct dm_snapshot *s) | |
517 | { | |
518 | struct block_device *bdev = s->origin->bdev; | |
519 | ||
520 | down_write(&_origins_lock); | |
521 | ||
522 | list_del(&s->list); | |
523 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
524 | |
525 | up_write(&_origins_lock); | |
1da177e4 LT |
526 | } |
527 | ||
528 | static void unregister_snapshot(struct dm_snapshot *s) | |
529 | { | |
530 | struct origin *o; | |
531 | ||
532 | down_write(&_origins_lock); | |
533 | o = __lookup_origin(s->origin->bdev); | |
534 | ||
535 | list_del(&s->list); | |
c1f0c183 | 536 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
537 | list_del(&o->hash_list); |
538 | kfree(o); | |
539 | } | |
540 | ||
541 | up_write(&_origins_lock); | |
542 | } | |
543 | ||
544 | /* | |
545 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
546 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
547 | * some consecutive chunks to be grouped together. | |
1da177e4 | 548 | */ |
3510cb94 JB |
549 | static int dm_exception_table_init(struct dm_exception_table *et, |
550 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
551 | { |
552 | unsigned int i; | |
553 | ||
d74f81f8 | 554 | et->hash_shift = hash_shift; |
1da177e4 LT |
555 | et->hash_mask = size - 1; |
556 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
557 | if (!et->table) | |
558 | return -ENOMEM; | |
559 | ||
560 | for (i = 0; i < size; i++) | |
561 | INIT_LIST_HEAD(et->table + i); | |
562 | ||
563 | return 0; | |
564 | } | |
565 | ||
3510cb94 JB |
566 | static void dm_exception_table_exit(struct dm_exception_table *et, |
567 | struct kmem_cache *mem) | |
1da177e4 LT |
568 | { |
569 | struct list_head *slot; | |
1d4989c8 | 570 | struct dm_exception *ex, *next; |
1da177e4 LT |
571 | int i, size; |
572 | ||
573 | size = et->hash_mask + 1; | |
574 | for (i = 0; i < size; i++) { | |
575 | slot = et->table + i; | |
576 | ||
577 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
578 | kmem_cache_free(mem, ex); | |
579 | } | |
580 | ||
581 | vfree(et->table); | |
582 | } | |
583 | ||
191437a5 | 584 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 585 | { |
d74f81f8 | 586 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
587 | } |
588 | ||
3510cb94 | 589 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
590 | { |
591 | list_del(&e->hash_list); | |
592 | } | |
593 | ||
594 | /* | |
595 | * Return the exception data for a sector, or NULL if not | |
596 | * remapped. | |
597 | */ | |
3510cb94 JB |
598 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
599 | chunk_t chunk) | |
1da177e4 LT |
600 | { |
601 | struct list_head *slot; | |
1d4989c8 | 602 | struct dm_exception *e; |
1da177e4 LT |
603 | |
604 | slot = &et->table[exception_hash(et, chunk)]; | |
605 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
606 | if (chunk >= e->old_chunk && |
607 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
608 | return e; |
609 | ||
610 | return NULL; | |
611 | } | |
612 | ||
119bc547 | 613 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 614 | { |
1d4989c8 | 615 | struct dm_exception *e; |
1da177e4 | 616 | |
119bc547 MP |
617 | e = kmem_cache_alloc(exception_cache, gfp); |
618 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
619 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
620 | ||
621 | return e; | |
622 | } | |
623 | ||
3510cb94 | 624 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
625 | { |
626 | kmem_cache_free(exception_cache, e); | |
627 | } | |
628 | ||
92e86812 | 629 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 630 | { |
92e86812 MP |
631 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
632 | GFP_NOIO); | |
633 | ||
879129d2 | 634 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
635 | pe->snap = s; |
636 | ||
637 | return pe; | |
1da177e4 LT |
638 | } |
639 | ||
028867ac | 640 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 641 | { |
879129d2 MP |
642 | struct dm_snapshot *s = pe->snap; |
643 | ||
644 | mempool_free(pe, s->pending_pool); | |
4e857c58 | 645 | smp_mb__before_atomic(); |
879129d2 | 646 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
647 | } |
648 | ||
3510cb94 JB |
649 | static void dm_insert_exception(struct dm_exception_table *eh, |
650 | struct dm_exception *new_e) | |
d74f81f8 | 651 | { |
d74f81f8 | 652 | struct list_head *l; |
1d4989c8 | 653 | struct dm_exception *e = NULL; |
d74f81f8 MB |
654 | |
655 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
656 | ||
657 | /* Add immediately if this table doesn't support consecutive chunks */ | |
658 | if (!eh->hash_shift) | |
659 | goto out; | |
660 | ||
661 | /* List is ordered by old_chunk */ | |
662 | list_for_each_entry_reverse(e, l, hash_list) { | |
663 | /* Insert after an existing chunk? */ | |
664 | if (new_e->old_chunk == (e->old_chunk + | |
665 | dm_consecutive_chunk_count(e) + 1) && | |
666 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
667 | dm_consecutive_chunk_count(e) + 1)) { | |
668 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 669 | free_completed_exception(new_e); |
d74f81f8 MB |
670 | return; |
671 | } | |
672 | ||
673 | /* Insert before an existing chunk? */ | |
674 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
675 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
676 | dm_consecutive_chunk_count_inc(e); | |
677 | e->old_chunk--; | |
678 | e->new_chunk--; | |
3510cb94 | 679 | free_completed_exception(new_e); |
d74f81f8 MB |
680 | return; |
681 | } | |
682 | ||
683 | if (new_e->old_chunk > e->old_chunk) | |
684 | break; | |
685 | } | |
686 | ||
687 | out: | |
688 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
689 | } | |
690 | ||
a159c1ac JB |
691 | /* |
692 | * Callback used by the exception stores to load exceptions when | |
693 | * initialising. | |
694 | */ | |
695 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 696 | { |
a159c1ac | 697 | struct dm_snapshot *s = context; |
1d4989c8 | 698 | struct dm_exception *e; |
1da177e4 | 699 | |
119bc547 | 700 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
701 | if (!e) |
702 | return -ENOMEM; | |
703 | ||
704 | e->old_chunk = old; | |
d74f81f8 MB |
705 | |
706 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 707 | e->new_chunk = new; |
d74f81f8 | 708 | |
3510cb94 | 709 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 710 | |
1da177e4 LT |
711 | return 0; |
712 | } | |
713 | ||
7e201b35 MP |
714 | /* |
715 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
716 | * Return zero if the origin has no snapshots. | |
717 | */ | |
542f9038 | 718 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
719 | { |
720 | struct dm_snapshot *snap; | |
721 | unsigned chunk_size = 0; | |
722 | ||
723 | if (o) | |
724 | list_for_each_entry(snap, &o->snapshots, list) | |
725 | chunk_size = min_not_zero(chunk_size, | |
726 | snap->store->chunk_size); | |
727 | ||
542f9038 | 728 | return (uint32_t) chunk_size; |
7e201b35 MP |
729 | } |
730 | ||
1da177e4 LT |
731 | /* |
732 | * Hard coded magic. | |
733 | */ | |
734 | static int calc_max_buckets(void) | |
735 | { | |
736 | /* use a fixed size of 2MB */ | |
737 | unsigned long mem = 2 * 1024 * 1024; | |
738 | mem /= sizeof(struct list_head); | |
739 | ||
740 | return mem; | |
741 | } | |
742 | ||
1da177e4 LT |
743 | /* |
744 | * Allocate room for a suitable hash table. | |
745 | */ | |
fee1998e | 746 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 747 | { |
60e356f3 | 748 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
749 | |
750 | /* | |
751 | * Calculate based on the size of the original volume or | |
752 | * the COW volume... | |
753 | */ | |
fc56f6fb | 754 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
755 | max_buckets = calc_max_buckets(); |
756 | ||
60e356f3 | 757 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
758 | hash_size = min(hash_size, max_buckets); |
759 | ||
8e87b9b8 MP |
760 | if (hash_size < 64) |
761 | hash_size = 64; | |
8defd830 | 762 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
763 | if (dm_exception_table_init(&s->complete, hash_size, |
764 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
765 | return -ENOMEM; |
766 | ||
767 | /* | |
768 | * Allocate hash table for in-flight exceptions | |
769 | * Make this smaller than the real hash table | |
770 | */ | |
771 | hash_size >>= 3; | |
772 | if (hash_size < 64) | |
773 | hash_size = 64; | |
774 | ||
3510cb94 JB |
775 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
776 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
777 | return -ENOMEM; |
778 | } | |
779 | ||
780 | return 0; | |
781 | } | |
782 | ||
1e03f97e MP |
783 | static void merge_shutdown(struct dm_snapshot *s) |
784 | { | |
785 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 786 | smp_mb__after_atomic(); |
1e03f97e MP |
787 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
788 | } | |
789 | ||
9fe86254 MP |
790 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
791 | { | |
792 | s->first_merging_chunk = 0; | |
793 | s->num_merging_chunks = 0; | |
794 | ||
795 | return bio_list_get(&s->bios_queued_during_merge); | |
796 | } | |
797 | ||
1e03f97e MP |
798 | /* |
799 | * Remove one chunk from the index of completed exceptions. | |
800 | */ | |
801 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
802 | chunk_t old_chunk) | |
803 | { | |
804 | struct dm_exception *e; | |
805 | ||
1e03f97e MP |
806 | e = dm_lookup_exception(&s->complete, old_chunk); |
807 | if (!e) { | |
808 | DMERR("Corruption detected: exception for block %llu is " | |
809 | "on disk but not in memory", | |
810 | (unsigned long long)old_chunk); | |
811 | return -EINVAL; | |
812 | } | |
813 | ||
814 | /* | |
815 | * If this is the only chunk using this exception, remove exception. | |
816 | */ | |
817 | if (!dm_consecutive_chunk_count(e)) { | |
818 | dm_remove_exception(e); | |
819 | free_completed_exception(e); | |
820 | return 0; | |
821 | } | |
822 | ||
823 | /* | |
824 | * The chunk may be either at the beginning or the end of a | |
825 | * group of consecutive chunks - never in the middle. We are | |
826 | * removing chunks in the opposite order to that in which they | |
827 | * were added, so this should always be true. | |
828 | * Decrement the consecutive chunk counter and adjust the | |
829 | * starting point if necessary. | |
830 | */ | |
831 | if (old_chunk == e->old_chunk) { | |
832 | e->old_chunk++; | |
833 | e->new_chunk++; | |
834 | } else if (old_chunk != e->old_chunk + | |
835 | dm_consecutive_chunk_count(e)) { | |
836 | DMERR("Attempt to merge block %llu from the " | |
837 | "middle of a chunk range [%llu - %llu]", | |
838 | (unsigned long long)old_chunk, | |
839 | (unsigned long long)e->old_chunk, | |
840 | (unsigned long long) | |
841 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
842 | return -EINVAL; | |
843 | } | |
844 | ||
845 | dm_consecutive_chunk_count_dec(e); | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
9fe86254 MP |
850 | static void flush_bios(struct bio *bio); |
851 | ||
852 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 853 | { |
9fe86254 MP |
854 | struct bio *b = NULL; |
855 | int r; | |
856 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
857 | |
858 | down_write(&s->lock); | |
9fe86254 MP |
859 | |
860 | /* | |
861 | * Process chunks (and associated exceptions) in reverse order | |
862 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
863 | */ | |
864 | do { | |
865 | r = __remove_single_exception_chunk(s, old_chunk); | |
866 | if (r) | |
867 | goto out; | |
868 | } while (old_chunk-- > s->first_merging_chunk); | |
869 | ||
870 | b = __release_queued_bios_after_merge(s); | |
871 | ||
872 | out: | |
1e03f97e | 873 | up_write(&s->lock); |
9fe86254 MP |
874 | if (b) |
875 | flush_bios(b); | |
1e03f97e MP |
876 | |
877 | return r; | |
878 | } | |
879 | ||
73dfd078 MP |
880 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
881 | sector_t sector, unsigned chunk_size); | |
882 | ||
1e03f97e MP |
883 | static void merge_callback(int read_err, unsigned long write_err, |
884 | void *context); | |
885 | ||
73dfd078 MP |
886 | static uint64_t read_pending_exceptions_done_count(void) |
887 | { | |
888 | uint64_t pending_exceptions_done; | |
889 | ||
890 | spin_lock(&_pending_exceptions_done_spinlock); | |
891 | pending_exceptions_done = _pending_exceptions_done_count; | |
892 | spin_unlock(&_pending_exceptions_done_spinlock); | |
893 | ||
894 | return pending_exceptions_done; | |
895 | } | |
896 | ||
897 | static void increment_pending_exceptions_done_count(void) | |
898 | { | |
899 | spin_lock(&_pending_exceptions_done_spinlock); | |
900 | _pending_exceptions_done_count++; | |
901 | spin_unlock(&_pending_exceptions_done_spinlock); | |
902 | ||
903 | wake_up_all(&_pending_exceptions_done); | |
904 | } | |
905 | ||
1e03f97e MP |
906 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
907 | { | |
8a2d5286 | 908 | int i, linear_chunks; |
1e03f97e MP |
909 | chunk_t old_chunk, new_chunk; |
910 | struct dm_io_region src, dest; | |
8a2d5286 | 911 | sector_t io_size; |
73dfd078 | 912 | uint64_t previous_count; |
1e03f97e MP |
913 | |
914 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
915 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
916 | goto shut; | |
917 | ||
918 | /* | |
919 | * valid flag never changes during merge, so no lock required. | |
920 | */ | |
921 | if (!s->valid) { | |
922 | DMERR("Snapshot is invalid: can't merge"); | |
923 | goto shut; | |
924 | } | |
925 | ||
8a2d5286 MS |
926 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
927 | &new_chunk); | |
928 | if (linear_chunks <= 0) { | |
d8ddb1cf | 929 | if (linear_chunks < 0) { |
1e03f97e MP |
930 | DMERR("Read error in exception store: " |
931 | "shutting down merge"); | |
d8ddb1cf MS |
932 | down_write(&s->lock); |
933 | s->merge_failed = 1; | |
934 | up_write(&s->lock); | |
935 | } | |
1e03f97e MP |
936 | goto shut; |
937 | } | |
938 | ||
8a2d5286 MS |
939 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
940 | old_chunk = old_chunk + 1 - linear_chunks; | |
941 | new_chunk = new_chunk + 1 - linear_chunks; | |
942 | ||
943 | /* | |
944 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
945 | * from the exception store to the origin | |
946 | */ | |
947 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 948 | |
1e03f97e MP |
949 | dest.bdev = s->origin->bdev; |
950 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 951 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
952 | |
953 | src.bdev = s->cow->bdev; | |
954 | src.sector = chunk_to_sector(s->store, new_chunk); | |
955 | src.count = dest.count; | |
956 | ||
73dfd078 MP |
957 | /* |
958 | * Reallocate any exceptions needed in other snapshots then | |
959 | * wait for the pending exceptions to complete. | |
960 | * Each time any pending exception (globally on the system) | |
961 | * completes we are woken and repeat the process to find out | |
962 | * if we can proceed. While this may not seem a particularly | |
963 | * efficient algorithm, it is not expected to have any | |
964 | * significant impact on performance. | |
965 | */ | |
966 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 967 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
968 | wait_event(_pending_exceptions_done, |
969 | (read_pending_exceptions_done_count() != | |
970 | previous_count)); | |
971 | /* Retry after the wait, until all exceptions are done. */ | |
972 | previous_count = read_pending_exceptions_done_count(); | |
973 | } | |
974 | ||
9fe86254 MP |
975 | down_write(&s->lock); |
976 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 977 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
978 | up_write(&s->lock); |
979 | ||
8a2d5286 MS |
980 | /* Wait until writes to all 'linear_chunks' drain */ |
981 | for (i = 0; i < linear_chunks; i++) | |
982 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 983 | |
1e03f97e MP |
984 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
985 | return; | |
986 | ||
987 | shut: | |
988 | merge_shutdown(s); | |
989 | } | |
990 | ||
9fe86254 MP |
991 | static void error_bios(struct bio *bio); |
992 | ||
1e03f97e MP |
993 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
994 | { | |
995 | struct dm_snapshot *s = context; | |
9fe86254 | 996 | struct bio *b = NULL; |
1e03f97e MP |
997 | |
998 | if (read_err || write_err) { | |
999 | if (read_err) | |
1000 | DMERR("Read error: shutting down merge."); | |
1001 | else | |
1002 | DMERR("Write error: shutting down merge."); | |
1003 | goto shut; | |
1004 | } | |
1005 | ||
9fe86254 MP |
1006 | if (s->store->type->commit_merge(s->store, |
1007 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1008 | DMERR("Write error in exception store: shutting down merge"); |
1009 | goto shut; | |
1010 | } | |
1011 | ||
9fe86254 MP |
1012 | if (remove_single_exception_chunk(s) < 0) |
1013 | goto shut; | |
1014 | ||
1e03f97e MP |
1015 | snapshot_merge_next_chunks(s); |
1016 | ||
1017 | return; | |
1018 | ||
1019 | shut: | |
9fe86254 | 1020 | down_write(&s->lock); |
d8ddb1cf | 1021 | s->merge_failed = 1; |
9fe86254 MP |
1022 | b = __release_queued_bios_after_merge(s); |
1023 | up_write(&s->lock); | |
1024 | error_bios(b); | |
1025 | ||
1e03f97e MP |
1026 | merge_shutdown(s); |
1027 | } | |
1028 | ||
1029 | static void start_merge(struct dm_snapshot *s) | |
1030 | { | |
1031 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1032 | snapshot_merge_next_chunks(s); | |
1033 | } | |
1034 | ||
1e03f97e MP |
1035 | /* |
1036 | * Stop the merging process and wait until it finishes. | |
1037 | */ | |
1038 | static void stop_merge(struct dm_snapshot *s) | |
1039 | { | |
1040 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1041 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1042 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1043 | } | |
1044 | ||
1da177e4 LT |
1045 | /* |
1046 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
1047 | */ | |
1048 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1049 | { | |
1050 | struct dm_snapshot *s; | |
cd45daff | 1051 | int i; |
1da177e4 | 1052 | int r = -EINVAL; |
fc56f6fb | 1053 | char *origin_path, *cow_path; |
55a62eef | 1054 | unsigned args_used, num_flush_bios = 1; |
10b8106a | 1055 | fmode_t origin_mode = FMODE_READ; |
1da177e4 | 1056 | |
4c7e3bf4 | 1057 | if (argc != 4) { |
72d94861 | 1058 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1059 | r = -EINVAL; |
fc56f6fb | 1060 | goto bad; |
1da177e4 LT |
1061 | } |
1062 | ||
10b8106a | 1063 | if (dm_target_is_snapshot_merge(ti)) { |
55a62eef | 1064 | num_flush_bios = 2; |
10b8106a MS |
1065 | origin_mode = FMODE_WRITE; |
1066 | } | |
1067 | ||
fc56f6fb MS |
1068 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1069 | if (!s) { | |
a2d2b034 | 1070 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1071 | r = -ENOMEM; |
1072 | goto bad; | |
1073 | } | |
1074 | ||
c2411045 MP |
1075 | origin_path = argv[0]; |
1076 | argv++; | |
1077 | argc--; | |
1078 | ||
1079 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1080 | if (r) { | |
1081 | ti->error = "Cannot get origin device"; | |
1082 | goto bad_origin; | |
1083 | } | |
1084 | ||
fc56f6fb MS |
1085 | cow_path = argv[0]; |
1086 | argv++; | |
1087 | argc--; | |
1088 | ||
024d37e9 | 1089 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1090 | if (r) { |
1091 | ti->error = "Cannot get COW device"; | |
1092 | goto bad_cow; | |
1093 | } | |
1094 | ||
1095 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1096 | if (r) { |
1097 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1098 | r = -EINVAL; |
fc56f6fb | 1099 | goto bad_store; |
1da177e4 LT |
1100 | } |
1101 | ||
fee1998e JB |
1102 | argv += args_used; |
1103 | argc -= args_used; | |
1104 | ||
fc56f6fb | 1105 | s->ti = ti; |
1da177e4 | 1106 | s->valid = 1; |
aa14edeb | 1107 | s->active = 0; |
879129d2 | 1108 | atomic_set(&s->pending_exceptions_count, 0); |
230c83af MP |
1109 | s->exception_start_sequence = 0; |
1110 | s->exception_complete_sequence = 0; | |
1111 | INIT_LIST_HEAD(&s->out_of_order_list); | |
1da177e4 | 1112 | init_rwsem(&s->lock); |
c1f0c183 | 1113 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1114 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1115 | s->state_bits = 0; |
d8ddb1cf | 1116 | s->merge_failed = 0; |
9fe86254 MP |
1117 | s->first_merging_chunk = 0; |
1118 | s->num_merging_chunks = 0; | |
1119 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1120 | |
1121 | /* Allocate hash table for COW data */ | |
fee1998e | 1122 | if (init_hash_tables(s)) { |
1da177e4 LT |
1123 | ti->error = "Unable to allocate hash table space"; |
1124 | r = -ENOMEM; | |
fee1998e | 1125 | goto bad_hash_tables; |
1da177e4 LT |
1126 | } |
1127 | ||
df5d2e90 | 1128 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1129 | if (IS_ERR(s->kcopyd_client)) { |
1130 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1131 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1132 | goto bad_kcopyd; |
1da177e4 LT |
1133 | } |
1134 | ||
92e86812 MP |
1135 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1136 | if (!s->pending_pool) { | |
1137 | ti->error = "Could not allocate mempool for pending exceptions"; | |
09e8b813 | 1138 | r = -ENOMEM; |
fee1998e | 1139 | goto bad_pending_pool; |
92e86812 MP |
1140 | } |
1141 | ||
cd45daff MP |
1142 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1143 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1144 | ||
1145 | spin_lock_init(&s->tracked_chunk_lock); | |
1146 | ||
c1f0c183 | 1147 | ti->private = s; |
55a62eef | 1148 | ti->num_flush_bios = num_flush_bios; |
42bc954f | 1149 | ti->per_bio_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1150 | |
1151 | /* Add snapshot to the list of snapshots for this origin */ | |
1152 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1153 | r = register_snapshot(s); | |
1154 | if (r == -ENOMEM) { | |
1155 | ti->error = "Snapshot origin struct allocation failed"; | |
1156 | goto bad_load_and_register; | |
1157 | } else if (r < 0) { | |
1158 | /* invalid handover, register_snapshot has set ti->error */ | |
1159 | goto bad_load_and_register; | |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * Metadata must only be loaded into one table at once, so skip this | |
1164 | * if metadata will be handed over during resume. | |
1165 | * Chunk size will be set during the handover - set it to zero to | |
1166 | * ensure it's ignored. | |
1167 | */ | |
1168 | if (r > 0) { | |
1169 | s->store->chunk_size = 0; | |
1170 | return 0; | |
1171 | } | |
1172 | ||
493df71c JB |
1173 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1174 | (void *)s); | |
0764147b | 1175 | if (r < 0) { |
f9cea4f7 | 1176 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1177 | goto bad_read_metadata; |
0764147b MB |
1178 | } else if (r > 0) { |
1179 | s->valid = 0; | |
1180 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1181 | } |
aa14edeb | 1182 | |
3f2412dc MP |
1183 | if (!s->store->chunk_size) { |
1184 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1185 | goto bad_read_metadata; |
1da177e4 | 1186 | } |
542f9038 MS |
1187 | |
1188 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1189 | if (r) | |
1190 | goto bad_read_metadata; | |
1da177e4 LT |
1191 | |
1192 | return 0; | |
1193 | ||
c1f0c183 MS |
1194 | bad_read_metadata: |
1195 | unregister_snapshot(s); | |
1196 | ||
fee1998e | 1197 | bad_load_and_register: |
92e86812 MP |
1198 | mempool_destroy(s->pending_pool); |
1199 | ||
fee1998e | 1200 | bad_pending_pool: |
eb69aca5 | 1201 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1202 | |
fee1998e | 1203 | bad_kcopyd: |
3510cb94 JB |
1204 | dm_exception_table_exit(&s->pending, pending_cache); |
1205 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1206 | |
fee1998e | 1207 | bad_hash_tables: |
fc56f6fb | 1208 | dm_exception_store_destroy(s->store); |
1da177e4 | 1209 | |
fc56f6fb MS |
1210 | bad_store: |
1211 | dm_put_device(ti, s->cow); | |
fee1998e | 1212 | |
fc56f6fb | 1213 | bad_cow: |
c2411045 MP |
1214 | dm_put_device(ti, s->origin); |
1215 | ||
1216 | bad_origin: | |
fc56f6fb MS |
1217 | kfree(s); |
1218 | ||
1219 | bad: | |
1da177e4 LT |
1220 | return r; |
1221 | } | |
1222 | ||
31c93a0c MB |
1223 | static void __free_exceptions(struct dm_snapshot *s) |
1224 | { | |
eb69aca5 | 1225 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1226 | s->kcopyd_client = NULL; |
1227 | ||
3510cb94 JB |
1228 | dm_exception_table_exit(&s->pending, pending_cache); |
1229 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1230 | } |
1231 | ||
c1f0c183 MS |
1232 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1233 | struct dm_snapshot *snap_dest) | |
1234 | { | |
1235 | union { | |
1236 | struct dm_exception_table table_swap; | |
1237 | struct dm_exception_store *store_swap; | |
1238 | } u; | |
1239 | ||
1240 | /* | |
1241 | * Swap all snapshot context information between the two instances. | |
1242 | */ | |
1243 | u.table_swap = snap_dest->complete; | |
1244 | snap_dest->complete = snap_src->complete; | |
1245 | snap_src->complete = u.table_swap; | |
1246 | ||
1247 | u.store_swap = snap_dest->store; | |
1248 | snap_dest->store = snap_src->store; | |
1249 | snap_src->store = u.store_swap; | |
1250 | ||
1251 | snap_dest->store->snap = snap_dest; | |
1252 | snap_src->store->snap = snap_src; | |
1253 | ||
542f9038 | 1254 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 MS |
1255 | snap_dest->valid = snap_src->valid; |
1256 | ||
1257 | /* | |
1258 | * Set source invalid to ensure it receives no further I/O. | |
1259 | */ | |
1260 | snap_src->valid = 0; | |
1261 | } | |
1262 | ||
1da177e4 LT |
1263 | static void snapshot_dtr(struct dm_target *ti) |
1264 | { | |
cd45daff MP |
1265 | #ifdef CONFIG_DM_DEBUG |
1266 | int i; | |
1267 | #endif | |
028867ac | 1268 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1269 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1270 | |
c1f0c183 MS |
1271 | down_read(&_origins_lock); |
1272 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1273 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1274 | if (snap_src && snap_dest && (s == snap_src)) { |
1275 | down_write(&snap_dest->lock); | |
1276 | snap_dest->valid = 0; | |
1277 | up_write(&snap_dest->lock); | |
1278 | DMERR("Cancelling snapshot handover."); | |
1279 | } | |
1280 | up_read(&_origins_lock); | |
1281 | ||
1e03f97e MP |
1282 | if (dm_target_is_snapshot_merge(ti)) |
1283 | stop_merge(s); | |
1284 | ||
138728dc AK |
1285 | /* Prevent further origin writes from using this snapshot. */ |
1286 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1287 | unregister_snapshot(s); |
1288 | ||
879129d2 | 1289 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1290 | msleep(1); |
879129d2 MP |
1291 | /* |
1292 | * Ensure instructions in mempool_destroy aren't reordered | |
1293 | * before atomic_read. | |
1294 | */ | |
1295 | smp_mb(); | |
1296 | ||
cd45daff MP |
1297 | #ifdef CONFIG_DM_DEBUG |
1298 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1299 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1300 | #endif | |
1301 | ||
31c93a0c | 1302 | __free_exceptions(s); |
1da177e4 | 1303 | |
92e86812 MP |
1304 | mempool_destroy(s->pending_pool); |
1305 | ||
fee1998e | 1306 | dm_exception_store_destroy(s->store); |
138728dc | 1307 | |
fc56f6fb MS |
1308 | dm_put_device(ti, s->cow); |
1309 | ||
c2411045 MP |
1310 | dm_put_device(ti, s->origin); |
1311 | ||
1da177e4 LT |
1312 | kfree(s); |
1313 | } | |
1314 | ||
1315 | /* | |
1316 | * Flush a list of buffers. | |
1317 | */ | |
1318 | static void flush_bios(struct bio *bio) | |
1319 | { | |
1320 | struct bio *n; | |
1321 | ||
1322 | while (bio) { | |
1323 | n = bio->bi_next; | |
1324 | bio->bi_next = NULL; | |
1325 | generic_make_request(bio); | |
1326 | bio = n; | |
1327 | } | |
1328 | } | |
1329 | ||
515ad66c MP |
1330 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1331 | ||
1332 | /* | |
1333 | * Flush a list of buffers. | |
1334 | */ | |
1335 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1336 | { | |
1337 | struct bio *n; | |
1338 | int r; | |
1339 | ||
1340 | while (bio) { | |
1341 | n = bio->bi_next; | |
1342 | bio->bi_next = NULL; | |
1343 | r = do_origin(s->origin, bio); | |
1344 | if (r == DM_MAPIO_REMAPPED) | |
1345 | generic_make_request(bio); | |
1346 | bio = n; | |
1347 | } | |
1348 | } | |
1349 | ||
1da177e4 LT |
1350 | /* |
1351 | * Error a list of buffers. | |
1352 | */ | |
1353 | static void error_bios(struct bio *bio) | |
1354 | { | |
1355 | struct bio *n; | |
1356 | ||
1357 | while (bio) { | |
1358 | n = bio->bi_next; | |
1359 | bio->bi_next = NULL; | |
6712ecf8 | 1360 | bio_io_error(bio); |
1da177e4 LT |
1361 | bio = n; |
1362 | } | |
1363 | } | |
1364 | ||
695368ac | 1365 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1366 | { |
1367 | if (!s->valid) | |
1368 | return; | |
1369 | ||
1370 | if (err == -EIO) | |
1371 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1372 | else if (err == -ENOMEM) | |
1373 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1374 | ||
493df71c JB |
1375 | if (s->store->type->drop_snapshot) |
1376 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1377 | |
1378 | s->valid = 0; | |
1379 | ||
fc56f6fb | 1380 | dm_table_event(s->ti->table); |
76df1c65 AK |
1381 | } |
1382 | ||
028867ac | 1383 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1384 | { |
1d4989c8 | 1385 | struct dm_exception *e; |
1da177e4 | 1386 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1387 | struct bio *origin_bios = NULL; |
1388 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1389 | struct bio *full_bio = NULL; |
9d493fa8 | 1390 | int error = 0; |
1da177e4 | 1391 | |
76df1c65 AK |
1392 | if (!success) { |
1393 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1394 | down_write(&s->lock); |
695368ac | 1395 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1396 | error = 1; |
76df1c65 AK |
1397 | goto out; |
1398 | } | |
1399 | ||
119bc547 | 1400 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1401 | if (!e) { |
1da177e4 | 1402 | down_write(&s->lock); |
695368ac | 1403 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1404 | error = 1; |
76df1c65 AK |
1405 | goto out; |
1406 | } | |
1407 | *e = pe->e; | |
1da177e4 | 1408 | |
76df1c65 AK |
1409 | down_write(&s->lock); |
1410 | if (!s->valid) { | |
3510cb94 | 1411 | free_completed_exception(e); |
9d493fa8 | 1412 | error = 1; |
76df1c65 | 1413 | goto out; |
1da177e4 LT |
1414 | } |
1415 | ||
615d1eb9 MS |
1416 | /* Check for conflicting reads */ |
1417 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1418 | |
9d493fa8 AK |
1419 | /* |
1420 | * Add a proper exception, and remove the | |
1421 | * in-flight exception from the list. | |
1422 | */ | |
3510cb94 | 1423 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1424 | |
a2d2b034 | 1425 | out: |
3510cb94 | 1426 | dm_remove_exception(&pe->e); |
9d493fa8 | 1427 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1428 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 MP |
1429 | full_bio = pe->full_bio; |
1430 | if (full_bio) { | |
1431 | full_bio->bi_end_io = pe->full_bio_end_io; | |
1432 | full_bio->bi_private = pe->full_bio_private; | |
196d38bc | 1433 | atomic_inc(&full_bio->bi_remaining); |
a6e50b40 | 1434 | } |
515ad66c | 1435 | free_pending_exception(pe); |
1da177e4 | 1436 | |
73dfd078 MP |
1437 | increment_pending_exceptions_done_count(); |
1438 | ||
9d493fa8 AK |
1439 | up_write(&s->lock); |
1440 | ||
1441 | /* Submit any pending write bios */ | |
a6e50b40 MP |
1442 | if (error) { |
1443 | if (full_bio) | |
1444 | bio_io_error(full_bio); | |
9d493fa8 | 1445 | error_bios(snapshot_bios); |
a6e50b40 MP |
1446 | } else { |
1447 | if (full_bio) | |
1448 | bio_endio(full_bio, 0); | |
9d493fa8 | 1449 | flush_bios(snapshot_bios); |
a6e50b40 | 1450 | } |
9d493fa8 | 1451 | |
515ad66c | 1452 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | static void commit_callback(void *context, int success) | |
1456 | { | |
028867ac AK |
1457 | struct dm_snap_pending_exception *pe = context; |
1458 | ||
1da177e4 LT |
1459 | pending_complete(pe, success); |
1460 | } | |
1461 | ||
230c83af MP |
1462 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1463 | { | |
1464 | struct dm_snapshot *s = pe->snap; | |
1465 | ||
1466 | if (unlikely(pe->copy_error)) | |
1467 | pending_complete(pe, 0); | |
1468 | ||
1469 | else | |
1470 | /* Update the metadata if we are persistent */ | |
1471 | s->store->type->commit_exception(s->store, &pe->e, | |
1472 | commit_callback, pe); | |
1473 | } | |
1474 | ||
1da177e4 LT |
1475 | /* |
1476 | * Called when the copy I/O has finished. kcopyd actually runs | |
1477 | * this code so don't block. | |
1478 | */ | |
4cdc1d1f | 1479 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1480 | { |
028867ac | 1481 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1482 | struct dm_snapshot *s = pe->snap; |
1483 | ||
230c83af | 1484 | pe->copy_error = read_err || write_err; |
1da177e4 | 1485 | |
230c83af MP |
1486 | if (pe->exception_sequence == s->exception_complete_sequence) { |
1487 | s->exception_complete_sequence++; | |
1488 | complete_exception(pe); | |
1489 | ||
1490 | while (!list_empty(&s->out_of_order_list)) { | |
1491 | pe = list_entry(s->out_of_order_list.next, | |
1492 | struct dm_snap_pending_exception, out_of_order_entry); | |
1493 | if (pe->exception_sequence != s->exception_complete_sequence) | |
1494 | break; | |
1495 | s->exception_complete_sequence++; | |
1496 | list_del(&pe->out_of_order_entry); | |
1497 | complete_exception(pe); | |
1498 | } | |
1499 | } else { | |
1500 | struct list_head *lh; | |
1501 | struct dm_snap_pending_exception *pe2; | |
1502 | ||
1503 | list_for_each_prev(lh, &s->out_of_order_list) { | |
1504 | pe2 = list_entry(lh, struct dm_snap_pending_exception, out_of_order_entry); | |
1505 | if (pe2->exception_sequence < pe->exception_sequence) | |
1506 | break; | |
1507 | } | |
1508 | list_add(&pe->out_of_order_entry, lh); | |
1509 | } | |
1da177e4 LT |
1510 | } |
1511 | ||
1512 | /* | |
1513 | * Dispatches the copy operation to kcopyd. | |
1514 | */ | |
028867ac | 1515 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1516 | { |
1517 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1518 | struct dm_io_region src, dest; |
1da177e4 LT |
1519 | struct block_device *bdev = s->origin->bdev; |
1520 | sector_t dev_size; | |
1521 | ||
1522 | dev_size = get_dev_size(bdev); | |
1523 | ||
1524 | src.bdev = bdev; | |
71fab00a | 1525 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1526 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1527 | |
fc56f6fb | 1528 | dest.bdev = s->cow->bdev; |
71fab00a | 1529 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1530 | dest.count = src.count; |
1531 | ||
1532 | /* Hand over to kcopyd */ | |
a2d2b034 | 1533 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1534 | } |
1535 | ||
a6e50b40 MP |
1536 | static void full_bio_end_io(struct bio *bio, int error) |
1537 | { | |
1538 | void *callback_data = bio->bi_private; | |
1539 | ||
1540 | dm_kcopyd_do_callback(callback_data, 0, error ? 1 : 0); | |
1541 | } | |
1542 | ||
1543 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1544 | struct bio *bio) | |
1545 | { | |
1546 | struct dm_snapshot *s = pe->snap; | |
1547 | void *callback_data; | |
1548 | ||
1549 | pe->full_bio = bio; | |
1550 | pe->full_bio_end_io = bio->bi_end_io; | |
1551 | pe->full_bio_private = bio->bi_private; | |
1552 | ||
1553 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, | |
1554 | copy_callback, pe); | |
1555 | ||
1556 | bio->bi_end_io = full_bio_end_io; | |
1557 | bio->bi_private = callback_data; | |
1558 | ||
1559 | generic_make_request(bio); | |
1560 | } | |
1561 | ||
2913808e MP |
1562 | static struct dm_snap_pending_exception * |
1563 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1564 | { | |
3510cb94 | 1565 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1566 | |
1567 | if (!e) | |
1568 | return NULL; | |
1569 | ||
1570 | return container_of(e, struct dm_snap_pending_exception, e); | |
1571 | } | |
1572 | ||
1da177e4 LT |
1573 | /* |
1574 | * Looks to see if this snapshot already has a pending exception | |
1575 | * for this chunk, otherwise it allocates a new one and inserts | |
1576 | * it into the pending table. | |
1577 | * | |
1578 | * NOTE: a write lock must be held on snap->lock before calling | |
1579 | * this. | |
1580 | */ | |
028867ac | 1581 | static struct dm_snap_pending_exception * |
c6621392 MP |
1582 | __find_pending_exception(struct dm_snapshot *s, |
1583 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1584 | { |
c6621392 | 1585 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1586 | |
2913808e MP |
1587 | pe2 = __lookup_pending_exception(s, chunk); |
1588 | if (pe2) { | |
76df1c65 | 1589 | free_pending_exception(pe); |
2913808e | 1590 | return pe2; |
1da177e4 LT |
1591 | } |
1592 | ||
76df1c65 AK |
1593 | pe->e.old_chunk = chunk; |
1594 | bio_list_init(&pe->origin_bios); | |
1595 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1596 | pe->started = 0; |
a6e50b40 | 1597 | pe->full_bio = NULL; |
76df1c65 | 1598 | |
493df71c | 1599 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1600 | free_pending_exception(pe); |
1601 | return NULL; | |
1602 | } | |
1603 | ||
230c83af MP |
1604 | pe->exception_sequence = s->exception_start_sequence++; |
1605 | ||
3510cb94 | 1606 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1607 | |
1da177e4 LT |
1608 | return pe; |
1609 | } | |
1610 | ||
1d4989c8 | 1611 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1612 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1613 | { |
fc56f6fb | 1614 | bio->bi_bdev = s->cow->bdev; |
4f024f37 KO |
1615 | bio->bi_iter.bi_sector = |
1616 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1617 | (chunk - e->old_chunk)) + | |
1618 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1619 | } |
1620 | ||
7de3ee57 | 1621 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1622 | { |
1d4989c8 | 1623 | struct dm_exception *e; |
028867ac | 1624 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1625 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1626 | chunk_t chunk; |
028867ac | 1627 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1628 | |
ee18026a MP |
1629 | init_tracked_chunk(bio); |
1630 | ||
d87f4c14 | 1631 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1632 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1633 | return DM_MAPIO_REMAPPED; |
1634 | } | |
1635 | ||
4f024f37 | 1636 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
1da177e4 LT |
1637 | |
1638 | /* Full snapshots are not usable */ | |
76df1c65 | 1639 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1640 | if (!s->valid) |
f6a80ea8 | 1641 | return -EIO; |
1da177e4 | 1642 | |
ba40a2aa AK |
1643 | /* FIXME: should only take write lock if we need |
1644 | * to copy an exception */ | |
1645 | down_write(&s->lock); | |
1646 | ||
1647 | if (!s->valid) { | |
1648 | r = -EIO; | |
1649 | goto out_unlock; | |
1650 | } | |
1651 | ||
1652 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1653 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1654 | if (e) { |
d74f81f8 | 1655 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1656 | goto out_unlock; |
1657 | } | |
1658 | ||
1da177e4 LT |
1659 | /* |
1660 | * Write to snapshot - higher level takes care of RW/RO | |
1661 | * flags so we should only get this if we are | |
1662 | * writeable. | |
1663 | */ | |
1664 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1665 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1666 | if (!pe) { |
c6621392 MP |
1667 | up_write(&s->lock); |
1668 | pe = alloc_pending_exception(s); | |
1669 | down_write(&s->lock); | |
1670 | ||
1671 | if (!s->valid) { | |
1672 | free_pending_exception(pe); | |
1673 | r = -EIO; | |
1674 | goto out_unlock; | |
1675 | } | |
1676 | ||
3510cb94 | 1677 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1678 | if (e) { |
1679 | free_pending_exception(pe); | |
1680 | remap_exception(s, e, bio, chunk); | |
1681 | goto out_unlock; | |
1682 | } | |
1683 | ||
c6621392 | 1684 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1685 | if (!pe) { |
1686 | __invalidate_snapshot(s, -ENOMEM); | |
1687 | r = -EIO; | |
1688 | goto out_unlock; | |
1689 | } | |
1da177e4 LT |
1690 | } |
1691 | ||
d74f81f8 | 1692 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 1693 | |
d2a7ad29 | 1694 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1695 | |
a6e50b40 | 1696 | if (!pe->started && |
4f024f37 KO |
1697 | bio->bi_iter.bi_size == |
1698 | (s->store->chunk_size << SECTOR_SHIFT)) { | |
a6e50b40 MP |
1699 | pe->started = 1; |
1700 | up_write(&s->lock); | |
1701 | start_full_bio(pe, bio); | |
1702 | goto out; | |
1703 | } | |
1704 | ||
1705 | bio_list_add(&pe->snapshot_bios, bio); | |
1706 | ||
76df1c65 AK |
1707 | if (!pe->started) { |
1708 | /* this is protected by snap->lock */ | |
1709 | pe->started = 1; | |
ba40a2aa | 1710 | up_write(&s->lock); |
76df1c65 | 1711 | start_copy(pe); |
ba40a2aa AK |
1712 | goto out; |
1713 | } | |
cd45daff | 1714 | } else { |
ba40a2aa | 1715 | bio->bi_bdev = s->origin->bdev; |
ee18026a | 1716 | track_chunk(s, bio, chunk); |
cd45daff | 1717 | } |
1da177e4 | 1718 | |
a2d2b034 | 1719 | out_unlock: |
ba40a2aa | 1720 | up_write(&s->lock); |
a2d2b034 | 1721 | out: |
1da177e4 LT |
1722 | return r; |
1723 | } | |
1724 | ||
3452c2a1 MP |
1725 | /* |
1726 | * A snapshot-merge target behaves like a combination of a snapshot | |
1727 | * target and a snapshot-origin target. It only generates new | |
1728 | * exceptions in other snapshots and not in the one that is being | |
1729 | * merged. | |
1730 | * | |
1731 | * For each chunk, if there is an existing exception, it is used to | |
1732 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1733 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1734 | * If merging is currently taking place on the chunk in question, the |
1735 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 1736 | */ |
7de3ee57 | 1737 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
1738 | { |
1739 | struct dm_exception *e; | |
1740 | struct dm_snapshot *s = ti->private; | |
1741 | int r = DM_MAPIO_REMAPPED; | |
1742 | chunk_t chunk; | |
1743 | ||
ee18026a MP |
1744 | init_tracked_chunk(bio); |
1745 | ||
d87f4c14 | 1746 | if (bio->bi_rw & REQ_FLUSH) { |
55a62eef | 1747 | if (!dm_bio_get_target_bio_nr(bio)) |
10b8106a MS |
1748 | bio->bi_bdev = s->origin->bdev; |
1749 | else | |
1750 | bio->bi_bdev = s->cow->bdev; | |
10b8106a MS |
1751 | return DM_MAPIO_REMAPPED; |
1752 | } | |
1753 | ||
4f024f37 | 1754 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 1755 | |
9fe86254 | 1756 | down_write(&s->lock); |
3452c2a1 | 1757 | |
d2fdb776 MP |
1758 | /* Full merging snapshots are redirected to the origin */ |
1759 | if (!s->valid) | |
1760 | goto redirect_to_origin; | |
3452c2a1 MP |
1761 | |
1762 | /* If the block is already remapped - use that */ | |
1763 | e = dm_lookup_exception(&s->complete, chunk); | |
1764 | if (e) { | |
9fe86254 MP |
1765 | /* Queue writes overlapping with chunks being merged */ |
1766 | if (bio_rw(bio) == WRITE && | |
1767 | chunk >= s->first_merging_chunk && | |
1768 | chunk < (s->first_merging_chunk + | |
1769 | s->num_merging_chunks)) { | |
1770 | bio->bi_bdev = s->origin->bdev; | |
1771 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1772 | r = DM_MAPIO_SUBMITTED; | |
1773 | goto out_unlock; | |
1774 | } | |
17aa0332 | 1775 | |
3452c2a1 | 1776 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1777 | |
1778 | if (bio_rw(bio) == WRITE) | |
ee18026a | 1779 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
1780 | goto out_unlock; |
1781 | } | |
1782 | ||
d2fdb776 | 1783 | redirect_to_origin: |
3452c2a1 MP |
1784 | bio->bi_bdev = s->origin->bdev; |
1785 | ||
1786 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1787 | up_write(&s->lock); |
3452c2a1 MP |
1788 | return do_origin(s->origin, bio); |
1789 | } | |
1790 | ||
1791 | out_unlock: | |
9fe86254 | 1792 | up_write(&s->lock); |
3452c2a1 MP |
1793 | |
1794 | return r; | |
1795 | } | |
1796 | ||
7de3ee57 | 1797 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, int error) |
cd45daff MP |
1798 | { |
1799 | struct dm_snapshot *s = ti->private; | |
cd45daff | 1800 | |
ee18026a MP |
1801 | if (is_bio_tracked(bio)) |
1802 | stop_tracking_chunk(s, bio); | |
cd45daff MP |
1803 | |
1804 | return 0; | |
1805 | } | |
1806 | ||
1e03f97e MP |
1807 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1808 | { | |
1809 | struct dm_snapshot *s = ti->private; | |
1810 | ||
1811 | stop_merge(s); | |
1812 | } | |
1813 | ||
c1f0c183 MS |
1814 | static int snapshot_preresume(struct dm_target *ti) |
1815 | { | |
1816 | int r = 0; | |
1817 | struct dm_snapshot *s = ti->private; | |
1818 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1819 | ||
1820 | down_read(&_origins_lock); | |
9d3b15c4 | 1821 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1822 | if (snap_src && snap_dest) { |
1823 | down_read(&snap_src->lock); | |
1824 | if (s == snap_src) { | |
1825 | DMERR("Unable to resume snapshot source until " | |
1826 | "handover completes."); | |
1827 | r = -EINVAL; | |
b83b2f29 | 1828 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
1829 | DMERR("Unable to perform snapshot handover until " |
1830 | "source is suspended."); | |
1831 | r = -EINVAL; | |
1832 | } | |
1833 | up_read(&snap_src->lock); | |
1834 | } | |
1835 | up_read(&_origins_lock); | |
1836 | ||
1837 | return r; | |
1838 | } | |
1839 | ||
1da177e4 LT |
1840 | static void snapshot_resume(struct dm_target *ti) |
1841 | { | |
028867ac | 1842 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1843 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1844 | ||
1845 | down_read(&_origins_lock); | |
9d3b15c4 | 1846 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1847 | if (snap_src && snap_dest) { |
1848 | down_write(&snap_src->lock); | |
1849 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1850 | __handover_exceptions(snap_src, snap_dest); | |
1851 | up_write(&snap_dest->lock); | |
1852 | up_write(&snap_src->lock); | |
1853 | } | |
1854 | up_read(&_origins_lock); | |
1855 | ||
1856 | /* Now we have correct chunk size, reregister */ | |
1857 | reregister_snapshot(s); | |
1da177e4 | 1858 | |
aa14edeb AK |
1859 | down_write(&s->lock); |
1860 | s->active = 1; | |
1861 | up_write(&s->lock); | |
1da177e4 LT |
1862 | } |
1863 | ||
542f9038 | 1864 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 1865 | { |
542f9038 | 1866 | uint32_t min_chunksize; |
1e03f97e MP |
1867 | |
1868 | down_read(&_origins_lock); | |
1869 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1870 | up_read(&_origins_lock); | |
1871 | ||
1872 | return min_chunksize; | |
1873 | } | |
1874 | ||
1875 | static void snapshot_merge_resume(struct dm_target *ti) | |
1876 | { | |
1877 | struct dm_snapshot *s = ti->private; | |
1878 | ||
1879 | /* | |
1880 | * Handover exceptions from existing snapshot. | |
1881 | */ | |
1882 | snapshot_resume(ti); | |
1883 | ||
1884 | /* | |
542f9038 | 1885 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 1886 | */ |
542f9038 | 1887 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
1888 | |
1889 | start_merge(s); | |
1890 | } | |
1891 | ||
fd7c092e MP |
1892 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
1893 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 1894 | { |
2e4a31df | 1895 | unsigned sz = 0; |
028867ac | 1896 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1897 | |
1898 | switch (type) { | |
1899 | case STATUSTYPE_INFO: | |
94e76572 MP |
1900 | |
1901 | down_write(&snap->lock); | |
1902 | ||
1da177e4 | 1903 | if (!snap->valid) |
2e4a31df | 1904 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1905 | else if (snap->merge_failed) |
1906 | DMEMIT("Merge failed"); | |
1da177e4 | 1907 | else { |
985903bb MS |
1908 | if (snap->store->type->usage) { |
1909 | sector_t total_sectors, sectors_allocated, | |
1910 | metadata_sectors; | |
1911 | snap->store->type->usage(snap->store, | |
1912 | &total_sectors, | |
1913 | §ors_allocated, | |
1914 | &metadata_sectors); | |
1915 | DMEMIT("%llu/%llu %llu", | |
1916 | (unsigned long long)sectors_allocated, | |
1917 | (unsigned long long)total_sectors, | |
1918 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1919 | } |
1920 | else | |
2e4a31df | 1921 | DMEMIT("Unknown"); |
1da177e4 | 1922 | } |
94e76572 MP |
1923 | |
1924 | up_write(&snap->lock); | |
1925 | ||
1da177e4 LT |
1926 | break; |
1927 | ||
1928 | case STATUSTYPE_TABLE: | |
1929 | /* | |
1930 | * kdevname returns a static pointer so we need | |
1931 | * to make private copies if the output is to | |
1932 | * make sense. | |
1933 | */ | |
fc56f6fb | 1934 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1935 | snap->store->type->status(snap->store, type, result + sz, |
1936 | maxlen - sz); | |
1da177e4 LT |
1937 | break; |
1938 | } | |
1da177e4 LT |
1939 | } |
1940 | ||
8811f46c MS |
1941 | static int snapshot_iterate_devices(struct dm_target *ti, |
1942 | iterate_devices_callout_fn fn, void *data) | |
1943 | { | |
1944 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
1945 | int r; |
1946 | ||
1947 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 1948 | |
1e5554c8 MP |
1949 | if (!r) |
1950 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
1951 | ||
1952 | return r; | |
8811f46c MS |
1953 | } |
1954 | ||
1955 | ||
1da177e4 LT |
1956 | /*----------------------------------------------------------------- |
1957 | * Origin methods | |
1958 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1959 | |
1960 | /* | |
1961 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1962 | * supplied bio was ignored. The caller may submit it immediately. | |
1963 | * (No remapping actually occurs as the origin is always a direct linear | |
1964 | * map.) | |
1965 | * | |
1966 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1967 | * and any supplied bio is added to a list to be submitted once all | |
1968 | * the necessary exceptions exist. | |
1969 | */ | |
1970 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1971 | struct bio *bio) | |
1da177e4 | 1972 | { |
515ad66c | 1973 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1974 | struct dm_snapshot *snap; |
1d4989c8 | 1975 | struct dm_exception *e; |
515ad66c MP |
1976 | struct dm_snap_pending_exception *pe; |
1977 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1978 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1979 | chunk_t chunk; |
1980 | ||
1981 | /* Do all the snapshots on this origin */ | |
1982 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
1983 | /* |
1984 | * Don't make new exceptions in a merging snapshot | |
1985 | * because it has effectively been deleted | |
1986 | */ | |
1987 | if (dm_target_is_snapshot_merge(snap->ti)) | |
1988 | continue; | |
1989 | ||
76df1c65 AK |
1990 | down_write(&snap->lock); |
1991 | ||
aa14edeb AK |
1992 | /* Only deal with valid and active snapshots */ |
1993 | if (!snap->valid || !snap->active) | |
76df1c65 | 1994 | goto next_snapshot; |
1da177e4 | 1995 | |
d5e404c1 | 1996 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1997 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1998 | goto next_snapshot; |
1da177e4 LT |
1999 | |
2000 | /* | |
2001 | * Remember, different snapshots can have | |
2002 | * different chunk sizes. | |
2003 | */ | |
9eaae8ff | 2004 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
2005 | |
2006 | /* | |
2007 | * Check exception table to see if block | |
2008 | * is already remapped in this snapshot | |
2009 | * and trigger an exception if not. | |
2010 | */ | |
3510cb94 | 2011 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
2012 | if (e) |
2013 | goto next_snapshot; | |
2014 | ||
2913808e | 2015 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2016 | if (!pe) { |
c6621392 MP |
2017 | up_write(&snap->lock); |
2018 | pe = alloc_pending_exception(snap); | |
2019 | down_write(&snap->lock); | |
2020 | ||
2021 | if (!snap->valid) { | |
2022 | free_pending_exception(pe); | |
2023 | goto next_snapshot; | |
2024 | } | |
2025 | ||
3510cb94 | 2026 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
2027 | if (e) { |
2028 | free_pending_exception(pe); | |
2029 | goto next_snapshot; | |
2030 | } | |
2031 | ||
c6621392 | 2032 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
2033 | if (!pe) { |
2034 | __invalidate_snapshot(snap, -ENOMEM); | |
2035 | goto next_snapshot; | |
2036 | } | |
76df1c65 AK |
2037 | } |
2038 | ||
515ad66c | 2039 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2040 | |
515ad66c MP |
2041 | /* |
2042 | * If an origin bio was supplied, queue it to wait for the | |
2043 | * completion of this exception, and start this one last, | |
2044 | * at the end of the function. | |
2045 | */ | |
2046 | if (bio) { | |
2047 | bio_list_add(&pe->origin_bios, bio); | |
2048 | bio = NULL; | |
76df1c65 | 2049 | |
515ad66c MP |
2050 | if (!pe->started) { |
2051 | pe->started = 1; | |
2052 | pe_to_start_last = pe; | |
2053 | } | |
76df1c65 AK |
2054 | } |
2055 | ||
2056 | if (!pe->started) { | |
2057 | pe->started = 1; | |
515ad66c | 2058 | pe_to_start_now = pe; |
1da177e4 LT |
2059 | } |
2060 | ||
a2d2b034 | 2061 | next_snapshot: |
1da177e4 | 2062 | up_write(&snap->lock); |
1da177e4 | 2063 | |
515ad66c MP |
2064 | if (pe_to_start_now) { |
2065 | start_copy(pe_to_start_now); | |
2066 | pe_to_start_now = NULL; | |
2067 | } | |
b4b610f6 AK |
2068 | } |
2069 | ||
1da177e4 | 2070 | /* |
515ad66c MP |
2071 | * Submit the exception against which the bio is queued last, |
2072 | * to give the other exceptions a head start. | |
1da177e4 | 2073 | */ |
515ad66c MP |
2074 | if (pe_to_start_last) |
2075 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2076 | |
2077 | return r; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * Called on a write from the origin driver. | |
2082 | */ | |
2083 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2084 | { | |
2085 | struct origin *o; | |
d2a7ad29 | 2086 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2087 | |
2088 | down_read(&_origins_lock); | |
2089 | o = __lookup_origin(origin->bdev); | |
2090 | if (o) | |
4f024f37 | 2091 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
1da177e4 LT |
2092 | up_read(&_origins_lock); |
2093 | ||
2094 | return r; | |
2095 | } | |
2096 | ||
73dfd078 MP |
2097 | /* |
2098 | * Trigger exceptions in all non-merging snapshots. | |
2099 | * | |
2100 | * The chunk size of the merging snapshot may be larger than the chunk | |
2101 | * size of some other snapshot so we may need to reallocate multiple | |
2102 | * chunks in other snapshots. | |
2103 | * | |
2104 | * We scan all the overlapping exceptions in the other snapshots. | |
2105 | * Returns 1 if anything was reallocated and must be waited for, | |
2106 | * otherwise returns 0. | |
2107 | * | |
2108 | * size must be a multiple of merging_snap's chunk_size. | |
2109 | */ | |
2110 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2111 | sector_t sector, unsigned size) | |
2112 | { | |
2113 | int must_wait = 0; | |
2114 | sector_t n; | |
2115 | struct origin *o; | |
2116 | ||
2117 | /* | |
542f9038 | 2118 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2119 | * by snapshot_merge_resume(). |
2120 | */ | |
2121 | down_read(&_origins_lock); | |
2122 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2123 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2124 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2125 | DM_MAPIO_SUBMITTED) | |
2126 | must_wait = 1; | |
2127 | up_read(&_origins_lock); | |
2128 | ||
2129 | return must_wait; | |
2130 | } | |
2131 | ||
1da177e4 LT |
2132 | /* |
2133 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2134 | */ | |
2135 | ||
599cdf3b MP |
2136 | struct dm_origin { |
2137 | struct dm_dev *dev; | |
298eaa89 | 2138 | unsigned split_boundary; |
599cdf3b MP |
2139 | }; |
2140 | ||
1da177e4 LT |
2141 | /* |
2142 | * Construct an origin mapping: <dev_path> | |
2143 | * The context for an origin is merely a 'struct dm_dev *' | |
2144 | * pointing to the real device. | |
2145 | */ | |
2146 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2147 | { | |
2148 | int r; | |
599cdf3b | 2149 | struct dm_origin *o; |
1da177e4 LT |
2150 | |
2151 | if (argc != 1) { | |
72d94861 | 2152 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2153 | return -EINVAL; |
2154 | } | |
2155 | ||
599cdf3b MP |
2156 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2157 | if (!o) { | |
2158 | ti->error = "Cannot allocate private origin structure"; | |
2159 | r = -ENOMEM; | |
2160 | goto bad_alloc; | |
2161 | } | |
2162 | ||
2163 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2164 | if (r) { |
2165 | ti->error = "Cannot get target device"; | |
599cdf3b | 2166 | goto bad_open; |
1da177e4 LT |
2167 | } |
2168 | ||
599cdf3b | 2169 | ti->private = o; |
55a62eef | 2170 | ti->num_flush_bios = 1; |
494b3ee7 | 2171 | |
1da177e4 | 2172 | return 0; |
599cdf3b MP |
2173 | |
2174 | bad_open: | |
2175 | kfree(o); | |
2176 | bad_alloc: | |
2177 | return r; | |
1da177e4 LT |
2178 | } |
2179 | ||
2180 | static void origin_dtr(struct dm_target *ti) | |
2181 | { | |
599cdf3b MP |
2182 | struct dm_origin *o = ti->private; |
2183 | dm_put_device(ti, o->dev); | |
2184 | kfree(o); | |
1da177e4 LT |
2185 | } |
2186 | ||
7de3ee57 | 2187 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2188 | { |
599cdf3b | 2189 | struct dm_origin *o = ti->private; |
298eaa89 | 2190 | unsigned available_sectors; |
1da177e4 | 2191 | |
599cdf3b | 2192 | bio->bi_bdev = o->dev->bdev; |
1da177e4 | 2193 | |
298eaa89 | 2194 | if (unlikely(bio->bi_rw & REQ_FLUSH)) |
494b3ee7 MP |
2195 | return DM_MAPIO_REMAPPED; |
2196 | ||
298eaa89 | 2197 | if (bio_rw(bio) != WRITE) |
494b3ee7 MP |
2198 | return DM_MAPIO_REMAPPED; |
2199 | ||
298eaa89 MP |
2200 | available_sectors = o->split_boundary - |
2201 | ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1)); | |
2202 | ||
2203 | if (bio_sectors(bio) > available_sectors) | |
2204 | dm_accept_partial_bio(bio, available_sectors); | |
2205 | ||
1da177e4 | 2206 | /* Only tell snapshots if this is a write */ |
298eaa89 | 2207 | return do_origin(o->dev, bio); |
1da177e4 LT |
2208 | } |
2209 | ||
1da177e4 | 2210 | /* |
542f9038 | 2211 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2212 | * chunk sizes. |
2213 | */ | |
2214 | static void origin_resume(struct dm_target *ti) | |
2215 | { | |
599cdf3b | 2216 | struct dm_origin *o = ti->private; |
1da177e4 | 2217 | |
298eaa89 | 2218 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
1da177e4 LT |
2219 | } |
2220 | ||
fd7c092e MP |
2221 | static void origin_status(struct dm_target *ti, status_type_t type, |
2222 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2223 | { |
599cdf3b | 2224 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2225 | |
2226 | switch (type) { | |
2227 | case STATUSTYPE_INFO: | |
2228 | result[0] = '\0'; | |
2229 | break; | |
2230 | ||
2231 | case STATUSTYPE_TABLE: | |
599cdf3b | 2232 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 LT |
2233 | break; |
2234 | } | |
1da177e4 LT |
2235 | } |
2236 | ||
b1d55528 MP |
2237 | static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
2238 | struct bio_vec *biovec, int max_size) | |
2239 | { | |
599cdf3b MP |
2240 | struct dm_origin *o = ti->private; |
2241 | struct request_queue *q = bdev_get_queue(o->dev->bdev); | |
b1d55528 MP |
2242 | |
2243 | if (!q->merge_bvec_fn) | |
2244 | return max_size; | |
2245 | ||
599cdf3b | 2246 | bvm->bi_bdev = o->dev->bdev; |
b1d55528 MP |
2247 | |
2248 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2249 | } | |
2250 | ||
8811f46c MS |
2251 | static int origin_iterate_devices(struct dm_target *ti, |
2252 | iterate_devices_callout_fn fn, void *data) | |
2253 | { | |
599cdf3b | 2254 | struct dm_origin *o = ti->private; |
8811f46c | 2255 | |
599cdf3b | 2256 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2257 | } |
2258 | ||
1da177e4 LT |
2259 | static struct target_type origin_target = { |
2260 | .name = "snapshot-origin", | |
fd7c092e | 2261 | .version = {1, 8, 1}, |
1da177e4 LT |
2262 | .module = THIS_MODULE, |
2263 | .ctr = origin_ctr, | |
2264 | .dtr = origin_dtr, | |
2265 | .map = origin_map, | |
2266 | .resume = origin_resume, | |
2267 | .status = origin_status, | |
b1d55528 | 2268 | .merge = origin_merge, |
8811f46c | 2269 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2270 | }; |
2271 | ||
2272 | static struct target_type snapshot_target = { | |
2273 | .name = "snapshot", | |
230c83af | 2274 | .version = {1, 12, 0}, |
1da177e4 LT |
2275 | .module = THIS_MODULE, |
2276 | .ctr = snapshot_ctr, | |
2277 | .dtr = snapshot_dtr, | |
2278 | .map = snapshot_map, | |
cd45daff | 2279 | .end_io = snapshot_end_io, |
c1f0c183 | 2280 | .preresume = snapshot_preresume, |
1da177e4 LT |
2281 | .resume = snapshot_resume, |
2282 | .status = snapshot_status, | |
8811f46c | 2283 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2284 | }; |
2285 | ||
d698aa45 MP |
2286 | static struct target_type merge_target = { |
2287 | .name = dm_snapshot_merge_target_name, | |
42bc954f | 2288 | .version = {1, 2, 0}, |
d698aa45 MP |
2289 | .module = THIS_MODULE, |
2290 | .ctr = snapshot_ctr, | |
2291 | .dtr = snapshot_dtr, | |
3452c2a1 | 2292 | .map = snapshot_merge_map, |
d698aa45 | 2293 | .end_io = snapshot_end_io, |
1e03f97e | 2294 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2295 | .preresume = snapshot_preresume, |
1e03f97e | 2296 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2297 | .status = snapshot_status, |
2298 | .iterate_devices = snapshot_iterate_devices, | |
2299 | }; | |
2300 | ||
1da177e4 LT |
2301 | static int __init dm_snapshot_init(void) |
2302 | { | |
2303 | int r; | |
2304 | ||
4db6bfe0 AK |
2305 | r = dm_exception_store_init(); |
2306 | if (r) { | |
2307 | DMERR("Failed to initialize exception stores"); | |
2308 | return r; | |
2309 | } | |
2310 | ||
1da177e4 | 2311 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2312 | if (r < 0) { |
1da177e4 | 2313 | DMERR("snapshot target register failed %d", r); |
034a186d | 2314 | goto bad_register_snapshot_target; |
1da177e4 LT |
2315 | } |
2316 | ||
2317 | r = dm_register_target(&origin_target); | |
2318 | if (r < 0) { | |
72d94861 | 2319 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2320 | goto bad_register_origin_target; |
2321 | } | |
2322 | ||
2323 | r = dm_register_target(&merge_target); | |
2324 | if (r < 0) { | |
2325 | DMERR("Merge target register failed %d", r); | |
2326 | goto bad_register_merge_target; | |
1da177e4 LT |
2327 | } |
2328 | ||
2329 | r = init_origin_hash(); | |
2330 | if (r) { | |
2331 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2332 | goto bad_origin_hash; |
1da177e4 LT |
2333 | } |
2334 | ||
1d4989c8 | 2335 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2336 | if (!exception_cache) { |
2337 | DMERR("Couldn't create exception cache."); | |
2338 | r = -ENOMEM; | |
d698aa45 | 2339 | goto bad_exception_cache; |
1da177e4 LT |
2340 | } |
2341 | ||
028867ac | 2342 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2343 | if (!pending_cache) { |
2344 | DMERR("Couldn't create pending cache."); | |
2345 | r = -ENOMEM; | |
d698aa45 | 2346 | goto bad_pending_cache; |
1da177e4 LT |
2347 | } |
2348 | ||
1da177e4 LT |
2349 | return 0; |
2350 | ||
d698aa45 | 2351 | bad_pending_cache: |
1da177e4 | 2352 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2353 | bad_exception_cache: |
1da177e4 | 2354 | exit_origin_hash(); |
d698aa45 MP |
2355 | bad_origin_hash: |
2356 | dm_unregister_target(&merge_target); | |
2357 | bad_register_merge_target: | |
1da177e4 | 2358 | dm_unregister_target(&origin_target); |
d698aa45 | 2359 | bad_register_origin_target: |
1da177e4 | 2360 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2361 | bad_register_snapshot_target: |
2362 | dm_exception_store_exit(); | |
d698aa45 | 2363 | |
1da177e4 LT |
2364 | return r; |
2365 | } | |
2366 | ||
2367 | static void __exit dm_snapshot_exit(void) | |
2368 | { | |
10d3bd09 MP |
2369 | dm_unregister_target(&snapshot_target); |
2370 | dm_unregister_target(&origin_target); | |
d698aa45 | 2371 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2372 | |
2373 | exit_origin_hash(); | |
1da177e4 LT |
2374 | kmem_cache_destroy(pending_cache); |
2375 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2376 | |
2377 | dm_exception_store_exit(); | |
1da177e4 LT |
2378 | } |
2379 | ||
2380 | /* Module hooks */ | |
2381 | module_init(dm_snapshot_init); | |
2382 | module_exit(dm_snapshot_exit); | |
2383 | ||
2384 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2385 | MODULE_AUTHOR("Joe Thornber"); | |
2386 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2387 | MODULE_ALIAS("dm-snapshot-origin"); |
2388 | MODULE_ALIAS("dm-snapshot-merge"); |