Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * dm-snapshot.c | |
3 | * | |
4 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. | |
5 | * | |
6 | * This file is released under the GPL. | |
7 | */ | |
8 | ||
9 | #include <linux/blkdev.h> | |
1da177e4 | 10 | #include <linux/device-mapper.h> |
90fa1527 | 11 | #include <linux/delay.h> |
1da177e4 LT |
12 | #include <linux/fs.h> |
13 | #include <linux/init.h> | |
14 | #include <linux/kdev_t.h> | |
15 | #include <linux/list.h> | |
16 | #include <linux/mempool.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
a765e20e | 21 | #include <linux/dm-kcopyd.h> |
ccc45ea8 | 22 | #include <linux/workqueue.h> |
1da177e4 | 23 | |
aea53d92 | 24 | #include "dm-exception-store.h" |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "snapshots" |
27 | ||
d698aa45 MP |
28 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
29 | ||
30 | #define dm_target_is_snapshot_merge(ti) \ | |
31 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
32 | ||
1da177e4 LT |
33 | /* |
34 | * The percentage increment we will wake up users at | |
35 | */ | |
36 | #define WAKE_UP_PERCENT 5 | |
37 | ||
38 | /* | |
39 | * kcopyd priority of snapshot operations | |
40 | */ | |
41 | #define SNAPSHOT_COPY_PRIORITY 2 | |
42 | ||
43 | /* | |
8ee2767a | 44 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 45 | */ |
8ee2767a | 46 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 47 | |
cd45daff MP |
48 | /* |
49 | * The size of the mempool used to track chunks in use. | |
50 | */ | |
51 | #define MIN_IOS 256 | |
52 | ||
ccc45ea8 JB |
53 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
54 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
55 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
56 | ||
191437a5 | 57 | struct dm_exception_table { |
ccc45ea8 JB |
58 | uint32_t hash_mask; |
59 | unsigned hash_shift; | |
60 | struct list_head *table; | |
61 | }; | |
62 | ||
63 | struct dm_snapshot { | |
64 | struct rw_semaphore lock; | |
65 | ||
66 | struct dm_dev *origin; | |
fc56f6fb MS |
67 | struct dm_dev *cow; |
68 | ||
69 | struct dm_target *ti; | |
ccc45ea8 JB |
70 | |
71 | /* List of snapshots per Origin */ | |
72 | struct list_head list; | |
73 | ||
d8ddb1cf MS |
74 | /* |
75 | * You can't use a snapshot if this is 0 (e.g. if full). | |
76 | * A snapshot-merge target never clears this. | |
77 | */ | |
ccc45ea8 JB |
78 | int valid; |
79 | ||
80 | /* Origin writes don't trigger exceptions until this is set */ | |
81 | int active; | |
82 | ||
c26655ca MS |
83 | /* Whether or not owning mapped_device is suspended */ |
84 | int suspended; | |
85 | ||
ccc45ea8 JB |
86 | atomic_t pending_exceptions_count; |
87 | ||
924e600d MS |
88 | mempool_t *pending_pool; |
89 | ||
191437a5 JB |
90 | struct dm_exception_table pending; |
91 | struct dm_exception_table complete; | |
ccc45ea8 JB |
92 | |
93 | /* | |
94 | * pe_lock protects all pending_exception operations and access | |
95 | * as well as the snapshot_bios list. | |
96 | */ | |
97 | spinlock_t pe_lock; | |
98 | ||
924e600d MS |
99 | /* Chunks with outstanding reads */ |
100 | spinlock_t tracked_chunk_lock; | |
101 | mempool_t *tracked_chunk_pool; | |
102 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
103 | ||
ccc45ea8 JB |
104 | /* The on disk metadata handler */ |
105 | struct dm_exception_store *store; | |
106 | ||
107 | struct dm_kcopyd_client *kcopyd_client; | |
108 | ||
109 | /* Queue of snapshot writes for ksnapd to flush */ | |
110 | struct bio_list queued_bios; | |
111 | struct work_struct queued_bios_work; | |
112 | ||
924e600d MS |
113 | /* Wait for events based on state_bits */ |
114 | unsigned long state_bits; | |
115 | ||
116 | /* Range of chunks currently being merged. */ | |
117 | chunk_t first_merging_chunk; | |
118 | int num_merging_chunks; | |
1e03f97e | 119 | |
d8ddb1cf MS |
120 | /* |
121 | * The merge operation failed if this flag is set. | |
122 | * Failure modes are handled as follows: | |
123 | * - I/O error reading the header | |
124 | * => don't load the target; abort. | |
125 | * - Header does not have "valid" flag set | |
126 | * => use the origin; forget about the snapshot. | |
127 | * - I/O error when reading exceptions | |
128 | * => don't load the target; abort. | |
129 | * (We can't use the intermediate origin state.) | |
130 | * - I/O error while merging | |
131 | * => stop merging; set merge_failed; process I/O normally. | |
132 | */ | |
133 | int merge_failed; | |
134 | ||
9fe86254 MP |
135 | /* |
136 | * Incoming bios that overlap with chunks being merged must wait | |
137 | * for them to be committed. | |
138 | */ | |
139 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
140 | }; |
141 | ||
1e03f97e MP |
142 | /* |
143 | * state_bits: | |
144 | * RUNNING_MERGE - Merge operation is in progress. | |
145 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
146 | * cleared afterwards. | |
147 | */ | |
148 | #define RUNNING_MERGE 0 | |
149 | #define SHUTDOWN_MERGE 1 | |
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 | ||
c642f9e0 | 163 | static struct workqueue_struct *ksnapd; |
c4028958 | 164 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 165 | |
ccc45ea8 JB |
166 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
167 | chunk_t chunk) | |
168 | { | |
169 | return chunk << store->chunk_shift; | |
170 | } | |
171 | ||
172 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
173 | { | |
174 | /* | |
175 | * There is only ever one instance of a particular block | |
176 | * device so we can compare pointers safely. | |
177 | */ | |
178 | return lhs == rhs; | |
179 | } | |
180 | ||
028867ac | 181 | struct dm_snap_pending_exception { |
1d4989c8 | 182 | struct dm_exception e; |
1da177e4 LT |
183 | |
184 | /* | |
185 | * Origin buffers waiting for this to complete are held | |
186 | * in a bio list | |
187 | */ | |
188 | struct bio_list origin_bios; | |
189 | struct bio_list snapshot_bios; | |
190 | ||
1da177e4 LT |
191 | /* Pointer back to snapshot context */ |
192 | struct dm_snapshot *snap; | |
193 | ||
194 | /* | |
195 | * 1 indicates the exception has already been sent to | |
196 | * kcopyd. | |
197 | */ | |
198 | int started; | |
199 | }; | |
200 | ||
201 | /* | |
202 | * Hash table mapping origin volumes to lists of snapshots and | |
203 | * a lock to protect it | |
204 | */ | |
e18b890b CL |
205 | static struct kmem_cache *exception_cache; |
206 | static struct kmem_cache *pending_cache; | |
1da177e4 | 207 | |
cd45daff MP |
208 | struct dm_snap_tracked_chunk { |
209 | struct hlist_node node; | |
210 | chunk_t chunk; | |
211 | }; | |
212 | ||
213 | static struct kmem_cache *tracked_chunk_cache; | |
214 | ||
215 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
216 | chunk_t chunk) | |
217 | { | |
218 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
219 | GFP_NOIO); | |
220 | unsigned long flags; | |
221 | ||
222 | c->chunk = chunk; | |
223 | ||
224 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
225 | hlist_add_head(&c->node, | |
226 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
227 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
228 | ||
229 | return c; | |
230 | } | |
231 | ||
232 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
233 | struct dm_snap_tracked_chunk *c) | |
234 | { | |
235 | unsigned long flags; | |
236 | ||
237 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
238 | hlist_del(&c->node); | |
239 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
240 | ||
241 | mempool_free(c, s->tracked_chunk_pool); | |
242 | } | |
243 | ||
a8d41b59 MP |
244 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
245 | { | |
246 | struct dm_snap_tracked_chunk *c; | |
247 | struct hlist_node *hn; | |
248 | int found = 0; | |
249 | ||
250 | spin_lock_irq(&s->tracked_chunk_lock); | |
251 | ||
252 | hlist_for_each_entry(c, hn, | |
253 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
254 | if (c->chunk == chunk) { | |
255 | found = 1; | |
256 | break; | |
257 | } | |
258 | } | |
259 | ||
260 | spin_unlock_irq(&s->tracked_chunk_lock); | |
261 | ||
262 | return found; | |
263 | } | |
264 | ||
615d1eb9 MS |
265 | /* |
266 | * This conflicting I/O is extremely improbable in the caller, | |
267 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
268 | */ | |
269 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
270 | { | |
271 | while (__chunk_is_tracked(s, chunk)) | |
272 | msleep(1); | |
273 | } | |
274 | ||
1da177e4 LT |
275 | /* |
276 | * One of these per registered origin, held in the snapshot_origins hash | |
277 | */ | |
278 | struct origin { | |
279 | /* The origin device */ | |
280 | struct block_device *bdev; | |
281 | ||
282 | struct list_head hash_list; | |
283 | ||
284 | /* List of snapshots for this origin */ | |
285 | struct list_head snapshots; | |
286 | }; | |
287 | ||
288 | /* | |
289 | * Size of the hash table for origin volumes. If we make this | |
290 | * the size of the minors list then it should be nearly perfect | |
291 | */ | |
292 | #define ORIGIN_HASH_SIZE 256 | |
293 | #define ORIGIN_MASK 0xFF | |
294 | static struct list_head *_origins; | |
295 | static struct rw_semaphore _origins_lock; | |
296 | ||
73dfd078 MP |
297 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
298 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
299 | static uint64_t _pending_exceptions_done_count; | |
300 | ||
1da177e4 LT |
301 | static int init_origin_hash(void) |
302 | { | |
303 | int i; | |
304 | ||
305 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
306 | GFP_KERNEL); | |
307 | if (!_origins) { | |
72d94861 | 308 | DMERR("unable to allocate memory"); |
1da177e4 LT |
309 | return -ENOMEM; |
310 | } | |
311 | ||
312 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
313 | INIT_LIST_HEAD(_origins + i); | |
314 | init_rwsem(&_origins_lock); | |
315 | ||
316 | return 0; | |
317 | } | |
318 | ||
319 | static void exit_origin_hash(void) | |
320 | { | |
321 | kfree(_origins); | |
322 | } | |
323 | ||
028867ac | 324 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
325 | { |
326 | return bdev->bd_dev & ORIGIN_MASK; | |
327 | } | |
328 | ||
329 | static struct origin *__lookup_origin(struct block_device *origin) | |
330 | { | |
331 | struct list_head *ol; | |
332 | struct origin *o; | |
333 | ||
334 | ol = &_origins[origin_hash(origin)]; | |
335 | list_for_each_entry (o, ol, hash_list) | |
336 | if (bdev_equal(o->bdev, origin)) | |
337 | return o; | |
338 | ||
339 | return NULL; | |
340 | } | |
341 | ||
342 | static void __insert_origin(struct origin *o) | |
343 | { | |
344 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
345 | list_add_tail(&o->hash_list, sl); | |
346 | } | |
347 | ||
c1f0c183 MS |
348 | /* |
349 | * _origins_lock must be held when calling this function. | |
350 | * Returns number of snapshots registered using the supplied cow device, plus: | |
351 | * snap_src - a snapshot suitable for use as a source of exception handover | |
352 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
353 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
354 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 355 | * |
9d3b15c4 | 356 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
357 | * 0: NULL, NULL - first new snapshot |
358 | * 1: snap_src, NULL - normal snapshot | |
359 | * 2: snap_src, snap_dest - waiting for handover | |
360 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
361 | * 1: NULL, snap_dest - source got destroyed without handover | |
362 | */ | |
363 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
364 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
365 | struct dm_snapshot **snap_dest, |
366 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
367 | { |
368 | struct dm_snapshot *s; | |
369 | struct origin *o; | |
370 | int count = 0; | |
371 | int active; | |
372 | ||
373 | o = __lookup_origin(snap->origin->bdev); | |
374 | if (!o) | |
375 | goto out; | |
376 | ||
377 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
378 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
379 | *snap_merge = s; | |
c1f0c183 MS |
380 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
381 | continue; | |
382 | ||
383 | down_read(&s->lock); | |
384 | active = s->active; | |
385 | up_read(&s->lock); | |
386 | ||
387 | if (active) { | |
388 | if (snap_src) | |
389 | *snap_src = s; | |
390 | } else if (snap_dest) | |
391 | *snap_dest = s; | |
392 | ||
393 | count++; | |
394 | } | |
395 | ||
396 | out: | |
397 | return count; | |
398 | } | |
399 | ||
400 | /* | |
401 | * On success, returns 1 if this snapshot is a handover destination, | |
402 | * otherwise returns 0. | |
403 | */ | |
404 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
405 | { | |
406 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 407 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
408 | |
409 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
410 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
411 | &snap_merge) == 2) || | |
c1f0c183 MS |
412 | snap_dest) { |
413 | snap->ti->error = "Snapshot cow pairing for exception " | |
414 | "table handover failed"; | |
415 | return -EINVAL; | |
416 | } | |
417 | ||
418 | /* | |
419 | * If no snap_src was found, snap cannot become a handover | |
420 | * destination. | |
421 | */ | |
422 | if (!snap_src) | |
423 | return 0; | |
424 | ||
9d3b15c4 MP |
425 | /* |
426 | * Non-snapshot-merge handover? | |
427 | */ | |
428 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
429 | return 1; | |
430 | ||
431 | /* | |
432 | * Do not allow more than one merging snapshot. | |
433 | */ | |
434 | if (snap_merge) { | |
435 | snap->ti->error = "A snapshot is already merging."; | |
436 | return -EINVAL; | |
437 | } | |
438 | ||
1e03f97e MP |
439 | if (!snap_src->store->type->prepare_merge || |
440 | !snap_src->store->type->commit_merge) { | |
441 | snap->ti->error = "Snapshot exception store does not " | |
442 | "support snapshot-merge."; | |
443 | return -EINVAL; | |
444 | } | |
445 | ||
c1f0c183 MS |
446 | return 1; |
447 | } | |
448 | ||
449 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
450 | { | |
451 | struct dm_snapshot *l; | |
452 | ||
453 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
454 | list_for_each_entry(l, &o->snapshots, list) | |
455 | if (l->store->chunk_size < s->store->chunk_size) | |
456 | break; | |
457 | list_add_tail(&s->list, &l->list); | |
458 | } | |
459 | ||
1da177e4 LT |
460 | /* |
461 | * Make a note of the snapshot and its origin so we can look it | |
462 | * up when the origin has a write on it. | |
c1f0c183 MS |
463 | * |
464 | * Also validate snapshot exception store handovers. | |
465 | * On success, returns 1 if this registration is a handover destination, | |
466 | * otherwise returns 0. | |
1da177e4 LT |
467 | */ |
468 | static int register_snapshot(struct dm_snapshot *snap) | |
469 | { | |
c1f0c183 | 470 | struct origin *o, *new_o = NULL; |
1da177e4 | 471 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 472 | int r = 0; |
1da177e4 | 473 | |
60c856c8 MP |
474 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
475 | if (!new_o) | |
476 | return -ENOMEM; | |
477 | ||
1da177e4 | 478 | down_write(&_origins_lock); |
1da177e4 | 479 | |
c1f0c183 MS |
480 | r = __validate_exception_handover(snap); |
481 | if (r < 0) { | |
482 | kfree(new_o); | |
483 | goto out; | |
484 | } | |
485 | ||
486 | o = __lookup_origin(bdev); | |
60c856c8 MP |
487 | if (o) |
488 | kfree(new_o); | |
489 | else { | |
1da177e4 | 490 | /* New origin */ |
60c856c8 | 491 | o = new_o; |
1da177e4 LT |
492 | |
493 | /* Initialise the struct */ | |
494 | INIT_LIST_HEAD(&o->snapshots); | |
495 | o->bdev = bdev; | |
496 | ||
497 | __insert_origin(o); | |
498 | } | |
499 | ||
c1f0c183 MS |
500 | __insert_snapshot(o, snap); |
501 | ||
502 | out: | |
503 | up_write(&_origins_lock); | |
504 | ||
505 | return r; | |
506 | } | |
507 | ||
508 | /* | |
509 | * Move snapshot to correct place in list according to chunk size. | |
510 | */ | |
511 | static void reregister_snapshot(struct dm_snapshot *s) | |
512 | { | |
513 | struct block_device *bdev = s->origin->bdev; | |
514 | ||
515 | down_write(&_origins_lock); | |
516 | ||
517 | list_del(&s->list); | |
518 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
519 | |
520 | up_write(&_origins_lock); | |
1da177e4 LT |
521 | } |
522 | ||
523 | static void unregister_snapshot(struct dm_snapshot *s) | |
524 | { | |
525 | struct origin *o; | |
526 | ||
527 | down_write(&_origins_lock); | |
528 | o = __lookup_origin(s->origin->bdev); | |
529 | ||
530 | list_del(&s->list); | |
c1f0c183 | 531 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
532 | list_del(&o->hash_list); |
533 | kfree(o); | |
534 | } | |
535 | ||
536 | up_write(&_origins_lock); | |
537 | } | |
538 | ||
539 | /* | |
540 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
541 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
542 | * some consecutive chunks to be grouped together. | |
1da177e4 | 543 | */ |
3510cb94 JB |
544 | static int dm_exception_table_init(struct dm_exception_table *et, |
545 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
546 | { |
547 | unsigned int i; | |
548 | ||
d74f81f8 | 549 | et->hash_shift = hash_shift; |
1da177e4 LT |
550 | et->hash_mask = size - 1; |
551 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
552 | if (!et->table) | |
553 | return -ENOMEM; | |
554 | ||
555 | for (i = 0; i < size; i++) | |
556 | INIT_LIST_HEAD(et->table + i); | |
557 | ||
558 | return 0; | |
559 | } | |
560 | ||
3510cb94 JB |
561 | static void dm_exception_table_exit(struct dm_exception_table *et, |
562 | struct kmem_cache *mem) | |
1da177e4 LT |
563 | { |
564 | struct list_head *slot; | |
1d4989c8 | 565 | struct dm_exception *ex, *next; |
1da177e4 LT |
566 | int i, size; |
567 | ||
568 | size = et->hash_mask + 1; | |
569 | for (i = 0; i < size; i++) { | |
570 | slot = et->table + i; | |
571 | ||
572 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
573 | kmem_cache_free(mem, ex); | |
574 | } | |
575 | ||
576 | vfree(et->table); | |
577 | } | |
578 | ||
191437a5 | 579 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 580 | { |
d74f81f8 | 581 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
582 | } |
583 | ||
3510cb94 | 584 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
585 | { |
586 | list_del(&e->hash_list); | |
587 | } | |
588 | ||
589 | /* | |
590 | * Return the exception data for a sector, or NULL if not | |
591 | * remapped. | |
592 | */ | |
3510cb94 JB |
593 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
594 | chunk_t chunk) | |
1da177e4 LT |
595 | { |
596 | struct list_head *slot; | |
1d4989c8 | 597 | struct dm_exception *e; |
1da177e4 LT |
598 | |
599 | slot = &et->table[exception_hash(et, chunk)]; | |
600 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
601 | if (chunk >= e->old_chunk && |
602 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
603 | return e; |
604 | ||
605 | return NULL; | |
606 | } | |
607 | ||
3510cb94 | 608 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 609 | { |
1d4989c8 | 610 | struct dm_exception *e; |
1da177e4 LT |
611 | |
612 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
613 | if (!e) | |
614 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
615 | ||
616 | return e; | |
617 | } | |
618 | ||
3510cb94 | 619 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
620 | { |
621 | kmem_cache_free(exception_cache, e); | |
622 | } | |
623 | ||
92e86812 | 624 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 625 | { |
92e86812 MP |
626 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
627 | GFP_NOIO); | |
628 | ||
879129d2 | 629 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
630 | pe->snap = s; |
631 | ||
632 | return pe; | |
1da177e4 LT |
633 | } |
634 | ||
028867ac | 635 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 636 | { |
879129d2 MP |
637 | struct dm_snapshot *s = pe->snap; |
638 | ||
639 | mempool_free(pe, s->pending_pool); | |
640 | smp_mb__before_atomic_dec(); | |
641 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
642 | } |
643 | ||
3510cb94 JB |
644 | static void dm_insert_exception(struct dm_exception_table *eh, |
645 | struct dm_exception *new_e) | |
d74f81f8 | 646 | { |
d74f81f8 | 647 | struct list_head *l; |
1d4989c8 | 648 | struct dm_exception *e = NULL; |
d74f81f8 MB |
649 | |
650 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
651 | ||
652 | /* Add immediately if this table doesn't support consecutive chunks */ | |
653 | if (!eh->hash_shift) | |
654 | goto out; | |
655 | ||
656 | /* List is ordered by old_chunk */ | |
657 | list_for_each_entry_reverse(e, l, hash_list) { | |
658 | /* Insert after an existing chunk? */ | |
659 | if (new_e->old_chunk == (e->old_chunk + | |
660 | dm_consecutive_chunk_count(e) + 1) && | |
661 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
662 | dm_consecutive_chunk_count(e) + 1)) { | |
663 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 664 | free_completed_exception(new_e); |
d74f81f8 MB |
665 | return; |
666 | } | |
667 | ||
668 | /* Insert before an existing chunk? */ | |
669 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
670 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
671 | dm_consecutive_chunk_count_inc(e); | |
672 | e->old_chunk--; | |
673 | e->new_chunk--; | |
3510cb94 | 674 | free_completed_exception(new_e); |
d74f81f8 MB |
675 | return; |
676 | } | |
677 | ||
678 | if (new_e->old_chunk > e->old_chunk) | |
679 | break; | |
680 | } | |
681 | ||
682 | out: | |
683 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
684 | } | |
685 | ||
a159c1ac JB |
686 | /* |
687 | * Callback used by the exception stores to load exceptions when | |
688 | * initialising. | |
689 | */ | |
690 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 691 | { |
a159c1ac | 692 | struct dm_snapshot *s = context; |
1d4989c8 | 693 | struct dm_exception *e; |
1da177e4 | 694 | |
3510cb94 | 695 | e = alloc_completed_exception(); |
1da177e4 LT |
696 | if (!e) |
697 | return -ENOMEM; | |
698 | ||
699 | e->old_chunk = old; | |
d74f81f8 MB |
700 | |
701 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 702 | e->new_chunk = new; |
d74f81f8 | 703 | |
3510cb94 | 704 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 705 | |
1da177e4 LT |
706 | return 0; |
707 | } | |
708 | ||
7e201b35 MP |
709 | /* |
710 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
711 | * Return zero if the origin has no snapshots. | |
712 | */ | |
713 | static sector_t __minimum_chunk_size(struct origin *o) | |
714 | { | |
715 | struct dm_snapshot *snap; | |
716 | unsigned chunk_size = 0; | |
717 | ||
718 | if (o) | |
719 | list_for_each_entry(snap, &o->snapshots, list) | |
720 | chunk_size = min_not_zero(chunk_size, | |
721 | snap->store->chunk_size); | |
722 | ||
723 | return chunk_size; | |
724 | } | |
725 | ||
1da177e4 LT |
726 | /* |
727 | * Hard coded magic. | |
728 | */ | |
729 | static int calc_max_buckets(void) | |
730 | { | |
731 | /* use a fixed size of 2MB */ | |
732 | unsigned long mem = 2 * 1024 * 1024; | |
733 | mem /= sizeof(struct list_head); | |
734 | ||
735 | return mem; | |
736 | } | |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Allocate room for a suitable hash table. | |
740 | */ | |
fee1998e | 741 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
742 | { |
743 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
744 | ||
745 | /* | |
746 | * Calculate based on the size of the original volume or | |
747 | * the COW volume... | |
748 | */ | |
fc56f6fb | 749 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
750 | origin_dev_size = get_dev_size(s->origin->bdev); |
751 | max_buckets = calc_max_buckets(); | |
752 | ||
fee1998e | 753 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
754 | hash_size = min(hash_size, max_buckets); |
755 | ||
8e87b9b8 MP |
756 | if (hash_size < 64) |
757 | hash_size = 64; | |
8defd830 | 758 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
759 | if (dm_exception_table_init(&s->complete, hash_size, |
760 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
761 | return -ENOMEM; |
762 | ||
763 | /* | |
764 | * Allocate hash table for in-flight exceptions | |
765 | * Make this smaller than the real hash table | |
766 | */ | |
767 | hash_size >>= 3; | |
768 | if (hash_size < 64) | |
769 | hash_size = 64; | |
770 | ||
3510cb94 JB |
771 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
772 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
773 | return -ENOMEM; |
774 | } | |
775 | ||
776 | return 0; | |
777 | } | |
778 | ||
1e03f97e MP |
779 | static void merge_shutdown(struct dm_snapshot *s) |
780 | { | |
781 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
782 | smp_mb__after_clear_bit(); | |
783 | wake_up_bit(&s->state_bits, RUNNING_MERGE); | |
784 | } | |
785 | ||
9fe86254 MP |
786 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
787 | { | |
788 | s->first_merging_chunk = 0; | |
789 | s->num_merging_chunks = 0; | |
790 | ||
791 | return bio_list_get(&s->bios_queued_during_merge); | |
792 | } | |
793 | ||
1e03f97e MP |
794 | /* |
795 | * Remove one chunk from the index of completed exceptions. | |
796 | */ | |
797 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
798 | chunk_t old_chunk) | |
799 | { | |
800 | struct dm_exception *e; | |
801 | ||
1e03f97e MP |
802 | e = dm_lookup_exception(&s->complete, old_chunk); |
803 | if (!e) { | |
804 | DMERR("Corruption detected: exception for block %llu is " | |
805 | "on disk but not in memory", | |
806 | (unsigned long long)old_chunk); | |
807 | return -EINVAL; | |
808 | } | |
809 | ||
810 | /* | |
811 | * If this is the only chunk using this exception, remove exception. | |
812 | */ | |
813 | if (!dm_consecutive_chunk_count(e)) { | |
814 | dm_remove_exception(e); | |
815 | free_completed_exception(e); | |
816 | return 0; | |
817 | } | |
818 | ||
819 | /* | |
820 | * The chunk may be either at the beginning or the end of a | |
821 | * group of consecutive chunks - never in the middle. We are | |
822 | * removing chunks in the opposite order to that in which they | |
823 | * were added, so this should always be true. | |
824 | * Decrement the consecutive chunk counter and adjust the | |
825 | * starting point if necessary. | |
826 | */ | |
827 | if (old_chunk == e->old_chunk) { | |
828 | e->old_chunk++; | |
829 | e->new_chunk++; | |
830 | } else if (old_chunk != e->old_chunk + | |
831 | dm_consecutive_chunk_count(e)) { | |
832 | DMERR("Attempt to merge block %llu from the " | |
833 | "middle of a chunk range [%llu - %llu]", | |
834 | (unsigned long long)old_chunk, | |
835 | (unsigned long long)e->old_chunk, | |
836 | (unsigned long long) | |
837 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
838 | return -EINVAL; | |
839 | } | |
840 | ||
841 | dm_consecutive_chunk_count_dec(e); | |
842 | ||
843 | return 0; | |
844 | } | |
845 | ||
9fe86254 MP |
846 | static void flush_bios(struct bio *bio); |
847 | ||
848 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 849 | { |
9fe86254 MP |
850 | struct bio *b = NULL; |
851 | int r; | |
852 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e MP |
853 | |
854 | down_write(&s->lock); | |
9fe86254 MP |
855 | |
856 | /* | |
857 | * Process chunks (and associated exceptions) in reverse order | |
858 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
859 | */ | |
860 | do { | |
861 | r = __remove_single_exception_chunk(s, old_chunk); | |
862 | if (r) | |
863 | goto out; | |
864 | } while (old_chunk-- > s->first_merging_chunk); | |
865 | ||
866 | b = __release_queued_bios_after_merge(s); | |
867 | ||
868 | out: | |
1e03f97e | 869 | up_write(&s->lock); |
9fe86254 MP |
870 | if (b) |
871 | flush_bios(b); | |
1e03f97e MP |
872 | |
873 | return r; | |
874 | } | |
875 | ||
73dfd078 MP |
876 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
877 | sector_t sector, unsigned chunk_size); | |
878 | ||
1e03f97e MP |
879 | static void merge_callback(int read_err, unsigned long write_err, |
880 | void *context); | |
881 | ||
73dfd078 MP |
882 | static uint64_t read_pending_exceptions_done_count(void) |
883 | { | |
884 | uint64_t pending_exceptions_done; | |
885 | ||
886 | spin_lock(&_pending_exceptions_done_spinlock); | |
887 | pending_exceptions_done = _pending_exceptions_done_count; | |
888 | spin_unlock(&_pending_exceptions_done_spinlock); | |
889 | ||
890 | return pending_exceptions_done; | |
891 | } | |
892 | ||
893 | static void increment_pending_exceptions_done_count(void) | |
894 | { | |
895 | spin_lock(&_pending_exceptions_done_spinlock); | |
896 | _pending_exceptions_done_count++; | |
897 | spin_unlock(&_pending_exceptions_done_spinlock); | |
898 | ||
899 | wake_up_all(&_pending_exceptions_done); | |
900 | } | |
901 | ||
1e03f97e MP |
902 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
903 | { | |
8a2d5286 | 904 | int i, linear_chunks; |
1e03f97e MP |
905 | chunk_t old_chunk, new_chunk; |
906 | struct dm_io_region src, dest; | |
8a2d5286 | 907 | sector_t io_size; |
73dfd078 | 908 | uint64_t previous_count; |
1e03f97e MP |
909 | |
910 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
911 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
912 | goto shut; | |
913 | ||
914 | /* | |
915 | * valid flag never changes during merge, so no lock required. | |
916 | */ | |
917 | if (!s->valid) { | |
918 | DMERR("Snapshot is invalid: can't merge"); | |
919 | goto shut; | |
920 | } | |
921 | ||
8a2d5286 MS |
922 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
923 | &new_chunk); | |
924 | if (linear_chunks <= 0) { | |
d8ddb1cf | 925 | if (linear_chunks < 0) { |
1e03f97e MP |
926 | DMERR("Read error in exception store: " |
927 | "shutting down merge"); | |
d8ddb1cf MS |
928 | down_write(&s->lock); |
929 | s->merge_failed = 1; | |
930 | up_write(&s->lock); | |
931 | } | |
1e03f97e MP |
932 | goto shut; |
933 | } | |
934 | ||
8a2d5286 MS |
935 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
936 | old_chunk = old_chunk + 1 - linear_chunks; | |
937 | new_chunk = new_chunk + 1 - linear_chunks; | |
938 | ||
939 | /* | |
940 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
941 | * from the exception store to the origin | |
942 | */ | |
943 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 944 | |
1e03f97e MP |
945 | dest.bdev = s->origin->bdev; |
946 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 947 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
948 | |
949 | src.bdev = s->cow->bdev; | |
950 | src.sector = chunk_to_sector(s->store, new_chunk); | |
951 | src.count = dest.count; | |
952 | ||
73dfd078 MP |
953 | /* |
954 | * Reallocate any exceptions needed in other snapshots then | |
955 | * wait for the pending exceptions to complete. | |
956 | * Each time any pending exception (globally on the system) | |
957 | * completes we are woken and repeat the process to find out | |
958 | * if we can proceed. While this may not seem a particularly | |
959 | * efficient algorithm, it is not expected to have any | |
960 | * significant impact on performance. | |
961 | */ | |
962 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 963 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
964 | wait_event(_pending_exceptions_done, |
965 | (read_pending_exceptions_done_count() != | |
966 | previous_count)); | |
967 | /* Retry after the wait, until all exceptions are done. */ | |
968 | previous_count = read_pending_exceptions_done_count(); | |
969 | } | |
970 | ||
9fe86254 MP |
971 | down_write(&s->lock); |
972 | s->first_merging_chunk = old_chunk; | |
8a2d5286 | 973 | s->num_merging_chunks = linear_chunks; |
9fe86254 MP |
974 | up_write(&s->lock); |
975 | ||
8a2d5286 MS |
976 | /* Wait until writes to all 'linear_chunks' drain */ |
977 | for (i = 0; i < linear_chunks; i++) | |
978 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 979 | |
1e03f97e MP |
980 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
981 | return; | |
982 | ||
983 | shut: | |
984 | merge_shutdown(s); | |
985 | } | |
986 | ||
9fe86254 MP |
987 | static void error_bios(struct bio *bio); |
988 | ||
1e03f97e MP |
989 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
990 | { | |
991 | struct dm_snapshot *s = context; | |
9fe86254 | 992 | struct bio *b = NULL; |
1e03f97e MP |
993 | |
994 | if (read_err || write_err) { | |
995 | if (read_err) | |
996 | DMERR("Read error: shutting down merge."); | |
997 | else | |
998 | DMERR("Write error: shutting down merge."); | |
999 | goto shut; | |
1000 | } | |
1001 | ||
9fe86254 MP |
1002 | if (s->store->type->commit_merge(s->store, |
1003 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1004 | DMERR("Write error in exception store: shutting down merge"); |
1005 | goto shut; | |
1006 | } | |
1007 | ||
9fe86254 MP |
1008 | if (remove_single_exception_chunk(s) < 0) |
1009 | goto shut; | |
1010 | ||
1e03f97e MP |
1011 | snapshot_merge_next_chunks(s); |
1012 | ||
1013 | return; | |
1014 | ||
1015 | shut: | |
9fe86254 | 1016 | down_write(&s->lock); |
d8ddb1cf | 1017 | s->merge_failed = 1; |
9fe86254 MP |
1018 | b = __release_queued_bios_after_merge(s); |
1019 | up_write(&s->lock); | |
1020 | error_bios(b); | |
1021 | ||
1e03f97e MP |
1022 | merge_shutdown(s); |
1023 | } | |
1024 | ||
1025 | static void start_merge(struct dm_snapshot *s) | |
1026 | { | |
1027 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1028 | snapshot_merge_next_chunks(s); | |
1029 | } | |
1030 | ||
1031 | static int wait_schedule(void *ptr) | |
1032 | { | |
1033 | schedule(); | |
1034 | ||
1035 | return 0; | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * Stop the merging process and wait until it finishes. | |
1040 | */ | |
1041 | static void stop_merge(struct dm_snapshot *s) | |
1042 | { | |
1043 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1044 | wait_on_bit(&s->state_bits, RUNNING_MERGE, wait_schedule, | |
1045 | TASK_UNINTERRUPTIBLE); | |
1046 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); | |
1047 | } | |
1048 | ||
1da177e4 LT |
1049 | /* |
1050 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
1051 | */ | |
1052 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1053 | { | |
1054 | struct dm_snapshot *s; | |
cd45daff | 1055 | int i; |
1da177e4 | 1056 | int r = -EINVAL; |
fc56f6fb | 1057 | char *origin_path, *cow_path; |
10b8106a MS |
1058 | unsigned args_used, num_flush_requests = 1; |
1059 | fmode_t origin_mode = FMODE_READ; | |
1da177e4 | 1060 | |
4c7e3bf4 | 1061 | if (argc != 4) { |
72d94861 | 1062 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 1063 | r = -EINVAL; |
fc56f6fb | 1064 | goto bad; |
1da177e4 LT |
1065 | } |
1066 | ||
10b8106a MS |
1067 | if (dm_target_is_snapshot_merge(ti)) { |
1068 | num_flush_requests = 2; | |
1069 | origin_mode = FMODE_WRITE; | |
1070 | } | |
1071 | ||
fc56f6fb MS |
1072 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
1073 | if (!s) { | |
1074 | ti->error = "Cannot allocate snapshot context private " | |
1075 | "structure"; | |
1076 | r = -ENOMEM; | |
1077 | goto bad; | |
1078 | } | |
1079 | ||
c2411045 MP |
1080 | origin_path = argv[0]; |
1081 | argv++; | |
1082 | argc--; | |
1083 | ||
1084 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1085 | if (r) { | |
1086 | ti->error = "Cannot get origin device"; | |
1087 | goto bad_origin; | |
1088 | } | |
1089 | ||
fc56f6fb MS |
1090 | cow_path = argv[0]; |
1091 | argv++; | |
1092 | argc--; | |
1093 | ||
8215d6ec | 1094 | r = dm_get_device(ti, cow_path, FMODE_READ | FMODE_WRITE, &s->cow); |
fc56f6fb MS |
1095 | if (r) { |
1096 | ti->error = "Cannot get COW device"; | |
1097 | goto bad_cow; | |
1098 | } | |
1099 | ||
1100 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1101 | if (r) { |
1102 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1103 | r = -EINVAL; |
fc56f6fb | 1104 | goto bad_store; |
1da177e4 LT |
1105 | } |
1106 | ||
fee1998e JB |
1107 | argv += args_used; |
1108 | argc -= args_used; | |
1109 | ||
fc56f6fb | 1110 | s->ti = ti; |
1da177e4 | 1111 | s->valid = 1; |
aa14edeb | 1112 | s->active = 0; |
c26655ca | 1113 | s->suspended = 0; |
879129d2 | 1114 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 1115 | init_rwsem(&s->lock); |
c1f0c183 | 1116 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1117 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1118 | s->state_bits = 0; |
d8ddb1cf | 1119 | s->merge_failed = 0; |
9fe86254 MP |
1120 | s->first_merging_chunk = 0; |
1121 | s->num_merging_chunks = 0; | |
1122 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1123 | |
1124 | /* Allocate hash table for COW data */ | |
fee1998e | 1125 | if (init_hash_tables(s)) { |
1da177e4 LT |
1126 | ti->error = "Unable to allocate hash table space"; |
1127 | r = -ENOMEM; | |
fee1998e | 1128 | goto bad_hash_tables; |
1da177e4 LT |
1129 | } |
1130 | ||
eb69aca5 | 1131 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
1132 | if (r) { |
1133 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 1134 | goto bad_kcopyd; |
1da177e4 LT |
1135 | } |
1136 | ||
92e86812 MP |
1137 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
1138 | if (!s->pending_pool) { | |
1139 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 1140 | goto bad_pending_pool; |
92e86812 MP |
1141 | } |
1142 | ||
cd45daff MP |
1143 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
1144 | tracked_chunk_cache); | |
1145 | if (!s->tracked_chunk_pool) { | |
1146 | ti->error = "Could not allocate tracked_chunk mempool for " | |
1147 | "tracking reads"; | |
92e86812 | 1148 | goto bad_tracked_chunk_pool; |
cd45daff MP |
1149 | } |
1150 | ||
1151 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1152 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1153 | ||
1154 | spin_lock_init(&s->tracked_chunk_lock); | |
1155 | ||
c1f0c183 MS |
1156 | bio_list_init(&s->queued_bios); |
1157 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); | |
1158 | ||
1159 | ti->private = s; | |
10b8106a | 1160 | ti->num_flush_requests = num_flush_requests; |
c1f0c183 MS |
1161 | |
1162 | /* Add snapshot to the list of snapshots for this origin */ | |
1163 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1164 | r = register_snapshot(s); | |
1165 | if (r == -ENOMEM) { | |
1166 | ti->error = "Snapshot origin struct allocation failed"; | |
1167 | goto bad_load_and_register; | |
1168 | } else if (r < 0) { | |
1169 | /* invalid handover, register_snapshot has set ti->error */ | |
1170 | goto bad_load_and_register; | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * Metadata must only be loaded into one table at once, so skip this | |
1175 | * if metadata will be handed over during resume. | |
1176 | * Chunk size will be set during the handover - set it to zero to | |
1177 | * ensure it's ignored. | |
1178 | */ | |
1179 | if (r > 0) { | |
1180 | s->store->chunk_size = 0; | |
1181 | return 0; | |
1182 | } | |
1183 | ||
493df71c JB |
1184 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1185 | (void *)s); | |
0764147b | 1186 | if (r < 0) { |
f9cea4f7 | 1187 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1188 | goto bad_read_metadata; |
0764147b MB |
1189 | } else if (r > 0) { |
1190 | s->valid = 0; | |
1191 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1192 | } |
aa14edeb | 1193 | |
3f2412dc MP |
1194 | if (!s->store->chunk_size) { |
1195 | ti->error = "Chunk size not set"; | |
c1f0c183 | 1196 | goto bad_read_metadata; |
1da177e4 | 1197 | } |
d0216849 | 1198 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
1199 | |
1200 | return 0; | |
1201 | ||
c1f0c183 MS |
1202 | bad_read_metadata: |
1203 | unregister_snapshot(s); | |
1204 | ||
fee1998e | 1205 | bad_load_and_register: |
cd45daff MP |
1206 | mempool_destroy(s->tracked_chunk_pool); |
1207 | ||
fee1998e | 1208 | bad_tracked_chunk_pool: |
92e86812 MP |
1209 | mempool_destroy(s->pending_pool); |
1210 | ||
fee1998e | 1211 | bad_pending_pool: |
eb69aca5 | 1212 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 1213 | |
fee1998e | 1214 | bad_kcopyd: |
3510cb94 JB |
1215 | dm_exception_table_exit(&s->pending, pending_cache); |
1216 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 1217 | |
fee1998e | 1218 | bad_hash_tables: |
fc56f6fb | 1219 | dm_exception_store_destroy(s->store); |
1da177e4 | 1220 | |
fc56f6fb MS |
1221 | bad_store: |
1222 | dm_put_device(ti, s->cow); | |
fee1998e | 1223 | |
fc56f6fb | 1224 | bad_cow: |
c2411045 MP |
1225 | dm_put_device(ti, s->origin); |
1226 | ||
1227 | bad_origin: | |
fc56f6fb MS |
1228 | kfree(s); |
1229 | ||
1230 | bad: | |
1da177e4 LT |
1231 | return r; |
1232 | } | |
1233 | ||
31c93a0c MB |
1234 | static void __free_exceptions(struct dm_snapshot *s) |
1235 | { | |
eb69aca5 | 1236 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1237 | s->kcopyd_client = NULL; |
1238 | ||
3510cb94 JB |
1239 | dm_exception_table_exit(&s->pending, pending_cache); |
1240 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1241 | } |
1242 | ||
c1f0c183 MS |
1243 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1244 | struct dm_snapshot *snap_dest) | |
1245 | { | |
1246 | union { | |
1247 | struct dm_exception_table table_swap; | |
1248 | struct dm_exception_store *store_swap; | |
1249 | } u; | |
1250 | ||
1251 | /* | |
1252 | * Swap all snapshot context information between the two instances. | |
1253 | */ | |
1254 | u.table_swap = snap_dest->complete; | |
1255 | snap_dest->complete = snap_src->complete; | |
1256 | snap_src->complete = u.table_swap; | |
1257 | ||
1258 | u.store_swap = snap_dest->store; | |
1259 | snap_dest->store = snap_src->store; | |
1260 | snap_src->store = u.store_swap; | |
1261 | ||
1262 | snap_dest->store->snap = snap_dest; | |
1263 | snap_src->store->snap = snap_src; | |
1264 | ||
1265 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
1266 | snap_dest->valid = snap_src->valid; | |
1267 | ||
1268 | /* | |
1269 | * Set source invalid to ensure it receives no further I/O. | |
1270 | */ | |
1271 | snap_src->valid = 0; | |
1272 | } | |
1273 | ||
1da177e4 LT |
1274 | static void snapshot_dtr(struct dm_target *ti) |
1275 | { | |
cd45daff MP |
1276 | #ifdef CONFIG_DM_DEBUG |
1277 | int i; | |
1278 | #endif | |
028867ac | 1279 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1280 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1281 | |
ca3a931f AK |
1282 | flush_workqueue(ksnapd); |
1283 | ||
c1f0c183 MS |
1284 | down_read(&_origins_lock); |
1285 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1286 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1287 | if (snap_src && snap_dest && (s == snap_src)) { |
1288 | down_write(&snap_dest->lock); | |
1289 | snap_dest->valid = 0; | |
1290 | up_write(&snap_dest->lock); | |
1291 | DMERR("Cancelling snapshot handover."); | |
1292 | } | |
1293 | up_read(&_origins_lock); | |
1294 | ||
1e03f97e MP |
1295 | if (dm_target_is_snapshot_merge(ti)) |
1296 | stop_merge(s); | |
1297 | ||
138728dc AK |
1298 | /* Prevent further origin writes from using this snapshot. */ |
1299 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1300 | unregister_snapshot(s); |
1301 | ||
879129d2 | 1302 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1303 | msleep(1); |
879129d2 MP |
1304 | /* |
1305 | * Ensure instructions in mempool_destroy aren't reordered | |
1306 | * before atomic_read. | |
1307 | */ | |
1308 | smp_mb(); | |
1309 | ||
cd45daff MP |
1310 | #ifdef CONFIG_DM_DEBUG |
1311 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1312 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1313 | #endif | |
1314 | ||
1315 | mempool_destroy(s->tracked_chunk_pool); | |
1316 | ||
31c93a0c | 1317 | __free_exceptions(s); |
1da177e4 | 1318 | |
92e86812 MP |
1319 | mempool_destroy(s->pending_pool); |
1320 | ||
fee1998e | 1321 | dm_exception_store_destroy(s->store); |
138728dc | 1322 | |
fc56f6fb MS |
1323 | dm_put_device(ti, s->cow); |
1324 | ||
c2411045 MP |
1325 | dm_put_device(ti, s->origin); |
1326 | ||
1da177e4 LT |
1327 | kfree(s); |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Flush a list of buffers. | |
1332 | */ | |
1333 | static void flush_bios(struct bio *bio) | |
1334 | { | |
1335 | struct bio *n; | |
1336 | ||
1337 | while (bio) { | |
1338 | n = bio->bi_next; | |
1339 | bio->bi_next = NULL; | |
1340 | generic_make_request(bio); | |
1341 | bio = n; | |
1342 | } | |
1343 | } | |
1344 | ||
c4028958 | 1345 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 1346 | { |
c4028958 DH |
1347 | struct dm_snapshot *s = |
1348 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
1349 | struct bio *queued_bios; |
1350 | unsigned long flags; | |
1351 | ||
1352 | spin_lock_irqsave(&s->pe_lock, flags); | |
1353 | queued_bios = bio_list_get(&s->queued_bios); | |
1354 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
1355 | ||
1356 | flush_bios(queued_bios); | |
1357 | } | |
1358 | ||
515ad66c MP |
1359 | static int do_origin(struct dm_dev *origin, struct bio *bio); |
1360 | ||
1361 | /* | |
1362 | * Flush a list of buffers. | |
1363 | */ | |
1364 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1365 | { | |
1366 | struct bio *n; | |
1367 | int r; | |
1368 | ||
1369 | while (bio) { | |
1370 | n = bio->bi_next; | |
1371 | bio->bi_next = NULL; | |
1372 | r = do_origin(s->origin, bio); | |
1373 | if (r == DM_MAPIO_REMAPPED) | |
1374 | generic_make_request(bio); | |
1375 | bio = n; | |
1376 | } | |
1377 | } | |
1378 | ||
1da177e4 LT |
1379 | /* |
1380 | * Error a list of buffers. | |
1381 | */ | |
1382 | static void error_bios(struct bio *bio) | |
1383 | { | |
1384 | struct bio *n; | |
1385 | ||
1386 | while (bio) { | |
1387 | n = bio->bi_next; | |
1388 | bio->bi_next = NULL; | |
6712ecf8 | 1389 | bio_io_error(bio); |
1da177e4 LT |
1390 | bio = n; |
1391 | } | |
1392 | } | |
1393 | ||
695368ac | 1394 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1395 | { |
1396 | if (!s->valid) | |
1397 | return; | |
1398 | ||
1399 | if (err == -EIO) | |
1400 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1401 | else if (err == -ENOMEM) | |
1402 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1403 | ||
493df71c JB |
1404 | if (s->store->type->drop_snapshot) |
1405 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1406 | |
1407 | s->valid = 0; | |
1408 | ||
fc56f6fb | 1409 | dm_table_event(s->ti->table); |
76df1c65 AK |
1410 | } |
1411 | ||
028867ac | 1412 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1413 | { |
1d4989c8 | 1414 | struct dm_exception *e; |
1da177e4 | 1415 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1416 | struct bio *origin_bios = NULL; |
1417 | struct bio *snapshot_bios = NULL; | |
1418 | int error = 0; | |
1da177e4 | 1419 | |
76df1c65 AK |
1420 | if (!success) { |
1421 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1422 | down_write(&s->lock); |
695368ac | 1423 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1424 | error = 1; |
76df1c65 AK |
1425 | goto out; |
1426 | } | |
1427 | ||
3510cb94 | 1428 | e = alloc_completed_exception(); |
76df1c65 | 1429 | if (!e) { |
1da177e4 | 1430 | down_write(&s->lock); |
695368ac | 1431 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1432 | error = 1; |
76df1c65 AK |
1433 | goto out; |
1434 | } | |
1435 | *e = pe->e; | |
1da177e4 | 1436 | |
76df1c65 AK |
1437 | down_write(&s->lock); |
1438 | if (!s->valid) { | |
3510cb94 | 1439 | free_completed_exception(e); |
9d493fa8 | 1440 | error = 1; |
76df1c65 | 1441 | goto out; |
1da177e4 LT |
1442 | } |
1443 | ||
615d1eb9 MS |
1444 | /* Check for conflicting reads */ |
1445 | __check_for_conflicting_io(s, pe->e.old_chunk); | |
a8d41b59 | 1446 | |
9d493fa8 AK |
1447 | /* |
1448 | * Add a proper exception, and remove the | |
1449 | * in-flight exception from the list. | |
1450 | */ | |
3510cb94 | 1451 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1452 | |
1da177e4 | 1453 | out: |
3510cb94 | 1454 | dm_remove_exception(&pe->e); |
9d493fa8 | 1455 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c MP |
1456 | origin_bios = bio_list_get(&pe->origin_bios); |
1457 | free_pending_exception(pe); | |
1da177e4 | 1458 | |
73dfd078 MP |
1459 | increment_pending_exceptions_done_count(); |
1460 | ||
9d493fa8 AK |
1461 | up_write(&s->lock); |
1462 | ||
1463 | /* Submit any pending write bios */ | |
1464 | if (error) | |
1465 | error_bios(snapshot_bios); | |
1466 | else | |
1467 | flush_bios(snapshot_bios); | |
1468 | ||
515ad66c | 1469 | retry_origin_bios(s, origin_bios); |
1da177e4 LT |
1470 | } |
1471 | ||
1472 | static void commit_callback(void *context, int success) | |
1473 | { | |
028867ac AK |
1474 | struct dm_snap_pending_exception *pe = context; |
1475 | ||
1da177e4 LT |
1476 | pending_complete(pe, success); |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Called when the copy I/O has finished. kcopyd actually runs | |
1481 | * this code so don't block. | |
1482 | */ | |
4cdc1d1f | 1483 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1484 | { |
028867ac | 1485 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1486 | struct dm_snapshot *s = pe->snap; |
1487 | ||
1488 | if (read_err || write_err) | |
1489 | pending_complete(pe, 0); | |
1490 | ||
1491 | else | |
1492 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1493 | s->store->type->commit_exception(s->store, &pe->e, |
1494 | commit_callback, pe); | |
1da177e4 LT |
1495 | } |
1496 | ||
1497 | /* | |
1498 | * Dispatches the copy operation to kcopyd. | |
1499 | */ | |
028867ac | 1500 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1501 | { |
1502 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1503 | struct dm_io_region src, dest; |
1da177e4 LT |
1504 | struct block_device *bdev = s->origin->bdev; |
1505 | sector_t dev_size; | |
1506 | ||
1507 | dev_size = get_dev_size(bdev); | |
1508 | ||
1509 | src.bdev = bdev; | |
71fab00a | 1510 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1511 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1512 | |
fc56f6fb | 1513 | dest.bdev = s->cow->bdev; |
71fab00a | 1514 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1515 | dest.count = src.count; |
1516 | ||
1517 | /* Hand over to kcopyd */ | |
eb69aca5 | 1518 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1519 | &src, 1, &dest, 0, copy_callback, pe); |
1520 | } | |
1521 | ||
2913808e MP |
1522 | static struct dm_snap_pending_exception * |
1523 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1524 | { | |
3510cb94 | 1525 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1526 | |
1527 | if (!e) | |
1528 | return NULL; | |
1529 | ||
1530 | return container_of(e, struct dm_snap_pending_exception, e); | |
1531 | } | |
1532 | ||
1da177e4 LT |
1533 | /* |
1534 | * Looks to see if this snapshot already has a pending exception | |
1535 | * for this chunk, otherwise it allocates a new one and inserts | |
1536 | * it into the pending table. | |
1537 | * | |
1538 | * NOTE: a write lock must be held on snap->lock before calling | |
1539 | * this. | |
1540 | */ | |
028867ac | 1541 | static struct dm_snap_pending_exception * |
c6621392 MP |
1542 | __find_pending_exception(struct dm_snapshot *s, |
1543 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1544 | { |
c6621392 | 1545 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1546 | |
2913808e MP |
1547 | pe2 = __lookup_pending_exception(s, chunk); |
1548 | if (pe2) { | |
76df1c65 | 1549 | free_pending_exception(pe); |
2913808e | 1550 | return pe2; |
1da177e4 LT |
1551 | } |
1552 | ||
76df1c65 AK |
1553 | pe->e.old_chunk = chunk; |
1554 | bio_list_init(&pe->origin_bios); | |
1555 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 AK |
1556 | pe->started = 0; |
1557 | ||
493df71c | 1558 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1559 | free_pending_exception(pe); |
1560 | return NULL; | |
1561 | } | |
1562 | ||
3510cb94 | 1563 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1564 | |
1da177e4 LT |
1565 | return pe; |
1566 | } | |
1567 | ||
1d4989c8 | 1568 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1569 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1570 | { |
fc56f6fb | 1571 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1572 | bio->bi_sector = chunk_to_sector(s->store, |
1573 | dm_chunk_number(e->new_chunk) + | |
1574 | (chunk - e->old_chunk)) + | |
1575 | (bio->bi_sector & | |
1576 | s->store->chunk_mask); | |
1da177e4 LT |
1577 | } |
1578 | ||
1579 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1580 | union map_info *map_context) | |
1581 | { | |
1d4989c8 | 1582 | struct dm_exception *e; |
028867ac | 1583 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1584 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1585 | chunk_t chunk; |
028867ac | 1586 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1587 | |
d87f4c14 | 1588 | if (bio->bi_rw & REQ_FLUSH) { |
fc56f6fb | 1589 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1590 | return DM_MAPIO_REMAPPED; |
1591 | } | |
1592 | ||
71fab00a | 1593 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1594 | |
1595 | /* Full snapshots are not usable */ | |
76df1c65 | 1596 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1597 | if (!s->valid) |
f6a80ea8 | 1598 | return -EIO; |
1da177e4 | 1599 | |
ba40a2aa AK |
1600 | /* FIXME: should only take write lock if we need |
1601 | * to copy an exception */ | |
1602 | down_write(&s->lock); | |
1603 | ||
1604 | if (!s->valid) { | |
1605 | r = -EIO; | |
1606 | goto out_unlock; | |
1607 | } | |
1608 | ||
1609 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1610 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1611 | if (e) { |
d74f81f8 | 1612 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1613 | goto out_unlock; |
1614 | } | |
1615 | ||
1da177e4 LT |
1616 | /* |
1617 | * Write to snapshot - higher level takes care of RW/RO | |
1618 | * flags so we should only get this if we are | |
1619 | * writeable. | |
1620 | */ | |
1621 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1622 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1623 | if (!pe) { |
c6621392 MP |
1624 | up_write(&s->lock); |
1625 | pe = alloc_pending_exception(s); | |
1626 | down_write(&s->lock); | |
1627 | ||
1628 | if (!s->valid) { | |
1629 | free_pending_exception(pe); | |
1630 | r = -EIO; | |
1631 | goto out_unlock; | |
1632 | } | |
1633 | ||
3510cb94 | 1634 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1635 | if (e) { |
1636 | free_pending_exception(pe); | |
1637 | remap_exception(s, e, bio, chunk); | |
1638 | goto out_unlock; | |
1639 | } | |
1640 | ||
c6621392 | 1641 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1642 | if (!pe) { |
1643 | __invalidate_snapshot(s, -ENOMEM); | |
1644 | r = -EIO; | |
1645 | goto out_unlock; | |
1646 | } | |
1da177e4 LT |
1647 | } |
1648 | ||
d74f81f8 | 1649 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1650 | bio_list_add(&pe->snapshot_bios, bio); |
1651 | ||
d2a7ad29 | 1652 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1653 | |
76df1c65 AK |
1654 | if (!pe->started) { |
1655 | /* this is protected by snap->lock */ | |
1656 | pe->started = 1; | |
ba40a2aa | 1657 | up_write(&s->lock); |
76df1c65 | 1658 | start_copy(pe); |
ba40a2aa AK |
1659 | goto out; |
1660 | } | |
cd45daff | 1661 | } else { |
ba40a2aa | 1662 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1663 | map_context->ptr = track_chunk(s, chunk); |
1664 | } | |
1da177e4 | 1665 | |
ba40a2aa AK |
1666 | out_unlock: |
1667 | up_write(&s->lock); | |
1668 | out: | |
1da177e4 LT |
1669 | return r; |
1670 | } | |
1671 | ||
3452c2a1 MP |
1672 | /* |
1673 | * A snapshot-merge target behaves like a combination of a snapshot | |
1674 | * target and a snapshot-origin target. It only generates new | |
1675 | * exceptions in other snapshots and not in the one that is being | |
1676 | * merged. | |
1677 | * | |
1678 | * For each chunk, if there is an existing exception, it is used to | |
1679 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
1680 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
1681 | * If merging is currently taking place on the chunk in question, the |
1682 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 MP |
1683 | */ |
1684 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio, | |
1685 | union map_info *map_context) | |
1686 | { | |
1687 | struct dm_exception *e; | |
1688 | struct dm_snapshot *s = ti->private; | |
1689 | int r = DM_MAPIO_REMAPPED; | |
1690 | chunk_t chunk; | |
1691 | ||
d87f4c14 | 1692 | if (bio->bi_rw & REQ_FLUSH) { |
57cba5d3 | 1693 | if (!map_context->target_request_nr) |
10b8106a MS |
1694 | bio->bi_bdev = s->origin->bdev; |
1695 | else | |
1696 | bio->bi_bdev = s->cow->bdev; | |
1697 | map_context->ptr = NULL; | |
1698 | return DM_MAPIO_REMAPPED; | |
1699 | } | |
1700 | ||
3452c2a1 MP |
1701 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1702 | ||
9fe86254 | 1703 | down_write(&s->lock); |
3452c2a1 | 1704 | |
d2fdb776 MP |
1705 | /* Full merging snapshots are redirected to the origin */ |
1706 | if (!s->valid) | |
1707 | goto redirect_to_origin; | |
3452c2a1 MP |
1708 | |
1709 | /* If the block is already remapped - use that */ | |
1710 | e = dm_lookup_exception(&s->complete, chunk); | |
1711 | if (e) { | |
9fe86254 MP |
1712 | /* Queue writes overlapping with chunks being merged */ |
1713 | if (bio_rw(bio) == WRITE && | |
1714 | chunk >= s->first_merging_chunk && | |
1715 | chunk < (s->first_merging_chunk + | |
1716 | s->num_merging_chunks)) { | |
1717 | bio->bi_bdev = s->origin->bdev; | |
1718 | bio_list_add(&s->bios_queued_during_merge, bio); | |
1719 | r = DM_MAPIO_SUBMITTED; | |
1720 | goto out_unlock; | |
1721 | } | |
17aa0332 | 1722 | |
3452c2a1 | 1723 | remap_exception(s, e, bio, chunk); |
17aa0332 MP |
1724 | |
1725 | if (bio_rw(bio) == WRITE) | |
1726 | map_context->ptr = track_chunk(s, chunk); | |
3452c2a1 MP |
1727 | goto out_unlock; |
1728 | } | |
1729 | ||
d2fdb776 | 1730 | redirect_to_origin: |
3452c2a1 MP |
1731 | bio->bi_bdev = s->origin->bdev; |
1732 | ||
1733 | if (bio_rw(bio) == WRITE) { | |
9fe86254 | 1734 | up_write(&s->lock); |
3452c2a1 MP |
1735 | return do_origin(s->origin, bio); |
1736 | } | |
1737 | ||
1738 | out_unlock: | |
9fe86254 | 1739 | up_write(&s->lock); |
3452c2a1 MP |
1740 | |
1741 | return r; | |
1742 | } | |
1743 | ||
cd45daff MP |
1744 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1745 | int error, union map_info *map_context) | |
1746 | { | |
1747 | struct dm_snapshot *s = ti->private; | |
1748 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1749 | ||
1750 | if (c) | |
1751 | stop_tracking_chunk(s, c); | |
1752 | ||
1753 | return 0; | |
1754 | } | |
1755 | ||
1e03f97e MP |
1756 | static void snapshot_merge_presuspend(struct dm_target *ti) |
1757 | { | |
1758 | struct dm_snapshot *s = ti->private; | |
1759 | ||
1760 | stop_merge(s); | |
1761 | } | |
1762 | ||
c26655ca MS |
1763 | static void snapshot_postsuspend(struct dm_target *ti) |
1764 | { | |
1765 | struct dm_snapshot *s = ti->private; | |
1766 | ||
1767 | down_write(&s->lock); | |
1768 | s->suspended = 1; | |
1769 | up_write(&s->lock); | |
1770 | } | |
1771 | ||
c1f0c183 MS |
1772 | static int snapshot_preresume(struct dm_target *ti) |
1773 | { | |
1774 | int r = 0; | |
1775 | struct dm_snapshot *s = ti->private; | |
1776 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1777 | ||
1778 | down_read(&_origins_lock); | |
9d3b15c4 | 1779 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1780 | if (snap_src && snap_dest) { |
1781 | down_read(&snap_src->lock); | |
1782 | if (s == snap_src) { | |
1783 | DMERR("Unable to resume snapshot source until " | |
1784 | "handover completes."); | |
1785 | r = -EINVAL; | |
1786 | } else if (!snap_src->suspended) { | |
1787 | DMERR("Unable to perform snapshot handover until " | |
1788 | "source is suspended."); | |
1789 | r = -EINVAL; | |
1790 | } | |
1791 | up_read(&snap_src->lock); | |
1792 | } | |
1793 | up_read(&_origins_lock); | |
1794 | ||
1795 | return r; | |
1796 | } | |
1797 | ||
1da177e4 LT |
1798 | static void snapshot_resume(struct dm_target *ti) |
1799 | { | |
028867ac | 1800 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1801 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1802 | ||
1803 | down_read(&_origins_lock); | |
9d3b15c4 | 1804 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 MS |
1805 | if (snap_src && snap_dest) { |
1806 | down_write(&snap_src->lock); | |
1807 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1808 | __handover_exceptions(snap_src, snap_dest); | |
1809 | up_write(&snap_dest->lock); | |
1810 | up_write(&snap_src->lock); | |
1811 | } | |
1812 | up_read(&_origins_lock); | |
1813 | ||
1814 | /* Now we have correct chunk size, reregister */ | |
1815 | reregister_snapshot(s); | |
1da177e4 | 1816 | |
aa14edeb AK |
1817 | down_write(&s->lock); |
1818 | s->active = 1; | |
c26655ca | 1819 | s->suspended = 0; |
aa14edeb | 1820 | up_write(&s->lock); |
1da177e4 LT |
1821 | } |
1822 | ||
1e03f97e MP |
1823 | static sector_t get_origin_minimum_chunksize(struct block_device *bdev) |
1824 | { | |
1825 | sector_t min_chunksize; | |
1826 | ||
1827 | down_read(&_origins_lock); | |
1828 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
1829 | up_read(&_origins_lock); | |
1830 | ||
1831 | return min_chunksize; | |
1832 | } | |
1833 | ||
1834 | static void snapshot_merge_resume(struct dm_target *ti) | |
1835 | { | |
1836 | struct dm_snapshot *s = ti->private; | |
1837 | ||
1838 | /* | |
1839 | * Handover exceptions from existing snapshot. | |
1840 | */ | |
1841 | snapshot_resume(ti); | |
1842 | ||
1843 | /* | |
1844 | * snapshot-merge acts as an origin, so set ti->split_io | |
1845 | */ | |
1846 | ti->split_io = get_origin_minimum_chunksize(s->origin->bdev); | |
1847 | ||
1848 | start_merge(s); | |
1849 | } | |
1850 | ||
1da177e4 LT |
1851 | static int snapshot_status(struct dm_target *ti, status_type_t type, |
1852 | char *result, unsigned int maxlen) | |
1853 | { | |
2e4a31df | 1854 | unsigned sz = 0; |
028867ac | 1855 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1856 | |
1857 | switch (type) { | |
1858 | case STATUSTYPE_INFO: | |
94e76572 MP |
1859 | |
1860 | down_write(&snap->lock); | |
1861 | ||
1da177e4 | 1862 | if (!snap->valid) |
2e4a31df | 1863 | DMEMIT("Invalid"); |
d8ddb1cf MS |
1864 | else if (snap->merge_failed) |
1865 | DMEMIT("Merge failed"); | |
1da177e4 | 1866 | else { |
985903bb MS |
1867 | if (snap->store->type->usage) { |
1868 | sector_t total_sectors, sectors_allocated, | |
1869 | metadata_sectors; | |
1870 | snap->store->type->usage(snap->store, | |
1871 | &total_sectors, | |
1872 | §ors_allocated, | |
1873 | &metadata_sectors); | |
1874 | DMEMIT("%llu/%llu %llu", | |
1875 | (unsigned long long)sectors_allocated, | |
1876 | (unsigned long long)total_sectors, | |
1877 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1878 | } |
1879 | else | |
2e4a31df | 1880 | DMEMIT("Unknown"); |
1da177e4 | 1881 | } |
94e76572 MP |
1882 | |
1883 | up_write(&snap->lock); | |
1884 | ||
1da177e4 LT |
1885 | break; |
1886 | ||
1887 | case STATUSTYPE_TABLE: | |
1888 | /* | |
1889 | * kdevname returns a static pointer so we need | |
1890 | * to make private copies if the output is to | |
1891 | * make sense. | |
1892 | */ | |
fc56f6fb | 1893 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1894 | snap->store->type->status(snap->store, type, result + sz, |
1895 | maxlen - sz); | |
1da177e4 LT |
1896 | break; |
1897 | } | |
1898 | ||
1899 | return 0; | |
1900 | } | |
1901 | ||
8811f46c MS |
1902 | static int snapshot_iterate_devices(struct dm_target *ti, |
1903 | iterate_devices_callout_fn fn, void *data) | |
1904 | { | |
1905 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
1906 | int r; |
1907 | ||
1908 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 1909 | |
1e5554c8 MP |
1910 | if (!r) |
1911 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
1912 | ||
1913 | return r; | |
8811f46c MS |
1914 | } |
1915 | ||
1916 | ||
1da177e4 LT |
1917 | /*----------------------------------------------------------------- |
1918 | * Origin methods | |
1919 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1920 | |
1921 | /* | |
1922 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1923 | * supplied bio was ignored. The caller may submit it immediately. | |
1924 | * (No remapping actually occurs as the origin is always a direct linear | |
1925 | * map.) | |
1926 | * | |
1927 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1928 | * and any supplied bio is added to a list to be submitted once all | |
1929 | * the necessary exceptions exist. | |
1930 | */ | |
1931 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1932 | struct bio *bio) | |
1da177e4 | 1933 | { |
515ad66c | 1934 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1935 | struct dm_snapshot *snap; |
1d4989c8 | 1936 | struct dm_exception *e; |
515ad66c MP |
1937 | struct dm_snap_pending_exception *pe; |
1938 | struct dm_snap_pending_exception *pe_to_start_now = NULL; | |
1939 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
1da177e4 LT |
1940 | chunk_t chunk; |
1941 | ||
1942 | /* Do all the snapshots on this origin */ | |
1943 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
1944 | /* |
1945 | * Don't make new exceptions in a merging snapshot | |
1946 | * because it has effectively been deleted | |
1947 | */ | |
1948 | if (dm_target_is_snapshot_merge(snap->ti)) | |
1949 | continue; | |
1950 | ||
76df1c65 AK |
1951 | down_write(&snap->lock); |
1952 | ||
aa14edeb AK |
1953 | /* Only deal with valid and active snapshots */ |
1954 | if (!snap->valid || !snap->active) | |
76df1c65 | 1955 | goto next_snapshot; |
1da177e4 | 1956 | |
d5e404c1 | 1957 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1958 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1959 | goto next_snapshot; |
1da177e4 LT |
1960 | |
1961 | /* | |
1962 | * Remember, different snapshots can have | |
1963 | * different chunk sizes. | |
1964 | */ | |
9eaae8ff | 1965 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1966 | |
1967 | /* | |
1968 | * Check exception table to see if block | |
1969 | * is already remapped in this snapshot | |
1970 | * and trigger an exception if not. | |
1971 | */ | |
3510cb94 | 1972 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1973 | if (e) |
1974 | goto next_snapshot; | |
1975 | ||
2913808e | 1976 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1977 | if (!pe) { |
c6621392 MP |
1978 | up_write(&snap->lock); |
1979 | pe = alloc_pending_exception(snap); | |
1980 | down_write(&snap->lock); | |
1981 | ||
1982 | if (!snap->valid) { | |
1983 | free_pending_exception(pe); | |
1984 | goto next_snapshot; | |
1985 | } | |
1986 | ||
3510cb94 | 1987 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1988 | if (e) { |
1989 | free_pending_exception(pe); | |
1990 | goto next_snapshot; | |
1991 | } | |
1992 | ||
c6621392 | 1993 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1994 | if (!pe) { |
1995 | __invalidate_snapshot(snap, -ENOMEM); | |
1996 | goto next_snapshot; | |
1997 | } | |
76df1c65 AK |
1998 | } |
1999 | ||
515ad66c | 2000 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2001 | |
515ad66c MP |
2002 | /* |
2003 | * If an origin bio was supplied, queue it to wait for the | |
2004 | * completion of this exception, and start this one last, | |
2005 | * at the end of the function. | |
2006 | */ | |
2007 | if (bio) { | |
2008 | bio_list_add(&pe->origin_bios, bio); | |
2009 | bio = NULL; | |
76df1c65 | 2010 | |
515ad66c MP |
2011 | if (!pe->started) { |
2012 | pe->started = 1; | |
2013 | pe_to_start_last = pe; | |
2014 | } | |
76df1c65 AK |
2015 | } |
2016 | ||
2017 | if (!pe->started) { | |
2018 | pe->started = 1; | |
515ad66c | 2019 | pe_to_start_now = pe; |
1da177e4 LT |
2020 | } |
2021 | ||
76df1c65 | 2022 | next_snapshot: |
1da177e4 | 2023 | up_write(&snap->lock); |
1da177e4 | 2024 | |
515ad66c MP |
2025 | if (pe_to_start_now) { |
2026 | start_copy(pe_to_start_now); | |
2027 | pe_to_start_now = NULL; | |
2028 | } | |
b4b610f6 AK |
2029 | } |
2030 | ||
1da177e4 | 2031 | /* |
515ad66c MP |
2032 | * Submit the exception against which the bio is queued last, |
2033 | * to give the other exceptions a head start. | |
1da177e4 | 2034 | */ |
515ad66c MP |
2035 | if (pe_to_start_last) |
2036 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2037 | |
2038 | return r; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Called on a write from the origin driver. | |
2043 | */ | |
2044 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
2045 | { | |
2046 | struct origin *o; | |
d2a7ad29 | 2047 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
2048 | |
2049 | down_read(&_origins_lock); | |
2050 | o = __lookup_origin(origin->bdev); | |
2051 | if (o) | |
9eaae8ff | 2052 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
2053 | up_read(&_origins_lock); |
2054 | ||
2055 | return r; | |
2056 | } | |
2057 | ||
73dfd078 MP |
2058 | /* |
2059 | * Trigger exceptions in all non-merging snapshots. | |
2060 | * | |
2061 | * The chunk size of the merging snapshot may be larger than the chunk | |
2062 | * size of some other snapshot so we may need to reallocate multiple | |
2063 | * chunks in other snapshots. | |
2064 | * | |
2065 | * We scan all the overlapping exceptions in the other snapshots. | |
2066 | * Returns 1 if anything was reallocated and must be waited for, | |
2067 | * otherwise returns 0. | |
2068 | * | |
2069 | * size must be a multiple of merging_snap's chunk_size. | |
2070 | */ | |
2071 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2072 | sector_t sector, unsigned size) | |
2073 | { | |
2074 | int must_wait = 0; | |
2075 | sector_t n; | |
2076 | struct origin *o; | |
2077 | ||
2078 | /* | |
2079 | * The origin's __minimum_chunk_size() got stored in split_io | |
2080 | * by snapshot_merge_resume(). | |
2081 | */ | |
2082 | down_read(&_origins_lock); | |
2083 | o = __lookup_origin(merging_snap->origin->bdev); | |
2084 | for (n = 0; n < size; n += merging_snap->ti->split_io) | |
2085 | if (__origin_write(&o->snapshots, sector + n, NULL) == | |
2086 | DM_MAPIO_SUBMITTED) | |
2087 | must_wait = 1; | |
2088 | up_read(&_origins_lock); | |
2089 | ||
2090 | return must_wait; | |
2091 | } | |
2092 | ||
1da177e4 LT |
2093 | /* |
2094 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2095 | */ | |
2096 | ||
2097 | /* | |
2098 | * Construct an origin mapping: <dev_path> | |
2099 | * The context for an origin is merely a 'struct dm_dev *' | |
2100 | * pointing to the real device. | |
2101 | */ | |
2102 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2103 | { | |
2104 | int r; | |
2105 | struct dm_dev *dev; | |
2106 | ||
2107 | if (argc != 1) { | |
72d94861 | 2108 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2109 | return -EINVAL; |
2110 | } | |
2111 | ||
8215d6ec | 2112 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev); |
1da177e4 LT |
2113 | if (r) { |
2114 | ti->error = "Cannot get target device"; | |
2115 | return r; | |
2116 | } | |
2117 | ||
2118 | ti->private = dev; | |
494b3ee7 MP |
2119 | ti->num_flush_requests = 1; |
2120 | ||
1da177e4 LT |
2121 | return 0; |
2122 | } | |
2123 | ||
2124 | static void origin_dtr(struct dm_target *ti) | |
2125 | { | |
028867ac | 2126 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2127 | dm_put_device(ti, dev); |
2128 | } | |
2129 | ||
2130 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
2131 | union map_info *map_context) | |
2132 | { | |
028867ac | 2133 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2134 | bio->bi_bdev = dev->bdev; |
2135 | ||
d87f4c14 | 2136 | if (bio->bi_rw & REQ_FLUSH) |
494b3ee7 MP |
2137 | return DM_MAPIO_REMAPPED; |
2138 | ||
1da177e4 | 2139 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 2140 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
2141 | } |
2142 | ||
1da177e4 LT |
2143 | /* |
2144 | * Set the target "split_io" field to the minimum of all the snapshots' | |
2145 | * chunk sizes. | |
2146 | */ | |
2147 | static void origin_resume(struct dm_target *ti) | |
2148 | { | |
028867ac | 2149 | struct dm_dev *dev = ti->private; |
1da177e4 | 2150 | |
1e03f97e | 2151 | ti->split_io = get_origin_minimum_chunksize(dev->bdev); |
1da177e4 LT |
2152 | } |
2153 | ||
2154 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
2155 | unsigned int maxlen) | |
2156 | { | |
028867ac | 2157 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
2158 | |
2159 | switch (type) { | |
2160 | case STATUSTYPE_INFO: | |
2161 | result[0] = '\0'; | |
2162 | break; | |
2163 | ||
2164 | case STATUSTYPE_TABLE: | |
2165 | snprintf(result, maxlen, "%s", dev->name); | |
2166 | break; | |
2167 | } | |
2168 | ||
2169 | return 0; | |
2170 | } | |
2171 | ||
b1d55528 MP |
2172 | static int origin_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
2173 | struct bio_vec *biovec, int max_size) | |
2174 | { | |
2175 | struct dm_dev *dev = ti->private; | |
2176 | struct request_queue *q = bdev_get_queue(dev->bdev); | |
2177 | ||
2178 | if (!q->merge_bvec_fn) | |
2179 | return max_size; | |
2180 | ||
2181 | bvm->bi_bdev = dev->bdev; | |
2182 | bvm->bi_sector = bvm->bi_sector; | |
2183 | ||
2184 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
2185 | } | |
2186 | ||
8811f46c MS |
2187 | static int origin_iterate_devices(struct dm_target *ti, |
2188 | iterate_devices_callout_fn fn, void *data) | |
2189 | { | |
2190 | struct dm_dev *dev = ti->private; | |
2191 | ||
2192 | return fn(ti, dev, 0, ti->len, data); | |
2193 | } | |
2194 | ||
1da177e4 LT |
2195 | static struct target_type origin_target = { |
2196 | .name = "snapshot-origin", | |
8811f46c | 2197 | .version = {1, 7, 0}, |
1da177e4 LT |
2198 | .module = THIS_MODULE, |
2199 | .ctr = origin_ctr, | |
2200 | .dtr = origin_dtr, | |
2201 | .map = origin_map, | |
2202 | .resume = origin_resume, | |
2203 | .status = origin_status, | |
b1d55528 | 2204 | .merge = origin_merge, |
8811f46c | 2205 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2206 | }; |
2207 | ||
2208 | static struct target_type snapshot_target = { | |
2209 | .name = "snapshot", | |
c26655ca | 2210 | .version = {1, 9, 0}, |
1da177e4 LT |
2211 | .module = THIS_MODULE, |
2212 | .ctr = snapshot_ctr, | |
2213 | .dtr = snapshot_dtr, | |
2214 | .map = snapshot_map, | |
cd45daff | 2215 | .end_io = snapshot_end_io, |
c26655ca | 2216 | .postsuspend = snapshot_postsuspend, |
c1f0c183 | 2217 | .preresume = snapshot_preresume, |
1da177e4 LT |
2218 | .resume = snapshot_resume, |
2219 | .status = snapshot_status, | |
8811f46c | 2220 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
2221 | }; |
2222 | ||
d698aa45 MP |
2223 | static struct target_type merge_target = { |
2224 | .name = dm_snapshot_merge_target_name, | |
2225 | .version = {1, 0, 0}, | |
2226 | .module = THIS_MODULE, | |
2227 | .ctr = snapshot_ctr, | |
2228 | .dtr = snapshot_dtr, | |
3452c2a1 | 2229 | .map = snapshot_merge_map, |
d698aa45 | 2230 | .end_io = snapshot_end_io, |
1e03f97e | 2231 | .presuspend = snapshot_merge_presuspend, |
d698aa45 MP |
2232 | .postsuspend = snapshot_postsuspend, |
2233 | .preresume = snapshot_preresume, | |
1e03f97e | 2234 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2235 | .status = snapshot_status, |
2236 | .iterate_devices = snapshot_iterate_devices, | |
2237 | }; | |
2238 | ||
1da177e4 LT |
2239 | static int __init dm_snapshot_init(void) |
2240 | { | |
2241 | int r; | |
2242 | ||
4db6bfe0 AK |
2243 | r = dm_exception_store_init(); |
2244 | if (r) { | |
2245 | DMERR("Failed to initialize exception stores"); | |
2246 | return r; | |
2247 | } | |
2248 | ||
1da177e4 | 2249 | r = dm_register_target(&snapshot_target); |
d698aa45 | 2250 | if (r < 0) { |
1da177e4 | 2251 | DMERR("snapshot target register failed %d", r); |
034a186d | 2252 | goto bad_register_snapshot_target; |
1da177e4 LT |
2253 | } |
2254 | ||
2255 | r = dm_register_target(&origin_target); | |
2256 | if (r < 0) { | |
72d94861 | 2257 | DMERR("Origin target register failed %d", r); |
d698aa45 MP |
2258 | goto bad_register_origin_target; |
2259 | } | |
2260 | ||
2261 | r = dm_register_target(&merge_target); | |
2262 | if (r < 0) { | |
2263 | DMERR("Merge target register failed %d", r); | |
2264 | goto bad_register_merge_target; | |
1da177e4 LT |
2265 | } |
2266 | ||
2267 | r = init_origin_hash(); | |
2268 | if (r) { | |
2269 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2270 | goto bad_origin_hash; |
1da177e4 LT |
2271 | } |
2272 | ||
1d4989c8 | 2273 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2274 | if (!exception_cache) { |
2275 | DMERR("Couldn't create exception cache."); | |
2276 | r = -ENOMEM; | |
d698aa45 | 2277 | goto bad_exception_cache; |
1da177e4 LT |
2278 | } |
2279 | ||
028867ac | 2280 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2281 | if (!pending_cache) { |
2282 | DMERR("Couldn't create pending cache."); | |
2283 | r = -ENOMEM; | |
d698aa45 | 2284 | goto bad_pending_cache; |
1da177e4 LT |
2285 | } |
2286 | ||
cd45daff MP |
2287 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
2288 | if (!tracked_chunk_cache) { | |
2289 | DMERR("Couldn't create cache to track chunks in use."); | |
2290 | r = -ENOMEM; | |
d698aa45 | 2291 | goto bad_tracked_chunk_cache; |
cd45daff MP |
2292 | } |
2293 | ||
ca3a931f AK |
2294 | ksnapd = create_singlethread_workqueue("ksnapd"); |
2295 | if (!ksnapd) { | |
2296 | DMERR("Failed to create ksnapd workqueue."); | |
2297 | r = -ENOMEM; | |
92e86812 | 2298 | goto bad_pending_pool; |
ca3a931f AK |
2299 | } |
2300 | ||
1da177e4 LT |
2301 | return 0; |
2302 | ||
4db6bfe0 | 2303 | bad_pending_pool: |
cd45daff | 2304 | kmem_cache_destroy(tracked_chunk_cache); |
d698aa45 | 2305 | bad_tracked_chunk_cache: |
1da177e4 | 2306 | kmem_cache_destroy(pending_cache); |
d698aa45 | 2307 | bad_pending_cache: |
1da177e4 | 2308 | kmem_cache_destroy(exception_cache); |
d698aa45 | 2309 | bad_exception_cache: |
1da177e4 | 2310 | exit_origin_hash(); |
d698aa45 MP |
2311 | bad_origin_hash: |
2312 | dm_unregister_target(&merge_target); | |
2313 | bad_register_merge_target: | |
1da177e4 | 2314 | dm_unregister_target(&origin_target); |
d698aa45 | 2315 | bad_register_origin_target: |
1da177e4 | 2316 | dm_unregister_target(&snapshot_target); |
034a186d JB |
2317 | bad_register_snapshot_target: |
2318 | dm_exception_store_exit(); | |
d698aa45 | 2319 | |
1da177e4 LT |
2320 | return r; |
2321 | } | |
2322 | ||
2323 | static void __exit dm_snapshot_exit(void) | |
2324 | { | |
ca3a931f AK |
2325 | destroy_workqueue(ksnapd); |
2326 | ||
10d3bd09 MP |
2327 | dm_unregister_target(&snapshot_target); |
2328 | dm_unregister_target(&origin_target); | |
d698aa45 | 2329 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2330 | |
2331 | exit_origin_hash(); | |
1da177e4 LT |
2332 | kmem_cache_destroy(pending_cache); |
2333 | kmem_cache_destroy(exception_cache); | |
cd45daff | 2334 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
2335 | |
2336 | dm_exception_store_exit(); | |
1da177e4 LT |
2337 | } |
2338 | ||
2339 | /* Module hooks */ | |
2340 | module_init(dm_snapshot_init); | |
2341 | module_exit(dm_snapshot_exit); | |
2342 | ||
2343 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2344 | MODULE_AUTHOR("Joe Thornber"); | |
2345 | MODULE_LICENSE("GPL"); |