]>
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 | ||
1da177e4 LT |
28 | /* |
29 | * The percentage increment we will wake up users at | |
30 | */ | |
31 | #define WAKE_UP_PERCENT 5 | |
32 | ||
33 | /* | |
34 | * kcopyd priority of snapshot operations | |
35 | */ | |
36 | #define SNAPSHOT_COPY_PRIORITY 2 | |
37 | ||
38 | /* | |
8ee2767a | 39 | * Reserve 1MB for each snapshot initially (with minimum of 1 page). |
1da177e4 | 40 | */ |
8ee2767a | 41 | #define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1) |
1da177e4 | 42 | |
cd45daff MP |
43 | /* |
44 | * The size of the mempool used to track chunks in use. | |
45 | */ | |
46 | #define MIN_IOS 256 | |
47 | ||
ccc45ea8 JB |
48 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
49 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
50 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
51 | ||
191437a5 | 52 | struct dm_exception_table { |
ccc45ea8 JB |
53 | uint32_t hash_mask; |
54 | unsigned hash_shift; | |
55 | struct list_head *table; | |
56 | }; | |
57 | ||
58 | struct dm_snapshot { | |
59 | struct rw_semaphore lock; | |
60 | ||
61 | struct dm_dev *origin; | |
fc56f6fb MS |
62 | struct dm_dev *cow; |
63 | ||
64 | struct dm_target *ti; | |
ccc45ea8 JB |
65 | |
66 | /* List of snapshots per Origin */ | |
67 | struct list_head list; | |
68 | ||
69 | /* You can't use a snapshot if this is 0 (e.g. if full) */ | |
70 | int valid; | |
71 | ||
72 | /* Origin writes don't trigger exceptions until this is set */ | |
73 | int active; | |
74 | ||
c26655ca MS |
75 | /* Whether or not owning mapped_device is suspended */ |
76 | int suspended; | |
77 | ||
ccc45ea8 JB |
78 | mempool_t *pending_pool; |
79 | ||
80 | atomic_t pending_exceptions_count; | |
81 | ||
191437a5 JB |
82 | struct dm_exception_table pending; |
83 | struct dm_exception_table complete; | |
ccc45ea8 JB |
84 | |
85 | /* | |
86 | * pe_lock protects all pending_exception operations and access | |
87 | * as well as the snapshot_bios list. | |
88 | */ | |
89 | spinlock_t pe_lock; | |
90 | ||
91 | /* The on disk metadata handler */ | |
92 | struct dm_exception_store *store; | |
93 | ||
94 | struct dm_kcopyd_client *kcopyd_client; | |
95 | ||
96 | /* Queue of snapshot writes for ksnapd to flush */ | |
97 | struct bio_list queued_bios; | |
98 | struct work_struct queued_bios_work; | |
99 | ||
100 | /* Chunks with outstanding reads */ | |
101 | mempool_t *tracked_chunk_pool; | |
102 | spinlock_t tracked_chunk_lock; | |
103 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; | |
104 | }; | |
105 | ||
fc56f6fb MS |
106 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
107 | { | |
108 | return s->cow; | |
109 | } | |
110 | EXPORT_SYMBOL(dm_snap_cow); | |
111 | ||
c642f9e0 | 112 | static struct workqueue_struct *ksnapd; |
c4028958 | 113 | static void flush_queued_bios(struct work_struct *work); |
ca3a931f | 114 | |
ccc45ea8 JB |
115 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
116 | chunk_t chunk) | |
117 | { | |
118 | return chunk << store->chunk_shift; | |
119 | } | |
120 | ||
121 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
122 | { | |
123 | /* | |
124 | * There is only ever one instance of a particular block | |
125 | * device so we can compare pointers safely. | |
126 | */ | |
127 | return lhs == rhs; | |
128 | } | |
129 | ||
028867ac | 130 | struct dm_snap_pending_exception { |
1d4989c8 | 131 | struct dm_exception e; |
1da177e4 LT |
132 | |
133 | /* | |
134 | * Origin buffers waiting for this to complete are held | |
135 | * in a bio list | |
136 | */ | |
137 | struct bio_list origin_bios; | |
138 | struct bio_list snapshot_bios; | |
139 | ||
eccf0817 AK |
140 | /* |
141 | * Short-term queue of pending exceptions prior to submission. | |
142 | */ | |
143 | struct list_head list; | |
144 | ||
1da177e4 | 145 | /* |
b4b610f6 | 146 | * The primary pending_exception is the one that holds |
4b832e8d | 147 | * the ref_count and the list of origin_bios for a |
b4b610f6 AK |
148 | * group of pending_exceptions. It is always last to get freed. |
149 | * These fields get set up when writing to the origin. | |
1da177e4 | 150 | */ |
028867ac | 151 | struct dm_snap_pending_exception *primary_pe; |
b4b610f6 AK |
152 | |
153 | /* | |
154 | * Number of pending_exceptions processing this chunk. | |
155 | * When this drops to zero we must complete the origin bios. | |
156 | * If incrementing or decrementing this, hold pe->snap->lock for | |
157 | * the sibling concerned and not pe->primary_pe->snap->lock unless | |
158 | * they are the same. | |
159 | */ | |
4b832e8d | 160 | atomic_t ref_count; |
1da177e4 LT |
161 | |
162 | /* Pointer back to snapshot context */ | |
163 | struct dm_snapshot *snap; | |
164 | ||
165 | /* | |
166 | * 1 indicates the exception has already been sent to | |
167 | * kcopyd. | |
168 | */ | |
169 | int started; | |
170 | }; | |
171 | ||
172 | /* | |
173 | * Hash table mapping origin volumes to lists of snapshots and | |
174 | * a lock to protect it | |
175 | */ | |
e18b890b CL |
176 | static struct kmem_cache *exception_cache; |
177 | static struct kmem_cache *pending_cache; | |
1da177e4 | 178 | |
cd45daff MP |
179 | struct dm_snap_tracked_chunk { |
180 | struct hlist_node node; | |
181 | chunk_t chunk; | |
182 | }; | |
183 | ||
184 | static struct kmem_cache *tracked_chunk_cache; | |
185 | ||
186 | static struct dm_snap_tracked_chunk *track_chunk(struct dm_snapshot *s, | |
187 | chunk_t chunk) | |
188 | { | |
189 | struct dm_snap_tracked_chunk *c = mempool_alloc(s->tracked_chunk_pool, | |
190 | GFP_NOIO); | |
191 | unsigned long flags; | |
192 | ||
193 | c->chunk = chunk; | |
194 | ||
195 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
196 | hlist_add_head(&c->node, | |
197 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
198 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
199 | ||
200 | return c; | |
201 | } | |
202 | ||
203 | static void stop_tracking_chunk(struct dm_snapshot *s, | |
204 | struct dm_snap_tracked_chunk *c) | |
205 | { | |
206 | unsigned long flags; | |
207 | ||
208 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
209 | hlist_del(&c->node); | |
210 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
211 | ||
212 | mempool_free(c, s->tracked_chunk_pool); | |
213 | } | |
214 | ||
a8d41b59 MP |
215 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
216 | { | |
217 | struct dm_snap_tracked_chunk *c; | |
218 | struct hlist_node *hn; | |
219 | int found = 0; | |
220 | ||
221 | spin_lock_irq(&s->tracked_chunk_lock); | |
222 | ||
223 | hlist_for_each_entry(c, hn, | |
224 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { | |
225 | if (c->chunk == chunk) { | |
226 | found = 1; | |
227 | break; | |
228 | } | |
229 | } | |
230 | ||
231 | spin_unlock_irq(&s->tracked_chunk_lock); | |
232 | ||
233 | return found; | |
234 | } | |
235 | ||
1da177e4 LT |
236 | /* |
237 | * One of these per registered origin, held in the snapshot_origins hash | |
238 | */ | |
239 | struct origin { | |
240 | /* The origin device */ | |
241 | struct block_device *bdev; | |
242 | ||
243 | struct list_head hash_list; | |
244 | ||
245 | /* List of snapshots for this origin */ | |
246 | struct list_head snapshots; | |
247 | }; | |
248 | ||
249 | /* | |
250 | * Size of the hash table for origin volumes. If we make this | |
251 | * the size of the minors list then it should be nearly perfect | |
252 | */ | |
253 | #define ORIGIN_HASH_SIZE 256 | |
254 | #define ORIGIN_MASK 0xFF | |
255 | static struct list_head *_origins; | |
256 | static struct rw_semaphore _origins_lock; | |
257 | ||
258 | static int init_origin_hash(void) | |
259 | { | |
260 | int i; | |
261 | ||
262 | _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), | |
263 | GFP_KERNEL); | |
264 | if (!_origins) { | |
72d94861 | 265 | DMERR("unable to allocate memory"); |
1da177e4 LT |
266 | return -ENOMEM; |
267 | } | |
268 | ||
269 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
270 | INIT_LIST_HEAD(_origins + i); | |
271 | init_rwsem(&_origins_lock); | |
272 | ||
273 | return 0; | |
274 | } | |
275 | ||
276 | static void exit_origin_hash(void) | |
277 | { | |
278 | kfree(_origins); | |
279 | } | |
280 | ||
028867ac | 281 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
282 | { |
283 | return bdev->bd_dev & ORIGIN_MASK; | |
284 | } | |
285 | ||
286 | static struct origin *__lookup_origin(struct block_device *origin) | |
287 | { | |
288 | struct list_head *ol; | |
289 | struct origin *o; | |
290 | ||
291 | ol = &_origins[origin_hash(origin)]; | |
292 | list_for_each_entry (o, ol, hash_list) | |
293 | if (bdev_equal(o->bdev, origin)) | |
294 | return o; | |
295 | ||
296 | return NULL; | |
297 | } | |
298 | ||
299 | static void __insert_origin(struct origin *o) | |
300 | { | |
301 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
302 | list_add_tail(&o->hash_list, sl); | |
303 | } | |
304 | ||
c1f0c183 MS |
305 | /* |
306 | * _origins_lock must be held when calling this function. | |
307 | * Returns number of snapshots registered using the supplied cow device, plus: | |
308 | * snap_src - a snapshot suitable for use as a source of exception handover | |
309 | * snap_dest - a snapshot capable of receiving exception handover. | |
310 | * | |
311 | * Possible return values and states: | |
312 | * 0: NULL, NULL - first new snapshot | |
313 | * 1: snap_src, NULL - normal snapshot | |
314 | * 2: snap_src, snap_dest - waiting for handover | |
315 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
316 | * 1: NULL, snap_dest - source got destroyed without handover | |
317 | */ | |
318 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
319 | struct dm_snapshot **snap_src, | |
320 | struct dm_snapshot **snap_dest) | |
321 | { | |
322 | struct dm_snapshot *s; | |
323 | struct origin *o; | |
324 | int count = 0; | |
325 | int active; | |
326 | ||
327 | o = __lookup_origin(snap->origin->bdev); | |
328 | if (!o) | |
329 | goto out; | |
330 | ||
331 | list_for_each_entry(s, &o->snapshots, list) { | |
332 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) | |
333 | continue; | |
334 | ||
335 | down_read(&s->lock); | |
336 | active = s->active; | |
337 | up_read(&s->lock); | |
338 | ||
339 | if (active) { | |
340 | if (snap_src) | |
341 | *snap_src = s; | |
342 | } else if (snap_dest) | |
343 | *snap_dest = s; | |
344 | ||
345 | count++; | |
346 | } | |
347 | ||
348 | out: | |
349 | return count; | |
350 | } | |
351 | ||
352 | /* | |
353 | * On success, returns 1 if this snapshot is a handover destination, | |
354 | * otherwise returns 0. | |
355 | */ | |
356 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
357 | { | |
358 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
359 | ||
360 | /* Does snapshot need exceptions handed over to it? */ | |
361 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest) == 2) || | |
362 | snap_dest) { | |
363 | snap->ti->error = "Snapshot cow pairing for exception " | |
364 | "table handover failed"; | |
365 | return -EINVAL; | |
366 | } | |
367 | ||
368 | /* | |
369 | * If no snap_src was found, snap cannot become a handover | |
370 | * destination. | |
371 | */ | |
372 | if (!snap_src) | |
373 | return 0; | |
374 | ||
375 | return 1; | |
376 | } | |
377 | ||
378 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
379 | { | |
380 | struct dm_snapshot *l; | |
381 | ||
382 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
383 | list_for_each_entry(l, &o->snapshots, list) | |
384 | if (l->store->chunk_size < s->store->chunk_size) | |
385 | break; | |
386 | list_add_tail(&s->list, &l->list); | |
387 | } | |
388 | ||
1da177e4 LT |
389 | /* |
390 | * Make a note of the snapshot and its origin so we can look it | |
391 | * up when the origin has a write on it. | |
c1f0c183 MS |
392 | * |
393 | * Also validate snapshot exception store handovers. | |
394 | * On success, returns 1 if this registration is a handover destination, | |
395 | * otherwise returns 0. | |
1da177e4 LT |
396 | */ |
397 | static int register_snapshot(struct dm_snapshot *snap) | |
398 | { | |
c1f0c183 | 399 | struct origin *o, *new_o = NULL; |
1da177e4 | 400 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 401 | int r = 0; |
1da177e4 | 402 | |
60c856c8 MP |
403 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
404 | if (!new_o) | |
405 | return -ENOMEM; | |
406 | ||
1da177e4 | 407 | down_write(&_origins_lock); |
1da177e4 | 408 | |
c1f0c183 MS |
409 | r = __validate_exception_handover(snap); |
410 | if (r < 0) { | |
411 | kfree(new_o); | |
412 | goto out; | |
413 | } | |
414 | ||
415 | o = __lookup_origin(bdev); | |
60c856c8 MP |
416 | if (o) |
417 | kfree(new_o); | |
418 | else { | |
1da177e4 | 419 | /* New origin */ |
60c856c8 | 420 | o = new_o; |
1da177e4 LT |
421 | |
422 | /* Initialise the struct */ | |
423 | INIT_LIST_HEAD(&o->snapshots); | |
424 | o->bdev = bdev; | |
425 | ||
426 | __insert_origin(o); | |
427 | } | |
428 | ||
c1f0c183 MS |
429 | __insert_snapshot(o, snap); |
430 | ||
431 | out: | |
432 | up_write(&_origins_lock); | |
433 | ||
434 | return r; | |
435 | } | |
436 | ||
437 | /* | |
438 | * Move snapshot to correct place in list according to chunk size. | |
439 | */ | |
440 | static void reregister_snapshot(struct dm_snapshot *s) | |
441 | { | |
442 | struct block_device *bdev = s->origin->bdev; | |
443 | ||
444 | down_write(&_origins_lock); | |
445 | ||
446 | list_del(&s->list); | |
447 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
448 | |
449 | up_write(&_origins_lock); | |
1da177e4 LT |
450 | } |
451 | ||
452 | static void unregister_snapshot(struct dm_snapshot *s) | |
453 | { | |
454 | struct origin *o; | |
455 | ||
456 | down_write(&_origins_lock); | |
457 | o = __lookup_origin(s->origin->bdev); | |
458 | ||
459 | list_del(&s->list); | |
c1f0c183 | 460 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
461 | list_del(&o->hash_list); |
462 | kfree(o); | |
463 | } | |
464 | ||
465 | up_write(&_origins_lock); | |
466 | } | |
467 | ||
468 | /* | |
469 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
470 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
471 | * some consecutive chunks to be grouped together. | |
1da177e4 | 472 | */ |
3510cb94 JB |
473 | static int dm_exception_table_init(struct dm_exception_table *et, |
474 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
475 | { |
476 | unsigned int i; | |
477 | ||
d74f81f8 | 478 | et->hash_shift = hash_shift; |
1da177e4 LT |
479 | et->hash_mask = size - 1; |
480 | et->table = dm_vcalloc(size, sizeof(struct list_head)); | |
481 | if (!et->table) | |
482 | return -ENOMEM; | |
483 | ||
484 | for (i = 0; i < size; i++) | |
485 | INIT_LIST_HEAD(et->table + i); | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
3510cb94 JB |
490 | static void dm_exception_table_exit(struct dm_exception_table *et, |
491 | struct kmem_cache *mem) | |
1da177e4 LT |
492 | { |
493 | struct list_head *slot; | |
1d4989c8 | 494 | struct dm_exception *ex, *next; |
1da177e4 LT |
495 | int i, size; |
496 | ||
497 | size = et->hash_mask + 1; | |
498 | for (i = 0; i < size; i++) { | |
499 | slot = et->table + i; | |
500 | ||
501 | list_for_each_entry_safe (ex, next, slot, hash_list) | |
502 | kmem_cache_free(mem, ex); | |
503 | } | |
504 | ||
505 | vfree(et->table); | |
506 | } | |
507 | ||
191437a5 | 508 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 509 | { |
d74f81f8 | 510 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
511 | } |
512 | ||
3510cb94 | 513 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 LT |
514 | { |
515 | list_del(&e->hash_list); | |
516 | } | |
517 | ||
518 | /* | |
519 | * Return the exception data for a sector, or NULL if not | |
520 | * remapped. | |
521 | */ | |
3510cb94 JB |
522 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
523 | chunk_t chunk) | |
1da177e4 LT |
524 | { |
525 | struct list_head *slot; | |
1d4989c8 | 526 | struct dm_exception *e; |
1da177e4 LT |
527 | |
528 | slot = &et->table[exception_hash(et, chunk)]; | |
529 | list_for_each_entry (e, slot, hash_list) | |
d74f81f8 MB |
530 | if (chunk >= e->old_chunk && |
531 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
532 | return e; |
533 | ||
534 | return NULL; | |
535 | } | |
536 | ||
3510cb94 | 537 | static struct dm_exception *alloc_completed_exception(void) |
1da177e4 | 538 | { |
1d4989c8 | 539 | struct dm_exception *e; |
1da177e4 LT |
540 | |
541 | e = kmem_cache_alloc(exception_cache, GFP_NOIO); | |
542 | if (!e) | |
543 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); | |
544 | ||
545 | return e; | |
546 | } | |
547 | ||
3510cb94 | 548 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
549 | { |
550 | kmem_cache_free(exception_cache, e); | |
551 | } | |
552 | ||
92e86812 | 553 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 554 | { |
92e86812 MP |
555 | struct dm_snap_pending_exception *pe = mempool_alloc(s->pending_pool, |
556 | GFP_NOIO); | |
557 | ||
879129d2 | 558 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
559 | pe->snap = s; |
560 | ||
561 | return pe; | |
1da177e4 LT |
562 | } |
563 | ||
028867ac | 564 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 565 | { |
879129d2 MP |
566 | struct dm_snapshot *s = pe->snap; |
567 | ||
568 | mempool_free(pe, s->pending_pool); | |
569 | smp_mb__before_atomic_dec(); | |
570 | atomic_dec(&s->pending_exceptions_count); | |
1da177e4 LT |
571 | } |
572 | ||
3510cb94 JB |
573 | static void dm_insert_exception(struct dm_exception_table *eh, |
574 | struct dm_exception *new_e) | |
d74f81f8 | 575 | { |
d74f81f8 | 576 | struct list_head *l; |
1d4989c8 | 577 | struct dm_exception *e = NULL; |
d74f81f8 MB |
578 | |
579 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
580 | ||
581 | /* Add immediately if this table doesn't support consecutive chunks */ | |
582 | if (!eh->hash_shift) | |
583 | goto out; | |
584 | ||
585 | /* List is ordered by old_chunk */ | |
586 | list_for_each_entry_reverse(e, l, hash_list) { | |
587 | /* Insert after an existing chunk? */ | |
588 | if (new_e->old_chunk == (e->old_chunk + | |
589 | dm_consecutive_chunk_count(e) + 1) && | |
590 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
591 | dm_consecutive_chunk_count(e) + 1)) { | |
592 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 593 | free_completed_exception(new_e); |
d74f81f8 MB |
594 | return; |
595 | } | |
596 | ||
597 | /* Insert before an existing chunk? */ | |
598 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
599 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
600 | dm_consecutive_chunk_count_inc(e); | |
601 | e->old_chunk--; | |
602 | e->new_chunk--; | |
3510cb94 | 603 | free_completed_exception(new_e); |
d74f81f8 MB |
604 | return; |
605 | } | |
606 | ||
607 | if (new_e->old_chunk > e->old_chunk) | |
608 | break; | |
609 | } | |
610 | ||
611 | out: | |
612 | list_add(&new_e->hash_list, e ? &e->hash_list : l); | |
613 | } | |
614 | ||
a159c1ac JB |
615 | /* |
616 | * Callback used by the exception stores to load exceptions when | |
617 | * initialising. | |
618 | */ | |
619 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 620 | { |
a159c1ac | 621 | struct dm_snapshot *s = context; |
1d4989c8 | 622 | struct dm_exception *e; |
1da177e4 | 623 | |
3510cb94 | 624 | e = alloc_completed_exception(); |
1da177e4 LT |
625 | if (!e) |
626 | return -ENOMEM; | |
627 | ||
628 | e->old_chunk = old; | |
d74f81f8 MB |
629 | |
630 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 631 | e->new_chunk = new; |
d74f81f8 | 632 | |
3510cb94 | 633 | dm_insert_exception(&s->complete, e); |
d74f81f8 | 634 | |
1da177e4 LT |
635 | return 0; |
636 | } | |
637 | ||
7e201b35 MP |
638 | #define min_not_zero(l, r) (((l) == 0) ? (r) : (((r) == 0) ? (l) : min(l, r))) |
639 | ||
640 | /* | |
641 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
642 | * Return zero if the origin has no snapshots. | |
643 | */ | |
644 | static sector_t __minimum_chunk_size(struct origin *o) | |
645 | { | |
646 | struct dm_snapshot *snap; | |
647 | unsigned chunk_size = 0; | |
648 | ||
649 | if (o) | |
650 | list_for_each_entry(snap, &o->snapshots, list) | |
651 | chunk_size = min_not_zero(chunk_size, | |
652 | snap->store->chunk_size); | |
653 | ||
654 | return chunk_size; | |
655 | } | |
656 | ||
1da177e4 LT |
657 | /* |
658 | * Hard coded magic. | |
659 | */ | |
660 | static int calc_max_buckets(void) | |
661 | { | |
662 | /* use a fixed size of 2MB */ | |
663 | unsigned long mem = 2 * 1024 * 1024; | |
664 | mem /= sizeof(struct list_head); | |
665 | ||
666 | return mem; | |
667 | } | |
668 | ||
1da177e4 LT |
669 | /* |
670 | * Allocate room for a suitable hash table. | |
671 | */ | |
fee1998e | 672 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 LT |
673 | { |
674 | sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; | |
675 | ||
676 | /* | |
677 | * Calculate based on the size of the original volume or | |
678 | * the COW volume... | |
679 | */ | |
fc56f6fb | 680 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
681 | origin_dev_size = get_dev_size(s->origin->bdev); |
682 | max_buckets = calc_max_buckets(); | |
683 | ||
fee1998e | 684 | hash_size = min(origin_dev_size, cow_dev_size) >> s->store->chunk_shift; |
1da177e4 LT |
685 | hash_size = min(hash_size, max_buckets); |
686 | ||
8e87b9b8 MP |
687 | if (hash_size < 64) |
688 | hash_size = 64; | |
8defd830 | 689 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
690 | if (dm_exception_table_init(&s->complete, hash_size, |
691 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
692 | return -ENOMEM; |
693 | ||
694 | /* | |
695 | * Allocate hash table for in-flight exceptions | |
696 | * Make this smaller than the real hash table | |
697 | */ | |
698 | hash_size >>= 3; | |
699 | if (hash_size < 64) | |
700 | hash_size = 64; | |
701 | ||
3510cb94 JB |
702 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
703 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
704 | return -ENOMEM; |
705 | } | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
1da177e4 LT |
710 | /* |
711 | * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> | |
712 | */ | |
713 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
714 | { | |
715 | struct dm_snapshot *s; | |
cd45daff | 716 | int i; |
1da177e4 | 717 | int r = -EINVAL; |
fc56f6fb | 718 | char *origin_path, *cow_path; |
fee1998e | 719 | unsigned args_used; |
1da177e4 | 720 | |
4c7e3bf4 | 721 | if (argc != 4) { |
72d94861 | 722 | ti->error = "requires exactly 4 arguments"; |
1da177e4 | 723 | r = -EINVAL; |
fc56f6fb | 724 | goto bad; |
1da177e4 LT |
725 | } |
726 | ||
727 | origin_path = argv[0]; | |
fee1998e JB |
728 | argv++; |
729 | argc--; | |
1da177e4 | 730 | |
fc56f6fb MS |
731 | s = kmalloc(sizeof(*s), GFP_KERNEL); |
732 | if (!s) { | |
733 | ti->error = "Cannot allocate snapshot context private " | |
734 | "structure"; | |
735 | r = -ENOMEM; | |
736 | goto bad; | |
737 | } | |
738 | ||
739 | cow_path = argv[0]; | |
740 | argv++; | |
741 | argc--; | |
742 | ||
743 | r = dm_get_device(ti, cow_path, 0, 0, | |
744 | FMODE_READ | FMODE_WRITE, &s->cow); | |
745 | if (r) { | |
746 | ti->error = "Cannot get COW device"; | |
747 | goto bad_cow; | |
748 | } | |
749 | ||
750 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
751 | if (r) { |
752 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 753 | r = -EINVAL; |
fc56f6fb | 754 | goto bad_store; |
1da177e4 LT |
755 | } |
756 | ||
fee1998e JB |
757 | argv += args_used; |
758 | argc -= args_used; | |
759 | ||
1da177e4 LT |
760 | r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); |
761 | if (r) { | |
762 | ti->error = "Cannot get origin device"; | |
fee1998e | 763 | goto bad_origin; |
1da177e4 LT |
764 | } |
765 | ||
fc56f6fb | 766 | s->ti = ti; |
1da177e4 | 767 | s->valid = 1; |
aa14edeb | 768 | s->active = 0; |
c26655ca | 769 | s->suspended = 0; |
879129d2 | 770 | atomic_set(&s->pending_exceptions_count, 0); |
1da177e4 | 771 | init_rwsem(&s->lock); |
c1f0c183 | 772 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 773 | spin_lock_init(&s->pe_lock); |
1da177e4 LT |
774 | |
775 | /* Allocate hash table for COW data */ | |
fee1998e | 776 | if (init_hash_tables(s)) { |
1da177e4 LT |
777 | ti->error = "Unable to allocate hash table space"; |
778 | r = -ENOMEM; | |
fee1998e | 779 | goto bad_hash_tables; |
1da177e4 LT |
780 | } |
781 | ||
eb69aca5 | 782 | r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); |
1da177e4 LT |
783 | if (r) { |
784 | ti->error = "Could not create kcopyd client"; | |
fee1998e | 785 | goto bad_kcopyd; |
1da177e4 LT |
786 | } |
787 | ||
92e86812 MP |
788 | s->pending_pool = mempool_create_slab_pool(MIN_IOS, pending_cache); |
789 | if (!s->pending_pool) { | |
790 | ti->error = "Could not allocate mempool for pending exceptions"; | |
fee1998e | 791 | goto bad_pending_pool; |
92e86812 MP |
792 | } |
793 | ||
cd45daff MP |
794 | s->tracked_chunk_pool = mempool_create_slab_pool(MIN_IOS, |
795 | tracked_chunk_cache); | |
796 | if (!s->tracked_chunk_pool) { | |
797 | ti->error = "Could not allocate tracked_chunk mempool for " | |
798 | "tracking reads"; | |
92e86812 | 799 | goto bad_tracked_chunk_pool; |
cd45daff MP |
800 | } |
801 | ||
802 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
803 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
804 | ||
805 | spin_lock_init(&s->tracked_chunk_lock); | |
806 | ||
c1f0c183 MS |
807 | bio_list_init(&s->queued_bios); |
808 | INIT_WORK(&s->queued_bios_work, flush_queued_bios); | |
809 | ||
810 | ti->private = s; | |
811 | ti->num_flush_requests = 1; | |
812 | ||
813 | /* Add snapshot to the list of snapshots for this origin */ | |
814 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
815 | r = register_snapshot(s); | |
816 | if (r == -ENOMEM) { | |
817 | ti->error = "Snapshot origin struct allocation failed"; | |
818 | goto bad_load_and_register; | |
819 | } else if (r < 0) { | |
820 | /* invalid handover, register_snapshot has set ti->error */ | |
821 | goto bad_load_and_register; | |
822 | } | |
823 | ||
824 | /* | |
825 | * Metadata must only be loaded into one table at once, so skip this | |
826 | * if metadata will be handed over during resume. | |
827 | * Chunk size will be set during the handover - set it to zero to | |
828 | * ensure it's ignored. | |
829 | */ | |
830 | if (r > 0) { | |
831 | s->store->chunk_size = 0; | |
832 | return 0; | |
833 | } | |
834 | ||
493df71c JB |
835 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
836 | (void *)s); | |
0764147b | 837 | if (r < 0) { |
f9cea4f7 | 838 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 839 | goto bad_read_metadata; |
0764147b MB |
840 | } else if (r > 0) { |
841 | s->valid = 0; | |
842 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 843 | } |
aa14edeb | 844 | |
3f2412dc MP |
845 | if (!s->store->chunk_size) { |
846 | ti->error = "Chunk size not set"; | |
c1f0c183 | 847 | goto bad_read_metadata; |
1da177e4 | 848 | } |
d0216849 | 849 | ti->split_io = s->store->chunk_size; |
1da177e4 LT |
850 | |
851 | return 0; | |
852 | ||
c1f0c183 MS |
853 | bad_read_metadata: |
854 | unregister_snapshot(s); | |
855 | ||
fee1998e | 856 | bad_load_and_register: |
cd45daff MP |
857 | mempool_destroy(s->tracked_chunk_pool); |
858 | ||
fee1998e | 859 | bad_tracked_chunk_pool: |
92e86812 MP |
860 | mempool_destroy(s->pending_pool); |
861 | ||
fee1998e | 862 | bad_pending_pool: |
eb69aca5 | 863 | dm_kcopyd_client_destroy(s->kcopyd_client); |
1da177e4 | 864 | |
fee1998e | 865 | bad_kcopyd: |
3510cb94 JB |
866 | dm_exception_table_exit(&s->pending, pending_cache); |
867 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 | 868 | |
fee1998e | 869 | bad_hash_tables: |
1da177e4 LT |
870 | dm_put_device(ti, s->origin); |
871 | ||
fee1998e | 872 | bad_origin: |
fc56f6fb | 873 | dm_exception_store_destroy(s->store); |
1da177e4 | 874 | |
fc56f6fb MS |
875 | bad_store: |
876 | dm_put_device(ti, s->cow); | |
fee1998e | 877 | |
fc56f6fb MS |
878 | bad_cow: |
879 | kfree(s); | |
880 | ||
881 | bad: | |
1da177e4 LT |
882 | return r; |
883 | } | |
884 | ||
31c93a0c MB |
885 | static void __free_exceptions(struct dm_snapshot *s) |
886 | { | |
eb69aca5 | 887 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
888 | s->kcopyd_client = NULL; |
889 | ||
3510cb94 JB |
890 | dm_exception_table_exit(&s->pending, pending_cache); |
891 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
892 | } |
893 | ||
c1f0c183 MS |
894 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
895 | struct dm_snapshot *snap_dest) | |
896 | { | |
897 | union { | |
898 | struct dm_exception_table table_swap; | |
899 | struct dm_exception_store *store_swap; | |
900 | } u; | |
901 | ||
902 | /* | |
903 | * Swap all snapshot context information between the two instances. | |
904 | */ | |
905 | u.table_swap = snap_dest->complete; | |
906 | snap_dest->complete = snap_src->complete; | |
907 | snap_src->complete = u.table_swap; | |
908 | ||
909 | u.store_swap = snap_dest->store; | |
910 | snap_dest->store = snap_src->store; | |
911 | snap_src->store = u.store_swap; | |
912 | ||
913 | snap_dest->store->snap = snap_dest; | |
914 | snap_src->store->snap = snap_src; | |
915 | ||
916 | snap_dest->ti->split_io = snap_dest->store->chunk_size; | |
917 | snap_dest->valid = snap_src->valid; | |
918 | ||
919 | /* | |
920 | * Set source invalid to ensure it receives no further I/O. | |
921 | */ | |
922 | snap_src->valid = 0; | |
923 | } | |
924 | ||
1da177e4 LT |
925 | static void snapshot_dtr(struct dm_target *ti) |
926 | { | |
cd45daff MP |
927 | #ifdef CONFIG_DM_DEBUG |
928 | int i; | |
929 | #endif | |
028867ac | 930 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 931 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 932 | |
ca3a931f AK |
933 | flush_workqueue(ksnapd); |
934 | ||
c1f0c183 MS |
935 | down_read(&_origins_lock); |
936 | /* Check whether exception handover must be cancelled */ | |
937 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
938 | if (snap_src && snap_dest && (s == snap_src)) { | |
939 | down_write(&snap_dest->lock); | |
940 | snap_dest->valid = 0; | |
941 | up_write(&snap_dest->lock); | |
942 | DMERR("Cancelling snapshot handover."); | |
943 | } | |
944 | up_read(&_origins_lock); | |
945 | ||
138728dc AK |
946 | /* Prevent further origin writes from using this snapshot. */ |
947 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
948 | unregister_snapshot(s); |
949 | ||
879129d2 | 950 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 951 | msleep(1); |
879129d2 MP |
952 | /* |
953 | * Ensure instructions in mempool_destroy aren't reordered | |
954 | * before atomic_read. | |
955 | */ | |
956 | smp_mb(); | |
957 | ||
cd45daff MP |
958 | #ifdef CONFIG_DM_DEBUG |
959 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
960 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
961 | #endif | |
962 | ||
963 | mempool_destroy(s->tracked_chunk_pool); | |
964 | ||
31c93a0c | 965 | __free_exceptions(s); |
1da177e4 | 966 | |
92e86812 MP |
967 | mempool_destroy(s->pending_pool); |
968 | ||
1da177e4 | 969 | dm_put_device(ti, s->origin); |
fee1998e JB |
970 | |
971 | dm_exception_store_destroy(s->store); | |
138728dc | 972 | |
fc56f6fb MS |
973 | dm_put_device(ti, s->cow); |
974 | ||
1da177e4 LT |
975 | kfree(s); |
976 | } | |
977 | ||
978 | /* | |
979 | * Flush a list of buffers. | |
980 | */ | |
981 | static void flush_bios(struct bio *bio) | |
982 | { | |
983 | struct bio *n; | |
984 | ||
985 | while (bio) { | |
986 | n = bio->bi_next; | |
987 | bio->bi_next = NULL; | |
988 | generic_make_request(bio); | |
989 | bio = n; | |
990 | } | |
991 | } | |
992 | ||
c4028958 | 993 | static void flush_queued_bios(struct work_struct *work) |
ca3a931f | 994 | { |
c4028958 DH |
995 | struct dm_snapshot *s = |
996 | container_of(work, struct dm_snapshot, queued_bios_work); | |
ca3a931f AK |
997 | struct bio *queued_bios; |
998 | unsigned long flags; | |
999 | ||
1000 | spin_lock_irqsave(&s->pe_lock, flags); | |
1001 | queued_bios = bio_list_get(&s->queued_bios); | |
1002 | spin_unlock_irqrestore(&s->pe_lock, flags); | |
1003 | ||
1004 | flush_bios(queued_bios); | |
1005 | } | |
1006 | ||
1da177e4 LT |
1007 | /* |
1008 | * Error a list of buffers. | |
1009 | */ | |
1010 | static void error_bios(struct bio *bio) | |
1011 | { | |
1012 | struct bio *n; | |
1013 | ||
1014 | while (bio) { | |
1015 | n = bio->bi_next; | |
1016 | bio->bi_next = NULL; | |
6712ecf8 | 1017 | bio_io_error(bio); |
1da177e4 LT |
1018 | bio = n; |
1019 | } | |
1020 | } | |
1021 | ||
695368ac | 1022 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1023 | { |
1024 | if (!s->valid) | |
1025 | return; | |
1026 | ||
1027 | if (err == -EIO) | |
1028 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1029 | else if (err == -ENOMEM) | |
1030 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1031 | ||
493df71c JB |
1032 | if (s->store->type->drop_snapshot) |
1033 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1034 | |
1035 | s->valid = 0; | |
1036 | ||
fc56f6fb | 1037 | dm_table_event(s->ti->table); |
76df1c65 AK |
1038 | } |
1039 | ||
028867ac | 1040 | static void get_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d AK |
1041 | { |
1042 | atomic_inc(&pe->ref_count); | |
1043 | } | |
1044 | ||
028867ac | 1045 | static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe) |
4b832e8d | 1046 | { |
028867ac | 1047 | struct dm_snap_pending_exception *primary_pe; |
4b832e8d AK |
1048 | struct bio *origin_bios = NULL; |
1049 | ||
1050 | primary_pe = pe->primary_pe; | |
1051 | ||
1052 | /* | |
1053 | * If this pe is involved in a write to the origin and | |
1054 | * it is the last sibling to complete then release | |
1055 | * the bios for the original write to the origin. | |
1056 | */ | |
1057 | if (primary_pe && | |
7c5f78b9 | 1058 | atomic_dec_and_test(&primary_pe->ref_count)) { |
4b832e8d | 1059 | origin_bios = bio_list_get(&primary_pe->origin_bios); |
7c5f78b9 MP |
1060 | free_pending_exception(primary_pe); |
1061 | } | |
4b832e8d AK |
1062 | |
1063 | /* | |
1064 | * Free the pe if it's not linked to an origin write or if | |
1065 | * it's not itself a primary pe. | |
1066 | */ | |
1067 | if (!primary_pe || primary_pe != pe) | |
1068 | free_pending_exception(pe); | |
1069 | ||
4b832e8d AK |
1070 | return origin_bios; |
1071 | } | |
1072 | ||
028867ac | 1073 | static void pending_complete(struct dm_snap_pending_exception *pe, int success) |
1da177e4 | 1074 | { |
1d4989c8 | 1075 | struct dm_exception *e; |
1da177e4 | 1076 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1077 | struct bio *origin_bios = NULL; |
1078 | struct bio *snapshot_bios = NULL; | |
1079 | int error = 0; | |
1da177e4 | 1080 | |
76df1c65 AK |
1081 | if (!success) { |
1082 | /* Read/write error - snapshot is unusable */ | |
1da177e4 | 1083 | down_write(&s->lock); |
695368ac | 1084 | __invalidate_snapshot(s, -EIO); |
9d493fa8 | 1085 | error = 1; |
76df1c65 AK |
1086 | goto out; |
1087 | } | |
1088 | ||
3510cb94 | 1089 | e = alloc_completed_exception(); |
76df1c65 | 1090 | if (!e) { |
1da177e4 | 1091 | down_write(&s->lock); |
695368ac | 1092 | __invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1093 | error = 1; |
76df1c65 AK |
1094 | goto out; |
1095 | } | |
1096 | *e = pe->e; | |
1da177e4 | 1097 | |
76df1c65 AK |
1098 | down_write(&s->lock); |
1099 | if (!s->valid) { | |
3510cb94 | 1100 | free_completed_exception(e); |
9d493fa8 | 1101 | error = 1; |
76df1c65 | 1102 | goto out; |
1da177e4 LT |
1103 | } |
1104 | ||
a8d41b59 MP |
1105 | /* |
1106 | * Check for conflicting reads. This is extremely improbable, | |
90fa1527 | 1107 | * so msleep(1) is sufficient and there is no need for a wait queue. |
a8d41b59 MP |
1108 | */ |
1109 | while (__chunk_is_tracked(s, pe->e.old_chunk)) | |
90fa1527 | 1110 | msleep(1); |
a8d41b59 | 1111 | |
9d493fa8 AK |
1112 | /* |
1113 | * Add a proper exception, and remove the | |
1114 | * in-flight exception from the list. | |
1115 | */ | |
3510cb94 | 1116 | dm_insert_exception(&s->complete, e); |
76df1c65 | 1117 | |
1da177e4 | 1118 | out: |
3510cb94 | 1119 | dm_remove_exception(&pe->e); |
9d493fa8 | 1120 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
4b832e8d | 1121 | origin_bios = put_pending_exception(pe); |
1da177e4 | 1122 | |
9d493fa8 AK |
1123 | up_write(&s->lock); |
1124 | ||
1125 | /* Submit any pending write bios */ | |
1126 | if (error) | |
1127 | error_bios(snapshot_bios); | |
1128 | else | |
1129 | flush_bios(snapshot_bios); | |
1130 | ||
1131 | flush_bios(origin_bios); | |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | static void commit_callback(void *context, int success) | |
1135 | { | |
028867ac AK |
1136 | struct dm_snap_pending_exception *pe = context; |
1137 | ||
1da177e4 LT |
1138 | pending_complete(pe, success); |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * Called when the copy I/O has finished. kcopyd actually runs | |
1143 | * this code so don't block. | |
1144 | */ | |
4cdc1d1f | 1145 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1146 | { |
028867ac | 1147 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1148 | struct dm_snapshot *s = pe->snap; |
1149 | ||
1150 | if (read_err || write_err) | |
1151 | pending_complete(pe, 0); | |
1152 | ||
1153 | else | |
1154 | /* Update the metadata if we are persistent */ | |
493df71c JB |
1155 | s->store->type->commit_exception(s->store, &pe->e, |
1156 | commit_callback, pe); | |
1da177e4 LT |
1157 | } |
1158 | ||
1159 | /* | |
1160 | * Dispatches the copy operation to kcopyd. | |
1161 | */ | |
028867ac | 1162 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1163 | { |
1164 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1165 | struct dm_io_region src, dest; |
1da177e4 LT |
1166 | struct block_device *bdev = s->origin->bdev; |
1167 | sector_t dev_size; | |
1168 | ||
1169 | dev_size = get_dev_size(bdev); | |
1170 | ||
1171 | src.bdev = bdev; | |
71fab00a | 1172 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1173 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1174 | |
fc56f6fb | 1175 | dest.bdev = s->cow->bdev; |
71fab00a | 1176 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1177 | dest.count = src.count; |
1178 | ||
1179 | /* Hand over to kcopyd */ | |
eb69aca5 | 1180 | dm_kcopyd_copy(s->kcopyd_client, |
1da177e4 LT |
1181 | &src, 1, &dest, 0, copy_callback, pe); |
1182 | } | |
1183 | ||
2913808e MP |
1184 | static struct dm_snap_pending_exception * |
1185 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1186 | { | |
3510cb94 | 1187 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1188 | |
1189 | if (!e) | |
1190 | return NULL; | |
1191 | ||
1192 | return container_of(e, struct dm_snap_pending_exception, e); | |
1193 | } | |
1194 | ||
1da177e4 LT |
1195 | /* |
1196 | * Looks to see if this snapshot already has a pending exception | |
1197 | * for this chunk, otherwise it allocates a new one and inserts | |
1198 | * it into the pending table. | |
1199 | * | |
1200 | * NOTE: a write lock must be held on snap->lock before calling | |
1201 | * this. | |
1202 | */ | |
028867ac | 1203 | static struct dm_snap_pending_exception * |
c6621392 MP |
1204 | __find_pending_exception(struct dm_snapshot *s, |
1205 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1206 | { |
c6621392 | 1207 | struct dm_snap_pending_exception *pe2; |
1da177e4 | 1208 | |
2913808e MP |
1209 | pe2 = __lookup_pending_exception(s, chunk); |
1210 | if (pe2) { | |
76df1c65 | 1211 | free_pending_exception(pe); |
2913808e | 1212 | return pe2; |
1da177e4 LT |
1213 | } |
1214 | ||
76df1c65 AK |
1215 | pe->e.old_chunk = chunk; |
1216 | bio_list_init(&pe->origin_bios); | |
1217 | bio_list_init(&pe->snapshot_bios); | |
1218 | pe->primary_pe = NULL; | |
4b832e8d | 1219 | atomic_set(&pe->ref_count, 0); |
76df1c65 AK |
1220 | pe->started = 0; |
1221 | ||
493df71c | 1222 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
76df1c65 AK |
1223 | free_pending_exception(pe); |
1224 | return NULL; | |
1225 | } | |
1226 | ||
4b832e8d | 1227 | get_pending_exception(pe); |
3510cb94 | 1228 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1229 | |
1da177e4 LT |
1230 | return pe; |
1231 | } | |
1232 | ||
1d4989c8 | 1233 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1234 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1235 | { |
fc56f6fb | 1236 | bio->bi_bdev = s->cow->bdev; |
71fab00a JB |
1237 | bio->bi_sector = chunk_to_sector(s->store, |
1238 | dm_chunk_number(e->new_chunk) + | |
1239 | (chunk - e->old_chunk)) + | |
1240 | (bio->bi_sector & | |
1241 | s->store->chunk_mask); | |
1da177e4 LT |
1242 | } |
1243 | ||
1244 | static int snapshot_map(struct dm_target *ti, struct bio *bio, | |
1245 | union map_info *map_context) | |
1246 | { | |
1d4989c8 | 1247 | struct dm_exception *e; |
028867ac | 1248 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1249 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1250 | chunk_t chunk; |
028867ac | 1251 | struct dm_snap_pending_exception *pe = NULL; |
1da177e4 | 1252 | |
494b3ee7 | 1253 | if (unlikely(bio_empty_barrier(bio))) { |
fc56f6fb | 1254 | bio->bi_bdev = s->cow->bdev; |
494b3ee7 MP |
1255 | return DM_MAPIO_REMAPPED; |
1256 | } | |
1257 | ||
71fab00a | 1258 | chunk = sector_to_chunk(s->store, bio->bi_sector); |
1da177e4 LT |
1259 | |
1260 | /* Full snapshots are not usable */ | |
76df1c65 | 1261 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1262 | if (!s->valid) |
f6a80ea8 | 1263 | return -EIO; |
1da177e4 | 1264 | |
ba40a2aa AK |
1265 | /* FIXME: should only take write lock if we need |
1266 | * to copy an exception */ | |
1267 | down_write(&s->lock); | |
1268 | ||
1269 | if (!s->valid) { | |
1270 | r = -EIO; | |
1271 | goto out_unlock; | |
1272 | } | |
1273 | ||
1274 | /* If the block is already remapped - use that, else remap it */ | |
3510cb94 | 1275 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 1276 | if (e) { |
d74f81f8 | 1277 | remap_exception(s, e, bio, chunk); |
ba40a2aa AK |
1278 | goto out_unlock; |
1279 | } | |
1280 | ||
1da177e4 LT |
1281 | /* |
1282 | * Write to snapshot - higher level takes care of RW/RO | |
1283 | * flags so we should only get this if we are | |
1284 | * writeable. | |
1285 | */ | |
1286 | if (bio_rw(bio) == WRITE) { | |
2913808e | 1287 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 1288 | if (!pe) { |
c6621392 MP |
1289 | up_write(&s->lock); |
1290 | pe = alloc_pending_exception(s); | |
1291 | down_write(&s->lock); | |
1292 | ||
1293 | if (!s->valid) { | |
1294 | free_pending_exception(pe); | |
1295 | r = -EIO; | |
1296 | goto out_unlock; | |
1297 | } | |
1298 | ||
3510cb94 | 1299 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
1300 | if (e) { |
1301 | free_pending_exception(pe); | |
1302 | remap_exception(s, e, bio, chunk); | |
1303 | goto out_unlock; | |
1304 | } | |
1305 | ||
c6621392 | 1306 | pe = __find_pending_exception(s, pe, chunk); |
2913808e MP |
1307 | if (!pe) { |
1308 | __invalidate_snapshot(s, -ENOMEM); | |
1309 | r = -EIO; | |
1310 | goto out_unlock; | |
1311 | } | |
1da177e4 LT |
1312 | } |
1313 | ||
d74f81f8 | 1314 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 AK |
1315 | bio_list_add(&pe->snapshot_bios, bio); |
1316 | ||
d2a7ad29 | 1317 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 1318 | |
76df1c65 AK |
1319 | if (!pe->started) { |
1320 | /* this is protected by snap->lock */ | |
1321 | pe->started = 1; | |
ba40a2aa | 1322 | up_write(&s->lock); |
76df1c65 | 1323 | start_copy(pe); |
ba40a2aa AK |
1324 | goto out; |
1325 | } | |
cd45daff | 1326 | } else { |
ba40a2aa | 1327 | bio->bi_bdev = s->origin->bdev; |
cd45daff MP |
1328 | map_context->ptr = track_chunk(s, chunk); |
1329 | } | |
1da177e4 | 1330 | |
ba40a2aa AK |
1331 | out_unlock: |
1332 | up_write(&s->lock); | |
1333 | out: | |
1da177e4 LT |
1334 | return r; |
1335 | } | |
1336 | ||
cd45daff MP |
1337 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
1338 | int error, union map_info *map_context) | |
1339 | { | |
1340 | struct dm_snapshot *s = ti->private; | |
1341 | struct dm_snap_tracked_chunk *c = map_context->ptr; | |
1342 | ||
1343 | if (c) | |
1344 | stop_tracking_chunk(s, c); | |
1345 | ||
1346 | return 0; | |
1347 | } | |
1348 | ||
c26655ca MS |
1349 | static void snapshot_postsuspend(struct dm_target *ti) |
1350 | { | |
1351 | struct dm_snapshot *s = ti->private; | |
1352 | ||
1353 | down_write(&s->lock); | |
1354 | s->suspended = 1; | |
1355 | up_write(&s->lock); | |
1356 | } | |
1357 | ||
c1f0c183 MS |
1358 | static int snapshot_preresume(struct dm_target *ti) |
1359 | { | |
1360 | int r = 0; | |
1361 | struct dm_snapshot *s = ti->private; | |
1362 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
1363 | ||
1364 | down_read(&_origins_lock); | |
1365 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1366 | if (snap_src && snap_dest) { | |
1367 | down_read(&snap_src->lock); | |
1368 | if (s == snap_src) { | |
1369 | DMERR("Unable to resume snapshot source until " | |
1370 | "handover completes."); | |
1371 | r = -EINVAL; | |
1372 | } else if (!snap_src->suspended) { | |
1373 | DMERR("Unable to perform snapshot handover until " | |
1374 | "source is suspended."); | |
1375 | r = -EINVAL; | |
1376 | } | |
1377 | up_read(&snap_src->lock); | |
1378 | } | |
1379 | up_read(&_origins_lock); | |
1380 | ||
1381 | return r; | |
1382 | } | |
1383 | ||
1da177e4 LT |
1384 | static void snapshot_resume(struct dm_target *ti) |
1385 | { | |
028867ac | 1386 | struct dm_snapshot *s = ti->private; |
c1f0c183 MS |
1387 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1388 | ||
1389 | down_read(&_origins_lock); | |
1390 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest); | |
1391 | if (snap_src && snap_dest) { | |
1392 | down_write(&snap_src->lock); | |
1393 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
1394 | __handover_exceptions(snap_src, snap_dest); | |
1395 | up_write(&snap_dest->lock); | |
1396 | up_write(&snap_src->lock); | |
1397 | } | |
1398 | up_read(&_origins_lock); | |
1399 | ||
1400 | /* Now we have correct chunk size, reregister */ | |
1401 | reregister_snapshot(s); | |
1da177e4 | 1402 | |
aa14edeb AK |
1403 | down_write(&s->lock); |
1404 | s->active = 1; | |
c26655ca | 1405 | s->suspended = 0; |
aa14edeb | 1406 | up_write(&s->lock); |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | static int snapshot_status(struct dm_target *ti, status_type_t type, | |
1410 | char *result, unsigned int maxlen) | |
1411 | { | |
2e4a31df | 1412 | unsigned sz = 0; |
028867ac | 1413 | struct dm_snapshot *snap = ti->private; |
1da177e4 LT |
1414 | |
1415 | switch (type) { | |
1416 | case STATUSTYPE_INFO: | |
94e76572 MP |
1417 | |
1418 | down_write(&snap->lock); | |
1419 | ||
1da177e4 | 1420 | if (!snap->valid) |
2e4a31df | 1421 | DMEMIT("Invalid"); |
1da177e4 | 1422 | else { |
985903bb MS |
1423 | if (snap->store->type->usage) { |
1424 | sector_t total_sectors, sectors_allocated, | |
1425 | metadata_sectors; | |
1426 | snap->store->type->usage(snap->store, | |
1427 | &total_sectors, | |
1428 | §ors_allocated, | |
1429 | &metadata_sectors); | |
1430 | DMEMIT("%llu/%llu %llu", | |
1431 | (unsigned long long)sectors_allocated, | |
1432 | (unsigned long long)total_sectors, | |
1433 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
1434 | } |
1435 | else | |
2e4a31df | 1436 | DMEMIT("Unknown"); |
1da177e4 | 1437 | } |
94e76572 MP |
1438 | |
1439 | up_write(&snap->lock); | |
1440 | ||
1da177e4 LT |
1441 | break; |
1442 | ||
1443 | case STATUSTYPE_TABLE: | |
1444 | /* | |
1445 | * kdevname returns a static pointer so we need | |
1446 | * to make private copies if the output is to | |
1447 | * make sense. | |
1448 | */ | |
fc56f6fb | 1449 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
1e302a92 JB |
1450 | snap->store->type->status(snap->store, type, result + sz, |
1451 | maxlen - sz); | |
1da177e4 LT |
1452 | break; |
1453 | } | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | ||
8811f46c MS |
1458 | static int snapshot_iterate_devices(struct dm_target *ti, |
1459 | iterate_devices_callout_fn fn, void *data) | |
1460 | { | |
1461 | struct dm_snapshot *snap = ti->private; | |
1462 | ||
1463 | return fn(ti, snap->origin, 0, ti->len, data); | |
1464 | } | |
1465 | ||
1466 | ||
1da177e4 LT |
1467 | /*----------------------------------------------------------------- |
1468 | * Origin methods | |
1469 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
1470 | |
1471 | /* | |
1472 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
1473 | * supplied bio was ignored. The caller may submit it immediately. | |
1474 | * (No remapping actually occurs as the origin is always a direct linear | |
1475 | * map.) | |
1476 | * | |
1477 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
1478 | * and any supplied bio is added to a list to be submitted once all | |
1479 | * the necessary exceptions exist. | |
1480 | */ | |
1481 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
1482 | struct bio *bio) | |
1da177e4 | 1483 | { |
d2a7ad29 | 1484 | int r = DM_MAPIO_REMAPPED, first = 0; |
1da177e4 | 1485 | struct dm_snapshot *snap; |
1d4989c8 | 1486 | struct dm_exception *e; |
028867ac | 1487 | struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL; |
1da177e4 | 1488 | chunk_t chunk; |
eccf0817 | 1489 | LIST_HEAD(pe_queue); |
1da177e4 LT |
1490 | |
1491 | /* Do all the snapshots on this origin */ | |
1492 | list_for_each_entry (snap, snapshots, list) { | |
1493 | ||
76df1c65 AK |
1494 | down_write(&snap->lock); |
1495 | ||
aa14edeb AK |
1496 | /* Only deal with valid and active snapshots */ |
1497 | if (!snap->valid || !snap->active) | |
76df1c65 | 1498 | goto next_snapshot; |
1da177e4 | 1499 | |
d5e404c1 | 1500 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 1501 | if (sector >= dm_table_get_size(snap->ti->table)) |
76df1c65 | 1502 | goto next_snapshot; |
1da177e4 LT |
1503 | |
1504 | /* | |
1505 | * Remember, different snapshots can have | |
1506 | * different chunk sizes. | |
1507 | */ | |
9eaae8ff | 1508 | chunk = sector_to_chunk(snap->store, sector); |
1da177e4 LT |
1509 | |
1510 | /* | |
1511 | * Check exception table to see if block | |
1512 | * is already remapped in this snapshot | |
1513 | * and trigger an exception if not. | |
b4b610f6 | 1514 | * |
4b832e8d | 1515 | * ref_count is initialised to 1 so pending_complete() |
b4b610f6 | 1516 | * won't destroy the primary_pe while we're inside this loop. |
1da177e4 | 1517 | */ |
3510cb94 | 1518 | e = dm_lookup_exception(&snap->complete, chunk); |
76df1c65 AK |
1519 | if (e) |
1520 | goto next_snapshot; | |
1521 | ||
2913808e | 1522 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 1523 | if (!pe) { |
c6621392 MP |
1524 | up_write(&snap->lock); |
1525 | pe = alloc_pending_exception(snap); | |
1526 | down_write(&snap->lock); | |
1527 | ||
1528 | if (!snap->valid) { | |
1529 | free_pending_exception(pe); | |
1530 | goto next_snapshot; | |
1531 | } | |
1532 | ||
3510cb94 | 1533 | e = dm_lookup_exception(&snap->complete, chunk); |
35bf659b MP |
1534 | if (e) { |
1535 | free_pending_exception(pe); | |
1536 | goto next_snapshot; | |
1537 | } | |
1538 | ||
c6621392 | 1539 | pe = __find_pending_exception(snap, pe, chunk); |
2913808e MP |
1540 | if (!pe) { |
1541 | __invalidate_snapshot(snap, -ENOMEM); | |
1542 | goto next_snapshot; | |
1543 | } | |
76df1c65 AK |
1544 | } |
1545 | ||
1546 | if (!primary_pe) { | |
1547 | /* | |
1548 | * Either every pe here has same | |
1549 | * primary_pe or none has one yet. | |
1550 | */ | |
1551 | if (pe->primary_pe) | |
1552 | primary_pe = pe->primary_pe; | |
1553 | else { | |
1554 | primary_pe = pe; | |
1555 | first = 1; | |
1da177e4 | 1556 | } |
76df1c65 | 1557 | |
9eaae8ff MP |
1558 | if (bio) |
1559 | bio_list_add(&primary_pe->origin_bios, bio); | |
76df1c65 | 1560 | |
d2a7ad29 | 1561 | r = DM_MAPIO_SUBMITTED; |
76df1c65 AK |
1562 | } |
1563 | ||
1564 | if (!pe->primary_pe) { | |
76df1c65 | 1565 | pe->primary_pe = primary_pe; |
4b832e8d | 1566 | get_pending_exception(primary_pe); |
76df1c65 AK |
1567 | } |
1568 | ||
1569 | if (!pe->started) { | |
1570 | pe->started = 1; | |
1571 | list_add_tail(&pe->list, &pe_queue); | |
1da177e4 LT |
1572 | } |
1573 | ||
76df1c65 | 1574 | next_snapshot: |
1da177e4 LT |
1575 | up_write(&snap->lock); |
1576 | } | |
1577 | ||
b4b610f6 | 1578 | if (!primary_pe) |
4b832e8d | 1579 | return r; |
b4b610f6 AK |
1580 | |
1581 | /* | |
1582 | * If this is the first time we're processing this chunk and | |
4b832e8d | 1583 | * ref_count is now 1 it means all the pending exceptions |
b4b610f6 AK |
1584 | * got completed while we were in the loop above, so it falls to |
1585 | * us here to remove the primary_pe and submit any origin_bios. | |
1586 | */ | |
1587 | ||
4b832e8d | 1588 | if (first && atomic_dec_and_test(&primary_pe->ref_count)) { |
b4b610f6 AK |
1589 | flush_bios(bio_list_get(&primary_pe->origin_bios)); |
1590 | free_pending_exception(primary_pe); | |
1591 | /* If we got here, pe_queue is necessarily empty. */ | |
4b832e8d | 1592 | return r; |
b4b610f6 AK |
1593 | } |
1594 | ||
1da177e4 LT |
1595 | /* |
1596 | * Now that we have a complete pe list we can start the copying. | |
1597 | */ | |
eccf0817 AK |
1598 | list_for_each_entry_safe(pe, next_pe, &pe_queue, list) |
1599 | start_copy(pe); | |
1da177e4 LT |
1600 | |
1601 | return r; | |
1602 | } | |
1603 | ||
1604 | /* | |
1605 | * Called on a write from the origin driver. | |
1606 | */ | |
1607 | static int do_origin(struct dm_dev *origin, struct bio *bio) | |
1608 | { | |
1609 | struct origin *o; | |
d2a7ad29 | 1610 | int r = DM_MAPIO_REMAPPED; |
1da177e4 LT |
1611 | |
1612 | down_read(&_origins_lock); | |
1613 | o = __lookup_origin(origin->bdev); | |
1614 | if (o) | |
9eaae8ff | 1615 | r = __origin_write(&o->snapshots, bio->bi_sector, bio); |
1da177e4 LT |
1616 | up_read(&_origins_lock); |
1617 | ||
1618 | return r; | |
1619 | } | |
1620 | ||
1621 | /* | |
1622 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
1623 | */ | |
1624 | ||
1625 | /* | |
1626 | * Construct an origin mapping: <dev_path> | |
1627 | * The context for an origin is merely a 'struct dm_dev *' | |
1628 | * pointing to the real device. | |
1629 | */ | |
1630 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1631 | { | |
1632 | int r; | |
1633 | struct dm_dev *dev; | |
1634 | ||
1635 | if (argc != 1) { | |
72d94861 | 1636 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
1637 | return -EINVAL; |
1638 | } | |
1639 | ||
1640 | r = dm_get_device(ti, argv[0], 0, ti->len, | |
1641 | dm_table_get_mode(ti->table), &dev); | |
1642 | if (r) { | |
1643 | ti->error = "Cannot get target device"; | |
1644 | return r; | |
1645 | } | |
1646 | ||
1647 | ti->private = dev; | |
494b3ee7 MP |
1648 | ti->num_flush_requests = 1; |
1649 | ||
1da177e4 LT |
1650 | return 0; |
1651 | } | |
1652 | ||
1653 | static void origin_dtr(struct dm_target *ti) | |
1654 | { | |
028867ac | 1655 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1656 | dm_put_device(ti, dev); |
1657 | } | |
1658 | ||
1659 | static int origin_map(struct dm_target *ti, struct bio *bio, | |
1660 | union map_info *map_context) | |
1661 | { | |
028867ac | 1662 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1663 | bio->bi_bdev = dev->bdev; |
1664 | ||
494b3ee7 MP |
1665 | if (unlikely(bio_empty_barrier(bio))) |
1666 | return DM_MAPIO_REMAPPED; | |
1667 | ||
1da177e4 | 1668 | /* Only tell snapshots if this is a write */ |
d2a7ad29 | 1669 | return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED; |
1da177e4 LT |
1670 | } |
1671 | ||
1da177e4 LT |
1672 | /* |
1673 | * Set the target "split_io" field to the minimum of all the snapshots' | |
1674 | * chunk sizes. | |
1675 | */ | |
1676 | static void origin_resume(struct dm_target *ti) | |
1677 | { | |
028867ac | 1678 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1679 | |
1680 | down_read(&_origins_lock); | |
1da177e4 | 1681 | |
7e201b35 MP |
1682 | ti->split_io = __minimum_chunk_size(__lookup_origin(dev->bdev)); |
1683 | ||
1684 | up_read(&_origins_lock); | |
1da177e4 LT |
1685 | } |
1686 | ||
1687 | static int origin_status(struct dm_target *ti, status_type_t type, char *result, | |
1688 | unsigned int maxlen) | |
1689 | { | |
028867ac | 1690 | struct dm_dev *dev = ti->private; |
1da177e4 LT |
1691 | |
1692 | switch (type) { | |
1693 | case STATUSTYPE_INFO: | |
1694 | result[0] = '\0'; | |
1695 | break; | |
1696 | ||
1697 | case STATUSTYPE_TABLE: | |
1698 | snprintf(result, maxlen, "%s", dev->name); | |
1699 | break; | |
1700 | } | |
1701 | ||
1702 | return 0; | |
1703 | } | |
1704 | ||
8811f46c MS |
1705 | static int origin_iterate_devices(struct dm_target *ti, |
1706 | iterate_devices_callout_fn fn, void *data) | |
1707 | { | |
1708 | struct dm_dev *dev = ti->private; | |
1709 | ||
1710 | return fn(ti, dev, 0, ti->len, data); | |
1711 | } | |
1712 | ||
1da177e4 LT |
1713 | static struct target_type origin_target = { |
1714 | .name = "snapshot-origin", | |
8811f46c | 1715 | .version = {1, 7, 0}, |
1da177e4 LT |
1716 | .module = THIS_MODULE, |
1717 | .ctr = origin_ctr, | |
1718 | .dtr = origin_dtr, | |
1719 | .map = origin_map, | |
1720 | .resume = origin_resume, | |
1721 | .status = origin_status, | |
8811f46c | 1722 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
1723 | }; |
1724 | ||
1725 | static struct target_type snapshot_target = { | |
1726 | .name = "snapshot", | |
c26655ca | 1727 | .version = {1, 9, 0}, |
1da177e4 LT |
1728 | .module = THIS_MODULE, |
1729 | .ctr = snapshot_ctr, | |
1730 | .dtr = snapshot_dtr, | |
1731 | .map = snapshot_map, | |
cd45daff | 1732 | .end_io = snapshot_end_io, |
c26655ca | 1733 | .postsuspend = snapshot_postsuspend, |
c1f0c183 | 1734 | .preresume = snapshot_preresume, |
1da177e4 LT |
1735 | .resume = snapshot_resume, |
1736 | .status = snapshot_status, | |
8811f46c | 1737 | .iterate_devices = snapshot_iterate_devices, |
1da177e4 LT |
1738 | }; |
1739 | ||
1740 | static int __init dm_snapshot_init(void) | |
1741 | { | |
1742 | int r; | |
1743 | ||
4db6bfe0 AK |
1744 | r = dm_exception_store_init(); |
1745 | if (r) { | |
1746 | DMERR("Failed to initialize exception stores"); | |
1747 | return r; | |
1748 | } | |
1749 | ||
1da177e4 LT |
1750 | r = dm_register_target(&snapshot_target); |
1751 | if (r) { | |
1752 | DMERR("snapshot target register failed %d", r); | |
034a186d | 1753 | goto bad_register_snapshot_target; |
1da177e4 LT |
1754 | } |
1755 | ||
1756 | r = dm_register_target(&origin_target); | |
1757 | if (r < 0) { | |
72d94861 | 1758 | DMERR("Origin target register failed %d", r); |
1da177e4 LT |
1759 | goto bad1; |
1760 | } | |
1761 | ||
1762 | r = init_origin_hash(); | |
1763 | if (r) { | |
1764 | DMERR("init_origin_hash failed."); | |
1765 | goto bad2; | |
1766 | } | |
1767 | ||
1d4989c8 | 1768 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
1769 | if (!exception_cache) { |
1770 | DMERR("Couldn't create exception cache."); | |
1771 | r = -ENOMEM; | |
1772 | goto bad3; | |
1773 | } | |
1774 | ||
028867ac | 1775 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
1776 | if (!pending_cache) { |
1777 | DMERR("Couldn't create pending cache."); | |
1778 | r = -ENOMEM; | |
1779 | goto bad4; | |
1780 | } | |
1781 | ||
cd45daff MP |
1782 | tracked_chunk_cache = KMEM_CACHE(dm_snap_tracked_chunk, 0); |
1783 | if (!tracked_chunk_cache) { | |
1784 | DMERR("Couldn't create cache to track chunks in use."); | |
1785 | r = -ENOMEM; | |
1786 | goto bad5; | |
1787 | } | |
1788 | ||
ca3a931f AK |
1789 | ksnapd = create_singlethread_workqueue("ksnapd"); |
1790 | if (!ksnapd) { | |
1791 | DMERR("Failed to create ksnapd workqueue."); | |
1792 | r = -ENOMEM; | |
92e86812 | 1793 | goto bad_pending_pool; |
ca3a931f AK |
1794 | } |
1795 | ||
1da177e4 LT |
1796 | return 0; |
1797 | ||
4db6bfe0 | 1798 | bad_pending_pool: |
cd45daff | 1799 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 | 1800 | bad5: |
1da177e4 | 1801 | kmem_cache_destroy(pending_cache); |
4db6bfe0 | 1802 | bad4: |
1da177e4 | 1803 | kmem_cache_destroy(exception_cache); |
4db6bfe0 | 1804 | bad3: |
1da177e4 | 1805 | exit_origin_hash(); |
4db6bfe0 | 1806 | bad2: |
1da177e4 | 1807 | dm_unregister_target(&origin_target); |
4db6bfe0 | 1808 | bad1: |
1da177e4 | 1809 | dm_unregister_target(&snapshot_target); |
034a186d JB |
1810 | |
1811 | bad_register_snapshot_target: | |
1812 | dm_exception_store_exit(); | |
1da177e4 LT |
1813 | return r; |
1814 | } | |
1815 | ||
1816 | static void __exit dm_snapshot_exit(void) | |
1817 | { | |
ca3a931f AK |
1818 | destroy_workqueue(ksnapd); |
1819 | ||
10d3bd09 MP |
1820 | dm_unregister_target(&snapshot_target); |
1821 | dm_unregister_target(&origin_target); | |
1da177e4 LT |
1822 | |
1823 | exit_origin_hash(); | |
1da177e4 LT |
1824 | kmem_cache_destroy(pending_cache); |
1825 | kmem_cache_destroy(exception_cache); | |
cd45daff | 1826 | kmem_cache_destroy(tracked_chunk_cache); |
4db6bfe0 AK |
1827 | |
1828 | dm_exception_store_exit(); | |
1da177e4 LT |
1829 | } |
1830 | ||
1831 | /* Module hooks */ | |
1832 | module_init(dm_snapshot_init); | |
1833 | module_exit(dm_snapshot_exit); | |
1834 | ||
1835 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
1836 | MODULE_AUTHOR("Joe Thornber"); | |
1837 | MODULE_LICENSE("GPL"); |