Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include <linux/blkdev.h> | |
1da177e4 | 8 | #include <linux/device-mapper.h> |
90fa1527 | 9 | #include <linux/delay.h> |
1da177e4 LT |
10 | #include <linux/fs.h> |
11 | #include <linux/init.h> | |
12 | #include <linux/kdev_t.h> | |
13 | #include <linux/list.h> | |
f79ae415 | 14 | #include <linux/list_bl.h> |
1da177e4 LT |
15 | #include <linux/mempool.h> |
16 | #include <linux/module.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/vmalloc.h> | |
6f3c3f0a | 19 | #include <linux/log2.h> |
a765e20e | 20 | #include <linux/dm-kcopyd.h> |
1da177e4 | 21 | |
b735fede MP |
22 | #include "dm.h" |
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 | ||
cd45daff MP |
33 | /* |
34 | * The size of the mempool used to track chunks in use. | |
35 | */ | |
36 | #define MIN_IOS 256 | |
37 | ||
ccc45ea8 JB |
38 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
39 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
40 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
41 | ||
191437a5 | 42 | struct dm_exception_table { |
ccc45ea8 JB |
43 | uint32_t hash_mask; |
44 | unsigned hash_shift; | |
f79ae415 | 45 | struct hlist_bl_head *table; |
ccc45ea8 JB |
46 | }; |
47 | ||
48 | struct dm_snapshot { | |
4ad8d880 | 49 | struct rw_semaphore lock; |
ccc45ea8 JB |
50 | |
51 | struct dm_dev *origin; | |
fc56f6fb MS |
52 | struct dm_dev *cow; |
53 | ||
54 | struct dm_target *ti; | |
ccc45ea8 JB |
55 | |
56 | /* List of snapshots per Origin */ | |
57 | struct list_head list; | |
58 | ||
d8ddb1cf MS |
59 | /* |
60 | * You can't use a snapshot if this is 0 (e.g. if full). | |
61 | * A snapshot-merge target never clears this. | |
62 | */ | |
ccc45ea8 JB |
63 | int valid; |
64 | ||
76c44f6d MP |
65 | /* |
66 | * The snapshot overflowed because of a write to the snapshot device. | |
67 | * We don't have to invalidate the snapshot in this case, but we need | |
68 | * to prevent further writes. | |
69 | */ | |
70 | int snapshot_overflowed; | |
71 | ||
ccc45ea8 JB |
72 | /* Origin writes don't trigger exceptions until this is set */ |
73 | int active; | |
74 | ||
ccc45ea8 JB |
75 | atomic_t pending_exceptions_count; |
76 | ||
3f1637f2 NT |
77 | spinlock_t pe_allocation_lock; |
78 | ||
79 | /* Protected by "pe_allocation_lock" */ | |
230c83af MP |
80 | sector_t exception_start_sequence; |
81 | ||
82 | /* Protected by kcopyd single-threaded callback */ | |
83 | sector_t exception_complete_sequence; | |
84 | ||
85 | /* | |
86 | * A list of pending exceptions that completed out of order. | |
87 | * Protected by kcopyd single-threaded callback. | |
88 | */ | |
3db2776d | 89 | struct rb_root out_of_order_tree; |
230c83af | 90 | |
6f1c819c | 91 | mempool_t pending_pool; |
924e600d | 92 | |
191437a5 JB |
93 | struct dm_exception_table pending; |
94 | struct dm_exception_table complete; | |
ccc45ea8 JB |
95 | |
96 | /* | |
97 | * pe_lock protects all pending_exception operations and access | |
98 | * as well as the snapshot_bios list. | |
99 | */ | |
100 | spinlock_t pe_lock; | |
101 | ||
924e600d MS |
102 | /* Chunks with outstanding reads */ |
103 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
104 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
105 | ||
ccc45ea8 JB |
106 | /* The on disk metadata handler */ |
107 | struct dm_exception_store *store; | |
108 | ||
b2155578 MP |
109 | unsigned in_progress; |
110 | struct wait_queue_head in_progress_wait; | |
721b1d98 | 111 | |
ccc45ea8 JB |
112 | struct dm_kcopyd_client *kcopyd_client; |
113 | ||
924e600d MS |
114 | /* Wait for events based on state_bits */ |
115 | unsigned long state_bits; | |
116 | ||
117 | /* Range of chunks currently being merged. */ | |
118 | chunk_t first_merging_chunk; | |
119 | int num_merging_chunks; | |
1e03f97e | 120 | |
d8ddb1cf MS |
121 | /* |
122 | * The merge operation failed if this flag is set. | |
123 | * Failure modes are handled as follows: | |
124 | * - I/O error reading the header | |
125 | * => don't load the target; abort. | |
126 | * - Header does not have "valid" flag set | |
127 | * => use the origin; forget about the snapshot. | |
128 | * - I/O error when reading exceptions | |
129 | * => don't load the target; abort. | |
130 | * (We can't use the intermediate origin state.) | |
131 | * - I/O error while merging | |
132 | * => stop merging; set merge_failed; process I/O normally. | |
133 | */ | |
2e602385 MS |
134 | bool merge_failed:1; |
135 | ||
136 | bool discard_zeroes_cow:1; | |
137 | bool discard_passdown_origin:1; | |
d8ddb1cf | 138 | |
9fe86254 MP |
139 | /* |
140 | * Incoming bios that overlap with chunks being merged must wait | |
141 | * for them to be committed. | |
142 | */ | |
143 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
144 | }; |
145 | ||
1e03f97e MP |
146 | /* |
147 | * state_bits: | |
148 | * RUNNING_MERGE - Merge operation is in progress. | |
149 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
150 | * cleared afterwards. | |
151 | */ | |
152 | #define RUNNING_MERGE 0 | |
153 | #define SHUTDOWN_MERGE 1 | |
154 | ||
721b1d98 NT |
155 | /* |
156 | * Maximum number of chunks being copied on write. | |
157 | * | |
158 | * The value was decided experimentally as a trade-off between memory | |
159 | * consumption, stalling the kernel's workqueues and maintaining a high enough | |
160 | * throughput. | |
161 | */ | |
162 | #define DEFAULT_COW_THRESHOLD 2048 | |
163 | ||
b2155578 MP |
164 | static unsigned cow_threshold = DEFAULT_COW_THRESHOLD; |
165 | module_param_named(snapshot_cow_threshold, cow_threshold, uint, 0644); | |
721b1d98 NT |
166 | MODULE_PARM_DESC(snapshot_cow_threshold, "Maximum number of chunks being copied on write"); |
167 | ||
df5d2e90 MP |
168 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
169 | "A percentage of time allocated for copy on write"); | |
170 | ||
c2411045 MP |
171 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
172 | { | |
173 | return s->origin; | |
174 | } | |
175 | EXPORT_SYMBOL(dm_snap_origin); | |
176 | ||
fc56f6fb MS |
177 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
178 | { | |
179 | return s->cow; | |
180 | } | |
181 | EXPORT_SYMBOL(dm_snap_cow); | |
182 | ||
ccc45ea8 JB |
183 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
184 | chunk_t chunk) | |
185 | { | |
186 | return chunk << store->chunk_shift; | |
187 | } | |
188 | ||
189 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
190 | { | |
191 | /* | |
192 | * There is only ever one instance of a particular block | |
193 | * device so we can compare pointers safely. | |
194 | */ | |
195 | return lhs == rhs; | |
196 | } | |
197 | ||
028867ac | 198 | struct dm_snap_pending_exception { |
1d4989c8 | 199 | struct dm_exception e; |
1da177e4 LT |
200 | |
201 | /* | |
202 | * Origin buffers waiting for this to complete are held | |
203 | * in a bio list | |
204 | */ | |
205 | struct bio_list origin_bios; | |
206 | struct bio_list snapshot_bios; | |
207 | ||
1da177e4 LT |
208 | /* Pointer back to snapshot context */ |
209 | struct dm_snapshot *snap; | |
210 | ||
211 | /* | |
212 | * 1 indicates the exception has already been sent to | |
213 | * kcopyd. | |
214 | */ | |
215 | int started; | |
a6e50b40 | 216 | |
230c83af MP |
217 | /* There was copying error. */ |
218 | int copy_error; | |
219 | ||
220 | /* A sequence number, it is used for in-order completion. */ | |
221 | sector_t exception_sequence; | |
222 | ||
3db2776d | 223 | struct rb_node out_of_order_node; |
230c83af | 224 | |
a6e50b40 MP |
225 | /* |
226 | * For writing a complete chunk, bypassing the copy. | |
227 | */ | |
228 | struct bio *full_bio; | |
229 | bio_end_io_t *full_bio_end_io; | |
1da177e4 LT |
230 | }; |
231 | ||
232 | /* | |
233 | * Hash table mapping origin volumes to lists of snapshots and | |
234 | * a lock to protect it | |
235 | */ | |
e18b890b CL |
236 | static struct kmem_cache *exception_cache; |
237 | static struct kmem_cache *pending_cache; | |
1da177e4 | 238 | |
cd45daff MP |
239 | struct dm_snap_tracked_chunk { |
240 | struct hlist_node node; | |
241 | chunk_t chunk; | |
242 | }; | |
243 | ||
ee18026a MP |
244 | static void init_tracked_chunk(struct bio *bio) |
245 | { | |
246 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
247 | INIT_HLIST_NODE(&c->node); | |
248 | } | |
249 | ||
250 | static bool is_bio_tracked(struct bio *bio) | |
251 | { | |
252 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
253 | return !hlist_unhashed(&c->node); | |
254 | } | |
255 | ||
256 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 257 | { |
42bc954f | 258 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
259 | |
260 | c->chunk = chunk; | |
261 | ||
9aa0c0e6 | 262 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
263 | hlist_add_head(&c->node, |
264 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 265 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
266 | } |
267 | ||
ee18026a | 268 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 269 | { |
ee18026a | 270 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
271 | unsigned long flags; |
272 | ||
273 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
274 | hlist_del(&c->node); | |
275 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
276 | } |
277 | ||
a8d41b59 MP |
278 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
279 | { | |
280 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
281 | int found = 0; |
282 | ||
283 | spin_lock_irq(&s->tracked_chunk_lock); | |
284 | ||
b67bfe0d | 285 | hlist_for_each_entry(c, |
a8d41b59 MP |
286 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
287 | if (c->chunk == chunk) { | |
288 | found = 1; | |
289 | break; | |
290 | } | |
291 | } | |
292 | ||
293 | spin_unlock_irq(&s->tracked_chunk_lock); | |
294 | ||
295 | return found; | |
296 | } | |
297 | ||
615d1eb9 MS |
298 | /* |
299 | * This conflicting I/O is extremely improbable in the caller, | |
300 | * so msleep(1) is sufficient and there is no need for a wait queue. | |
301 | */ | |
302 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
303 | { | |
304 | while (__chunk_is_tracked(s, chunk)) | |
305 | msleep(1); | |
306 | } | |
307 | ||
1da177e4 LT |
308 | /* |
309 | * One of these per registered origin, held in the snapshot_origins hash | |
310 | */ | |
311 | struct origin { | |
312 | /* The origin device */ | |
313 | struct block_device *bdev; | |
314 | ||
315 | struct list_head hash_list; | |
316 | ||
317 | /* List of snapshots for this origin */ | |
318 | struct list_head snapshots; | |
319 | }; | |
320 | ||
b735fede MP |
321 | /* |
322 | * This structure is allocated for each origin target | |
323 | */ | |
324 | struct dm_origin { | |
325 | struct dm_dev *dev; | |
326 | struct dm_target *ti; | |
327 | unsigned split_boundary; | |
328 | struct list_head hash_list; | |
329 | }; | |
330 | ||
1da177e4 LT |
331 | /* |
332 | * Size of the hash table for origin volumes. If we make this | |
333 | * the size of the minors list then it should be nearly perfect | |
334 | */ | |
335 | #define ORIGIN_HASH_SIZE 256 | |
336 | #define ORIGIN_MASK 0xFF | |
337 | static struct list_head *_origins; | |
b735fede | 338 | static struct list_head *_dm_origins; |
1da177e4 LT |
339 | static struct rw_semaphore _origins_lock; |
340 | ||
73dfd078 MP |
341 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
342 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
343 | static uint64_t _pending_exceptions_done_count; | |
344 | ||
1da177e4 LT |
345 | static int init_origin_hash(void) |
346 | { | |
347 | int i; | |
348 | ||
6da2ec56 KC |
349 | _origins = kmalloc_array(ORIGIN_HASH_SIZE, sizeof(struct list_head), |
350 | GFP_KERNEL); | |
1da177e4 | 351 | if (!_origins) { |
b735fede | 352 | DMERR("unable to allocate memory for _origins"); |
1da177e4 LT |
353 | return -ENOMEM; |
354 | } | |
1da177e4 LT |
355 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
356 | INIT_LIST_HEAD(_origins + i); | |
b735fede | 357 | |
6da2ec56 KC |
358 | _dm_origins = kmalloc_array(ORIGIN_HASH_SIZE, |
359 | sizeof(struct list_head), | |
360 | GFP_KERNEL); | |
b735fede MP |
361 | if (!_dm_origins) { |
362 | DMERR("unable to allocate memory for _dm_origins"); | |
363 | kfree(_origins); | |
364 | return -ENOMEM; | |
365 | } | |
366 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
367 | INIT_LIST_HEAD(_dm_origins + i); | |
368 | ||
1da177e4 LT |
369 | init_rwsem(&_origins_lock); |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void exit_origin_hash(void) | |
375 | { | |
376 | kfree(_origins); | |
b735fede | 377 | kfree(_dm_origins); |
1da177e4 LT |
378 | } |
379 | ||
028867ac | 380 | static unsigned origin_hash(struct block_device *bdev) |
1da177e4 LT |
381 | { |
382 | return bdev->bd_dev & ORIGIN_MASK; | |
383 | } | |
384 | ||
385 | static struct origin *__lookup_origin(struct block_device *origin) | |
386 | { | |
387 | struct list_head *ol; | |
388 | struct origin *o; | |
389 | ||
390 | ol = &_origins[origin_hash(origin)]; | |
391 | list_for_each_entry (o, ol, hash_list) | |
392 | if (bdev_equal(o->bdev, origin)) | |
393 | return o; | |
394 | ||
395 | return NULL; | |
396 | } | |
397 | ||
398 | static void __insert_origin(struct origin *o) | |
399 | { | |
400 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
401 | list_add_tail(&o->hash_list, sl); | |
402 | } | |
403 | ||
b735fede MP |
404 | static struct dm_origin *__lookup_dm_origin(struct block_device *origin) |
405 | { | |
406 | struct list_head *ol; | |
407 | struct dm_origin *o; | |
408 | ||
409 | ol = &_dm_origins[origin_hash(origin)]; | |
410 | list_for_each_entry (o, ol, hash_list) | |
411 | if (bdev_equal(o->dev->bdev, origin)) | |
412 | return o; | |
413 | ||
414 | return NULL; | |
415 | } | |
416 | ||
417 | static void __insert_dm_origin(struct dm_origin *o) | |
418 | { | |
419 | struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)]; | |
420 | list_add_tail(&o->hash_list, sl); | |
421 | } | |
422 | ||
423 | static void __remove_dm_origin(struct dm_origin *o) | |
424 | { | |
425 | list_del(&o->hash_list); | |
426 | } | |
427 | ||
c1f0c183 MS |
428 | /* |
429 | * _origins_lock must be held when calling this function. | |
430 | * Returns number of snapshots registered using the supplied cow device, plus: | |
431 | * snap_src - a snapshot suitable for use as a source of exception handover | |
432 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
433 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
434 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 435 | * |
9d3b15c4 | 436 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
437 | * 0: NULL, NULL - first new snapshot |
438 | * 1: snap_src, NULL - normal snapshot | |
439 | * 2: snap_src, snap_dest - waiting for handover | |
440 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
441 | * 1: NULL, snap_dest - source got destroyed without handover | |
442 | */ | |
443 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
444 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
445 | struct dm_snapshot **snap_dest, |
446 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
447 | { |
448 | struct dm_snapshot *s; | |
449 | struct origin *o; | |
450 | int count = 0; | |
451 | int active; | |
452 | ||
453 | o = __lookup_origin(snap->origin->bdev); | |
454 | if (!o) | |
455 | goto out; | |
456 | ||
457 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
458 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
459 | *snap_merge = s; | |
c1f0c183 MS |
460 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
461 | continue; | |
462 | ||
4ad8d880 | 463 | down_read(&s->lock); |
c1f0c183 | 464 | active = s->active; |
4ad8d880 | 465 | up_read(&s->lock); |
c1f0c183 MS |
466 | |
467 | if (active) { | |
468 | if (snap_src) | |
469 | *snap_src = s; | |
470 | } else if (snap_dest) | |
471 | *snap_dest = s; | |
472 | ||
473 | count++; | |
474 | } | |
475 | ||
476 | out: | |
477 | return count; | |
478 | } | |
479 | ||
480 | /* | |
481 | * On success, returns 1 if this snapshot is a handover destination, | |
482 | * otherwise returns 0. | |
483 | */ | |
484 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
485 | { | |
486 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 487 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
488 | |
489 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
490 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
491 | &snap_merge) == 2) || | |
c1f0c183 MS |
492 | snap_dest) { |
493 | snap->ti->error = "Snapshot cow pairing for exception " | |
494 | "table handover failed"; | |
495 | return -EINVAL; | |
496 | } | |
497 | ||
498 | /* | |
499 | * If no snap_src was found, snap cannot become a handover | |
500 | * destination. | |
501 | */ | |
502 | if (!snap_src) | |
503 | return 0; | |
504 | ||
9d3b15c4 MP |
505 | /* |
506 | * Non-snapshot-merge handover? | |
507 | */ | |
508 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
509 | return 1; | |
510 | ||
511 | /* | |
512 | * Do not allow more than one merging snapshot. | |
513 | */ | |
514 | if (snap_merge) { | |
515 | snap->ti->error = "A snapshot is already merging."; | |
516 | return -EINVAL; | |
517 | } | |
518 | ||
1e03f97e MP |
519 | if (!snap_src->store->type->prepare_merge || |
520 | !snap_src->store->type->commit_merge) { | |
521 | snap->ti->error = "Snapshot exception store does not " | |
522 | "support snapshot-merge."; | |
523 | return -EINVAL; | |
524 | } | |
525 | ||
c1f0c183 MS |
526 | return 1; |
527 | } | |
528 | ||
529 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
530 | { | |
531 | struct dm_snapshot *l; | |
532 | ||
533 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
534 | list_for_each_entry(l, &o->snapshots, list) | |
535 | if (l->store->chunk_size < s->store->chunk_size) | |
536 | break; | |
537 | list_add_tail(&s->list, &l->list); | |
538 | } | |
539 | ||
1da177e4 LT |
540 | /* |
541 | * Make a note of the snapshot and its origin so we can look it | |
542 | * up when the origin has a write on it. | |
c1f0c183 MS |
543 | * |
544 | * Also validate snapshot exception store handovers. | |
545 | * On success, returns 1 if this registration is a handover destination, | |
546 | * otherwise returns 0. | |
1da177e4 LT |
547 | */ |
548 | static int register_snapshot(struct dm_snapshot *snap) | |
549 | { | |
c1f0c183 | 550 | struct origin *o, *new_o = NULL; |
1da177e4 | 551 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 552 | int r = 0; |
1da177e4 | 553 | |
60c856c8 MP |
554 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
555 | if (!new_o) | |
556 | return -ENOMEM; | |
557 | ||
1da177e4 | 558 | down_write(&_origins_lock); |
1da177e4 | 559 | |
c1f0c183 MS |
560 | r = __validate_exception_handover(snap); |
561 | if (r < 0) { | |
562 | kfree(new_o); | |
563 | goto out; | |
564 | } | |
565 | ||
566 | o = __lookup_origin(bdev); | |
60c856c8 MP |
567 | if (o) |
568 | kfree(new_o); | |
569 | else { | |
1da177e4 | 570 | /* New origin */ |
60c856c8 | 571 | o = new_o; |
1da177e4 LT |
572 | |
573 | /* Initialise the struct */ | |
574 | INIT_LIST_HEAD(&o->snapshots); | |
575 | o->bdev = bdev; | |
576 | ||
577 | __insert_origin(o); | |
578 | } | |
579 | ||
c1f0c183 MS |
580 | __insert_snapshot(o, snap); |
581 | ||
582 | out: | |
583 | up_write(&_origins_lock); | |
584 | ||
585 | return r; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Move snapshot to correct place in list according to chunk size. | |
590 | */ | |
591 | static void reregister_snapshot(struct dm_snapshot *s) | |
592 | { | |
593 | struct block_device *bdev = s->origin->bdev; | |
594 | ||
595 | down_write(&_origins_lock); | |
596 | ||
597 | list_del(&s->list); | |
598 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
599 | |
600 | up_write(&_origins_lock); | |
1da177e4 LT |
601 | } |
602 | ||
603 | static void unregister_snapshot(struct dm_snapshot *s) | |
604 | { | |
605 | struct origin *o; | |
606 | ||
607 | down_write(&_origins_lock); | |
608 | o = __lookup_origin(s->origin->bdev); | |
609 | ||
610 | list_del(&s->list); | |
c1f0c183 | 611 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
612 | list_del(&o->hash_list); |
613 | kfree(o); | |
614 | } | |
615 | ||
616 | up_write(&_origins_lock); | |
617 | } | |
618 | ||
619 | /* | |
620 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
621 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
622 | * some consecutive chunks to be grouped together. | |
1da177e4 | 623 | */ |
f79ae415 NT |
624 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk); |
625 | ||
626 | /* Lock to protect access to the completed and pending exception hash tables. */ | |
627 | struct dm_exception_table_lock { | |
628 | struct hlist_bl_head *complete_slot; | |
629 | struct hlist_bl_head *pending_slot; | |
630 | }; | |
631 | ||
632 | static void dm_exception_table_lock_init(struct dm_snapshot *s, chunk_t chunk, | |
633 | struct dm_exception_table_lock *lock) | |
634 | { | |
635 | struct dm_exception_table *complete = &s->complete; | |
636 | struct dm_exception_table *pending = &s->pending; | |
637 | ||
638 | lock->complete_slot = &complete->table[exception_hash(complete, chunk)]; | |
639 | lock->pending_slot = &pending->table[exception_hash(pending, chunk)]; | |
640 | } | |
641 | ||
642 | static void dm_exception_table_lock(struct dm_exception_table_lock *lock) | |
643 | { | |
644 | hlist_bl_lock(lock->complete_slot); | |
645 | hlist_bl_lock(lock->pending_slot); | |
646 | } | |
647 | ||
648 | static void dm_exception_table_unlock(struct dm_exception_table_lock *lock) | |
649 | { | |
650 | hlist_bl_unlock(lock->pending_slot); | |
651 | hlist_bl_unlock(lock->complete_slot); | |
652 | } | |
653 | ||
3510cb94 JB |
654 | static int dm_exception_table_init(struct dm_exception_table *et, |
655 | uint32_t size, unsigned hash_shift) | |
1da177e4 LT |
656 | { |
657 | unsigned int i; | |
658 | ||
d74f81f8 | 659 | et->hash_shift = hash_shift; |
1da177e4 | 660 | et->hash_mask = size - 1; |
7a35693a MWO |
661 | et->table = kvmalloc_array(size, sizeof(struct hlist_bl_head), |
662 | GFP_KERNEL); | |
1da177e4 LT |
663 | if (!et->table) |
664 | return -ENOMEM; | |
665 | ||
666 | for (i = 0; i < size; i++) | |
f79ae415 | 667 | INIT_HLIST_BL_HEAD(et->table + i); |
1da177e4 LT |
668 | |
669 | return 0; | |
670 | } | |
671 | ||
3510cb94 JB |
672 | static void dm_exception_table_exit(struct dm_exception_table *et, |
673 | struct kmem_cache *mem) | |
1da177e4 | 674 | { |
f79ae415 NT |
675 | struct hlist_bl_head *slot; |
676 | struct dm_exception *ex; | |
677 | struct hlist_bl_node *pos, *n; | |
1da177e4 LT |
678 | int i, size; |
679 | ||
680 | size = et->hash_mask + 1; | |
681 | for (i = 0; i < size; i++) { | |
682 | slot = et->table + i; | |
683 | ||
f79ae415 | 684 | hlist_bl_for_each_entry_safe(ex, pos, n, slot, hash_list) |
1da177e4 LT |
685 | kmem_cache_free(mem, ex); |
686 | } | |
687 | ||
7a35693a | 688 | kvfree(et->table); |
1da177e4 LT |
689 | } |
690 | ||
191437a5 | 691 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 692 | { |
d74f81f8 | 693 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
694 | } |
695 | ||
3510cb94 | 696 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 | 697 | { |
f79ae415 | 698 | hlist_bl_del(&e->hash_list); |
1da177e4 LT |
699 | } |
700 | ||
701 | /* | |
702 | * Return the exception data for a sector, or NULL if not | |
703 | * remapped. | |
704 | */ | |
3510cb94 JB |
705 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
706 | chunk_t chunk) | |
1da177e4 | 707 | { |
f79ae415 NT |
708 | struct hlist_bl_head *slot; |
709 | struct hlist_bl_node *pos; | |
1d4989c8 | 710 | struct dm_exception *e; |
1da177e4 LT |
711 | |
712 | slot = &et->table[exception_hash(et, chunk)]; | |
f79ae415 | 713 | hlist_bl_for_each_entry(e, pos, slot, hash_list) |
d74f81f8 MB |
714 | if (chunk >= e->old_chunk && |
715 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
716 | return e; |
717 | ||
718 | return NULL; | |
719 | } | |
720 | ||
119bc547 | 721 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 722 | { |
1d4989c8 | 723 | struct dm_exception *e; |
1da177e4 | 724 | |
119bc547 MP |
725 | e = kmem_cache_alloc(exception_cache, gfp); |
726 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
727 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
728 | ||
729 | return e; | |
730 | } | |
731 | ||
3510cb94 | 732 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
733 | { |
734 | kmem_cache_free(exception_cache, e); | |
735 | } | |
736 | ||
92e86812 | 737 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 738 | { |
6f1c819c | 739 | struct dm_snap_pending_exception *pe = mempool_alloc(&s->pending_pool, |
92e86812 MP |
740 | GFP_NOIO); |
741 | ||
879129d2 | 742 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
743 | pe->snap = s; |
744 | ||
745 | return pe; | |
1da177e4 LT |
746 | } |
747 | ||
028867ac | 748 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 749 | { |
879129d2 MP |
750 | struct dm_snapshot *s = pe->snap; |
751 | ||
6f1c819c | 752 | mempool_free(pe, &s->pending_pool); |
4e857c58 | 753 | smp_mb__before_atomic(); |
879129d2 | 754 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
755 | } |
756 | ||
3510cb94 JB |
757 | static void dm_insert_exception(struct dm_exception_table *eh, |
758 | struct dm_exception *new_e) | |
d74f81f8 | 759 | { |
f79ae415 NT |
760 | struct hlist_bl_head *l; |
761 | struct hlist_bl_node *pos; | |
1d4989c8 | 762 | struct dm_exception *e = NULL; |
d74f81f8 MB |
763 | |
764 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
765 | ||
766 | /* Add immediately if this table doesn't support consecutive chunks */ | |
767 | if (!eh->hash_shift) | |
768 | goto out; | |
769 | ||
770 | /* List is ordered by old_chunk */ | |
f79ae415 | 771 | hlist_bl_for_each_entry(e, pos, l, hash_list) { |
d74f81f8 MB |
772 | /* Insert after an existing chunk? */ |
773 | if (new_e->old_chunk == (e->old_chunk + | |
774 | dm_consecutive_chunk_count(e) + 1) && | |
775 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
776 | dm_consecutive_chunk_count(e) + 1)) { | |
777 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 778 | free_completed_exception(new_e); |
d74f81f8 MB |
779 | return; |
780 | } | |
781 | ||
782 | /* Insert before an existing chunk? */ | |
783 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
784 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
785 | dm_consecutive_chunk_count_inc(e); | |
786 | e->old_chunk--; | |
787 | e->new_chunk--; | |
3510cb94 | 788 | free_completed_exception(new_e); |
d74f81f8 MB |
789 | return; |
790 | } | |
791 | ||
f79ae415 | 792 | if (new_e->old_chunk < e->old_chunk) |
d74f81f8 MB |
793 | break; |
794 | } | |
795 | ||
796 | out: | |
f79ae415 NT |
797 | if (!e) { |
798 | /* | |
799 | * Either the table doesn't support consecutive chunks or slot | |
800 | * l is empty. | |
801 | */ | |
802 | hlist_bl_add_head(&new_e->hash_list, l); | |
803 | } else if (new_e->old_chunk < e->old_chunk) { | |
804 | /* Add before an existing exception */ | |
805 | hlist_bl_add_before(&new_e->hash_list, &e->hash_list); | |
806 | } else { | |
807 | /* Add to l's tail: e is the last exception in this slot */ | |
808 | hlist_bl_add_behind(&new_e->hash_list, &e->hash_list); | |
809 | } | |
d74f81f8 MB |
810 | } |
811 | ||
a159c1ac JB |
812 | /* |
813 | * Callback used by the exception stores to load exceptions when | |
814 | * initialising. | |
815 | */ | |
816 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 817 | { |
f79ae415 | 818 | struct dm_exception_table_lock lock; |
a159c1ac | 819 | struct dm_snapshot *s = context; |
1d4989c8 | 820 | struct dm_exception *e; |
1da177e4 | 821 | |
119bc547 | 822 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
823 | if (!e) |
824 | return -ENOMEM; | |
825 | ||
826 | e->old_chunk = old; | |
d74f81f8 MB |
827 | |
828 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 829 | e->new_chunk = new; |
d74f81f8 | 830 | |
f79ae415 NT |
831 | /* |
832 | * Although there is no need to lock access to the exception tables | |
833 | * here, if we don't then hlist_bl_add_head(), called by | |
834 | * dm_insert_exception(), will complain about accessing the | |
835 | * corresponding list without locking it first. | |
836 | */ | |
837 | dm_exception_table_lock_init(s, old, &lock); | |
838 | ||
839 | dm_exception_table_lock(&lock); | |
3510cb94 | 840 | dm_insert_exception(&s->complete, e); |
f79ae415 | 841 | dm_exception_table_unlock(&lock); |
d74f81f8 | 842 | |
1da177e4 LT |
843 | return 0; |
844 | } | |
845 | ||
7e201b35 MP |
846 | /* |
847 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
848 | * Return zero if the origin has no snapshots. | |
849 | */ | |
542f9038 | 850 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
851 | { |
852 | struct dm_snapshot *snap; | |
7e768532 | 853 | unsigned chunk_size = rounddown_pow_of_two(UINT_MAX); |
7e201b35 MP |
854 | |
855 | if (o) | |
856 | list_for_each_entry(snap, &o->snapshots, list) | |
f16dba5d MP |
857 | chunk_size = min_not_zero(chunk_size, |
858 | snap->store->chunk_size); | |
7e201b35 | 859 | |
542f9038 | 860 | return (uint32_t) chunk_size; |
7e201b35 MP |
861 | } |
862 | ||
1da177e4 LT |
863 | /* |
864 | * Hard coded magic. | |
865 | */ | |
866 | static int calc_max_buckets(void) | |
867 | { | |
868 | /* use a fixed size of 2MB */ | |
869 | unsigned long mem = 2 * 1024 * 1024; | |
f79ae415 | 870 | mem /= sizeof(struct hlist_bl_head); |
1da177e4 LT |
871 | |
872 | return mem; | |
873 | } | |
874 | ||
1da177e4 LT |
875 | /* |
876 | * Allocate room for a suitable hash table. | |
877 | */ | |
fee1998e | 878 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 879 | { |
60e356f3 | 880 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
881 | |
882 | /* | |
883 | * Calculate based on the size of the original volume or | |
884 | * the COW volume... | |
885 | */ | |
fc56f6fb | 886 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
887 | max_buckets = calc_max_buckets(); |
888 | ||
60e356f3 | 889 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
890 | hash_size = min(hash_size, max_buckets); |
891 | ||
8e87b9b8 MP |
892 | if (hash_size < 64) |
893 | hash_size = 64; | |
8defd830 | 894 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
895 | if (dm_exception_table_init(&s->complete, hash_size, |
896 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
897 | return -ENOMEM; |
898 | ||
899 | /* | |
900 | * Allocate hash table for in-flight exceptions | |
901 | * Make this smaller than the real hash table | |
902 | */ | |
903 | hash_size >>= 3; | |
904 | if (hash_size < 64) | |
905 | hash_size = 64; | |
906 | ||
3510cb94 JB |
907 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
908 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
909 | return -ENOMEM; |
910 | } | |
911 | ||
912 | return 0; | |
913 | } | |
914 | ||
1e03f97e MP |
915 | static void merge_shutdown(struct dm_snapshot *s) |
916 | { | |
917 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 918 | smp_mb__after_atomic(); |
1e03f97e MP |
919 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
920 | } | |
921 | ||
9fe86254 MP |
922 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
923 | { | |
924 | s->first_merging_chunk = 0; | |
925 | s->num_merging_chunks = 0; | |
926 | ||
927 | return bio_list_get(&s->bios_queued_during_merge); | |
928 | } | |
929 | ||
1e03f97e MP |
930 | /* |
931 | * Remove one chunk from the index of completed exceptions. | |
932 | */ | |
933 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
934 | chunk_t old_chunk) | |
935 | { | |
936 | struct dm_exception *e; | |
937 | ||
1e03f97e MP |
938 | e = dm_lookup_exception(&s->complete, old_chunk); |
939 | if (!e) { | |
940 | DMERR("Corruption detected: exception for block %llu is " | |
941 | "on disk but not in memory", | |
942 | (unsigned long long)old_chunk); | |
943 | return -EINVAL; | |
944 | } | |
945 | ||
946 | /* | |
947 | * If this is the only chunk using this exception, remove exception. | |
948 | */ | |
949 | if (!dm_consecutive_chunk_count(e)) { | |
950 | dm_remove_exception(e); | |
951 | free_completed_exception(e); | |
952 | return 0; | |
953 | } | |
954 | ||
955 | /* | |
956 | * The chunk may be either at the beginning or the end of a | |
957 | * group of consecutive chunks - never in the middle. We are | |
958 | * removing chunks in the opposite order to that in which they | |
959 | * were added, so this should always be true. | |
960 | * Decrement the consecutive chunk counter and adjust the | |
961 | * starting point if necessary. | |
962 | */ | |
963 | if (old_chunk == e->old_chunk) { | |
964 | e->old_chunk++; | |
965 | e->new_chunk++; | |
966 | } else if (old_chunk != e->old_chunk + | |
967 | dm_consecutive_chunk_count(e)) { | |
968 | DMERR("Attempt to merge block %llu from the " | |
969 | "middle of a chunk range [%llu - %llu]", | |
970 | (unsigned long long)old_chunk, | |
971 | (unsigned long long)e->old_chunk, | |
972 | (unsigned long long) | |
973 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
974 | return -EINVAL; | |
975 | } | |
976 | ||
977 | dm_consecutive_chunk_count_dec(e); | |
978 | ||
979 | return 0; | |
980 | } | |
981 | ||
9fe86254 MP |
982 | static void flush_bios(struct bio *bio); |
983 | ||
984 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 985 | { |
9fe86254 MP |
986 | struct bio *b = NULL; |
987 | int r; | |
988 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e | 989 | |
4ad8d880 | 990 | down_write(&s->lock); |
9fe86254 MP |
991 | |
992 | /* | |
993 | * Process chunks (and associated exceptions) in reverse order | |
994 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
995 | */ | |
996 | do { | |
997 | r = __remove_single_exception_chunk(s, old_chunk); | |
998 | if (r) | |
999 | goto out; | |
1000 | } while (old_chunk-- > s->first_merging_chunk); | |
1001 | ||
1002 | b = __release_queued_bios_after_merge(s); | |
1003 | ||
1004 | out: | |
4ad8d880 | 1005 | up_write(&s->lock); |
9fe86254 MP |
1006 | if (b) |
1007 | flush_bios(b); | |
1e03f97e MP |
1008 | |
1009 | return r; | |
1010 | } | |
1011 | ||
73dfd078 MP |
1012 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
1013 | sector_t sector, unsigned chunk_size); | |
1014 | ||
1e03f97e MP |
1015 | static void merge_callback(int read_err, unsigned long write_err, |
1016 | void *context); | |
1017 | ||
73dfd078 MP |
1018 | static uint64_t read_pending_exceptions_done_count(void) |
1019 | { | |
1020 | uint64_t pending_exceptions_done; | |
1021 | ||
1022 | spin_lock(&_pending_exceptions_done_spinlock); | |
1023 | pending_exceptions_done = _pending_exceptions_done_count; | |
1024 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1025 | ||
1026 | return pending_exceptions_done; | |
1027 | } | |
1028 | ||
1029 | static void increment_pending_exceptions_done_count(void) | |
1030 | { | |
1031 | spin_lock(&_pending_exceptions_done_spinlock); | |
1032 | _pending_exceptions_done_count++; | |
1033 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1034 | ||
1035 | wake_up_all(&_pending_exceptions_done); | |
1036 | } | |
1037 | ||
1e03f97e MP |
1038 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
1039 | { | |
8a2d5286 | 1040 | int i, linear_chunks; |
1e03f97e MP |
1041 | chunk_t old_chunk, new_chunk; |
1042 | struct dm_io_region src, dest; | |
8a2d5286 | 1043 | sector_t io_size; |
73dfd078 | 1044 | uint64_t previous_count; |
1e03f97e MP |
1045 | |
1046 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
1047 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
1048 | goto shut; | |
1049 | ||
1050 | /* | |
1051 | * valid flag never changes during merge, so no lock required. | |
1052 | */ | |
1053 | if (!s->valid) { | |
1054 | DMERR("Snapshot is invalid: can't merge"); | |
1055 | goto shut; | |
1056 | } | |
1057 | ||
8a2d5286 MS |
1058 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
1059 | &new_chunk); | |
1060 | if (linear_chunks <= 0) { | |
d8ddb1cf | 1061 | if (linear_chunks < 0) { |
1e03f97e MP |
1062 | DMERR("Read error in exception store: " |
1063 | "shutting down merge"); | |
4ad8d880 | 1064 | down_write(&s->lock); |
1d1dda8c | 1065 | s->merge_failed = true; |
4ad8d880 | 1066 | up_write(&s->lock); |
d8ddb1cf | 1067 | } |
1e03f97e MP |
1068 | goto shut; |
1069 | } | |
1070 | ||
8a2d5286 MS |
1071 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
1072 | old_chunk = old_chunk + 1 - linear_chunks; | |
1073 | new_chunk = new_chunk + 1 - linear_chunks; | |
1074 | ||
1075 | /* | |
1076 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
1077 | * from the exception store to the origin | |
1078 | */ | |
1079 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 1080 | |
1e03f97e MP |
1081 | dest.bdev = s->origin->bdev; |
1082 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 1083 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
1084 | |
1085 | src.bdev = s->cow->bdev; | |
1086 | src.sector = chunk_to_sector(s->store, new_chunk); | |
1087 | src.count = dest.count; | |
1088 | ||
73dfd078 MP |
1089 | /* |
1090 | * Reallocate any exceptions needed in other snapshots then | |
1091 | * wait for the pending exceptions to complete. | |
1092 | * Each time any pending exception (globally on the system) | |
1093 | * completes we are woken and repeat the process to find out | |
1094 | * if we can proceed. While this may not seem a particularly | |
1095 | * efficient algorithm, it is not expected to have any | |
1096 | * significant impact on performance. | |
1097 | */ | |
1098 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 1099 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
1100 | wait_event(_pending_exceptions_done, |
1101 | (read_pending_exceptions_done_count() != | |
1102 | previous_count)); | |
1103 | /* Retry after the wait, until all exceptions are done. */ | |
1104 | previous_count = read_pending_exceptions_done_count(); | |
1105 | } | |
1106 | ||
4ad8d880 | 1107 | down_write(&s->lock); |
9fe86254 | 1108 | s->first_merging_chunk = old_chunk; |
8a2d5286 | 1109 | s->num_merging_chunks = linear_chunks; |
4ad8d880 | 1110 | up_write(&s->lock); |
9fe86254 | 1111 | |
8a2d5286 MS |
1112 | /* Wait until writes to all 'linear_chunks' drain */ |
1113 | for (i = 0; i < linear_chunks; i++) | |
1114 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 1115 | |
1e03f97e MP |
1116 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
1117 | return; | |
1118 | ||
1119 | shut: | |
1120 | merge_shutdown(s); | |
1121 | } | |
1122 | ||
9fe86254 MP |
1123 | static void error_bios(struct bio *bio); |
1124 | ||
1e03f97e MP |
1125 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
1126 | { | |
1127 | struct dm_snapshot *s = context; | |
9fe86254 | 1128 | struct bio *b = NULL; |
1e03f97e MP |
1129 | |
1130 | if (read_err || write_err) { | |
1131 | if (read_err) | |
1132 | DMERR("Read error: shutting down merge."); | |
1133 | else | |
1134 | DMERR("Write error: shutting down merge."); | |
1135 | goto shut; | |
1136 | } | |
1137 | ||
eba33b8e | 1138 | if (blkdev_issue_flush(s->origin->bdev) < 0) { |
fcc42338 AK |
1139 | DMERR("Flush after merge failed: shutting down merge"); |
1140 | goto shut; | |
1141 | } | |
1142 | ||
9fe86254 MP |
1143 | if (s->store->type->commit_merge(s->store, |
1144 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1145 | DMERR("Write error in exception store: shutting down merge"); |
1146 | goto shut; | |
1147 | } | |
1148 | ||
9fe86254 MP |
1149 | if (remove_single_exception_chunk(s) < 0) |
1150 | goto shut; | |
1151 | ||
1e03f97e MP |
1152 | snapshot_merge_next_chunks(s); |
1153 | ||
1154 | return; | |
1155 | ||
1156 | shut: | |
4ad8d880 | 1157 | down_write(&s->lock); |
1d1dda8c | 1158 | s->merge_failed = true; |
9fe86254 | 1159 | b = __release_queued_bios_after_merge(s); |
4ad8d880 | 1160 | up_write(&s->lock); |
9fe86254 MP |
1161 | error_bios(b); |
1162 | ||
1e03f97e MP |
1163 | merge_shutdown(s); |
1164 | } | |
1165 | ||
1166 | static void start_merge(struct dm_snapshot *s) | |
1167 | { | |
1168 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1169 | snapshot_merge_next_chunks(s); | |
1170 | } | |
1171 | ||
1e03f97e MP |
1172 | /* |
1173 | * Stop the merging process and wait until it finishes. | |
1174 | */ | |
1175 | static void stop_merge(struct dm_snapshot *s) | |
1176 | { | |
1177 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1178 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1179 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1180 | } | |
1181 | ||
2e602385 MS |
1182 | static int parse_snapshot_features(struct dm_arg_set *as, struct dm_snapshot *s, |
1183 | struct dm_target *ti) | |
1184 | { | |
1185 | int r; | |
1186 | unsigned argc; | |
1187 | const char *arg_name; | |
1188 | ||
1189 | static const struct dm_arg _args[] = { | |
1190 | {0, 2, "Invalid number of feature arguments"}, | |
1191 | }; | |
1192 | ||
1193 | /* | |
1194 | * No feature arguments supplied. | |
1195 | */ | |
1196 | if (!as->argc) | |
1197 | return 0; | |
1198 | ||
1199 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1200 | if (r) | |
1201 | return -EINVAL; | |
1202 | ||
1203 | while (argc && !r) { | |
1204 | arg_name = dm_shift_arg(as); | |
1205 | argc--; | |
1206 | ||
1207 | if (!strcasecmp(arg_name, "discard_zeroes_cow")) | |
1208 | s->discard_zeroes_cow = true; | |
1209 | ||
1210 | else if (!strcasecmp(arg_name, "discard_passdown_origin")) | |
1211 | s->discard_passdown_origin = true; | |
1212 | ||
1213 | else { | |
1214 | ti->error = "Unrecognised feature requested"; | |
1215 | r = -EINVAL; | |
1216 | break; | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | if (!s->discard_zeroes_cow && s->discard_passdown_origin) { | |
1221 | /* | |
1222 | * TODO: really these are disjoint.. but ti->num_discard_bios | |
1223 | * and dm_bio_get_target_bio_nr() require rigid constraints. | |
1224 | */ | |
1225 | ti->error = "discard_passdown_origin feature depends on discard_zeroes_cow"; | |
1226 | r = -EINVAL; | |
1227 | } | |
1228 | ||
1229 | return r; | |
1230 | } | |
1231 | ||
1da177e4 | 1232 | /* |
2e602385 MS |
1233 | * Construct a snapshot mapping: |
1234 | * <origin_dev> <COW-dev> <p|po|n> <chunk-size> [<# feature args> [<arg>]*] | |
1da177e4 LT |
1235 | */ |
1236 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1237 | { | |
1238 | struct dm_snapshot *s; | |
2e602385 | 1239 | struct dm_arg_set as; |
cd45daff | 1240 | int i; |
1da177e4 | 1241 | int r = -EINVAL; |
fc56f6fb | 1242 | char *origin_path, *cow_path; |
4df2bf46 | 1243 | dev_t origin_dev, cow_dev; |
55a62eef | 1244 | unsigned args_used, num_flush_bios = 1; |
10b8106a | 1245 | fmode_t origin_mode = FMODE_READ; |
1da177e4 | 1246 | |
2e602385 MS |
1247 | if (argc < 4) { |
1248 | ti->error = "requires 4 or more arguments"; | |
1da177e4 | 1249 | r = -EINVAL; |
fc56f6fb | 1250 | goto bad; |
1da177e4 LT |
1251 | } |
1252 | ||
10b8106a | 1253 | if (dm_target_is_snapshot_merge(ti)) { |
55a62eef | 1254 | num_flush_bios = 2; |
10b8106a MS |
1255 | origin_mode = FMODE_WRITE; |
1256 | } | |
1257 | ||
d3775354 | 1258 | s = kzalloc(sizeof(*s), GFP_KERNEL); |
fc56f6fb | 1259 | if (!s) { |
a2d2b034 | 1260 | ti->error = "Cannot allocate private snapshot structure"; |
fc56f6fb MS |
1261 | r = -ENOMEM; |
1262 | goto bad; | |
1263 | } | |
1264 | ||
2e602385 MS |
1265 | as.argc = argc; |
1266 | as.argv = argv; | |
1267 | dm_consume_args(&as, 4); | |
1268 | r = parse_snapshot_features(&as, s, ti); | |
1269 | if (r) | |
1270 | goto bad_features; | |
1271 | ||
c2411045 MP |
1272 | origin_path = argv[0]; |
1273 | argv++; | |
1274 | argc--; | |
1275 | ||
1276 | r = dm_get_device(ti, origin_path, origin_mode, &s->origin); | |
1277 | if (r) { | |
1278 | ti->error = "Cannot get origin device"; | |
1279 | goto bad_origin; | |
1280 | } | |
4df2bf46 | 1281 | origin_dev = s->origin->bdev->bd_dev; |
c2411045 | 1282 | |
fc56f6fb MS |
1283 | cow_path = argv[0]; |
1284 | argv++; | |
1285 | argc--; | |
1286 | ||
4df2bf46 D |
1287 | cow_dev = dm_get_dev_t(cow_path); |
1288 | if (cow_dev && cow_dev == origin_dev) { | |
1289 | ti->error = "COW device cannot be the same as origin device"; | |
1290 | r = -EINVAL; | |
1291 | goto bad_cow; | |
1292 | } | |
1293 | ||
024d37e9 | 1294 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1295 | if (r) { |
1296 | ti->error = "Cannot get COW device"; | |
1297 | goto bad_cow; | |
1298 | } | |
1299 | ||
1300 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1301 | if (r) { |
1302 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1303 | r = -EINVAL; |
fc56f6fb | 1304 | goto bad_store; |
1da177e4 LT |
1305 | } |
1306 | ||
fee1998e JB |
1307 | argv += args_used; |
1308 | argc -= args_used; | |
1309 | ||
fc56f6fb | 1310 | s->ti = ti; |
1da177e4 | 1311 | s->valid = 1; |
76c44f6d | 1312 | s->snapshot_overflowed = 0; |
aa14edeb | 1313 | s->active = 0; |
879129d2 | 1314 | atomic_set(&s->pending_exceptions_count, 0); |
3f1637f2 | 1315 | spin_lock_init(&s->pe_allocation_lock); |
230c83af MP |
1316 | s->exception_start_sequence = 0; |
1317 | s->exception_complete_sequence = 0; | |
3db2776d | 1318 | s->out_of_order_tree = RB_ROOT; |
4ad8d880 | 1319 | init_rwsem(&s->lock); |
c1f0c183 | 1320 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1321 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1322 | s->state_bits = 0; |
1d1dda8c | 1323 | s->merge_failed = false; |
9fe86254 MP |
1324 | s->first_merging_chunk = 0; |
1325 | s->num_merging_chunks = 0; | |
1326 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1327 | |
1328 | /* Allocate hash table for COW data */ | |
fee1998e | 1329 | if (init_hash_tables(s)) { |
1da177e4 LT |
1330 | ti->error = "Unable to allocate hash table space"; |
1331 | r = -ENOMEM; | |
fee1998e | 1332 | goto bad_hash_tables; |
1da177e4 LT |
1333 | } |
1334 | ||
b2155578 | 1335 | init_waitqueue_head(&s->in_progress_wait); |
721b1d98 | 1336 | |
df5d2e90 | 1337 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1338 | if (IS_ERR(s->kcopyd_client)) { |
1339 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1340 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1341 | goto bad_kcopyd; |
1da177e4 LT |
1342 | } |
1343 | ||
6f1c819c KO |
1344 | r = mempool_init_slab_pool(&s->pending_pool, MIN_IOS, pending_cache); |
1345 | if (r) { | |
92e86812 | 1346 | ti->error = "Could not allocate mempool for pending exceptions"; |
fee1998e | 1347 | goto bad_pending_pool; |
92e86812 MP |
1348 | } |
1349 | ||
cd45daff MP |
1350 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1351 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1352 | ||
1353 | spin_lock_init(&s->tracked_chunk_lock); | |
1354 | ||
c1f0c183 | 1355 | ti->private = s; |
55a62eef | 1356 | ti->num_flush_bios = num_flush_bios; |
2e602385 MS |
1357 | if (s->discard_zeroes_cow) |
1358 | ti->num_discard_bios = (s->discard_passdown_origin ? 2 : 1); | |
30187e1d | 1359 | ti->per_io_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1360 | |
1361 | /* Add snapshot to the list of snapshots for this origin */ | |
1362 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1363 | r = register_snapshot(s); | |
1364 | if (r == -ENOMEM) { | |
1365 | ti->error = "Snapshot origin struct allocation failed"; | |
1366 | goto bad_load_and_register; | |
1367 | } else if (r < 0) { | |
1368 | /* invalid handover, register_snapshot has set ti->error */ | |
1369 | goto bad_load_and_register; | |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * Metadata must only be loaded into one table at once, so skip this | |
1374 | * if metadata will be handed over during resume. | |
1375 | * Chunk size will be set during the handover - set it to zero to | |
1376 | * ensure it's ignored. | |
1377 | */ | |
1378 | if (r > 0) { | |
1379 | s->store->chunk_size = 0; | |
1380 | return 0; | |
1381 | } | |
1382 | ||
493df71c JB |
1383 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1384 | (void *)s); | |
0764147b | 1385 | if (r < 0) { |
f9cea4f7 | 1386 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1387 | goto bad_read_metadata; |
0764147b MB |
1388 | } else if (r > 0) { |
1389 | s->valid = 0; | |
1390 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1391 | } |
aa14edeb | 1392 | |
3f2412dc MP |
1393 | if (!s->store->chunk_size) { |
1394 | ti->error = "Chunk size not set"; | |
c699a0db | 1395 | r = -EINVAL; |
c1f0c183 | 1396 | goto bad_read_metadata; |
1da177e4 | 1397 | } |
542f9038 MS |
1398 | |
1399 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1400 | if (r) | |
1401 | goto bad_read_metadata; | |
1da177e4 LT |
1402 | |
1403 | return 0; | |
1404 | ||
c1f0c183 MS |
1405 | bad_read_metadata: |
1406 | unregister_snapshot(s); | |
fee1998e | 1407 | bad_load_and_register: |
6f1c819c | 1408 | mempool_exit(&s->pending_pool); |
fee1998e | 1409 | bad_pending_pool: |
eb69aca5 | 1410 | dm_kcopyd_client_destroy(s->kcopyd_client); |
fee1998e | 1411 | bad_kcopyd: |
3510cb94 JB |
1412 | dm_exception_table_exit(&s->pending, pending_cache); |
1413 | dm_exception_table_exit(&s->complete, exception_cache); | |
fee1998e | 1414 | bad_hash_tables: |
fc56f6fb | 1415 | dm_exception_store_destroy(s->store); |
fc56f6fb MS |
1416 | bad_store: |
1417 | dm_put_device(ti, s->cow); | |
fc56f6fb | 1418 | bad_cow: |
c2411045 | 1419 | dm_put_device(ti, s->origin); |
c2411045 | 1420 | bad_origin: |
2e602385 | 1421 | bad_features: |
fc56f6fb | 1422 | kfree(s); |
fc56f6fb | 1423 | bad: |
1da177e4 LT |
1424 | return r; |
1425 | } | |
1426 | ||
31c93a0c MB |
1427 | static void __free_exceptions(struct dm_snapshot *s) |
1428 | { | |
eb69aca5 | 1429 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1430 | s->kcopyd_client = NULL; |
1431 | ||
3510cb94 JB |
1432 | dm_exception_table_exit(&s->pending, pending_cache); |
1433 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1434 | } |
1435 | ||
c1f0c183 MS |
1436 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1437 | struct dm_snapshot *snap_dest) | |
1438 | { | |
1439 | union { | |
1440 | struct dm_exception_table table_swap; | |
1441 | struct dm_exception_store *store_swap; | |
1442 | } u; | |
1443 | ||
1444 | /* | |
1445 | * Swap all snapshot context information between the two instances. | |
1446 | */ | |
1447 | u.table_swap = snap_dest->complete; | |
1448 | snap_dest->complete = snap_src->complete; | |
1449 | snap_src->complete = u.table_swap; | |
1450 | ||
1451 | u.store_swap = snap_dest->store; | |
1452 | snap_dest->store = snap_src->store; | |
b0d3cc01 | 1453 | snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow; |
c1f0c183 MS |
1454 | snap_src->store = u.store_swap; |
1455 | ||
1456 | snap_dest->store->snap = snap_dest; | |
1457 | snap_src->store->snap = snap_src; | |
1458 | ||
542f9038 | 1459 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 | 1460 | snap_dest->valid = snap_src->valid; |
76c44f6d | 1461 | snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed; |
c1f0c183 MS |
1462 | |
1463 | /* | |
1464 | * Set source invalid to ensure it receives no further I/O. | |
1465 | */ | |
1466 | snap_src->valid = 0; | |
1467 | } | |
1468 | ||
1da177e4 LT |
1469 | static void snapshot_dtr(struct dm_target *ti) |
1470 | { | |
cd45daff MP |
1471 | #ifdef CONFIG_DM_DEBUG |
1472 | int i; | |
1473 | #endif | |
028867ac | 1474 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1475 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1476 | |
c1f0c183 MS |
1477 | down_read(&_origins_lock); |
1478 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1479 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 1480 | if (snap_src && snap_dest && (s == snap_src)) { |
4ad8d880 | 1481 | down_write(&snap_dest->lock); |
c1f0c183 | 1482 | snap_dest->valid = 0; |
4ad8d880 | 1483 | up_write(&snap_dest->lock); |
c1f0c183 MS |
1484 | DMERR("Cancelling snapshot handover."); |
1485 | } | |
1486 | up_read(&_origins_lock); | |
1487 | ||
1e03f97e MP |
1488 | if (dm_target_is_snapshot_merge(ti)) |
1489 | stop_merge(s); | |
1490 | ||
138728dc AK |
1491 | /* Prevent further origin writes from using this snapshot. */ |
1492 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1493 | unregister_snapshot(s); |
1494 | ||
879129d2 | 1495 | while (atomic_read(&s->pending_exceptions_count)) |
90fa1527 | 1496 | msleep(1); |
879129d2 | 1497 | /* |
6f1c819c | 1498 | * Ensure instructions in mempool_exit aren't reordered |
879129d2 MP |
1499 | * before atomic_read. |
1500 | */ | |
1501 | smp_mb(); | |
1502 | ||
cd45daff MP |
1503 | #ifdef CONFIG_DM_DEBUG |
1504 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1505 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1506 | #endif | |
1507 | ||
31c93a0c | 1508 | __free_exceptions(s); |
1da177e4 | 1509 | |
6f1c819c | 1510 | mempool_exit(&s->pending_pool); |
92e86812 | 1511 | |
fee1998e | 1512 | dm_exception_store_destroy(s->store); |
138728dc | 1513 | |
fc56f6fb MS |
1514 | dm_put_device(ti, s->cow); |
1515 | ||
c2411045 MP |
1516 | dm_put_device(ti, s->origin); |
1517 | ||
b2155578 MP |
1518 | WARN_ON(s->in_progress); |
1519 | ||
1da177e4 LT |
1520 | kfree(s); |
1521 | } | |
1522 | ||
a2f83e8b MP |
1523 | static void account_start_copy(struct dm_snapshot *s) |
1524 | { | |
b2155578 MP |
1525 | spin_lock(&s->in_progress_wait.lock); |
1526 | s->in_progress++; | |
1527 | spin_unlock(&s->in_progress_wait.lock); | |
a2f83e8b MP |
1528 | } |
1529 | ||
1530 | static void account_end_copy(struct dm_snapshot *s) | |
1531 | { | |
b2155578 MP |
1532 | spin_lock(&s->in_progress_wait.lock); |
1533 | BUG_ON(!s->in_progress); | |
1534 | s->in_progress--; | |
1535 | if (likely(s->in_progress <= cow_threshold) && | |
1536 | unlikely(waitqueue_active(&s->in_progress_wait))) | |
1537 | wake_up_locked(&s->in_progress_wait); | |
1538 | spin_unlock(&s->in_progress_wait.lock); | |
1539 | } | |
1540 | ||
1541 | static bool wait_for_in_progress(struct dm_snapshot *s, bool unlock_origins) | |
1542 | { | |
1543 | if (unlikely(s->in_progress > cow_threshold)) { | |
1544 | spin_lock(&s->in_progress_wait.lock); | |
1545 | if (likely(s->in_progress > cow_threshold)) { | |
1546 | /* | |
1547 | * NOTE: this throttle doesn't account for whether | |
1548 | * the caller is servicing an IO that will trigger a COW | |
1549 | * so excess throttling may result for chunks not required | |
1550 | * to be COW'd. But if cow_threshold was reached, extra | |
1551 | * throttling is unlikely to negatively impact performance. | |
1552 | */ | |
1553 | DECLARE_WAITQUEUE(wait, current); | |
1554 | __add_wait_queue(&s->in_progress_wait, &wait); | |
1555 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1556 | spin_unlock(&s->in_progress_wait.lock); | |
1557 | if (unlock_origins) | |
1558 | up_read(&_origins_lock); | |
1559 | io_schedule(); | |
1560 | remove_wait_queue(&s->in_progress_wait, &wait); | |
1561 | return false; | |
1562 | } | |
1563 | spin_unlock(&s->in_progress_wait.lock); | |
1564 | } | |
1565 | return true; | |
a2f83e8b MP |
1566 | } |
1567 | ||
1da177e4 LT |
1568 | /* |
1569 | * Flush a list of buffers. | |
1570 | */ | |
1571 | static void flush_bios(struct bio *bio) | |
1572 | { | |
1573 | struct bio *n; | |
1574 | ||
1575 | while (bio) { | |
1576 | n = bio->bi_next; | |
1577 | bio->bi_next = NULL; | |
ed00aabd | 1578 | submit_bio_noacct(bio); |
1da177e4 LT |
1579 | bio = n; |
1580 | } | |
1581 | } | |
1582 | ||
b2155578 | 1583 | static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit); |
515ad66c MP |
1584 | |
1585 | /* | |
1586 | * Flush a list of buffers. | |
1587 | */ | |
1588 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1589 | { | |
1590 | struct bio *n; | |
1591 | int r; | |
1592 | ||
1593 | while (bio) { | |
1594 | n = bio->bi_next; | |
1595 | bio->bi_next = NULL; | |
b2155578 | 1596 | r = do_origin(s->origin, bio, false); |
515ad66c | 1597 | if (r == DM_MAPIO_REMAPPED) |
ed00aabd | 1598 | submit_bio_noacct(bio); |
515ad66c MP |
1599 | bio = n; |
1600 | } | |
1601 | } | |
1602 | ||
1da177e4 LT |
1603 | /* |
1604 | * Error a list of buffers. | |
1605 | */ | |
1606 | static void error_bios(struct bio *bio) | |
1607 | { | |
1608 | struct bio *n; | |
1609 | ||
1610 | while (bio) { | |
1611 | n = bio->bi_next; | |
1612 | bio->bi_next = NULL; | |
6712ecf8 | 1613 | bio_io_error(bio); |
1da177e4 LT |
1614 | bio = n; |
1615 | } | |
1616 | } | |
1617 | ||
695368ac | 1618 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1619 | { |
1620 | if (!s->valid) | |
1621 | return; | |
1622 | ||
1623 | if (err == -EIO) | |
1624 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1625 | else if (err == -ENOMEM) | |
1626 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1627 | ||
493df71c JB |
1628 | if (s->store->type->drop_snapshot) |
1629 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1630 | |
1631 | s->valid = 0; | |
1632 | ||
fc56f6fb | 1633 | dm_table_event(s->ti->table); |
76df1c65 AK |
1634 | } |
1635 | ||
3f1637f2 NT |
1636 | static void invalidate_snapshot(struct dm_snapshot *s, int err) |
1637 | { | |
1638 | down_write(&s->lock); | |
1639 | __invalidate_snapshot(s, err); | |
1640 | up_write(&s->lock); | |
1641 | } | |
1642 | ||
385277bf | 1643 | static void pending_complete(void *context, int success) |
1da177e4 | 1644 | { |
385277bf | 1645 | struct dm_snap_pending_exception *pe = context; |
1d4989c8 | 1646 | struct dm_exception *e; |
1da177e4 | 1647 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1648 | struct bio *origin_bios = NULL; |
1649 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1650 | struct bio *full_bio = NULL; |
f79ae415 | 1651 | struct dm_exception_table_lock lock; |
9d493fa8 | 1652 | int error = 0; |
1da177e4 | 1653 | |
f79ae415 NT |
1654 | dm_exception_table_lock_init(s, pe->e.old_chunk, &lock); |
1655 | ||
76df1c65 AK |
1656 | if (!success) { |
1657 | /* Read/write error - snapshot is unusable */ | |
3f1637f2 | 1658 | invalidate_snapshot(s, -EIO); |
9d493fa8 | 1659 | error = 1; |
f79ae415 NT |
1660 | |
1661 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1662 | goto out; |
1663 | } | |
1664 | ||
119bc547 | 1665 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1666 | if (!e) { |
3f1637f2 | 1667 | invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1668 | error = 1; |
f79ae415 NT |
1669 | |
1670 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1671 | goto out; |
1672 | } | |
1673 | *e = pe->e; | |
1da177e4 | 1674 | |
3f1637f2 | 1675 | down_read(&s->lock); |
f79ae415 | 1676 | dm_exception_table_lock(&lock); |
76df1c65 | 1677 | if (!s->valid) { |
3f1637f2 | 1678 | up_read(&s->lock); |
3510cb94 | 1679 | free_completed_exception(e); |
9d493fa8 | 1680 | error = 1; |
3f1637f2 | 1681 | |
76df1c65 | 1682 | goto out; |
1da177e4 LT |
1683 | } |
1684 | ||
9d493fa8 | 1685 | /* |
65fc7c37 NT |
1686 | * Add a proper exception. After inserting the completed exception all |
1687 | * subsequent snapshot reads to this chunk will be redirected to the | |
1688 | * COW device. This ensures that we do not starve. Moreover, as long | |
1689 | * as the pending exception exists, neither origin writes nor snapshot | |
1690 | * merging can overwrite the chunk in origin. | |
9d493fa8 | 1691 | */ |
3510cb94 | 1692 | dm_insert_exception(&s->complete, e); |
3f1637f2 | 1693 | up_read(&s->lock); |
76df1c65 | 1694 | |
65fc7c37 NT |
1695 | /* Wait for conflicting reads to drain */ |
1696 | if (__chunk_is_tracked(s, pe->e.old_chunk)) { | |
f79ae415 | 1697 | dm_exception_table_unlock(&lock); |
65fc7c37 | 1698 | __check_for_conflicting_io(s, pe->e.old_chunk); |
f79ae415 | 1699 | dm_exception_table_lock(&lock); |
65fc7c37 NT |
1700 | } |
1701 | ||
a2d2b034 | 1702 | out: |
65fc7c37 | 1703 | /* Remove the in-flight exception from the list */ |
3510cb94 | 1704 | dm_remove_exception(&pe->e); |
f79ae415 NT |
1705 | |
1706 | dm_exception_table_unlock(&lock); | |
1707 | ||
9d493fa8 | 1708 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1709 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 | 1710 | full_bio = pe->full_bio; |
fe3265b1 | 1711 | if (full_bio) |
a6e50b40 | 1712 | full_bio->bi_end_io = pe->full_bio_end_io; |
73dfd078 MP |
1713 | increment_pending_exceptions_done_count(); |
1714 | ||
9d493fa8 | 1715 | /* Submit any pending write bios */ |
a6e50b40 MP |
1716 | if (error) { |
1717 | if (full_bio) | |
1718 | bio_io_error(full_bio); | |
9d493fa8 | 1719 | error_bios(snapshot_bios); |
a6e50b40 MP |
1720 | } else { |
1721 | if (full_bio) | |
4246a0b6 | 1722 | bio_endio(full_bio); |
9d493fa8 | 1723 | flush_bios(snapshot_bios); |
a6e50b40 | 1724 | } |
9d493fa8 | 1725 | |
515ad66c | 1726 | retry_origin_bios(s, origin_bios); |
22aa66a3 MP |
1727 | |
1728 | free_pending_exception(pe); | |
1da177e4 LT |
1729 | } |
1730 | ||
230c83af MP |
1731 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1732 | { | |
1733 | struct dm_snapshot *s = pe->snap; | |
1734 | ||
385277bf MP |
1735 | /* Update the metadata if we are persistent */ |
1736 | s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error, | |
1737 | pending_complete, pe); | |
230c83af MP |
1738 | } |
1739 | ||
1da177e4 LT |
1740 | /* |
1741 | * Called when the copy I/O has finished. kcopyd actually runs | |
1742 | * this code so don't block. | |
1743 | */ | |
4cdc1d1f | 1744 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1745 | { |
028867ac | 1746 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1747 | struct dm_snapshot *s = pe->snap; |
1748 | ||
230c83af | 1749 | pe->copy_error = read_err || write_err; |
1da177e4 | 1750 | |
230c83af | 1751 | if (pe->exception_sequence == s->exception_complete_sequence) { |
3db2776d DJ |
1752 | struct rb_node *next; |
1753 | ||
230c83af MP |
1754 | s->exception_complete_sequence++; |
1755 | complete_exception(pe); | |
1756 | ||
3db2776d DJ |
1757 | next = rb_first(&s->out_of_order_tree); |
1758 | while (next) { | |
1759 | pe = rb_entry(next, struct dm_snap_pending_exception, | |
1760 | out_of_order_node); | |
230c83af MP |
1761 | if (pe->exception_sequence != s->exception_complete_sequence) |
1762 | break; | |
3db2776d | 1763 | next = rb_next(next); |
230c83af | 1764 | s->exception_complete_sequence++; |
3db2776d | 1765 | rb_erase(&pe->out_of_order_node, &s->out_of_order_tree); |
230c83af | 1766 | complete_exception(pe); |
3db2776d | 1767 | cond_resched(); |
230c83af MP |
1768 | } |
1769 | } else { | |
3db2776d DJ |
1770 | struct rb_node *parent = NULL; |
1771 | struct rb_node **p = &s->out_of_order_tree.rb_node; | |
230c83af MP |
1772 | struct dm_snap_pending_exception *pe2; |
1773 | ||
3db2776d DJ |
1774 | while (*p) { |
1775 | pe2 = rb_entry(*p, struct dm_snap_pending_exception, out_of_order_node); | |
1776 | parent = *p; | |
1777 | ||
1778 | BUG_ON(pe->exception_sequence == pe2->exception_sequence); | |
1779 | if (pe->exception_sequence < pe2->exception_sequence) | |
1780 | p = &((*p)->rb_left); | |
1781 | else | |
1782 | p = &((*p)->rb_right); | |
230c83af | 1783 | } |
3db2776d DJ |
1784 | |
1785 | rb_link_node(&pe->out_of_order_node, parent, p); | |
1786 | rb_insert_color(&pe->out_of_order_node, &s->out_of_order_tree); | |
230c83af | 1787 | } |
a2f83e8b | 1788 | account_end_copy(s); |
1da177e4 LT |
1789 | } |
1790 | ||
1791 | /* | |
1792 | * Dispatches the copy operation to kcopyd. | |
1793 | */ | |
028867ac | 1794 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1795 | { |
1796 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1797 | struct dm_io_region src, dest; |
1da177e4 LT |
1798 | struct block_device *bdev = s->origin->bdev; |
1799 | sector_t dev_size; | |
1800 | ||
1801 | dev_size = get_dev_size(bdev); | |
1802 | ||
1803 | src.bdev = bdev; | |
71fab00a | 1804 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1805 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1806 | |
fc56f6fb | 1807 | dest.bdev = s->cow->bdev; |
71fab00a | 1808 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1809 | dest.count = src.count; |
1810 | ||
1811 | /* Hand over to kcopyd */ | |
a2f83e8b | 1812 | account_start_copy(s); |
a2d2b034 | 1813 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1814 | } |
1815 | ||
4246a0b6 | 1816 | static void full_bio_end_io(struct bio *bio) |
a6e50b40 MP |
1817 | { |
1818 | void *callback_data = bio->bi_private; | |
1819 | ||
4e4cbee9 | 1820 | dm_kcopyd_do_callback(callback_data, 0, bio->bi_status ? 1 : 0); |
a6e50b40 MP |
1821 | } |
1822 | ||
1823 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1824 | struct bio *bio) | |
1825 | { | |
1826 | struct dm_snapshot *s = pe->snap; | |
1827 | void *callback_data; | |
1828 | ||
1829 | pe->full_bio = bio; | |
1830 | pe->full_bio_end_io = bio->bi_end_io; | |
a6e50b40 | 1831 | |
a2f83e8b | 1832 | account_start_copy(s); |
a6e50b40 MP |
1833 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, |
1834 | copy_callback, pe); | |
1835 | ||
1836 | bio->bi_end_io = full_bio_end_io; | |
1837 | bio->bi_private = callback_data; | |
1838 | ||
ed00aabd | 1839 | submit_bio_noacct(bio); |
a6e50b40 MP |
1840 | } |
1841 | ||
2913808e MP |
1842 | static struct dm_snap_pending_exception * |
1843 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1844 | { | |
3510cb94 | 1845 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1846 | |
1847 | if (!e) | |
1848 | return NULL; | |
1849 | ||
1850 | return container_of(e, struct dm_snap_pending_exception, e); | |
1851 | } | |
1852 | ||
1da177e4 | 1853 | /* |
65fc7c37 | 1854 | * Inserts a pending exception into the pending table. |
1da177e4 | 1855 | * |
3f1637f2 NT |
1856 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1857 | * before calling this. | |
1da177e4 | 1858 | */ |
028867ac | 1859 | static struct dm_snap_pending_exception * |
65fc7c37 NT |
1860 | __insert_pending_exception(struct dm_snapshot *s, |
1861 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1862 | { |
76df1c65 AK |
1863 | pe->e.old_chunk = chunk; |
1864 | bio_list_init(&pe->origin_bios); | |
1865 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1866 | pe->started = 0; |
a6e50b40 | 1867 | pe->full_bio = NULL; |
76df1c65 | 1868 | |
3f1637f2 | 1869 | spin_lock(&s->pe_allocation_lock); |
493df71c | 1870 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
3f1637f2 | 1871 | spin_unlock(&s->pe_allocation_lock); |
76df1c65 AK |
1872 | free_pending_exception(pe); |
1873 | return NULL; | |
1874 | } | |
1875 | ||
230c83af | 1876 | pe->exception_sequence = s->exception_start_sequence++; |
3f1637f2 | 1877 | spin_unlock(&s->pe_allocation_lock); |
230c83af | 1878 | |
3510cb94 | 1879 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1880 | |
1da177e4 LT |
1881 | return pe; |
1882 | } | |
1883 | ||
65fc7c37 NT |
1884 | /* |
1885 | * Looks to see if this snapshot already has a pending exception | |
1886 | * for this chunk, otherwise it allocates a new one and inserts | |
1887 | * it into the pending table. | |
1888 | * | |
3f1637f2 NT |
1889 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1890 | * before calling this. | |
65fc7c37 NT |
1891 | */ |
1892 | static struct dm_snap_pending_exception * | |
1893 | __find_pending_exception(struct dm_snapshot *s, | |
1894 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1895 | { | |
1896 | struct dm_snap_pending_exception *pe2; | |
1897 | ||
1898 | pe2 = __lookup_pending_exception(s, chunk); | |
1899 | if (pe2) { | |
1900 | free_pending_exception(pe); | |
1901 | return pe2; | |
1902 | } | |
1903 | ||
1904 | return __insert_pending_exception(s, pe, chunk); | |
1905 | } | |
1906 | ||
1d4989c8 | 1907 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1908 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1909 | { |
74d46992 | 1910 | bio_set_dev(bio, s->cow->bdev); |
4f024f37 KO |
1911 | bio->bi_iter.bi_sector = |
1912 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1913 | (chunk - e->old_chunk)) + | |
1914 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1915 | } |
1916 | ||
2e602385 MS |
1917 | static void zero_callback(int read_err, unsigned long write_err, void *context) |
1918 | { | |
1919 | struct bio *bio = context; | |
1920 | struct dm_snapshot *s = bio->bi_private; | |
1921 | ||
a2f83e8b | 1922 | account_end_copy(s); |
2e602385 MS |
1923 | bio->bi_status = write_err ? BLK_STS_IOERR : 0; |
1924 | bio_endio(bio); | |
1925 | } | |
1926 | ||
1927 | static void zero_exception(struct dm_snapshot *s, struct dm_exception *e, | |
1928 | struct bio *bio, chunk_t chunk) | |
1929 | { | |
1930 | struct dm_io_region dest; | |
1931 | ||
1932 | dest.bdev = s->cow->bdev; | |
1933 | dest.sector = bio->bi_iter.bi_sector; | |
1934 | dest.count = s->store->chunk_size; | |
1935 | ||
a2f83e8b | 1936 | account_start_copy(s); |
2e602385 MS |
1937 | WARN_ON_ONCE(bio->bi_private); |
1938 | bio->bi_private = s; | |
1939 | dm_kcopyd_zero(s->kcopyd_client, 1, &dest, 0, zero_callback, bio); | |
1940 | } | |
1941 | ||
1942 | static bool io_overlaps_chunk(struct dm_snapshot *s, struct bio *bio) | |
1943 | { | |
1944 | return bio->bi_iter.bi_size == | |
1945 | (s->store->chunk_size << SECTOR_SHIFT); | |
1946 | } | |
1947 | ||
7de3ee57 | 1948 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1949 | { |
1d4989c8 | 1950 | struct dm_exception *e; |
028867ac | 1951 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1952 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1953 | chunk_t chunk; |
028867ac | 1954 | struct dm_snap_pending_exception *pe = NULL; |
f79ae415 | 1955 | struct dm_exception_table_lock lock; |
1da177e4 | 1956 | |
ee18026a MP |
1957 | init_tracked_chunk(bio); |
1958 | ||
1eff9d32 | 1959 | if (bio->bi_opf & REQ_PREFLUSH) { |
74d46992 | 1960 | bio_set_dev(bio, s->cow->bdev); |
494b3ee7 MP |
1961 | return DM_MAPIO_REMAPPED; |
1962 | } | |
1963 | ||
4f024f37 | 1964 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
f79ae415 | 1965 | dm_exception_table_lock_init(s, chunk, &lock); |
1da177e4 LT |
1966 | |
1967 | /* Full snapshots are not usable */ | |
76df1c65 | 1968 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1969 | if (!s->valid) |
846785e6 | 1970 | return DM_MAPIO_KILL; |
1da177e4 | 1971 | |
b2155578 MP |
1972 | if (bio_data_dir(bio) == WRITE) { |
1973 | while (unlikely(!wait_for_in_progress(s, false))) | |
1974 | ; /* wait_for_in_progress() has slept */ | |
1975 | } | |
1976 | ||
3f1637f2 | 1977 | down_read(&s->lock); |
f79ae415 | 1978 | dm_exception_table_lock(&lock); |
ba40a2aa | 1979 | |
70246286 CH |
1980 | if (!s->valid || (unlikely(s->snapshot_overflowed) && |
1981 | bio_data_dir(bio) == WRITE)) { | |
846785e6 | 1982 | r = DM_MAPIO_KILL; |
ba40a2aa AK |
1983 | goto out_unlock; |
1984 | } | |
1985 | ||
2e602385 MS |
1986 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
1987 | if (s->discard_passdown_origin && dm_bio_get_target_bio_nr(bio)) { | |
1988 | /* | |
1989 | * passdown discard to origin (without triggering | |
1990 | * snapshot exceptions via do_origin; doing so would | |
1991 | * defeat the goal of freeing space in origin that is | |
1992 | * implied by the "discard_passdown_origin" feature) | |
1993 | */ | |
1994 | bio_set_dev(bio, s->origin->bdev); | |
1995 | track_chunk(s, bio, chunk); | |
1996 | goto out_unlock; | |
1997 | } | |
1998 | /* discard to snapshot (target_bio_nr == 0) zeroes exceptions */ | |
1999 | } | |
2000 | ||
ba40a2aa | 2001 | /* If the block is already remapped - use that, else remap it */ |
3510cb94 | 2002 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 2003 | if (e) { |
d74f81f8 | 2004 | remap_exception(s, e, bio, chunk); |
2e602385 MS |
2005 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD) && |
2006 | io_overlaps_chunk(s, bio)) { | |
2007 | dm_exception_table_unlock(&lock); | |
2008 | up_read(&s->lock); | |
2009 | zero_exception(s, e, bio, chunk); | |
2010 | r = DM_MAPIO_SUBMITTED; /* discard is not issued */ | |
2011 | goto out; | |
2012 | } | |
2013 | goto out_unlock; | |
2014 | } | |
2015 | ||
2016 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { | |
2017 | /* | |
2018 | * If no exception exists, complete discard immediately | |
2019 | * otherwise it'll trigger copy-out. | |
2020 | */ | |
2021 | bio_endio(bio); | |
2022 | r = DM_MAPIO_SUBMITTED; | |
ba40a2aa AK |
2023 | goto out_unlock; |
2024 | } | |
2025 | ||
1da177e4 LT |
2026 | /* |
2027 | * Write to snapshot - higher level takes care of RW/RO | |
2028 | * flags so we should only get this if we are | |
2029 | * writeable. | |
2030 | */ | |
70246286 | 2031 | if (bio_data_dir(bio) == WRITE) { |
2913808e | 2032 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 2033 | if (!pe) { |
f79ae415 | 2034 | dm_exception_table_unlock(&lock); |
c6621392 | 2035 | pe = alloc_pending_exception(s); |
f79ae415 | 2036 | dm_exception_table_lock(&lock); |
c6621392 | 2037 | |
3510cb94 | 2038 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
2039 | if (e) { |
2040 | free_pending_exception(pe); | |
2041 | remap_exception(s, e, bio, chunk); | |
2042 | goto out_unlock; | |
2043 | } | |
2044 | ||
c6621392 | 2045 | pe = __find_pending_exception(s, pe, chunk); |
2913808e | 2046 | if (!pe) { |
f79ae415 | 2047 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2048 | up_read(&s->lock); |
2049 | ||
2050 | down_write(&s->lock); | |
f79ae415 | 2051 | |
b0d3cc01 | 2052 | if (s->store->userspace_supports_overflow) { |
3f1637f2 NT |
2053 | if (s->valid && !s->snapshot_overflowed) { |
2054 | s->snapshot_overflowed = 1; | |
2055 | DMERR("Snapshot overflowed: Unable to allocate exception."); | |
2056 | } | |
b0d3cc01 MS |
2057 | } else |
2058 | __invalidate_snapshot(s, -ENOMEM); | |
f79ae415 NT |
2059 | up_write(&s->lock); |
2060 | ||
846785e6 | 2061 | r = DM_MAPIO_KILL; |
f79ae415 | 2062 | goto out; |
2913808e | 2063 | } |
1da177e4 LT |
2064 | } |
2065 | ||
d74f81f8 | 2066 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 2067 | |
d2a7ad29 | 2068 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 2069 | |
2e602385 | 2070 | if (!pe->started && io_overlaps_chunk(s, bio)) { |
a6e50b40 | 2071 | pe->started = 1; |
3f1637f2 | 2072 | |
f79ae415 | 2073 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2074 | up_read(&s->lock); |
2075 | ||
a6e50b40 MP |
2076 | start_full_bio(pe, bio); |
2077 | goto out; | |
2078 | } | |
2079 | ||
2080 | bio_list_add(&pe->snapshot_bios, bio); | |
2081 | ||
76df1c65 | 2082 | if (!pe->started) { |
3f1637f2 | 2083 | /* this is protected by the exception table lock */ |
76df1c65 | 2084 | pe->started = 1; |
3f1637f2 | 2085 | |
f79ae415 | 2086 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2087 | up_read(&s->lock); |
2088 | ||
76df1c65 | 2089 | start_copy(pe); |
ba40a2aa AK |
2090 | goto out; |
2091 | } | |
cd45daff | 2092 | } else { |
74d46992 | 2093 | bio_set_dev(bio, s->origin->bdev); |
ee18026a | 2094 | track_chunk(s, bio, chunk); |
cd45daff | 2095 | } |
1da177e4 | 2096 | |
a2d2b034 | 2097 | out_unlock: |
f79ae415 | 2098 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2099 | up_read(&s->lock); |
a2d2b034 | 2100 | out: |
1da177e4 LT |
2101 | return r; |
2102 | } | |
2103 | ||
3452c2a1 MP |
2104 | /* |
2105 | * A snapshot-merge target behaves like a combination of a snapshot | |
2106 | * target and a snapshot-origin target. It only generates new | |
2107 | * exceptions in other snapshots and not in the one that is being | |
2108 | * merged. | |
2109 | * | |
2110 | * For each chunk, if there is an existing exception, it is used to | |
2111 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
2112 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
2113 | * If merging is currently taking place on the chunk in question, the |
2114 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 2115 | */ |
7de3ee57 | 2116 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
2117 | { |
2118 | struct dm_exception *e; | |
2119 | struct dm_snapshot *s = ti->private; | |
2120 | int r = DM_MAPIO_REMAPPED; | |
2121 | chunk_t chunk; | |
2122 | ||
ee18026a MP |
2123 | init_tracked_chunk(bio); |
2124 | ||
1eff9d32 | 2125 | if (bio->bi_opf & REQ_PREFLUSH) { |
55a62eef | 2126 | if (!dm_bio_get_target_bio_nr(bio)) |
74d46992 | 2127 | bio_set_dev(bio, s->origin->bdev); |
10b8106a | 2128 | else |
74d46992 | 2129 | bio_set_dev(bio, s->cow->bdev); |
10b8106a MS |
2130 | return DM_MAPIO_REMAPPED; |
2131 | } | |
2132 | ||
3ee25485 MS |
2133 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
2134 | /* Once merging, discards no longer effect change */ | |
2135 | bio_endio(bio); | |
2136 | return DM_MAPIO_SUBMITTED; | |
2137 | } | |
2138 | ||
4f024f37 | 2139 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 2140 | |
4ad8d880 | 2141 | down_write(&s->lock); |
3452c2a1 | 2142 | |
d2fdb776 MP |
2143 | /* Full merging snapshots are redirected to the origin */ |
2144 | if (!s->valid) | |
2145 | goto redirect_to_origin; | |
3452c2a1 MP |
2146 | |
2147 | /* If the block is already remapped - use that */ | |
2148 | e = dm_lookup_exception(&s->complete, chunk); | |
2149 | if (e) { | |
9fe86254 | 2150 | /* Queue writes overlapping with chunks being merged */ |
70246286 | 2151 | if (bio_data_dir(bio) == WRITE && |
9fe86254 MP |
2152 | chunk >= s->first_merging_chunk && |
2153 | chunk < (s->first_merging_chunk + | |
2154 | s->num_merging_chunks)) { | |
74d46992 | 2155 | bio_set_dev(bio, s->origin->bdev); |
9fe86254 MP |
2156 | bio_list_add(&s->bios_queued_during_merge, bio); |
2157 | r = DM_MAPIO_SUBMITTED; | |
2158 | goto out_unlock; | |
2159 | } | |
17aa0332 | 2160 | |
3452c2a1 | 2161 | remap_exception(s, e, bio, chunk); |
17aa0332 | 2162 | |
70246286 | 2163 | if (bio_data_dir(bio) == WRITE) |
ee18026a | 2164 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
2165 | goto out_unlock; |
2166 | } | |
2167 | ||
d2fdb776 | 2168 | redirect_to_origin: |
74d46992 | 2169 | bio_set_dev(bio, s->origin->bdev); |
3452c2a1 | 2170 | |
70246286 | 2171 | if (bio_data_dir(bio) == WRITE) { |
4ad8d880 | 2172 | up_write(&s->lock); |
b2155578 | 2173 | return do_origin(s->origin, bio, false); |
3452c2a1 MP |
2174 | } |
2175 | ||
2176 | out_unlock: | |
4ad8d880 | 2177 | up_write(&s->lock); |
3452c2a1 MP |
2178 | |
2179 | return r; | |
2180 | } | |
2181 | ||
4e4cbee9 CH |
2182 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
2183 | blk_status_t *error) | |
cd45daff MP |
2184 | { |
2185 | struct dm_snapshot *s = ti->private; | |
cd45daff | 2186 | |
ee18026a MP |
2187 | if (is_bio_tracked(bio)) |
2188 | stop_tracking_chunk(s, bio); | |
cd45daff | 2189 | |
1be56909 | 2190 | return DM_ENDIO_DONE; |
cd45daff MP |
2191 | } |
2192 | ||
1e03f97e MP |
2193 | static void snapshot_merge_presuspend(struct dm_target *ti) |
2194 | { | |
2195 | struct dm_snapshot *s = ti->private; | |
2196 | ||
2197 | stop_merge(s); | |
2198 | } | |
2199 | ||
c1f0c183 MS |
2200 | static int snapshot_preresume(struct dm_target *ti) |
2201 | { | |
2202 | int r = 0; | |
2203 | struct dm_snapshot *s = ti->private; | |
2204 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
2205 | ||
2206 | down_read(&_origins_lock); | |
9d3b15c4 | 2207 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2208 | if (snap_src && snap_dest) { |
4ad8d880 | 2209 | down_read(&snap_src->lock); |
c1f0c183 MS |
2210 | if (s == snap_src) { |
2211 | DMERR("Unable to resume snapshot source until " | |
2212 | "handover completes."); | |
2213 | r = -EINVAL; | |
b83b2f29 | 2214 | } else if (!dm_suspended(snap_src->ti)) { |
c1f0c183 MS |
2215 | DMERR("Unable to perform snapshot handover until " |
2216 | "source is suspended."); | |
2217 | r = -EINVAL; | |
2218 | } | |
4ad8d880 | 2219 | up_read(&snap_src->lock); |
c1f0c183 MS |
2220 | } |
2221 | up_read(&_origins_lock); | |
2222 | ||
2223 | return r; | |
2224 | } | |
2225 | ||
1da177e4 LT |
2226 | static void snapshot_resume(struct dm_target *ti) |
2227 | { | |
028867ac | 2228 | struct dm_snapshot *s = ti->private; |
09ee96b2 | 2229 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL; |
b735fede MP |
2230 | struct dm_origin *o; |
2231 | struct mapped_device *origin_md = NULL; | |
09ee96b2 | 2232 | bool must_restart_merging = false; |
c1f0c183 MS |
2233 | |
2234 | down_read(&_origins_lock); | |
b735fede MP |
2235 | |
2236 | o = __lookup_dm_origin(s->origin->bdev); | |
2237 | if (o) | |
2238 | origin_md = dm_table_get_md(o->ti->table); | |
09ee96b2 MP |
2239 | if (!origin_md) { |
2240 | (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging); | |
2241 | if (snap_merging) | |
2242 | origin_md = dm_table_get_md(snap_merging->ti->table); | |
2243 | } | |
b735fede MP |
2244 | if (origin_md == dm_table_get_md(ti->table)) |
2245 | origin_md = NULL; | |
09ee96b2 MP |
2246 | if (origin_md) { |
2247 | if (dm_hold(origin_md)) | |
2248 | origin_md = NULL; | |
2249 | } | |
b735fede | 2250 | |
09ee96b2 MP |
2251 | up_read(&_origins_lock); |
2252 | ||
2253 | if (origin_md) { | |
b735fede | 2254 | dm_internal_suspend_fast(origin_md); |
09ee96b2 MP |
2255 | if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) { |
2256 | must_restart_merging = true; | |
2257 | stop_merge(snap_merging); | |
2258 | } | |
2259 | } | |
2260 | ||
2261 | down_read(&_origins_lock); | |
b735fede | 2262 | |
9d3b15c4 | 2263 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2264 | if (snap_src && snap_dest) { |
4ad8d880 NT |
2265 | down_write(&snap_src->lock); |
2266 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
c1f0c183 | 2267 | __handover_exceptions(snap_src, snap_dest); |
4ad8d880 NT |
2268 | up_write(&snap_dest->lock); |
2269 | up_write(&snap_src->lock); | |
c1f0c183 | 2270 | } |
b735fede | 2271 | |
c1f0c183 MS |
2272 | up_read(&_origins_lock); |
2273 | ||
09ee96b2 MP |
2274 | if (origin_md) { |
2275 | if (must_restart_merging) | |
2276 | start_merge(snap_merging); | |
2277 | dm_internal_resume_fast(origin_md); | |
2278 | dm_put(origin_md); | |
2279 | } | |
2280 | ||
c1f0c183 MS |
2281 | /* Now we have correct chunk size, reregister */ |
2282 | reregister_snapshot(s); | |
1da177e4 | 2283 | |
4ad8d880 | 2284 | down_write(&s->lock); |
aa14edeb | 2285 | s->active = 1; |
4ad8d880 | 2286 | up_write(&s->lock); |
1da177e4 LT |
2287 | } |
2288 | ||
542f9038 | 2289 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 2290 | { |
542f9038 | 2291 | uint32_t min_chunksize; |
1e03f97e MP |
2292 | |
2293 | down_read(&_origins_lock); | |
2294 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
2295 | up_read(&_origins_lock); | |
2296 | ||
2297 | return min_chunksize; | |
2298 | } | |
2299 | ||
2300 | static void snapshot_merge_resume(struct dm_target *ti) | |
2301 | { | |
2302 | struct dm_snapshot *s = ti->private; | |
2303 | ||
2304 | /* | |
2305 | * Handover exceptions from existing snapshot. | |
2306 | */ | |
2307 | snapshot_resume(ti); | |
2308 | ||
2309 | /* | |
542f9038 | 2310 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 2311 | */ |
542f9038 | 2312 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
2313 | |
2314 | start_merge(s); | |
2315 | } | |
2316 | ||
fd7c092e MP |
2317 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
2318 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2319 | { |
2e4a31df | 2320 | unsigned sz = 0; |
028867ac | 2321 | struct dm_snapshot *snap = ti->private; |
2e602385 | 2322 | unsigned num_features; |
1da177e4 LT |
2323 | |
2324 | switch (type) { | |
2325 | case STATUSTYPE_INFO: | |
94e76572 | 2326 | |
4ad8d880 | 2327 | down_write(&snap->lock); |
94e76572 | 2328 | |
1da177e4 | 2329 | if (!snap->valid) |
2e4a31df | 2330 | DMEMIT("Invalid"); |
d8ddb1cf MS |
2331 | else if (snap->merge_failed) |
2332 | DMEMIT("Merge failed"); | |
76c44f6d MP |
2333 | else if (snap->snapshot_overflowed) |
2334 | DMEMIT("Overflow"); | |
1da177e4 | 2335 | else { |
985903bb MS |
2336 | if (snap->store->type->usage) { |
2337 | sector_t total_sectors, sectors_allocated, | |
2338 | metadata_sectors; | |
2339 | snap->store->type->usage(snap->store, | |
2340 | &total_sectors, | |
2341 | §ors_allocated, | |
2342 | &metadata_sectors); | |
2343 | DMEMIT("%llu/%llu %llu", | |
2344 | (unsigned long long)sectors_allocated, | |
2345 | (unsigned long long)total_sectors, | |
2346 | (unsigned long long)metadata_sectors); | |
1da177e4 LT |
2347 | } |
2348 | else | |
2e4a31df | 2349 | DMEMIT("Unknown"); |
1da177e4 | 2350 | } |
94e76572 | 2351 | |
4ad8d880 | 2352 | up_write(&snap->lock); |
94e76572 | 2353 | |
1da177e4 LT |
2354 | break; |
2355 | ||
2356 | case STATUSTYPE_TABLE: | |
2357 | /* | |
2358 | * kdevname returns a static pointer so we need | |
2359 | * to make private copies if the output is to | |
2360 | * make sense. | |
2361 | */ | |
fc56f6fb | 2362 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
2e602385 MS |
2363 | sz += snap->store->type->status(snap->store, type, result + sz, |
2364 | maxlen - sz); | |
2365 | num_features = snap->discard_zeroes_cow + snap->discard_passdown_origin; | |
2366 | if (num_features) { | |
2367 | DMEMIT(" %u", num_features); | |
2368 | if (snap->discard_zeroes_cow) | |
2369 | DMEMIT(" discard_zeroes_cow"); | |
2370 | if (snap->discard_passdown_origin) | |
2371 | DMEMIT(" discard_passdown_origin"); | |
2372 | } | |
1da177e4 | 2373 | break; |
8ec45662 TS |
2374 | |
2375 | case STATUSTYPE_IMA: | |
2376 | DMEMIT_TARGET_NAME_VERSION(ti->type); | |
2377 | DMEMIT(",snap_origin_name=%s", snap->origin->name); | |
2378 | DMEMIT(",snap_cow_name=%s", snap->cow->name); | |
2379 | DMEMIT(",snap_valid=%c", snap->valid ? 'y' : 'n'); | |
2380 | DMEMIT(",snap_merge_failed=%c", snap->merge_failed ? 'y' : 'n'); | |
2381 | DMEMIT(",snapshot_overflowed=%c", snap->snapshot_overflowed ? 'y' : 'n'); | |
2382 | DMEMIT(";"); | |
2383 | break; | |
1da177e4 | 2384 | } |
1da177e4 LT |
2385 | } |
2386 | ||
8811f46c MS |
2387 | static int snapshot_iterate_devices(struct dm_target *ti, |
2388 | iterate_devices_callout_fn fn, void *data) | |
2389 | { | |
2390 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
2391 | int r; |
2392 | ||
2393 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 2394 | |
1e5554c8 MP |
2395 | if (!r) |
2396 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
2397 | ||
2398 | return r; | |
8811f46c MS |
2399 | } |
2400 | ||
2e602385 MS |
2401 | static void snapshot_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2402 | { | |
2403 | struct dm_snapshot *snap = ti->private; | |
2404 | ||
2405 | if (snap->discard_zeroes_cow) { | |
2406 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
2407 | ||
3ee25485 MS |
2408 | down_read(&_origins_lock); |
2409 | ||
2e602385 MS |
2410 | (void) __find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, NULL); |
2411 | if (snap_src && snap_dest) | |
2412 | snap = snap_src; | |
2413 | ||
2414 | /* All discards are split on chunk_size boundary */ | |
2415 | limits->discard_granularity = snap->store->chunk_size; | |
2416 | limits->max_discard_sectors = snap->store->chunk_size; | |
3ee25485 MS |
2417 | |
2418 | up_read(&_origins_lock); | |
2e602385 MS |
2419 | } |
2420 | } | |
8811f46c | 2421 | |
1da177e4 LT |
2422 | /*----------------------------------------------------------------- |
2423 | * Origin methods | |
2424 | *---------------------------------------------------------------*/ | |
9eaae8ff MP |
2425 | |
2426 | /* | |
2427 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
2428 | * supplied bio was ignored. The caller may submit it immediately. | |
2429 | * (No remapping actually occurs as the origin is always a direct linear | |
2430 | * map.) | |
2431 | * | |
2432 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
2433 | * and any supplied bio is added to a list to be submitted once all | |
2434 | * the necessary exceptions exist. | |
2435 | */ | |
2436 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
2437 | struct bio *bio) | |
1da177e4 | 2438 | { |
515ad66c | 2439 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2440 | struct dm_snapshot *snap; |
1d4989c8 | 2441 | struct dm_exception *e; |
65fc7c37 | 2442 | struct dm_snap_pending_exception *pe, *pe2; |
515ad66c MP |
2443 | struct dm_snap_pending_exception *pe_to_start_now = NULL; |
2444 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
f79ae415 | 2445 | struct dm_exception_table_lock lock; |
1da177e4 LT |
2446 | chunk_t chunk; |
2447 | ||
2448 | /* Do all the snapshots on this origin */ | |
2449 | list_for_each_entry (snap, snapshots, list) { | |
3452c2a1 MP |
2450 | /* |
2451 | * Don't make new exceptions in a merging snapshot | |
2452 | * because it has effectively been deleted | |
2453 | */ | |
2454 | if (dm_target_is_snapshot_merge(snap->ti)) | |
2455 | continue; | |
2456 | ||
d5e404c1 | 2457 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 2458 | if (sector >= dm_table_get_size(snap->ti->table)) |
f79ae415 | 2459 | continue; |
1da177e4 LT |
2460 | |
2461 | /* | |
2462 | * Remember, different snapshots can have | |
2463 | * different chunk sizes. | |
2464 | */ | |
9eaae8ff | 2465 | chunk = sector_to_chunk(snap->store, sector); |
f79ae415 NT |
2466 | dm_exception_table_lock_init(snap, chunk, &lock); |
2467 | ||
3f1637f2 | 2468 | down_read(&snap->lock); |
f79ae415 NT |
2469 | dm_exception_table_lock(&lock); |
2470 | ||
2471 | /* Only deal with valid and active snapshots */ | |
2472 | if (!snap->valid || !snap->active) | |
2473 | goto next_snapshot; | |
1da177e4 | 2474 | |
2913808e | 2475 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2476 | if (!pe) { |
65fc7c37 NT |
2477 | /* |
2478 | * Check exception table to see if block is already | |
2479 | * remapped in this snapshot and trigger an exception | |
2480 | * if not. | |
2481 | */ | |
2482 | e = dm_lookup_exception(&snap->complete, chunk); | |
2483 | if (e) | |
2484 | goto next_snapshot; | |
2485 | ||
f79ae415 | 2486 | dm_exception_table_unlock(&lock); |
c6621392 | 2487 | pe = alloc_pending_exception(snap); |
f79ae415 | 2488 | dm_exception_table_lock(&lock); |
c6621392 | 2489 | |
65fc7c37 NT |
2490 | pe2 = __lookup_pending_exception(snap, chunk); |
2491 | ||
2492 | if (!pe2) { | |
2493 | e = dm_lookup_exception(&snap->complete, chunk); | |
2494 | if (e) { | |
2495 | free_pending_exception(pe); | |
2496 | goto next_snapshot; | |
2497 | } | |
2498 | ||
2499 | pe = __insert_pending_exception(snap, pe, chunk); | |
2500 | if (!pe) { | |
f79ae415 | 2501 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2502 | up_read(&snap->lock); |
f79ae415 | 2503 | |
3f1637f2 | 2504 | invalidate_snapshot(snap, -ENOMEM); |
f79ae415 | 2505 | continue; |
65fc7c37 NT |
2506 | } |
2507 | } else { | |
35bf659b | 2508 | free_pending_exception(pe); |
65fc7c37 | 2509 | pe = pe2; |
2913808e | 2510 | } |
76df1c65 AK |
2511 | } |
2512 | ||
515ad66c | 2513 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2514 | |
515ad66c MP |
2515 | /* |
2516 | * If an origin bio was supplied, queue it to wait for the | |
2517 | * completion of this exception, and start this one last, | |
2518 | * at the end of the function. | |
2519 | */ | |
2520 | if (bio) { | |
2521 | bio_list_add(&pe->origin_bios, bio); | |
2522 | bio = NULL; | |
76df1c65 | 2523 | |
515ad66c MP |
2524 | if (!pe->started) { |
2525 | pe->started = 1; | |
2526 | pe_to_start_last = pe; | |
2527 | } | |
76df1c65 AK |
2528 | } |
2529 | ||
2530 | if (!pe->started) { | |
2531 | pe->started = 1; | |
515ad66c | 2532 | pe_to_start_now = pe; |
1da177e4 LT |
2533 | } |
2534 | ||
a2d2b034 | 2535 | next_snapshot: |
f79ae415 | 2536 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2537 | up_read(&snap->lock); |
1da177e4 | 2538 | |
515ad66c MP |
2539 | if (pe_to_start_now) { |
2540 | start_copy(pe_to_start_now); | |
2541 | pe_to_start_now = NULL; | |
2542 | } | |
b4b610f6 AK |
2543 | } |
2544 | ||
1da177e4 | 2545 | /* |
515ad66c MP |
2546 | * Submit the exception against which the bio is queued last, |
2547 | * to give the other exceptions a head start. | |
1da177e4 | 2548 | */ |
515ad66c MP |
2549 | if (pe_to_start_last) |
2550 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2551 | |
2552 | return r; | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * Called on a write from the origin driver. | |
2557 | */ | |
b2155578 | 2558 | static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit) |
1da177e4 LT |
2559 | { |
2560 | struct origin *o; | |
d2a7ad29 | 2561 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2562 | |
b2155578 | 2563 | again: |
1da177e4 LT |
2564 | down_read(&_origins_lock); |
2565 | o = __lookup_origin(origin->bdev); | |
b2155578 MP |
2566 | if (o) { |
2567 | if (limit) { | |
2568 | struct dm_snapshot *s; | |
2569 | list_for_each_entry(s, &o->snapshots, list) | |
2570 | if (unlikely(!wait_for_in_progress(s, true))) | |
2571 | goto again; | |
2572 | } | |
2573 | ||
4f024f37 | 2574 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
b2155578 | 2575 | } |
1da177e4 LT |
2576 | up_read(&_origins_lock); |
2577 | ||
2578 | return r; | |
2579 | } | |
2580 | ||
73dfd078 MP |
2581 | /* |
2582 | * Trigger exceptions in all non-merging snapshots. | |
2583 | * | |
2584 | * The chunk size of the merging snapshot may be larger than the chunk | |
2585 | * size of some other snapshot so we may need to reallocate multiple | |
2586 | * chunks in other snapshots. | |
2587 | * | |
2588 | * We scan all the overlapping exceptions in the other snapshots. | |
2589 | * Returns 1 if anything was reallocated and must be waited for, | |
2590 | * otherwise returns 0. | |
2591 | * | |
2592 | * size must be a multiple of merging_snap's chunk_size. | |
2593 | */ | |
2594 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
2595 | sector_t sector, unsigned size) | |
2596 | { | |
2597 | int must_wait = 0; | |
2598 | sector_t n; | |
2599 | struct origin *o; | |
2600 | ||
2601 | /* | |
542f9038 | 2602 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2603 | * by snapshot_merge_resume(). |
2604 | */ | |
2605 | down_read(&_origins_lock); | |
2606 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2607 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2608 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2609 | DM_MAPIO_SUBMITTED) | |
2610 | must_wait = 1; | |
2611 | up_read(&_origins_lock); | |
2612 | ||
2613 | return must_wait; | |
2614 | } | |
2615 | ||
1da177e4 LT |
2616 | /* |
2617 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2618 | */ | |
2619 | ||
2620 | /* | |
2621 | * Construct an origin mapping: <dev_path> | |
2622 | * The context for an origin is merely a 'struct dm_dev *' | |
2623 | * pointing to the real device. | |
2624 | */ | |
2625 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2626 | { | |
2627 | int r; | |
599cdf3b | 2628 | struct dm_origin *o; |
1da177e4 LT |
2629 | |
2630 | if (argc != 1) { | |
72d94861 | 2631 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2632 | return -EINVAL; |
2633 | } | |
2634 | ||
599cdf3b MP |
2635 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2636 | if (!o) { | |
2637 | ti->error = "Cannot allocate private origin structure"; | |
2638 | r = -ENOMEM; | |
2639 | goto bad_alloc; | |
2640 | } | |
2641 | ||
2642 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2643 | if (r) { |
2644 | ti->error = "Cannot get target device"; | |
599cdf3b | 2645 | goto bad_open; |
1da177e4 LT |
2646 | } |
2647 | ||
b735fede | 2648 | o->ti = ti; |
599cdf3b | 2649 | ti->private = o; |
55a62eef | 2650 | ti->num_flush_bios = 1; |
494b3ee7 | 2651 | |
1da177e4 | 2652 | return 0; |
599cdf3b MP |
2653 | |
2654 | bad_open: | |
2655 | kfree(o); | |
2656 | bad_alloc: | |
2657 | return r; | |
1da177e4 LT |
2658 | } |
2659 | ||
2660 | static void origin_dtr(struct dm_target *ti) | |
2661 | { | |
599cdf3b | 2662 | struct dm_origin *o = ti->private; |
b735fede | 2663 | |
599cdf3b MP |
2664 | dm_put_device(ti, o->dev); |
2665 | kfree(o); | |
1da177e4 LT |
2666 | } |
2667 | ||
7de3ee57 | 2668 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2669 | { |
599cdf3b | 2670 | struct dm_origin *o = ti->private; |
298eaa89 | 2671 | unsigned available_sectors; |
1da177e4 | 2672 | |
74d46992 | 2673 | bio_set_dev(bio, o->dev->bdev); |
1da177e4 | 2674 | |
1eff9d32 | 2675 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) |
494b3ee7 MP |
2676 | return DM_MAPIO_REMAPPED; |
2677 | ||
70246286 | 2678 | if (bio_data_dir(bio) != WRITE) |
494b3ee7 MP |
2679 | return DM_MAPIO_REMAPPED; |
2680 | ||
298eaa89 MP |
2681 | available_sectors = o->split_boundary - |
2682 | ((unsigned)bio->bi_iter.bi_sector & (o->split_boundary - 1)); | |
2683 | ||
2684 | if (bio_sectors(bio) > available_sectors) | |
2685 | dm_accept_partial_bio(bio, available_sectors); | |
2686 | ||
1da177e4 | 2687 | /* Only tell snapshots if this is a write */ |
b2155578 | 2688 | return do_origin(o->dev, bio, true); |
1da177e4 LT |
2689 | } |
2690 | ||
1da177e4 | 2691 | /* |
542f9038 | 2692 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2693 | * chunk sizes. |
2694 | */ | |
2695 | static void origin_resume(struct dm_target *ti) | |
2696 | { | |
599cdf3b | 2697 | struct dm_origin *o = ti->private; |
1da177e4 | 2698 | |
298eaa89 | 2699 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
b735fede MP |
2700 | |
2701 | down_write(&_origins_lock); | |
2702 | __insert_dm_origin(o); | |
2703 | up_write(&_origins_lock); | |
2704 | } | |
2705 | ||
2706 | static void origin_postsuspend(struct dm_target *ti) | |
2707 | { | |
2708 | struct dm_origin *o = ti->private; | |
2709 | ||
2710 | down_write(&_origins_lock); | |
2711 | __remove_dm_origin(o); | |
2712 | up_write(&_origins_lock); | |
1da177e4 LT |
2713 | } |
2714 | ||
fd7c092e MP |
2715 | static void origin_status(struct dm_target *ti, status_type_t type, |
2716 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 | 2717 | { |
599cdf3b | 2718 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2719 | |
2720 | switch (type) { | |
2721 | case STATUSTYPE_INFO: | |
2722 | result[0] = '\0'; | |
2723 | break; | |
2724 | ||
2725 | case STATUSTYPE_TABLE: | |
599cdf3b | 2726 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 | 2727 | break; |
8ec45662 TS |
2728 | case STATUSTYPE_IMA: |
2729 | result[0] = '\0'; | |
2730 | break; | |
1da177e4 | 2731 | } |
1da177e4 LT |
2732 | } |
2733 | ||
8811f46c MS |
2734 | static int origin_iterate_devices(struct dm_target *ti, |
2735 | iterate_devices_callout_fn fn, void *data) | |
2736 | { | |
599cdf3b | 2737 | struct dm_origin *o = ti->private; |
8811f46c | 2738 | |
599cdf3b | 2739 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2740 | } |
2741 | ||
1da177e4 LT |
2742 | static struct target_type origin_target = { |
2743 | .name = "snapshot-origin", | |
b735fede | 2744 | .version = {1, 9, 0}, |
1da177e4 LT |
2745 | .module = THIS_MODULE, |
2746 | .ctr = origin_ctr, | |
2747 | .dtr = origin_dtr, | |
2748 | .map = origin_map, | |
2749 | .resume = origin_resume, | |
b735fede | 2750 | .postsuspend = origin_postsuspend, |
1da177e4 | 2751 | .status = origin_status, |
8811f46c | 2752 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2753 | }; |
2754 | ||
2755 | static struct target_type snapshot_target = { | |
2756 | .name = "snapshot", | |
2e602385 | 2757 | .version = {1, 16, 0}, |
1da177e4 LT |
2758 | .module = THIS_MODULE, |
2759 | .ctr = snapshot_ctr, | |
2760 | .dtr = snapshot_dtr, | |
2761 | .map = snapshot_map, | |
cd45daff | 2762 | .end_io = snapshot_end_io, |
c1f0c183 | 2763 | .preresume = snapshot_preresume, |
1da177e4 LT |
2764 | .resume = snapshot_resume, |
2765 | .status = snapshot_status, | |
8811f46c | 2766 | .iterate_devices = snapshot_iterate_devices, |
2e602385 | 2767 | .io_hints = snapshot_io_hints, |
1da177e4 LT |
2768 | }; |
2769 | ||
d698aa45 MP |
2770 | static struct target_type merge_target = { |
2771 | .name = dm_snapshot_merge_target_name, | |
2e602385 | 2772 | .version = {1, 5, 0}, |
d698aa45 MP |
2773 | .module = THIS_MODULE, |
2774 | .ctr = snapshot_ctr, | |
2775 | .dtr = snapshot_dtr, | |
3452c2a1 | 2776 | .map = snapshot_merge_map, |
d698aa45 | 2777 | .end_io = snapshot_end_io, |
1e03f97e | 2778 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2779 | .preresume = snapshot_preresume, |
1e03f97e | 2780 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2781 | .status = snapshot_status, |
2782 | .iterate_devices = snapshot_iterate_devices, | |
2e602385 | 2783 | .io_hints = snapshot_io_hints, |
d698aa45 MP |
2784 | }; |
2785 | ||
1da177e4 LT |
2786 | static int __init dm_snapshot_init(void) |
2787 | { | |
2788 | int r; | |
2789 | ||
4db6bfe0 AK |
2790 | r = dm_exception_store_init(); |
2791 | if (r) { | |
2792 | DMERR("Failed to initialize exception stores"); | |
2793 | return r; | |
2794 | } | |
2795 | ||
1da177e4 LT |
2796 | r = init_origin_hash(); |
2797 | if (r) { | |
2798 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2799 | goto bad_origin_hash; |
1da177e4 LT |
2800 | } |
2801 | ||
1d4989c8 | 2802 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2803 | if (!exception_cache) { |
2804 | DMERR("Couldn't create exception cache."); | |
2805 | r = -ENOMEM; | |
d698aa45 | 2806 | goto bad_exception_cache; |
1da177e4 LT |
2807 | } |
2808 | ||
028867ac | 2809 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2810 | if (!pending_cache) { |
2811 | DMERR("Couldn't create pending cache."); | |
2812 | r = -ENOMEM; | |
d698aa45 | 2813 | goto bad_pending_cache; |
1da177e4 LT |
2814 | } |
2815 | ||
7e6358d2 | 2816 | r = dm_register_target(&snapshot_target); |
2817 | if (r < 0) { | |
2818 | DMERR("snapshot target register failed %d", r); | |
2819 | goto bad_register_snapshot_target; | |
2820 | } | |
2821 | ||
2822 | r = dm_register_target(&origin_target); | |
2823 | if (r < 0) { | |
2824 | DMERR("Origin target register failed %d", r); | |
2825 | goto bad_register_origin_target; | |
2826 | } | |
2827 | ||
2828 | r = dm_register_target(&merge_target); | |
2829 | if (r < 0) { | |
2830 | DMERR("Merge target register failed %d", r); | |
2831 | goto bad_register_merge_target; | |
2832 | } | |
2833 | ||
1da177e4 LT |
2834 | return 0; |
2835 | ||
d698aa45 | 2836 | bad_register_merge_target: |
1da177e4 | 2837 | dm_unregister_target(&origin_target); |
d698aa45 | 2838 | bad_register_origin_target: |
1da177e4 | 2839 | dm_unregister_target(&snapshot_target); |
034a186d | 2840 | bad_register_snapshot_target: |
7e6358d2 | 2841 | kmem_cache_destroy(pending_cache); |
2842 | bad_pending_cache: | |
2843 | kmem_cache_destroy(exception_cache); | |
2844 | bad_exception_cache: | |
2845 | exit_origin_hash(); | |
2846 | bad_origin_hash: | |
034a186d | 2847 | dm_exception_store_exit(); |
d698aa45 | 2848 | |
1da177e4 LT |
2849 | return r; |
2850 | } | |
2851 | ||
2852 | static void __exit dm_snapshot_exit(void) | |
2853 | { | |
10d3bd09 MP |
2854 | dm_unregister_target(&snapshot_target); |
2855 | dm_unregister_target(&origin_target); | |
d698aa45 | 2856 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2857 | |
2858 | exit_origin_hash(); | |
1da177e4 LT |
2859 | kmem_cache_destroy(pending_cache); |
2860 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2861 | |
2862 | dm_exception_store_exit(); | |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /* Module hooks */ | |
2866 | module_init(dm_snapshot_init); | |
2867 | module_exit(dm_snapshot_exit); | |
2868 | ||
2869 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2870 | MODULE_AUTHOR("Joe Thornber"); | |
2871 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2872 | MODULE_ALIAS("dm-snapshot-origin"); |
2873 | MODULE_ALIAS("dm-snapshot-merge"); |