Commit | Line | Data |
---|---|---|
3bd94003 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Copyright (C) 2001-2002 Sistina Software (UK) Limited. |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include <linux/blkdev.h> | |
1da177e4 | 9 | #include <linux/device-mapper.h> |
90fa1527 | 10 | #include <linux/delay.h> |
1da177e4 LT |
11 | #include <linux/fs.h> |
12 | #include <linux/init.h> | |
13 | #include <linux/kdev_t.h> | |
14 | #include <linux/list.h> | |
f79ae415 | 15 | #include <linux/list_bl.h> |
1da177e4 LT |
16 | #include <linux/mempool.h> |
17 | #include <linux/module.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/vmalloc.h> | |
6f3c3f0a | 20 | #include <linux/log2.h> |
a765e20e | 21 | #include <linux/dm-kcopyd.h> |
1da177e4 | 22 | |
b735fede MP |
23 | #include "dm.h" |
24 | ||
aea53d92 | 25 | #include "dm-exception-store.h" |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "snapshots" |
28 | ||
d698aa45 MP |
29 | static const char dm_snapshot_merge_target_name[] = "snapshot-merge"; |
30 | ||
31 | #define dm_target_is_snapshot_merge(ti) \ | |
32 | ((ti)->type->name == dm_snapshot_merge_target_name) | |
33 | ||
cd45daff MP |
34 | /* |
35 | * The size of the mempool used to track chunks in use. | |
36 | */ | |
37 | #define MIN_IOS 256 | |
38 | ||
ccc45ea8 JB |
39 | #define DM_TRACKED_CHUNK_HASH_SIZE 16 |
40 | #define DM_TRACKED_CHUNK_HASH(x) ((unsigned long)(x) & \ | |
41 | (DM_TRACKED_CHUNK_HASH_SIZE - 1)) | |
42 | ||
191437a5 | 43 | struct dm_exception_table { |
ccc45ea8 | 44 | uint32_t hash_mask; |
86a3238c | 45 | unsigned int hash_shift; |
f79ae415 | 46 | struct hlist_bl_head *table; |
ccc45ea8 JB |
47 | }; |
48 | ||
49 | struct dm_snapshot { | |
4ad8d880 | 50 | struct rw_semaphore lock; |
ccc45ea8 JB |
51 | |
52 | struct dm_dev *origin; | |
fc56f6fb MS |
53 | struct dm_dev *cow; |
54 | ||
55 | struct dm_target *ti; | |
ccc45ea8 JB |
56 | |
57 | /* List of snapshots per Origin */ | |
58 | struct list_head list; | |
59 | ||
d8ddb1cf MS |
60 | /* |
61 | * You can't use a snapshot if this is 0 (e.g. if full). | |
62 | * A snapshot-merge target never clears this. | |
63 | */ | |
ccc45ea8 JB |
64 | int valid; |
65 | ||
76c44f6d MP |
66 | /* |
67 | * The snapshot overflowed because of a write to the snapshot device. | |
68 | * We don't have to invalidate the snapshot in this case, but we need | |
69 | * to prevent further writes. | |
70 | */ | |
71 | int snapshot_overflowed; | |
72 | ||
ccc45ea8 JB |
73 | /* Origin writes don't trigger exceptions until this is set */ |
74 | int active; | |
75 | ||
ccc45ea8 JB |
76 | atomic_t pending_exceptions_count; |
77 | ||
3f1637f2 NT |
78 | spinlock_t pe_allocation_lock; |
79 | ||
80 | /* Protected by "pe_allocation_lock" */ | |
230c83af MP |
81 | sector_t exception_start_sequence; |
82 | ||
83 | /* Protected by kcopyd single-threaded callback */ | |
84 | sector_t exception_complete_sequence; | |
85 | ||
86 | /* | |
87 | * A list of pending exceptions that completed out of order. | |
88 | * Protected by kcopyd single-threaded callback. | |
89 | */ | |
3db2776d | 90 | struct rb_root out_of_order_tree; |
230c83af | 91 | |
6f1c819c | 92 | mempool_t pending_pool; |
924e600d | 93 | |
191437a5 JB |
94 | struct dm_exception_table pending; |
95 | struct dm_exception_table complete; | |
ccc45ea8 JB |
96 | |
97 | /* | |
98 | * pe_lock protects all pending_exception operations and access | |
99 | * as well as the snapshot_bios list. | |
100 | */ | |
101 | spinlock_t pe_lock; | |
102 | ||
924e600d MS |
103 | /* Chunks with outstanding reads */ |
104 | spinlock_t tracked_chunk_lock; | |
924e600d MS |
105 | struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE]; |
106 | ||
ccc45ea8 JB |
107 | /* The on disk metadata handler */ |
108 | struct dm_exception_store *store; | |
109 | ||
86a3238c | 110 | unsigned int in_progress; |
b2155578 | 111 | struct wait_queue_head in_progress_wait; |
721b1d98 | 112 | |
ccc45ea8 JB |
113 | struct dm_kcopyd_client *kcopyd_client; |
114 | ||
924e600d MS |
115 | /* Wait for events based on state_bits */ |
116 | unsigned long state_bits; | |
117 | ||
118 | /* Range of chunks currently being merged. */ | |
119 | chunk_t first_merging_chunk; | |
120 | int num_merging_chunks; | |
1e03f97e | 121 | |
d8ddb1cf MS |
122 | /* |
123 | * The merge operation failed if this flag is set. | |
124 | * Failure modes are handled as follows: | |
125 | * - I/O error reading the header | |
8ca817c4 | 126 | * => don't load the target; abort. |
d8ddb1cf | 127 | * - Header does not have "valid" flag set |
8ca817c4 | 128 | * => use the origin; forget about the snapshot. |
d8ddb1cf | 129 | * - I/O error when reading exceptions |
8ca817c4 | 130 | * => don't load the target; abort. |
d8ddb1cf MS |
131 | * (We can't use the intermediate origin state.) |
132 | * - I/O error while merging | |
133 | * => stop merging; set merge_failed; process I/O normally. | |
134 | */ | |
2e602385 MS |
135 | bool merge_failed:1; |
136 | ||
137 | bool discard_zeroes_cow:1; | |
138 | bool discard_passdown_origin:1; | |
d8ddb1cf | 139 | |
9fe86254 MP |
140 | /* |
141 | * Incoming bios that overlap with chunks being merged must wait | |
142 | * for them to be committed. | |
143 | */ | |
144 | struct bio_list bios_queued_during_merge; | |
ccc45ea8 JB |
145 | }; |
146 | ||
1e03f97e MP |
147 | /* |
148 | * state_bits: | |
149 | * RUNNING_MERGE - Merge operation is in progress. | |
150 | * SHUTDOWN_MERGE - Set to signal that merge needs to be stopped; | |
151 | * cleared afterwards. | |
152 | */ | |
153 | #define RUNNING_MERGE 0 | |
154 | #define SHUTDOWN_MERGE 1 | |
155 | ||
721b1d98 NT |
156 | /* |
157 | * Maximum number of chunks being copied on write. | |
158 | * | |
159 | * The value was decided experimentally as a trade-off between memory | |
160 | * consumption, stalling the kernel's workqueues and maintaining a high enough | |
161 | * throughput. | |
162 | */ | |
163 | #define DEFAULT_COW_THRESHOLD 2048 | |
164 | ||
86a3238c | 165 | static unsigned int cow_threshold = DEFAULT_COW_THRESHOLD; |
b2155578 | 166 | module_param_named(snapshot_cow_threshold, cow_threshold, uint, 0644); |
721b1d98 NT |
167 | MODULE_PARM_DESC(snapshot_cow_threshold, "Maximum number of chunks being copied on write"); |
168 | ||
df5d2e90 MP |
169 | DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(snapshot_copy_throttle, |
170 | "A percentage of time allocated for copy on write"); | |
171 | ||
c2411045 MP |
172 | struct dm_dev *dm_snap_origin(struct dm_snapshot *s) |
173 | { | |
174 | return s->origin; | |
175 | } | |
176 | EXPORT_SYMBOL(dm_snap_origin); | |
177 | ||
fc56f6fb MS |
178 | struct dm_dev *dm_snap_cow(struct dm_snapshot *s) |
179 | { | |
180 | return s->cow; | |
181 | } | |
182 | EXPORT_SYMBOL(dm_snap_cow); | |
183 | ||
ccc45ea8 JB |
184 | static sector_t chunk_to_sector(struct dm_exception_store *store, |
185 | chunk_t chunk) | |
186 | { | |
187 | return chunk << store->chunk_shift; | |
188 | } | |
189 | ||
190 | static int bdev_equal(struct block_device *lhs, struct block_device *rhs) | |
191 | { | |
192 | /* | |
193 | * There is only ever one instance of a particular block | |
194 | * device so we can compare pointers safely. | |
195 | */ | |
196 | return lhs == rhs; | |
197 | } | |
198 | ||
028867ac | 199 | struct dm_snap_pending_exception { |
1d4989c8 | 200 | struct dm_exception e; |
1da177e4 LT |
201 | |
202 | /* | |
203 | * Origin buffers waiting for this to complete are held | |
204 | * in a bio list | |
205 | */ | |
206 | struct bio_list origin_bios; | |
207 | struct bio_list snapshot_bios; | |
208 | ||
1da177e4 LT |
209 | /* Pointer back to snapshot context */ |
210 | struct dm_snapshot *snap; | |
211 | ||
212 | /* | |
213 | * 1 indicates the exception has already been sent to | |
214 | * kcopyd. | |
215 | */ | |
216 | int started; | |
a6e50b40 | 217 | |
230c83af MP |
218 | /* There was copying error. */ |
219 | int copy_error; | |
220 | ||
221 | /* A sequence number, it is used for in-order completion. */ | |
222 | sector_t exception_sequence; | |
223 | ||
3db2776d | 224 | struct rb_node out_of_order_node; |
230c83af | 225 | |
a6e50b40 MP |
226 | /* |
227 | * For writing a complete chunk, bypassing the copy. | |
228 | */ | |
229 | struct bio *full_bio; | |
230 | bio_end_io_t *full_bio_end_io; | |
1da177e4 LT |
231 | }; |
232 | ||
233 | /* | |
234 | * Hash table mapping origin volumes to lists of snapshots and | |
235 | * a lock to protect it | |
236 | */ | |
e18b890b CL |
237 | static struct kmem_cache *exception_cache; |
238 | static struct kmem_cache *pending_cache; | |
1da177e4 | 239 | |
cd45daff MP |
240 | struct dm_snap_tracked_chunk { |
241 | struct hlist_node node; | |
242 | chunk_t chunk; | |
243 | }; | |
244 | ||
ee18026a MP |
245 | static void init_tracked_chunk(struct bio *bio) |
246 | { | |
247 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
0ef0b471 | 248 | |
ee18026a MP |
249 | INIT_HLIST_NODE(&c->node); |
250 | } | |
251 | ||
252 | static bool is_bio_tracked(struct bio *bio) | |
253 | { | |
254 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); | |
0ef0b471 | 255 | |
ee18026a MP |
256 | return !hlist_unhashed(&c->node); |
257 | } | |
258 | ||
259 | static void track_chunk(struct dm_snapshot *s, struct bio *bio, chunk_t chunk) | |
cd45daff | 260 | { |
42bc954f | 261 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
262 | |
263 | c->chunk = chunk; | |
264 | ||
9aa0c0e6 | 265 | spin_lock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
266 | hlist_add_head(&c->node, |
267 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)]); | |
9aa0c0e6 | 268 | spin_unlock_irq(&s->tracked_chunk_lock); |
cd45daff MP |
269 | } |
270 | ||
ee18026a | 271 | static void stop_tracking_chunk(struct dm_snapshot *s, struct bio *bio) |
cd45daff | 272 | { |
ee18026a | 273 | struct dm_snap_tracked_chunk *c = dm_per_bio_data(bio, sizeof(struct dm_snap_tracked_chunk)); |
cd45daff MP |
274 | unsigned long flags; |
275 | ||
276 | spin_lock_irqsave(&s->tracked_chunk_lock, flags); | |
277 | hlist_del(&c->node); | |
278 | spin_unlock_irqrestore(&s->tracked_chunk_lock, flags); | |
cd45daff MP |
279 | } |
280 | ||
a8d41b59 MP |
281 | static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk) |
282 | { | |
283 | struct dm_snap_tracked_chunk *c; | |
a8d41b59 MP |
284 | int found = 0; |
285 | ||
286 | spin_lock_irq(&s->tracked_chunk_lock); | |
287 | ||
b67bfe0d | 288 | hlist_for_each_entry(c, |
a8d41b59 MP |
289 | &s->tracked_chunk_hash[DM_TRACKED_CHUNK_HASH(chunk)], node) { |
290 | if (c->chunk == chunk) { | |
291 | found = 1; | |
292 | break; | |
293 | } | |
294 | } | |
295 | ||
296 | spin_unlock_irq(&s->tracked_chunk_lock); | |
297 | ||
298 | return found; | |
299 | } | |
300 | ||
615d1eb9 MS |
301 | /* |
302 | * This conflicting I/O is extremely improbable in the caller, | |
238d991f | 303 | * so fsleep(1000) is sufficient and there is no need for a wait queue. |
615d1eb9 MS |
304 | */ |
305 | static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk) | |
306 | { | |
307 | while (__chunk_is_tracked(s, chunk)) | |
238d991f | 308 | fsleep(1000); |
615d1eb9 MS |
309 | } |
310 | ||
1da177e4 LT |
311 | /* |
312 | * One of these per registered origin, held in the snapshot_origins hash | |
313 | */ | |
314 | struct origin { | |
315 | /* The origin device */ | |
316 | struct block_device *bdev; | |
317 | ||
318 | struct list_head hash_list; | |
319 | ||
320 | /* List of snapshots for this origin */ | |
321 | struct list_head snapshots; | |
322 | }; | |
323 | ||
b735fede MP |
324 | /* |
325 | * This structure is allocated for each origin target | |
326 | */ | |
327 | struct dm_origin { | |
328 | struct dm_dev *dev; | |
329 | struct dm_target *ti; | |
86a3238c | 330 | unsigned int split_boundary; |
b735fede MP |
331 | struct list_head hash_list; |
332 | }; | |
333 | ||
1da177e4 LT |
334 | /* |
335 | * Size of the hash table for origin volumes. If we make this | |
336 | * the size of the minors list then it should be nearly perfect | |
337 | */ | |
338 | #define ORIGIN_HASH_SIZE 256 | |
339 | #define ORIGIN_MASK 0xFF | |
340 | static struct list_head *_origins; | |
b735fede | 341 | static struct list_head *_dm_origins; |
1da177e4 LT |
342 | static struct rw_semaphore _origins_lock; |
343 | ||
73dfd078 MP |
344 | static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done); |
345 | static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock); | |
346 | static uint64_t _pending_exceptions_done_count; | |
347 | ||
1da177e4 LT |
348 | static int init_origin_hash(void) |
349 | { | |
350 | int i; | |
351 | ||
6da2ec56 KC |
352 | _origins = kmalloc_array(ORIGIN_HASH_SIZE, sizeof(struct list_head), |
353 | GFP_KERNEL); | |
1da177e4 | 354 | if (!_origins) { |
b735fede | 355 | DMERR("unable to allocate memory for _origins"); |
1da177e4 LT |
356 | return -ENOMEM; |
357 | } | |
1da177e4 LT |
358 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) |
359 | INIT_LIST_HEAD(_origins + i); | |
b735fede | 360 | |
6da2ec56 KC |
361 | _dm_origins = kmalloc_array(ORIGIN_HASH_SIZE, |
362 | sizeof(struct list_head), | |
363 | GFP_KERNEL); | |
b735fede MP |
364 | if (!_dm_origins) { |
365 | DMERR("unable to allocate memory for _dm_origins"); | |
366 | kfree(_origins); | |
367 | return -ENOMEM; | |
368 | } | |
369 | for (i = 0; i < ORIGIN_HASH_SIZE; i++) | |
370 | INIT_LIST_HEAD(_dm_origins + i); | |
371 | ||
1da177e4 LT |
372 | init_rwsem(&_origins_lock); |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
377 | static void exit_origin_hash(void) | |
378 | { | |
379 | kfree(_origins); | |
b735fede | 380 | kfree(_dm_origins); |
1da177e4 LT |
381 | } |
382 | ||
86a3238c | 383 | static unsigned int origin_hash(struct block_device *bdev) |
1da177e4 LT |
384 | { |
385 | return bdev->bd_dev & ORIGIN_MASK; | |
386 | } | |
387 | ||
388 | static struct origin *__lookup_origin(struct block_device *origin) | |
389 | { | |
390 | struct list_head *ol; | |
391 | struct origin *o; | |
392 | ||
393 | ol = &_origins[origin_hash(origin)]; | |
43be9c74 | 394 | list_for_each_entry(o, ol, hash_list) |
1da177e4 LT |
395 | if (bdev_equal(o->bdev, origin)) |
396 | return o; | |
397 | ||
398 | return NULL; | |
399 | } | |
400 | ||
401 | static void __insert_origin(struct origin *o) | |
402 | { | |
403 | struct list_head *sl = &_origins[origin_hash(o->bdev)]; | |
0ef0b471 | 404 | |
1da177e4 LT |
405 | list_add_tail(&o->hash_list, sl); |
406 | } | |
407 | ||
b735fede MP |
408 | static struct dm_origin *__lookup_dm_origin(struct block_device *origin) |
409 | { | |
410 | struct list_head *ol; | |
411 | struct dm_origin *o; | |
412 | ||
413 | ol = &_dm_origins[origin_hash(origin)]; | |
43be9c74 | 414 | list_for_each_entry(o, ol, hash_list) |
b735fede MP |
415 | if (bdev_equal(o->dev->bdev, origin)) |
416 | return o; | |
417 | ||
418 | return NULL; | |
419 | } | |
420 | ||
421 | static void __insert_dm_origin(struct dm_origin *o) | |
422 | { | |
423 | struct list_head *sl = &_dm_origins[origin_hash(o->dev->bdev)]; | |
0ef0b471 | 424 | |
b735fede MP |
425 | list_add_tail(&o->hash_list, sl); |
426 | } | |
427 | ||
428 | static void __remove_dm_origin(struct dm_origin *o) | |
429 | { | |
430 | list_del(&o->hash_list); | |
431 | } | |
432 | ||
c1f0c183 MS |
433 | /* |
434 | * _origins_lock must be held when calling this function. | |
435 | * Returns number of snapshots registered using the supplied cow device, plus: | |
436 | * snap_src - a snapshot suitable for use as a source of exception handover | |
437 | * snap_dest - a snapshot capable of receiving exception handover. | |
9d3b15c4 MP |
438 | * snap_merge - an existing snapshot-merge target linked to the same origin. |
439 | * There can be at most one snapshot-merge target. The parameter is optional. | |
c1f0c183 | 440 | * |
9d3b15c4 | 441 | * Possible return values and states of snap_src and snap_dest. |
c1f0c183 MS |
442 | * 0: NULL, NULL - first new snapshot |
443 | * 1: snap_src, NULL - normal snapshot | |
444 | * 2: snap_src, snap_dest - waiting for handover | |
445 | * 2: snap_src, NULL - handed over, waiting for old to be deleted | |
446 | * 1: NULL, snap_dest - source got destroyed without handover | |
447 | */ | |
448 | static int __find_snapshots_sharing_cow(struct dm_snapshot *snap, | |
449 | struct dm_snapshot **snap_src, | |
9d3b15c4 MP |
450 | struct dm_snapshot **snap_dest, |
451 | struct dm_snapshot **snap_merge) | |
c1f0c183 MS |
452 | { |
453 | struct dm_snapshot *s; | |
454 | struct origin *o; | |
455 | int count = 0; | |
456 | int active; | |
457 | ||
458 | o = __lookup_origin(snap->origin->bdev); | |
459 | if (!o) | |
460 | goto out; | |
461 | ||
462 | list_for_each_entry(s, &o->snapshots, list) { | |
9d3b15c4 MP |
463 | if (dm_target_is_snapshot_merge(s->ti) && snap_merge) |
464 | *snap_merge = s; | |
c1f0c183 MS |
465 | if (!bdev_equal(s->cow->bdev, snap->cow->bdev)) |
466 | continue; | |
467 | ||
4ad8d880 | 468 | down_read(&s->lock); |
c1f0c183 | 469 | active = s->active; |
4ad8d880 | 470 | up_read(&s->lock); |
c1f0c183 MS |
471 | |
472 | if (active) { | |
473 | if (snap_src) | |
474 | *snap_src = s; | |
475 | } else if (snap_dest) | |
476 | *snap_dest = s; | |
477 | ||
478 | count++; | |
479 | } | |
480 | ||
481 | out: | |
482 | return count; | |
483 | } | |
484 | ||
485 | /* | |
486 | * On success, returns 1 if this snapshot is a handover destination, | |
487 | * otherwise returns 0. | |
488 | */ | |
489 | static int __validate_exception_handover(struct dm_snapshot *snap) | |
490 | { | |
491 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
9d3b15c4 | 492 | struct dm_snapshot *snap_merge = NULL; |
c1f0c183 MS |
493 | |
494 | /* Does snapshot need exceptions handed over to it? */ | |
9d3b15c4 MP |
495 | if ((__find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, |
496 | &snap_merge) == 2) || | |
c1f0c183 | 497 | snap_dest) { |
2e84fecf | 498 | snap->ti->error = "Snapshot cow pairing for exception table handover failed"; |
c1f0c183 MS |
499 | return -EINVAL; |
500 | } | |
501 | ||
502 | /* | |
503 | * If no snap_src was found, snap cannot become a handover | |
504 | * destination. | |
505 | */ | |
506 | if (!snap_src) | |
507 | return 0; | |
508 | ||
9d3b15c4 MP |
509 | /* |
510 | * Non-snapshot-merge handover? | |
511 | */ | |
512 | if (!dm_target_is_snapshot_merge(snap->ti)) | |
513 | return 1; | |
514 | ||
515 | /* | |
516 | * Do not allow more than one merging snapshot. | |
517 | */ | |
518 | if (snap_merge) { | |
519 | snap->ti->error = "A snapshot is already merging."; | |
520 | return -EINVAL; | |
521 | } | |
522 | ||
1e03f97e MP |
523 | if (!snap_src->store->type->prepare_merge || |
524 | !snap_src->store->type->commit_merge) { | |
2e84fecf | 525 | snap->ti->error = "Snapshot exception store does not support snapshot-merge."; |
1e03f97e MP |
526 | return -EINVAL; |
527 | } | |
528 | ||
c1f0c183 MS |
529 | return 1; |
530 | } | |
531 | ||
532 | static void __insert_snapshot(struct origin *o, struct dm_snapshot *s) | |
533 | { | |
534 | struct dm_snapshot *l; | |
535 | ||
536 | /* Sort the list according to chunk size, largest-first smallest-last */ | |
537 | list_for_each_entry(l, &o->snapshots, list) | |
538 | if (l->store->chunk_size < s->store->chunk_size) | |
539 | break; | |
540 | list_add_tail(&s->list, &l->list); | |
541 | } | |
542 | ||
1da177e4 LT |
543 | /* |
544 | * Make a note of the snapshot and its origin so we can look it | |
545 | * up when the origin has a write on it. | |
c1f0c183 MS |
546 | * |
547 | * Also validate snapshot exception store handovers. | |
548 | * On success, returns 1 if this registration is a handover destination, | |
549 | * otherwise returns 0. | |
1da177e4 LT |
550 | */ |
551 | static int register_snapshot(struct dm_snapshot *snap) | |
552 | { | |
c1f0c183 | 553 | struct origin *o, *new_o = NULL; |
1da177e4 | 554 | struct block_device *bdev = snap->origin->bdev; |
c1f0c183 | 555 | int r = 0; |
1da177e4 | 556 | |
60c856c8 MP |
557 | new_o = kmalloc(sizeof(*new_o), GFP_KERNEL); |
558 | if (!new_o) | |
559 | return -ENOMEM; | |
560 | ||
1da177e4 | 561 | down_write(&_origins_lock); |
1da177e4 | 562 | |
c1f0c183 MS |
563 | r = __validate_exception_handover(snap); |
564 | if (r < 0) { | |
565 | kfree(new_o); | |
566 | goto out; | |
567 | } | |
568 | ||
569 | o = __lookup_origin(bdev); | |
60c856c8 MP |
570 | if (o) |
571 | kfree(new_o); | |
572 | else { | |
1da177e4 | 573 | /* New origin */ |
60c856c8 | 574 | o = new_o; |
1da177e4 LT |
575 | |
576 | /* Initialise the struct */ | |
577 | INIT_LIST_HEAD(&o->snapshots); | |
578 | o->bdev = bdev; | |
579 | ||
580 | __insert_origin(o); | |
581 | } | |
582 | ||
c1f0c183 MS |
583 | __insert_snapshot(o, snap); |
584 | ||
585 | out: | |
586 | up_write(&_origins_lock); | |
587 | ||
588 | return r; | |
589 | } | |
590 | ||
591 | /* | |
592 | * Move snapshot to correct place in list according to chunk size. | |
593 | */ | |
594 | static void reregister_snapshot(struct dm_snapshot *s) | |
595 | { | |
596 | struct block_device *bdev = s->origin->bdev; | |
597 | ||
598 | down_write(&_origins_lock); | |
599 | ||
600 | list_del(&s->list); | |
601 | __insert_snapshot(__lookup_origin(bdev), s); | |
1da177e4 LT |
602 | |
603 | up_write(&_origins_lock); | |
1da177e4 LT |
604 | } |
605 | ||
606 | static void unregister_snapshot(struct dm_snapshot *s) | |
607 | { | |
608 | struct origin *o; | |
609 | ||
610 | down_write(&_origins_lock); | |
611 | o = __lookup_origin(s->origin->bdev); | |
612 | ||
613 | list_del(&s->list); | |
c1f0c183 | 614 | if (o && list_empty(&o->snapshots)) { |
1da177e4 LT |
615 | list_del(&o->hash_list); |
616 | kfree(o); | |
617 | } | |
618 | ||
619 | up_write(&_origins_lock); | |
620 | } | |
621 | ||
622 | /* | |
623 | * Implementation of the exception hash tables. | |
d74f81f8 MB |
624 | * The lowest hash_shift bits of the chunk number are ignored, allowing |
625 | * some consecutive chunks to be grouped together. | |
1da177e4 | 626 | */ |
f79ae415 NT |
627 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk); |
628 | ||
629 | /* Lock to protect access to the completed and pending exception hash tables. */ | |
630 | struct dm_exception_table_lock { | |
631 | struct hlist_bl_head *complete_slot; | |
632 | struct hlist_bl_head *pending_slot; | |
633 | }; | |
634 | ||
635 | static void dm_exception_table_lock_init(struct dm_snapshot *s, chunk_t chunk, | |
636 | struct dm_exception_table_lock *lock) | |
637 | { | |
638 | struct dm_exception_table *complete = &s->complete; | |
639 | struct dm_exception_table *pending = &s->pending; | |
640 | ||
641 | lock->complete_slot = &complete->table[exception_hash(complete, chunk)]; | |
642 | lock->pending_slot = &pending->table[exception_hash(pending, chunk)]; | |
643 | } | |
644 | ||
645 | static void dm_exception_table_lock(struct dm_exception_table_lock *lock) | |
646 | { | |
647 | hlist_bl_lock(lock->complete_slot); | |
648 | hlist_bl_lock(lock->pending_slot); | |
649 | } | |
650 | ||
651 | static void dm_exception_table_unlock(struct dm_exception_table_lock *lock) | |
652 | { | |
653 | hlist_bl_unlock(lock->pending_slot); | |
654 | hlist_bl_unlock(lock->complete_slot); | |
655 | } | |
656 | ||
3510cb94 | 657 | static int dm_exception_table_init(struct dm_exception_table *et, |
86a3238c | 658 | uint32_t size, unsigned int hash_shift) |
1da177e4 LT |
659 | { |
660 | unsigned int i; | |
661 | ||
d74f81f8 | 662 | et->hash_shift = hash_shift; |
1da177e4 | 663 | et->hash_mask = size - 1; |
7a35693a MWO |
664 | et->table = kvmalloc_array(size, sizeof(struct hlist_bl_head), |
665 | GFP_KERNEL); | |
1da177e4 LT |
666 | if (!et->table) |
667 | return -ENOMEM; | |
668 | ||
669 | for (i = 0; i < size; i++) | |
f79ae415 | 670 | INIT_HLIST_BL_HEAD(et->table + i); |
1da177e4 LT |
671 | |
672 | return 0; | |
673 | } | |
674 | ||
3510cb94 JB |
675 | static void dm_exception_table_exit(struct dm_exception_table *et, |
676 | struct kmem_cache *mem) | |
1da177e4 | 677 | { |
f79ae415 NT |
678 | struct hlist_bl_head *slot; |
679 | struct dm_exception *ex; | |
680 | struct hlist_bl_node *pos, *n; | |
1da177e4 LT |
681 | int i, size; |
682 | ||
683 | size = et->hash_mask + 1; | |
684 | for (i = 0; i < size; i++) { | |
685 | slot = et->table + i; | |
686 | ||
f79ae415 | 687 | hlist_bl_for_each_entry_safe(ex, pos, n, slot, hash_list) |
1da177e4 LT |
688 | kmem_cache_free(mem, ex); |
689 | } | |
690 | ||
7a35693a | 691 | kvfree(et->table); |
1da177e4 LT |
692 | } |
693 | ||
191437a5 | 694 | static uint32_t exception_hash(struct dm_exception_table *et, chunk_t chunk) |
1da177e4 | 695 | { |
d74f81f8 | 696 | return (chunk >> et->hash_shift) & et->hash_mask; |
1da177e4 LT |
697 | } |
698 | ||
3510cb94 | 699 | static void dm_remove_exception(struct dm_exception *e) |
1da177e4 | 700 | { |
f79ae415 | 701 | hlist_bl_del(&e->hash_list); |
1da177e4 LT |
702 | } |
703 | ||
704 | /* | |
705 | * Return the exception data for a sector, or NULL if not | |
706 | * remapped. | |
707 | */ | |
3510cb94 JB |
708 | static struct dm_exception *dm_lookup_exception(struct dm_exception_table *et, |
709 | chunk_t chunk) | |
1da177e4 | 710 | { |
f79ae415 NT |
711 | struct hlist_bl_head *slot; |
712 | struct hlist_bl_node *pos; | |
1d4989c8 | 713 | struct dm_exception *e; |
1da177e4 LT |
714 | |
715 | slot = &et->table[exception_hash(et, chunk)]; | |
f79ae415 | 716 | hlist_bl_for_each_entry(e, pos, slot, hash_list) |
d74f81f8 MB |
717 | if (chunk >= e->old_chunk && |
718 | chunk <= e->old_chunk + dm_consecutive_chunk_count(e)) | |
1da177e4 LT |
719 | return e; |
720 | ||
721 | return NULL; | |
722 | } | |
723 | ||
119bc547 | 724 | static struct dm_exception *alloc_completed_exception(gfp_t gfp) |
1da177e4 | 725 | { |
1d4989c8 | 726 | struct dm_exception *e; |
1da177e4 | 727 | |
119bc547 MP |
728 | e = kmem_cache_alloc(exception_cache, gfp); |
729 | if (!e && gfp == GFP_NOIO) | |
1da177e4 LT |
730 | e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); |
731 | ||
732 | return e; | |
733 | } | |
734 | ||
3510cb94 | 735 | static void free_completed_exception(struct dm_exception *e) |
1da177e4 LT |
736 | { |
737 | kmem_cache_free(exception_cache, e); | |
738 | } | |
739 | ||
92e86812 | 740 | static struct dm_snap_pending_exception *alloc_pending_exception(struct dm_snapshot *s) |
1da177e4 | 741 | { |
6f1c819c | 742 | struct dm_snap_pending_exception *pe = mempool_alloc(&s->pending_pool, |
92e86812 MP |
743 | GFP_NOIO); |
744 | ||
879129d2 | 745 | atomic_inc(&s->pending_exceptions_count); |
92e86812 MP |
746 | pe->snap = s; |
747 | ||
748 | return pe; | |
1da177e4 LT |
749 | } |
750 | ||
028867ac | 751 | static void free_pending_exception(struct dm_snap_pending_exception *pe) |
1da177e4 | 752 | { |
879129d2 MP |
753 | struct dm_snapshot *s = pe->snap; |
754 | ||
6f1c819c | 755 | mempool_free(pe, &s->pending_pool); |
4e857c58 | 756 | smp_mb__before_atomic(); |
879129d2 | 757 | atomic_dec(&s->pending_exceptions_count); |
1da177e4 LT |
758 | } |
759 | ||
3510cb94 JB |
760 | static void dm_insert_exception(struct dm_exception_table *eh, |
761 | struct dm_exception *new_e) | |
d74f81f8 | 762 | { |
f79ae415 NT |
763 | struct hlist_bl_head *l; |
764 | struct hlist_bl_node *pos; | |
1d4989c8 | 765 | struct dm_exception *e = NULL; |
d74f81f8 MB |
766 | |
767 | l = &eh->table[exception_hash(eh, new_e->old_chunk)]; | |
768 | ||
769 | /* Add immediately if this table doesn't support consecutive chunks */ | |
770 | if (!eh->hash_shift) | |
771 | goto out; | |
772 | ||
773 | /* List is ordered by old_chunk */ | |
f79ae415 | 774 | hlist_bl_for_each_entry(e, pos, l, hash_list) { |
d74f81f8 MB |
775 | /* Insert after an existing chunk? */ |
776 | if (new_e->old_chunk == (e->old_chunk + | |
777 | dm_consecutive_chunk_count(e) + 1) && | |
778 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) + | |
779 | dm_consecutive_chunk_count(e) + 1)) { | |
780 | dm_consecutive_chunk_count_inc(e); | |
3510cb94 | 781 | free_completed_exception(new_e); |
d74f81f8 MB |
782 | return; |
783 | } | |
784 | ||
785 | /* Insert before an existing chunk? */ | |
786 | if (new_e->old_chunk == (e->old_chunk - 1) && | |
787 | new_e->new_chunk == (dm_chunk_number(e->new_chunk) - 1)) { | |
788 | dm_consecutive_chunk_count_inc(e); | |
789 | e->old_chunk--; | |
790 | e->new_chunk--; | |
3510cb94 | 791 | free_completed_exception(new_e); |
d74f81f8 MB |
792 | return; |
793 | } | |
794 | ||
f79ae415 | 795 | if (new_e->old_chunk < e->old_chunk) |
d74f81f8 MB |
796 | break; |
797 | } | |
798 | ||
799 | out: | |
f79ae415 NT |
800 | if (!e) { |
801 | /* | |
802 | * Either the table doesn't support consecutive chunks or slot | |
803 | * l is empty. | |
804 | */ | |
805 | hlist_bl_add_head(&new_e->hash_list, l); | |
806 | } else if (new_e->old_chunk < e->old_chunk) { | |
807 | /* Add before an existing exception */ | |
808 | hlist_bl_add_before(&new_e->hash_list, &e->hash_list); | |
809 | } else { | |
810 | /* Add to l's tail: e is the last exception in this slot */ | |
811 | hlist_bl_add_behind(&new_e->hash_list, &e->hash_list); | |
812 | } | |
d74f81f8 MB |
813 | } |
814 | ||
a159c1ac JB |
815 | /* |
816 | * Callback used by the exception stores to load exceptions when | |
817 | * initialising. | |
818 | */ | |
819 | static int dm_add_exception(void *context, chunk_t old, chunk_t new) | |
1da177e4 | 820 | { |
f79ae415 | 821 | struct dm_exception_table_lock lock; |
a159c1ac | 822 | struct dm_snapshot *s = context; |
1d4989c8 | 823 | struct dm_exception *e; |
1da177e4 | 824 | |
119bc547 | 825 | e = alloc_completed_exception(GFP_KERNEL); |
1da177e4 LT |
826 | if (!e) |
827 | return -ENOMEM; | |
828 | ||
829 | e->old_chunk = old; | |
d74f81f8 MB |
830 | |
831 | /* Consecutive_count is implicitly initialised to zero */ | |
1da177e4 | 832 | e->new_chunk = new; |
d74f81f8 | 833 | |
f79ae415 NT |
834 | /* |
835 | * Although there is no need to lock access to the exception tables | |
836 | * here, if we don't then hlist_bl_add_head(), called by | |
837 | * dm_insert_exception(), will complain about accessing the | |
838 | * corresponding list without locking it first. | |
839 | */ | |
840 | dm_exception_table_lock_init(s, old, &lock); | |
841 | ||
842 | dm_exception_table_lock(&lock); | |
3510cb94 | 843 | dm_insert_exception(&s->complete, e); |
f79ae415 | 844 | dm_exception_table_unlock(&lock); |
d74f81f8 | 845 | |
1da177e4 LT |
846 | return 0; |
847 | } | |
848 | ||
7e201b35 MP |
849 | /* |
850 | * Return a minimum chunk size of all snapshots that have the specified origin. | |
851 | * Return zero if the origin has no snapshots. | |
852 | */ | |
542f9038 | 853 | static uint32_t __minimum_chunk_size(struct origin *o) |
7e201b35 MP |
854 | { |
855 | struct dm_snapshot *snap; | |
86a3238c | 856 | unsigned int chunk_size = rounddown_pow_of_two(UINT_MAX); |
7e201b35 MP |
857 | |
858 | if (o) | |
859 | list_for_each_entry(snap, &o->snapshots, list) | |
f16dba5d MP |
860 | chunk_size = min_not_zero(chunk_size, |
861 | snap->store->chunk_size); | |
7e201b35 | 862 | |
542f9038 | 863 | return (uint32_t) chunk_size; |
7e201b35 MP |
864 | } |
865 | ||
1da177e4 LT |
866 | /* |
867 | * Hard coded magic. | |
868 | */ | |
869 | static int calc_max_buckets(void) | |
870 | { | |
871 | /* use a fixed size of 2MB */ | |
872 | unsigned long mem = 2 * 1024 * 1024; | |
0ef0b471 | 873 | |
f79ae415 | 874 | mem /= sizeof(struct hlist_bl_head); |
1da177e4 LT |
875 | |
876 | return mem; | |
877 | } | |
878 | ||
1da177e4 LT |
879 | /* |
880 | * Allocate room for a suitable hash table. | |
881 | */ | |
fee1998e | 882 | static int init_hash_tables(struct dm_snapshot *s) |
1da177e4 | 883 | { |
60e356f3 | 884 | sector_t hash_size, cow_dev_size, max_buckets; |
1da177e4 LT |
885 | |
886 | /* | |
887 | * Calculate based on the size of the original volume or | |
888 | * the COW volume... | |
889 | */ | |
fc56f6fb | 890 | cow_dev_size = get_dev_size(s->cow->bdev); |
1da177e4 LT |
891 | max_buckets = calc_max_buckets(); |
892 | ||
60e356f3 | 893 | hash_size = cow_dev_size >> s->store->chunk_shift; |
1da177e4 LT |
894 | hash_size = min(hash_size, max_buckets); |
895 | ||
8e87b9b8 MP |
896 | if (hash_size < 64) |
897 | hash_size = 64; | |
8defd830 | 898 | hash_size = rounddown_pow_of_two(hash_size); |
3510cb94 JB |
899 | if (dm_exception_table_init(&s->complete, hash_size, |
900 | DM_CHUNK_CONSECUTIVE_BITS)) | |
1da177e4 LT |
901 | return -ENOMEM; |
902 | ||
903 | /* | |
904 | * Allocate hash table for in-flight exceptions | |
905 | * Make this smaller than the real hash table | |
906 | */ | |
907 | hash_size >>= 3; | |
908 | if (hash_size < 64) | |
909 | hash_size = 64; | |
910 | ||
3510cb94 JB |
911 | if (dm_exception_table_init(&s->pending, hash_size, 0)) { |
912 | dm_exception_table_exit(&s->complete, exception_cache); | |
1da177e4 LT |
913 | return -ENOMEM; |
914 | } | |
915 | ||
916 | return 0; | |
917 | } | |
918 | ||
1e03f97e MP |
919 | static void merge_shutdown(struct dm_snapshot *s) |
920 | { | |
921 | clear_bit_unlock(RUNNING_MERGE, &s->state_bits); | |
4e857c58 | 922 | smp_mb__after_atomic(); |
1e03f97e MP |
923 | wake_up_bit(&s->state_bits, RUNNING_MERGE); |
924 | } | |
925 | ||
9fe86254 MP |
926 | static struct bio *__release_queued_bios_after_merge(struct dm_snapshot *s) |
927 | { | |
928 | s->first_merging_chunk = 0; | |
929 | s->num_merging_chunks = 0; | |
930 | ||
931 | return bio_list_get(&s->bios_queued_during_merge); | |
932 | } | |
933 | ||
1e03f97e MP |
934 | /* |
935 | * Remove one chunk from the index of completed exceptions. | |
936 | */ | |
937 | static int __remove_single_exception_chunk(struct dm_snapshot *s, | |
938 | chunk_t old_chunk) | |
939 | { | |
940 | struct dm_exception *e; | |
941 | ||
1e03f97e MP |
942 | e = dm_lookup_exception(&s->complete, old_chunk); |
943 | if (!e) { | |
2e84fecf | 944 | DMERR("Corruption detected: exception for block %llu is on disk but not in memory", |
1e03f97e MP |
945 | (unsigned long long)old_chunk); |
946 | return -EINVAL; | |
947 | } | |
948 | ||
949 | /* | |
950 | * If this is the only chunk using this exception, remove exception. | |
951 | */ | |
952 | if (!dm_consecutive_chunk_count(e)) { | |
953 | dm_remove_exception(e); | |
954 | free_completed_exception(e); | |
955 | return 0; | |
956 | } | |
957 | ||
958 | /* | |
959 | * The chunk may be either at the beginning or the end of a | |
960 | * group of consecutive chunks - never in the middle. We are | |
961 | * removing chunks in the opposite order to that in which they | |
962 | * were added, so this should always be true. | |
963 | * Decrement the consecutive chunk counter and adjust the | |
964 | * starting point if necessary. | |
965 | */ | |
966 | if (old_chunk == e->old_chunk) { | |
967 | e->old_chunk++; | |
968 | e->new_chunk++; | |
969 | } else if (old_chunk != e->old_chunk + | |
970 | dm_consecutive_chunk_count(e)) { | |
2e84fecf | 971 | DMERR("Attempt to merge block %llu from the middle of a chunk range [%llu - %llu]", |
1e03f97e MP |
972 | (unsigned long long)old_chunk, |
973 | (unsigned long long)e->old_chunk, | |
974 | (unsigned long long) | |
975 | e->old_chunk + dm_consecutive_chunk_count(e)); | |
976 | return -EINVAL; | |
977 | } | |
978 | ||
979 | dm_consecutive_chunk_count_dec(e); | |
980 | ||
981 | return 0; | |
982 | } | |
983 | ||
9fe86254 MP |
984 | static void flush_bios(struct bio *bio); |
985 | ||
986 | static int remove_single_exception_chunk(struct dm_snapshot *s) | |
1e03f97e | 987 | { |
9fe86254 MP |
988 | struct bio *b = NULL; |
989 | int r; | |
990 | chunk_t old_chunk = s->first_merging_chunk + s->num_merging_chunks - 1; | |
1e03f97e | 991 | |
4ad8d880 | 992 | down_write(&s->lock); |
9fe86254 MP |
993 | |
994 | /* | |
995 | * Process chunks (and associated exceptions) in reverse order | |
996 | * so that dm_consecutive_chunk_count_dec() accounting works. | |
997 | */ | |
998 | do { | |
999 | r = __remove_single_exception_chunk(s, old_chunk); | |
1000 | if (r) | |
1001 | goto out; | |
1002 | } while (old_chunk-- > s->first_merging_chunk); | |
1003 | ||
1004 | b = __release_queued_bios_after_merge(s); | |
1005 | ||
1006 | out: | |
4ad8d880 | 1007 | up_write(&s->lock); |
9fe86254 MP |
1008 | if (b) |
1009 | flush_bios(b); | |
1e03f97e MP |
1010 | |
1011 | return r; | |
1012 | } | |
1013 | ||
73dfd078 | 1014 | static int origin_write_extent(struct dm_snapshot *merging_snap, |
86a3238c | 1015 | sector_t sector, unsigned int chunk_size); |
73dfd078 | 1016 | |
1e03f97e MP |
1017 | static void merge_callback(int read_err, unsigned long write_err, |
1018 | void *context); | |
1019 | ||
73dfd078 MP |
1020 | static uint64_t read_pending_exceptions_done_count(void) |
1021 | { | |
1022 | uint64_t pending_exceptions_done; | |
1023 | ||
1024 | spin_lock(&_pending_exceptions_done_spinlock); | |
1025 | pending_exceptions_done = _pending_exceptions_done_count; | |
1026 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1027 | ||
1028 | return pending_exceptions_done; | |
1029 | } | |
1030 | ||
1031 | static void increment_pending_exceptions_done_count(void) | |
1032 | { | |
1033 | spin_lock(&_pending_exceptions_done_spinlock); | |
1034 | _pending_exceptions_done_count++; | |
1035 | spin_unlock(&_pending_exceptions_done_spinlock); | |
1036 | ||
1037 | wake_up_all(&_pending_exceptions_done); | |
1038 | } | |
1039 | ||
1e03f97e MP |
1040 | static void snapshot_merge_next_chunks(struct dm_snapshot *s) |
1041 | { | |
8a2d5286 | 1042 | int i, linear_chunks; |
1e03f97e MP |
1043 | chunk_t old_chunk, new_chunk; |
1044 | struct dm_io_region src, dest; | |
8a2d5286 | 1045 | sector_t io_size; |
73dfd078 | 1046 | uint64_t previous_count; |
1e03f97e MP |
1047 | |
1048 | BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits)); | |
1049 | if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits))) | |
1050 | goto shut; | |
1051 | ||
1052 | /* | |
1053 | * valid flag never changes during merge, so no lock required. | |
1054 | */ | |
1055 | if (!s->valid) { | |
1056 | DMERR("Snapshot is invalid: can't merge"); | |
1057 | goto shut; | |
1058 | } | |
1059 | ||
8a2d5286 MS |
1060 | linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk, |
1061 | &new_chunk); | |
1062 | if (linear_chunks <= 0) { | |
d8ddb1cf | 1063 | if (linear_chunks < 0) { |
2e84fecf | 1064 | DMERR("Read error in exception store: shutting down merge"); |
4ad8d880 | 1065 | down_write(&s->lock); |
1d1dda8c | 1066 | s->merge_failed = true; |
4ad8d880 | 1067 | up_write(&s->lock); |
d8ddb1cf | 1068 | } |
1e03f97e MP |
1069 | goto shut; |
1070 | } | |
1071 | ||
8a2d5286 MS |
1072 | /* Adjust old_chunk and new_chunk to reflect start of linear region */ |
1073 | old_chunk = old_chunk + 1 - linear_chunks; | |
1074 | new_chunk = new_chunk + 1 - linear_chunks; | |
1075 | ||
1076 | /* | |
1077 | * Use one (potentially large) I/O to copy all 'linear_chunks' | |
1078 | * from the exception store to the origin | |
1079 | */ | |
1080 | io_size = linear_chunks * s->store->chunk_size; | |
1e03f97e | 1081 | |
1e03f97e MP |
1082 | dest.bdev = s->origin->bdev; |
1083 | dest.sector = chunk_to_sector(s->store, old_chunk); | |
8a2d5286 | 1084 | dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector); |
1e03f97e MP |
1085 | |
1086 | src.bdev = s->cow->bdev; | |
1087 | src.sector = chunk_to_sector(s->store, new_chunk); | |
1088 | src.count = dest.count; | |
1089 | ||
73dfd078 MP |
1090 | /* |
1091 | * Reallocate any exceptions needed in other snapshots then | |
1092 | * wait for the pending exceptions to complete. | |
1093 | * Each time any pending exception (globally on the system) | |
1094 | * completes we are woken and repeat the process to find out | |
1095 | * if we can proceed. While this may not seem a particularly | |
1096 | * efficient algorithm, it is not expected to have any | |
1097 | * significant impact on performance. | |
1098 | */ | |
1099 | previous_count = read_pending_exceptions_done_count(); | |
8a2d5286 | 1100 | while (origin_write_extent(s, dest.sector, io_size)) { |
73dfd078 MP |
1101 | wait_event(_pending_exceptions_done, |
1102 | (read_pending_exceptions_done_count() != | |
1103 | previous_count)); | |
1104 | /* Retry after the wait, until all exceptions are done. */ | |
1105 | previous_count = read_pending_exceptions_done_count(); | |
1106 | } | |
1107 | ||
4ad8d880 | 1108 | down_write(&s->lock); |
9fe86254 | 1109 | s->first_merging_chunk = old_chunk; |
8a2d5286 | 1110 | s->num_merging_chunks = linear_chunks; |
4ad8d880 | 1111 | up_write(&s->lock); |
9fe86254 | 1112 | |
8a2d5286 MS |
1113 | /* Wait until writes to all 'linear_chunks' drain */ |
1114 | for (i = 0; i < linear_chunks; i++) | |
1115 | __check_for_conflicting_io(s, old_chunk + i); | |
9fe86254 | 1116 | |
1e03f97e MP |
1117 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s); |
1118 | return; | |
1119 | ||
1120 | shut: | |
1121 | merge_shutdown(s); | |
1122 | } | |
1123 | ||
9fe86254 MP |
1124 | static void error_bios(struct bio *bio); |
1125 | ||
1e03f97e MP |
1126 | static void merge_callback(int read_err, unsigned long write_err, void *context) |
1127 | { | |
1128 | struct dm_snapshot *s = context; | |
9fe86254 | 1129 | struct bio *b = NULL; |
1e03f97e MP |
1130 | |
1131 | if (read_err || write_err) { | |
1132 | if (read_err) | |
1133 | DMERR("Read error: shutting down merge."); | |
1134 | else | |
1135 | DMERR("Write error: shutting down merge."); | |
1136 | goto shut; | |
1137 | } | |
1138 | ||
eba33b8e | 1139 | if (blkdev_issue_flush(s->origin->bdev) < 0) { |
fcc42338 AK |
1140 | DMERR("Flush after merge failed: shutting down merge"); |
1141 | goto shut; | |
1142 | } | |
1143 | ||
9fe86254 MP |
1144 | if (s->store->type->commit_merge(s->store, |
1145 | s->num_merging_chunks) < 0) { | |
1e03f97e MP |
1146 | DMERR("Write error in exception store: shutting down merge"); |
1147 | goto shut; | |
1148 | } | |
1149 | ||
9fe86254 MP |
1150 | if (remove_single_exception_chunk(s) < 0) |
1151 | goto shut; | |
1152 | ||
1e03f97e MP |
1153 | snapshot_merge_next_chunks(s); |
1154 | ||
1155 | return; | |
1156 | ||
1157 | shut: | |
4ad8d880 | 1158 | down_write(&s->lock); |
1d1dda8c | 1159 | s->merge_failed = true; |
9fe86254 | 1160 | b = __release_queued_bios_after_merge(s); |
4ad8d880 | 1161 | up_write(&s->lock); |
9fe86254 MP |
1162 | error_bios(b); |
1163 | ||
1e03f97e MP |
1164 | merge_shutdown(s); |
1165 | } | |
1166 | ||
1167 | static void start_merge(struct dm_snapshot *s) | |
1168 | { | |
1169 | if (!test_and_set_bit(RUNNING_MERGE, &s->state_bits)) | |
1170 | snapshot_merge_next_chunks(s); | |
1171 | } | |
1172 | ||
1e03f97e MP |
1173 | /* |
1174 | * Stop the merging process and wait until it finishes. | |
1175 | */ | |
1176 | static void stop_merge(struct dm_snapshot *s) | |
1177 | { | |
1178 | set_bit(SHUTDOWN_MERGE, &s->state_bits); | |
74316201 | 1179 | wait_on_bit(&s->state_bits, RUNNING_MERGE, TASK_UNINTERRUPTIBLE); |
1e03f97e MP |
1180 | clear_bit(SHUTDOWN_MERGE, &s->state_bits); |
1181 | } | |
1182 | ||
2e602385 MS |
1183 | static int parse_snapshot_features(struct dm_arg_set *as, struct dm_snapshot *s, |
1184 | struct dm_target *ti) | |
1185 | { | |
1186 | int r; | |
86a3238c | 1187 | unsigned int argc; |
2e602385 MS |
1188 | const char *arg_name; |
1189 | ||
1190 | static const struct dm_arg _args[] = { | |
1191 | {0, 2, "Invalid number of feature arguments"}, | |
1192 | }; | |
1193 | ||
1194 | /* | |
1195 | * No feature arguments supplied. | |
1196 | */ | |
1197 | if (!as->argc) | |
1198 | return 0; | |
1199 | ||
1200 | r = dm_read_arg_group(_args, as, &argc, &ti->error); | |
1201 | if (r) | |
1202 | return -EINVAL; | |
1203 | ||
1204 | while (argc && !r) { | |
1205 | arg_name = dm_shift_arg(as); | |
1206 | argc--; | |
1207 | ||
1208 | if (!strcasecmp(arg_name, "discard_zeroes_cow")) | |
1209 | s->discard_zeroes_cow = true; | |
1210 | ||
1211 | else if (!strcasecmp(arg_name, "discard_passdown_origin")) | |
1212 | s->discard_passdown_origin = true; | |
1213 | ||
1214 | else { | |
1215 | ti->error = "Unrecognised feature requested"; | |
1216 | r = -EINVAL; | |
1217 | break; | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | if (!s->discard_zeroes_cow && s->discard_passdown_origin) { | |
1222 | /* | |
1223 | * TODO: really these are disjoint.. but ti->num_discard_bios | |
1224 | * and dm_bio_get_target_bio_nr() require rigid constraints. | |
1225 | */ | |
1226 | ti->error = "discard_passdown_origin feature depends on discard_zeroes_cow"; | |
1227 | r = -EINVAL; | |
1228 | } | |
1229 | ||
1230 | return r; | |
1231 | } | |
1232 | ||
1da177e4 | 1233 | /* |
2e602385 MS |
1234 | * Construct a snapshot mapping: |
1235 | * <origin_dev> <COW-dev> <p|po|n> <chunk-size> [<# feature args> [<arg>]*] | |
1da177e4 LT |
1236 | */ |
1237 | static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
1238 | { | |
1239 | struct dm_snapshot *s; | |
2e602385 | 1240 | struct dm_arg_set as; |
cd45daff | 1241 | int i; |
1da177e4 | 1242 | int r = -EINVAL; |
fc56f6fb | 1243 | char *origin_path, *cow_path; |
86a3238c | 1244 | unsigned int 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 | } | |
1281 | ||
fc56f6fb MS |
1282 | cow_path = argv[0]; |
1283 | argv++; | |
1284 | argc--; | |
1285 | ||
024d37e9 | 1286 | r = dm_get_device(ti, cow_path, dm_table_get_mode(ti->table), &s->cow); |
fc56f6fb MS |
1287 | if (r) { |
1288 | ti->error = "Cannot get COW device"; | |
1289 | goto bad_cow; | |
1290 | } | |
26110d5a CH |
1291 | if (s->cow->bdev && s->cow->bdev == s->origin->bdev) { |
1292 | ti->error = "COW device cannot be the same as origin device"; | |
1293 | r = -EINVAL; | |
1294 | goto bad_store; | |
1295 | } | |
fc56f6fb MS |
1296 | |
1297 | r = dm_exception_store_create(ti, argc, argv, s, &args_used, &s->store); | |
fee1998e JB |
1298 | if (r) { |
1299 | ti->error = "Couldn't create exception store"; | |
1da177e4 | 1300 | r = -EINVAL; |
fc56f6fb | 1301 | goto bad_store; |
1da177e4 LT |
1302 | } |
1303 | ||
fee1998e JB |
1304 | argv += args_used; |
1305 | argc -= args_used; | |
1306 | ||
fc56f6fb | 1307 | s->ti = ti; |
1da177e4 | 1308 | s->valid = 1; |
76c44f6d | 1309 | s->snapshot_overflowed = 0; |
aa14edeb | 1310 | s->active = 0; |
879129d2 | 1311 | atomic_set(&s->pending_exceptions_count, 0); |
3f1637f2 | 1312 | spin_lock_init(&s->pe_allocation_lock); |
230c83af MP |
1313 | s->exception_start_sequence = 0; |
1314 | s->exception_complete_sequence = 0; | |
3db2776d | 1315 | s->out_of_order_tree = RB_ROOT; |
4ad8d880 | 1316 | init_rwsem(&s->lock); |
c1f0c183 | 1317 | INIT_LIST_HEAD(&s->list); |
ca3a931f | 1318 | spin_lock_init(&s->pe_lock); |
1e03f97e | 1319 | s->state_bits = 0; |
1d1dda8c | 1320 | s->merge_failed = false; |
9fe86254 MP |
1321 | s->first_merging_chunk = 0; |
1322 | s->num_merging_chunks = 0; | |
1323 | bio_list_init(&s->bios_queued_during_merge); | |
1da177e4 LT |
1324 | |
1325 | /* Allocate hash table for COW data */ | |
fee1998e | 1326 | if (init_hash_tables(s)) { |
1da177e4 LT |
1327 | ti->error = "Unable to allocate hash table space"; |
1328 | r = -ENOMEM; | |
fee1998e | 1329 | goto bad_hash_tables; |
1da177e4 LT |
1330 | } |
1331 | ||
b2155578 | 1332 | init_waitqueue_head(&s->in_progress_wait); |
721b1d98 | 1333 | |
df5d2e90 | 1334 | s->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle); |
fa34ce73 MP |
1335 | if (IS_ERR(s->kcopyd_client)) { |
1336 | r = PTR_ERR(s->kcopyd_client); | |
1da177e4 | 1337 | ti->error = "Could not create kcopyd client"; |
fee1998e | 1338 | goto bad_kcopyd; |
1da177e4 LT |
1339 | } |
1340 | ||
6f1c819c KO |
1341 | r = mempool_init_slab_pool(&s->pending_pool, MIN_IOS, pending_cache); |
1342 | if (r) { | |
92e86812 | 1343 | ti->error = "Could not allocate mempool for pending exceptions"; |
fee1998e | 1344 | goto bad_pending_pool; |
92e86812 MP |
1345 | } |
1346 | ||
cd45daff MP |
1347 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) |
1348 | INIT_HLIST_HEAD(&s->tracked_chunk_hash[i]); | |
1349 | ||
1350 | spin_lock_init(&s->tracked_chunk_lock); | |
1351 | ||
c1f0c183 | 1352 | ti->private = s; |
55a62eef | 1353 | ti->num_flush_bios = num_flush_bios; |
2e602385 MS |
1354 | if (s->discard_zeroes_cow) |
1355 | ti->num_discard_bios = (s->discard_passdown_origin ? 2 : 1); | |
30187e1d | 1356 | ti->per_io_data_size = sizeof(struct dm_snap_tracked_chunk); |
c1f0c183 MS |
1357 | |
1358 | /* Add snapshot to the list of snapshots for this origin */ | |
1359 | /* Exceptions aren't triggered till snapshot_resume() is called */ | |
1360 | r = register_snapshot(s); | |
1361 | if (r == -ENOMEM) { | |
1362 | ti->error = "Snapshot origin struct allocation failed"; | |
1363 | goto bad_load_and_register; | |
1364 | } else if (r < 0) { | |
1365 | /* invalid handover, register_snapshot has set ti->error */ | |
1366 | goto bad_load_and_register; | |
1367 | } | |
1368 | ||
1369 | /* | |
1370 | * Metadata must only be loaded into one table at once, so skip this | |
1371 | * if metadata will be handed over during resume. | |
1372 | * Chunk size will be set during the handover - set it to zero to | |
1373 | * ensure it's ignored. | |
1374 | */ | |
1375 | if (r > 0) { | |
1376 | s->store->chunk_size = 0; | |
1377 | return 0; | |
1378 | } | |
1379 | ||
493df71c JB |
1380 | r = s->store->type->read_metadata(s->store, dm_add_exception, |
1381 | (void *)s); | |
0764147b | 1382 | if (r < 0) { |
f9cea4f7 | 1383 | ti->error = "Failed to read snapshot metadata"; |
c1f0c183 | 1384 | goto bad_read_metadata; |
0764147b MB |
1385 | } else if (r > 0) { |
1386 | s->valid = 0; | |
1387 | DMWARN("Snapshot is marked invalid."); | |
f9cea4f7 | 1388 | } |
aa14edeb | 1389 | |
3f2412dc MP |
1390 | if (!s->store->chunk_size) { |
1391 | ti->error = "Chunk size not set"; | |
c699a0db | 1392 | r = -EINVAL; |
c1f0c183 | 1393 | goto bad_read_metadata; |
1da177e4 | 1394 | } |
542f9038 MS |
1395 | |
1396 | r = dm_set_target_max_io_len(ti, s->store->chunk_size); | |
1397 | if (r) | |
1398 | goto bad_read_metadata; | |
1da177e4 LT |
1399 | |
1400 | return 0; | |
1401 | ||
c1f0c183 MS |
1402 | bad_read_metadata: |
1403 | unregister_snapshot(s); | |
fee1998e | 1404 | bad_load_and_register: |
6f1c819c | 1405 | mempool_exit(&s->pending_pool); |
fee1998e | 1406 | bad_pending_pool: |
eb69aca5 | 1407 | dm_kcopyd_client_destroy(s->kcopyd_client); |
fee1998e | 1408 | bad_kcopyd: |
3510cb94 JB |
1409 | dm_exception_table_exit(&s->pending, pending_cache); |
1410 | dm_exception_table_exit(&s->complete, exception_cache); | |
fee1998e | 1411 | bad_hash_tables: |
fc56f6fb | 1412 | dm_exception_store_destroy(s->store); |
fc56f6fb MS |
1413 | bad_store: |
1414 | dm_put_device(ti, s->cow); | |
fc56f6fb | 1415 | bad_cow: |
c2411045 | 1416 | dm_put_device(ti, s->origin); |
c2411045 | 1417 | bad_origin: |
2e602385 | 1418 | bad_features: |
fc56f6fb | 1419 | kfree(s); |
fc56f6fb | 1420 | bad: |
1da177e4 LT |
1421 | return r; |
1422 | } | |
1423 | ||
31c93a0c MB |
1424 | static void __free_exceptions(struct dm_snapshot *s) |
1425 | { | |
eb69aca5 | 1426 | dm_kcopyd_client_destroy(s->kcopyd_client); |
31c93a0c MB |
1427 | s->kcopyd_client = NULL; |
1428 | ||
3510cb94 JB |
1429 | dm_exception_table_exit(&s->pending, pending_cache); |
1430 | dm_exception_table_exit(&s->complete, exception_cache); | |
31c93a0c MB |
1431 | } |
1432 | ||
c1f0c183 MS |
1433 | static void __handover_exceptions(struct dm_snapshot *snap_src, |
1434 | struct dm_snapshot *snap_dest) | |
1435 | { | |
1436 | union { | |
1437 | struct dm_exception_table table_swap; | |
1438 | struct dm_exception_store *store_swap; | |
1439 | } u; | |
1440 | ||
1441 | /* | |
1442 | * Swap all snapshot context information between the two instances. | |
1443 | */ | |
1444 | u.table_swap = snap_dest->complete; | |
1445 | snap_dest->complete = snap_src->complete; | |
1446 | snap_src->complete = u.table_swap; | |
1447 | ||
1448 | u.store_swap = snap_dest->store; | |
1449 | snap_dest->store = snap_src->store; | |
b0d3cc01 | 1450 | snap_dest->store->userspace_supports_overflow = u.store_swap->userspace_supports_overflow; |
c1f0c183 MS |
1451 | snap_src->store = u.store_swap; |
1452 | ||
1453 | snap_dest->store->snap = snap_dest; | |
1454 | snap_src->store->snap = snap_src; | |
1455 | ||
542f9038 | 1456 | snap_dest->ti->max_io_len = snap_dest->store->chunk_size; |
c1f0c183 | 1457 | snap_dest->valid = snap_src->valid; |
76c44f6d | 1458 | snap_dest->snapshot_overflowed = snap_src->snapshot_overflowed; |
c1f0c183 MS |
1459 | |
1460 | /* | |
1461 | * Set source invalid to ensure it receives no further I/O. | |
1462 | */ | |
1463 | snap_src->valid = 0; | |
1464 | } | |
1465 | ||
1da177e4 LT |
1466 | static void snapshot_dtr(struct dm_target *ti) |
1467 | { | |
cd45daff MP |
1468 | #ifdef CONFIG_DM_DEBUG |
1469 | int i; | |
1470 | #endif | |
028867ac | 1471 | struct dm_snapshot *s = ti->private; |
c1f0c183 | 1472 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; |
1da177e4 | 1473 | |
c1f0c183 MS |
1474 | down_read(&_origins_lock); |
1475 | /* Check whether exception handover must be cancelled */ | |
9d3b15c4 | 1476 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 1477 | if (snap_src && snap_dest && (s == snap_src)) { |
4ad8d880 | 1478 | down_write(&snap_dest->lock); |
c1f0c183 | 1479 | snap_dest->valid = 0; |
4ad8d880 | 1480 | up_write(&snap_dest->lock); |
c1f0c183 MS |
1481 | DMERR("Cancelling snapshot handover."); |
1482 | } | |
1483 | up_read(&_origins_lock); | |
1484 | ||
1e03f97e MP |
1485 | if (dm_target_is_snapshot_merge(ti)) |
1486 | stop_merge(s); | |
1487 | ||
138728dc AK |
1488 | /* Prevent further origin writes from using this snapshot. */ |
1489 | /* After this returns there can be no new kcopyd jobs. */ | |
1da177e4 LT |
1490 | unregister_snapshot(s); |
1491 | ||
879129d2 | 1492 | while (atomic_read(&s->pending_exceptions_count)) |
238d991f | 1493 | fsleep(1000); |
879129d2 | 1494 | /* |
6f1c819c | 1495 | * Ensure instructions in mempool_exit aren't reordered |
879129d2 MP |
1496 | * before atomic_read. |
1497 | */ | |
1498 | smp_mb(); | |
1499 | ||
cd45daff MP |
1500 | #ifdef CONFIG_DM_DEBUG |
1501 | for (i = 0; i < DM_TRACKED_CHUNK_HASH_SIZE; i++) | |
1502 | BUG_ON(!hlist_empty(&s->tracked_chunk_hash[i])); | |
1503 | #endif | |
1504 | ||
31c93a0c | 1505 | __free_exceptions(s); |
1da177e4 | 1506 | |
6f1c819c | 1507 | mempool_exit(&s->pending_pool); |
92e86812 | 1508 | |
fee1998e | 1509 | dm_exception_store_destroy(s->store); |
138728dc | 1510 | |
fc56f6fb MS |
1511 | dm_put_device(ti, s->cow); |
1512 | ||
c2411045 MP |
1513 | dm_put_device(ti, s->origin); |
1514 | ||
b2155578 MP |
1515 | WARN_ON(s->in_progress); |
1516 | ||
1da177e4 LT |
1517 | kfree(s); |
1518 | } | |
1519 | ||
a2f83e8b MP |
1520 | static void account_start_copy(struct dm_snapshot *s) |
1521 | { | |
b2155578 MP |
1522 | spin_lock(&s->in_progress_wait.lock); |
1523 | s->in_progress++; | |
1524 | spin_unlock(&s->in_progress_wait.lock); | |
a2f83e8b MP |
1525 | } |
1526 | ||
1527 | static void account_end_copy(struct dm_snapshot *s) | |
1528 | { | |
b2155578 MP |
1529 | spin_lock(&s->in_progress_wait.lock); |
1530 | BUG_ON(!s->in_progress); | |
1531 | s->in_progress--; | |
1532 | if (likely(s->in_progress <= cow_threshold) && | |
1533 | unlikely(waitqueue_active(&s->in_progress_wait))) | |
1534 | wake_up_locked(&s->in_progress_wait); | |
1535 | spin_unlock(&s->in_progress_wait.lock); | |
1536 | } | |
1537 | ||
1538 | static bool wait_for_in_progress(struct dm_snapshot *s, bool unlock_origins) | |
1539 | { | |
1540 | if (unlikely(s->in_progress > cow_threshold)) { | |
1541 | spin_lock(&s->in_progress_wait.lock); | |
1542 | if (likely(s->in_progress > cow_threshold)) { | |
1543 | /* | |
1544 | * NOTE: this throttle doesn't account for whether | |
1545 | * the caller is servicing an IO that will trigger a COW | |
1546 | * so excess throttling may result for chunks not required | |
1547 | * to be COW'd. But if cow_threshold was reached, extra | |
1548 | * throttling is unlikely to negatively impact performance. | |
1549 | */ | |
1550 | DECLARE_WAITQUEUE(wait, current); | |
0ef0b471 | 1551 | |
b2155578 MP |
1552 | __add_wait_queue(&s->in_progress_wait, &wait); |
1553 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
1554 | spin_unlock(&s->in_progress_wait.lock); | |
1555 | if (unlock_origins) | |
1556 | up_read(&_origins_lock); | |
1557 | io_schedule(); | |
1558 | remove_wait_queue(&s->in_progress_wait, &wait); | |
1559 | return false; | |
1560 | } | |
1561 | spin_unlock(&s->in_progress_wait.lock); | |
1562 | } | |
1563 | return true; | |
a2f83e8b MP |
1564 | } |
1565 | ||
1da177e4 LT |
1566 | /* |
1567 | * Flush a list of buffers. | |
1568 | */ | |
1569 | static void flush_bios(struct bio *bio) | |
1570 | { | |
1571 | struct bio *n; | |
1572 | ||
1573 | while (bio) { | |
1574 | n = bio->bi_next; | |
1575 | bio->bi_next = NULL; | |
ed00aabd | 1576 | submit_bio_noacct(bio); |
1da177e4 LT |
1577 | bio = n; |
1578 | } | |
1579 | } | |
1580 | ||
b2155578 | 1581 | static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit); |
515ad66c MP |
1582 | |
1583 | /* | |
1584 | * Flush a list of buffers. | |
1585 | */ | |
1586 | static void retry_origin_bios(struct dm_snapshot *s, struct bio *bio) | |
1587 | { | |
1588 | struct bio *n; | |
1589 | int r; | |
1590 | ||
1591 | while (bio) { | |
1592 | n = bio->bi_next; | |
1593 | bio->bi_next = NULL; | |
b2155578 | 1594 | r = do_origin(s->origin, bio, false); |
515ad66c | 1595 | if (r == DM_MAPIO_REMAPPED) |
ed00aabd | 1596 | submit_bio_noacct(bio); |
515ad66c MP |
1597 | bio = n; |
1598 | } | |
1599 | } | |
1600 | ||
1da177e4 LT |
1601 | /* |
1602 | * Error a list of buffers. | |
1603 | */ | |
1604 | static void error_bios(struct bio *bio) | |
1605 | { | |
1606 | struct bio *n; | |
1607 | ||
1608 | while (bio) { | |
1609 | n = bio->bi_next; | |
1610 | bio->bi_next = NULL; | |
6712ecf8 | 1611 | bio_io_error(bio); |
1da177e4 LT |
1612 | bio = n; |
1613 | } | |
1614 | } | |
1615 | ||
695368ac | 1616 | static void __invalidate_snapshot(struct dm_snapshot *s, int err) |
76df1c65 AK |
1617 | { |
1618 | if (!s->valid) | |
1619 | return; | |
1620 | ||
1621 | if (err == -EIO) | |
1622 | DMERR("Invalidating snapshot: Error reading/writing."); | |
1623 | else if (err == -ENOMEM) | |
1624 | DMERR("Invalidating snapshot: Unable to allocate exception."); | |
1625 | ||
493df71c JB |
1626 | if (s->store->type->drop_snapshot) |
1627 | s->store->type->drop_snapshot(s->store); | |
76df1c65 AK |
1628 | |
1629 | s->valid = 0; | |
1630 | ||
fc56f6fb | 1631 | dm_table_event(s->ti->table); |
76df1c65 AK |
1632 | } |
1633 | ||
3f1637f2 NT |
1634 | static void invalidate_snapshot(struct dm_snapshot *s, int err) |
1635 | { | |
1636 | down_write(&s->lock); | |
1637 | __invalidate_snapshot(s, err); | |
1638 | up_write(&s->lock); | |
1639 | } | |
1640 | ||
385277bf | 1641 | static void pending_complete(void *context, int success) |
1da177e4 | 1642 | { |
385277bf | 1643 | struct dm_snap_pending_exception *pe = context; |
1d4989c8 | 1644 | struct dm_exception *e; |
1da177e4 | 1645 | struct dm_snapshot *s = pe->snap; |
9d493fa8 AK |
1646 | struct bio *origin_bios = NULL; |
1647 | struct bio *snapshot_bios = NULL; | |
a6e50b40 | 1648 | struct bio *full_bio = NULL; |
f79ae415 | 1649 | struct dm_exception_table_lock lock; |
9d493fa8 | 1650 | int error = 0; |
1da177e4 | 1651 | |
f79ae415 NT |
1652 | dm_exception_table_lock_init(s, pe->e.old_chunk, &lock); |
1653 | ||
76df1c65 AK |
1654 | if (!success) { |
1655 | /* Read/write error - snapshot is unusable */ | |
3f1637f2 | 1656 | invalidate_snapshot(s, -EIO); |
9d493fa8 | 1657 | error = 1; |
f79ae415 NT |
1658 | |
1659 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1660 | goto out; |
1661 | } | |
1662 | ||
119bc547 | 1663 | e = alloc_completed_exception(GFP_NOIO); |
76df1c65 | 1664 | if (!e) { |
3f1637f2 | 1665 | invalidate_snapshot(s, -ENOMEM); |
9d493fa8 | 1666 | error = 1; |
f79ae415 NT |
1667 | |
1668 | dm_exception_table_lock(&lock); | |
76df1c65 AK |
1669 | goto out; |
1670 | } | |
1671 | *e = pe->e; | |
1da177e4 | 1672 | |
3f1637f2 | 1673 | down_read(&s->lock); |
f79ae415 | 1674 | dm_exception_table_lock(&lock); |
76df1c65 | 1675 | if (!s->valid) { |
3f1637f2 | 1676 | up_read(&s->lock); |
3510cb94 | 1677 | free_completed_exception(e); |
9d493fa8 | 1678 | error = 1; |
3f1637f2 | 1679 | |
76df1c65 | 1680 | goto out; |
1da177e4 LT |
1681 | } |
1682 | ||
9d493fa8 | 1683 | /* |
65fc7c37 NT |
1684 | * Add a proper exception. After inserting the completed exception all |
1685 | * subsequent snapshot reads to this chunk will be redirected to the | |
1686 | * COW device. This ensures that we do not starve. Moreover, as long | |
1687 | * as the pending exception exists, neither origin writes nor snapshot | |
1688 | * merging can overwrite the chunk in origin. | |
9d493fa8 | 1689 | */ |
3510cb94 | 1690 | dm_insert_exception(&s->complete, e); |
3f1637f2 | 1691 | up_read(&s->lock); |
76df1c65 | 1692 | |
65fc7c37 NT |
1693 | /* Wait for conflicting reads to drain */ |
1694 | if (__chunk_is_tracked(s, pe->e.old_chunk)) { | |
f79ae415 | 1695 | dm_exception_table_unlock(&lock); |
65fc7c37 | 1696 | __check_for_conflicting_io(s, pe->e.old_chunk); |
f79ae415 | 1697 | dm_exception_table_lock(&lock); |
65fc7c37 NT |
1698 | } |
1699 | ||
a2d2b034 | 1700 | out: |
65fc7c37 | 1701 | /* Remove the in-flight exception from the list */ |
3510cb94 | 1702 | dm_remove_exception(&pe->e); |
f79ae415 NT |
1703 | |
1704 | dm_exception_table_unlock(&lock); | |
1705 | ||
9d493fa8 | 1706 | snapshot_bios = bio_list_get(&pe->snapshot_bios); |
515ad66c | 1707 | origin_bios = bio_list_get(&pe->origin_bios); |
a6e50b40 | 1708 | full_bio = pe->full_bio; |
fe3265b1 | 1709 | if (full_bio) |
a6e50b40 | 1710 | full_bio->bi_end_io = pe->full_bio_end_io; |
73dfd078 MP |
1711 | increment_pending_exceptions_done_count(); |
1712 | ||
9d493fa8 | 1713 | /* Submit any pending write bios */ |
a6e50b40 MP |
1714 | if (error) { |
1715 | if (full_bio) | |
1716 | bio_io_error(full_bio); | |
9d493fa8 | 1717 | error_bios(snapshot_bios); |
a6e50b40 MP |
1718 | } else { |
1719 | if (full_bio) | |
4246a0b6 | 1720 | bio_endio(full_bio); |
9d493fa8 | 1721 | flush_bios(snapshot_bios); |
a6e50b40 | 1722 | } |
9d493fa8 | 1723 | |
515ad66c | 1724 | retry_origin_bios(s, origin_bios); |
22aa66a3 MP |
1725 | |
1726 | free_pending_exception(pe); | |
1da177e4 LT |
1727 | } |
1728 | ||
230c83af MP |
1729 | static void complete_exception(struct dm_snap_pending_exception *pe) |
1730 | { | |
1731 | struct dm_snapshot *s = pe->snap; | |
1732 | ||
385277bf MP |
1733 | /* Update the metadata if we are persistent */ |
1734 | s->store->type->commit_exception(s->store, &pe->e, !pe->copy_error, | |
1735 | pending_complete, pe); | |
230c83af MP |
1736 | } |
1737 | ||
1da177e4 LT |
1738 | /* |
1739 | * Called when the copy I/O has finished. kcopyd actually runs | |
1740 | * this code so don't block. | |
1741 | */ | |
4cdc1d1f | 1742 | static void copy_callback(int read_err, unsigned long write_err, void *context) |
1da177e4 | 1743 | { |
028867ac | 1744 | struct dm_snap_pending_exception *pe = context; |
1da177e4 LT |
1745 | struct dm_snapshot *s = pe->snap; |
1746 | ||
230c83af | 1747 | pe->copy_error = read_err || write_err; |
1da177e4 | 1748 | |
230c83af | 1749 | if (pe->exception_sequence == s->exception_complete_sequence) { |
3db2776d DJ |
1750 | struct rb_node *next; |
1751 | ||
230c83af MP |
1752 | s->exception_complete_sequence++; |
1753 | complete_exception(pe); | |
1754 | ||
3db2776d DJ |
1755 | next = rb_first(&s->out_of_order_tree); |
1756 | while (next) { | |
1757 | pe = rb_entry(next, struct dm_snap_pending_exception, | |
1758 | out_of_order_node); | |
230c83af MP |
1759 | if (pe->exception_sequence != s->exception_complete_sequence) |
1760 | break; | |
3db2776d | 1761 | next = rb_next(next); |
230c83af | 1762 | s->exception_complete_sequence++; |
3db2776d | 1763 | rb_erase(&pe->out_of_order_node, &s->out_of_order_tree); |
230c83af | 1764 | complete_exception(pe); |
3db2776d | 1765 | cond_resched(); |
230c83af MP |
1766 | } |
1767 | } else { | |
3db2776d DJ |
1768 | struct rb_node *parent = NULL; |
1769 | struct rb_node **p = &s->out_of_order_tree.rb_node; | |
230c83af MP |
1770 | struct dm_snap_pending_exception *pe2; |
1771 | ||
3db2776d DJ |
1772 | while (*p) { |
1773 | pe2 = rb_entry(*p, struct dm_snap_pending_exception, out_of_order_node); | |
1774 | parent = *p; | |
1775 | ||
1776 | BUG_ON(pe->exception_sequence == pe2->exception_sequence); | |
1777 | if (pe->exception_sequence < pe2->exception_sequence) | |
1778 | p = &((*p)->rb_left); | |
1779 | else | |
1780 | p = &((*p)->rb_right); | |
230c83af | 1781 | } |
3db2776d DJ |
1782 | |
1783 | rb_link_node(&pe->out_of_order_node, parent, p); | |
1784 | rb_insert_color(&pe->out_of_order_node, &s->out_of_order_tree); | |
230c83af | 1785 | } |
a2f83e8b | 1786 | account_end_copy(s); |
1da177e4 LT |
1787 | } |
1788 | ||
1789 | /* | |
1790 | * Dispatches the copy operation to kcopyd. | |
1791 | */ | |
028867ac | 1792 | static void start_copy(struct dm_snap_pending_exception *pe) |
1da177e4 LT |
1793 | { |
1794 | struct dm_snapshot *s = pe->snap; | |
22a1ceb1 | 1795 | struct dm_io_region src, dest; |
1da177e4 LT |
1796 | struct block_device *bdev = s->origin->bdev; |
1797 | sector_t dev_size; | |
1798 | ||
1799 | dev_size = get_dev_size(bdev); | |
1800 | ||
1801 | src.bdev = bdev; | |
71fab00a | 1802 | src.sector = chunk_to_sector(s->store, pe->e.old_chunk); |
df96eee6 | 1803 | src.count = min((sector_t)s->store->chunk_size, dev_size - src.sector); |
1da177e4 | 1804 | |
fc56f6fb | 1805 | dest.bdev = s->cow->bdev; |
71fab00a | 1806 | dest.sector = chunk_to_sector(s->store, pe->e.new_chunk); |
1da177e4 LT |
1807 | dest.count = src.count; |
1808 | ||
1809 | /* Hand over to kcopyd */ | |
a2f83e8b | 1810 | account_start_copy(s); |
a2d2b034 | 1811 | dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, copy_callback, pe); |
1da177e4 LT |
1812 | } |
1813 | ||
4246a0b6 | 1814 | static void full_bio_end_io(struct bio *bio) |
a6e50b40 MP |
1815 | { |
1816 | void *callback_data = bio->bi_private; | |
1817 | ||
4e4cbee9 | 1818 | dm_kcopyd_do_callback(callback_data, 0, bio->bi_status ? 1 : 0); |
a6e50b40 MP |
1819 | } |
1820 | ||
1821 | static void start_full_bio(struct dm_snap_pending_exception *pe, | |
1822 | struct bio *bio) | |
1823 | { | |
1824 | struct dm_snapshot *s = pe->snap; | |
1825 | void *callback_data; | |
1826 | ||
1827 | pe->full_bio = bio; | |
1828 | pe->full_bio_end_io = bio->bi_end_io; | |
a6e50b40 | 1829 | |
a2f83e8b | 1830 | account_start_copy(s); |
a6e50b40 MP |
1831 | callback_data = dm_kcopyd_prepare_callback(s->kcopyd_client, |
1832 | copy_callback, pe); | |
1833 | ||
1834 | bio->bi_end_io = full_bio_end_io; | |
1835 | bio->bi_private = callback_data; | |
1836 | ||
ed00aabd | 1837 | submit_bio_noacct(bio); |
a6e50b40 MP |
1838 | } |
1839 | ||
2913808e MP |
1840 | static struct dm_snap_pending_exception * |
1841 | __lookup_pending_exception(struct dm_snapshot *s, chunk_t chunk) | |
1842 | { | |
3510cb94 | 1843 | struct dm_exception *e = dm_lookup_exception(&s->pending, chunk); |
2913808e MP |
1844 | |
1845 | if (!e) | |
1846 | return NULL; | |
1847 | ||
1848 | return container_of(e, struct dm_snap_pending_exception, e); | |
1849 | } | |
1850 | ||
1da177e4 | 1851 | /* |
65fc7c37 | 1852 | * Inserts a pending exception into the pending table. |
1da177e4 | 1853 | * |
3f1637f2 NT |
1854 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1855 | * before calling this. | |
1da177e4 | 1856 | */ |
028867ac | 1857 | static struct dm_snap_pending_exception * |
65fc7c37 NT |
1858 | __insert_pending_exception(struct dm_snapshot *s, |
1859 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1da177e4 | 1860 | { |
76df1c65 AK |
1861 | pe->e.old_chunk = chunk; |
1862 | bio_list_init(&pe->origin_bios); | |
1863 | bio_list_init(&pe->snapshot_bios); | |
76df1c65 | 1864 | pe->started = 0; |
a6e50b40 | 1865 | pe->full_bio = NULL; |
76df1c65 | 1866 | |
3f1637f2 | 1867 | spin_lock(&s->pe_allocation_lock); |
493df71c | 1868 | if (s->store->type->prepare_exception(s->store, &pe->e)) { |
3f1637f2 | 1869 | spin_unlock(&s->pe_allocation_lock); |
76df1c65 AK |
1870 | free_pending_exception(pe); |
1871 | return NULL; | |
1872 | } | |
1873 | ||
230c83af | 1874 | pe->exception_sequence = s->exception_start_sequence++; |
3f1637f2 | 1875 | spin_unlock(&s->pe_allocation_lock); |
230c83af | 1876 | |
3510cb94 | 1877 | dm_insert_exception(&s->pending, &pe->e); |
76df1c65 | 1878 | |
1da177e4 LT |
1879 | return pe; |
1880 | } | |
1881 | ||
65fc7c37 NT |
1882 | /* |
1883 | * Looks to see if this snapshot already has a pending exception | |
1884 | * for this chunk, otherwise it allocates a new one and inserts | |
1885 | * it into the pending table. | |
1886 | * | |
3f1637f2 NT |
1887 | * NOTE: a write lock must be held on the chunk's pending exception table slot |
1888 | * before calling this. | |
65fc7c37 NT |
1889 | */ |
1890 | static struct dm_snap_pending_exception * | |
1891 | __find_pending_exception(struct dm_snapshot *s, | |
1892 | struct dm_snap_pending_exception *pe, chunk_t chunk) | |
1893 | { | |
1894 | struct dm_snap_pending_exception *pe2; | |
1895 | ||
1896 | pe2 = __lookup_pending_exception(s, chunk); | |
1897 | if (pe2) { | |
1898 | free_pending_exception(pe); | |
1899 | return pe2; | |
1900 | } | |
1901 | ||
1902 | return __insert_pending_exception(s, pe, chunk); | |
1903 | } | |
1904 | ||
1d4989c8 | 1905 | static void remap_exception(struct dm_snapshot *s, struct dm_exception *e, |
d74f81f8 | 1906 | struct bio *bio, chunk_t chunk) |
1da177e4 | 1907 | { |
74d46992 | 1908 | bio_set_dev(bio, s->cow->bdev); |
4f024f37 KO |
1909 | bio->bi_iter.bi_sector = |
1910 | chunk_to_sector(s->store, dm_chunk_number(e->new_chunk) + | |
1911 | (chunk - e->old_chunk)) + | |
1912 | (bio->bi_iter.bi_sector & s->store->chunk_mask); | |
1da177e4 LT |
1913 | } |
1914 | ||
2e602385 MS |
1915 | static void zero_callback(int read_err, unsigned long write_err, void *context) |
1916 | { | |
1917 | struct bio *bio = context; | |
1918 | struct dm_snapshot *s = bio->bi_private; | |
1919 | ||
a2f83e8b | 1920 | account_end_copy(s); |
2e602385 MS |
1921 | bio->bi_status = write_err ? BLK_STS_IOERR : 0; |
1922 | bio_endio(bio); | |
1923 | } | |
1924 | ||
1925 | static void zero_exception(struct dm_snapshot *s, struct dm_exception *e, | |
1926 | struct bio *bio, chunk_t chunk) | |
1927 | { | |
1928 | struct dm_io_region dest; | |
1929 | ||
1930 | dest.bdev = s->cow->bdev; | |
1931 | dest.sector = bio->bi_iter.bi_sector; | |
1932 | dest.count = s->store->chunk_size; | |
1933 | ||
a2f83e8b | 1934 | account_start_copy(s); |
2e602385 MS |
1935 | WARN_ON_ONCE(bio->bi_private); |
1936 | bio->bi_private = s; | |
1937 | dm_kcopyd_zero(s->kcopyd_client, 1, &dest, 0, zero_callback, bio); | |
1938 | } | |
1939 | ||
1940 | static bool io_overlaps_chunk(struct dm_snapshot *s, struct bio *bio) | |
1941 | { | |
1942 | return bio->bi_iter.bi_size == | |
1943 | (s->store->chunk_size << SECTOR_SHIFT); | |
1944 | } | |
1945 | ||
7de3ee57 | 1946 | static int snapshot_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 1947 | { |
1d4989c8 | 1948 | struct dm_exception *e; |
028867ac | 1949 | struct dm_snapshot *s = ti->private; |
d2a7ad29 | 1950 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 1951 | chunk_t chunk; |
028867ac | 1952 | struct dm_snap_pending_exception *pe = NULL; |
f79ae415 | 1953 | struct dm_exception_table_lock lock; |
1da177e4 | 1954 | |
ee18026a MP |
1955 | init_tracked_chunk(bio); |
1956 | ||
1eff9d32 | 1957 | if (bio->bi_opf & REQ_PREFLUSH) { |
74d46992 | 1958 | bio_set_dev(bio, s->cow->bdev); |
494b3ee7 MP |
1959 | return DM_MAPIO_REMAPPED; |
1960 | } | |
1961 | ||
4f024f37 | 1962 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
f79ae415 | 1963 | dm_exception_table_lock_init(s, chunk, &lock); |
1da177e4 LT |
1964 | |
1965 | /* Full snapshots are not usable */ | |
76df1c65 | 1966 | /* To get here the table must be live so s->active is always set. */ |
1da177e4 | 1967 | if (!s->valid) |
846785e6 | 1968 | return DM_MAPIO_KILL; |
1da177e4 | 1969 | |
b2155578 MP |
1970 | if (bio_data_dir(bio) == WRITE) { |
1971 | while (unlikely(!wait_for_in_progress(s, false))) | |
1972 | ; /* wait_for_in_progress() has slept */ | |
1973 | } | |
1974 | ||
3f1637f2 | 1975 | down_read(&s->lock); |
f79ae415 | 1976 | dm_exception_table_lock(&lock); |
ba40a2aa | 1977 | |
70246286 CH |
1978 | if (!s->valid || (unlikely(s->snapshot_overflowed) && |
1979 | bio_data_dir(bio) == WRITE)) { | |
846785e6 | 1980 | r = DM_MAPIO_KILL; |
ba40a2aa AK |
1981 | goto out_unlock; |
1982 | } | |
1983 | ||
2e602385 MS |
1984 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
1985 | if (s->discard_passdown_origin && dm_bio_get_target_bio_nr(bio)) { | |
1986 | /* | |
1987 | * passdown discard to origin (without triggering | |
1988 | * snapshot exceptions via do_origin; doing so would | |
1989 | * defeat the goal of freeing space in origin that is | |
1990 | * implied by the "discard_passdown_origin" feature) | |
1991 | */ | |
1992 | bio_set_dev(bio, s->origin->bdev); | |
1993 | track_chunk(s, bio, chunk); | |
1994 | goto out_unlock; | |
1995 | } | |
1996 | /* discard to snapshot (target_bio_nr == 0) zeroes exceptions */ | |
1997 | } | |
1998 | ||
ba40a2aa | 1999 | /* If the block is already remapped - use that, else remap it */ |
3510cb94 | 2000 | e = dm_lookup_exception(&s->complete, chunk); |
ba40a2aa | 2001 | if (e) { |
d74f81f8 | 2002 | remap_exception(s, e, bio, chunk); |
2e602385 MS |
2003 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD) && |
2004 | io_overlaps_chunk(s, bio)) { | |
2005 | dm_exception_table_unlock(&lock); | |
2006 | up_read(&s->lock); | |
2007 | zero_exception(s, e, bio, chunk); | |
2008 | r = DM_MAPIO_SUBMITTED; /* discard is not issued */ | |
2009 | goto out; | |
2010 | } | |
2011 | goto out_unlock; | |
2012 | } | |
2013 | ||
2014 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { | |
2015 | /* | |
2016 | * If no exception exists, complete discard immediately | |
2017 | * otherwise it'll trigger copy-out. | |
2018 | */ | |
2019 | bio_endio(bio); | |
2020 | r = DM_MAPIO_SUBMITTED; | |
ba40a2aa AK |
2021 | goto out_unlock; |
2022 | } | |
2023 | ||
1da177e4 LT |
2024 | /* |
2025 | * Write to snapshot - higher level takes care of RW/RO | |
2026 | * flags so we should only get this if we are | |
962c6296 | 2027 | * writable. |
1da177e4 | 2028 | */ |
70246286 | 2029 | if (bio_data_dir(bio) == WRITE) { |
2913808e | 2030 | pe = __lookup_pending_exception(s, chunk); |
76df1c65 | 2031 | if (!pe) { |
f79ae415 | 2032 | dm_exception_table_unlock(&lock); |
c6621392 | 2033 | pe = alloc_pending_exception(s); |
f79ae415 | 2034 | dm_exception_table_lock(&lock); |
c6621392 | 2035 | |
3510cb94 | 2036 | e = dm_lookup_exception(&s->complete, chunk); |
35bf659b MP |
2037 | if (e) { |
2038 | free_pending_exception(pe); | |
2039 | remap_exception(s, e, bio, chunk); | |
2040 | goto out_unlock; | |
2041 | } | |
2042 | ||
c6621392 | 2043 | pe = __find_pending_exception(s, pe, chunk); |
2913808e | 2044 | if (!pe) { |
f79ae415 | 2045 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2046 | up_read(&s->lock); |
2047 | ||
2048 | down_write(&s->lock); | |
f79ae415 | 2049 | |
b0d3cc01 | 2050 | if (s->store->userspace_supports_overflow) { |
3f1637f2 NT |
2051 | if (s->valid && !s->snapshot_overflowed) { |
2052 | s->snapshot_overflowed = 1; | |
2053 | DMERR("Snapshot overflowed: Unable to allocate exception."); | |
2054 | } | |
b0d3cc01 MS |
2055 | } else |
2056 | __invalidate_snapshot(s, -ENOMEM); | |
f79ae415 NT |
2057 | up_write(&s->lock); |
2058 | ||
846785e6 | 2059 | r = DM_MAPIO_KILL; |
f79ae415 | 2060 | goto out; |
2913808e | 2061 | } |
1da177e4 LT |
2062 | } |
2063 | ||
d74f81f8 | 2064 | remap_exception(s, &pe->e, bio, chunk); |
76df1c65 | 2065 | |
d2a7ad29 | 2066 | r = DM_MAPIO_SUBMITTED; |
ba40a2aa | 2067 | |
2e602385 | 2068 | if (!pe->started && io_overlaps_chunk(s, bio)) { |
a6e50b40 | 2069 | pe->started = 1; |
3f1637f2 | 2070 | |
f79ae415 | 2071 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2072 | up_read(&s->lock); |
2073 | ||
a6e50b40 MP |
2074 | start_full_bio(pe, bio); |
2075 | goto out; | |
2076 | } | |
2077 | ||
2078 | bio_list_add(&pe->snapshot_bios, bio); | |
2079 | ||
76df1c65 | 2080 | if (!pe->started) { |
3f1637f2 | 2081 | /* this is protected by the exception table lock */ |
76df1c65 | 2082 | pe->started = 1; |
3f1637f2 | 2083 | |
f79ae415 | 2084 | dm_exception_table_unlock(&lock); |
3f1637f2 NT |
2085 | up_read(&s->lock); |
2086 | ||
76df1c65 | 2087 | start_copy(pe); |
ba40a2aa AK |
2088 | goto out; |
2089 | } | |
cd45daff | 2090 | } else { |
74d46992 | 2091 | bio_set_dev(bio, s->origin->bdev); |
ee18026a | 2092 | track_chunk(s, bio, chunk); |
cd45daff | 2093 | } |
1da177e4 | 2094 | |
a2d2b034 | 2095 | out_unlock: |
f79ae415 | 2096 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2097 | up_read(&s->lock); |
a2d2b034 | 2098 | out: |
1da177e4 LT |
2099 | return r; |
2100 | } | |
2101 | ||
3452c2a1 MP |
2102 | /* |
2103 | * A snapshot-merge target behaves like a combination of a snapshot | |
2104 | * target and a snapshot-origin target. It only generates new | |
2105 | * exceptions in other snapshots and not in the one that is being | |
2106 | * merged. | |
2107 | * | |
2108 | * For each chunk, if there is an existing exception, it is used to | |
2109 | * redirect I/O to the cow device. Otherwise I/O is sent to the origin, | |
2110 | * which in turn might generate exceptions in other snapshots. | |
9fe86254 MP |
2111 | * If merging is currently taking place on the chunk in question, the |
2112 | * I/O is deferred by adding it to s->bios_queued_during_merge. | |
3452c2a1 | 2113 | */ |
7de3ee57 | 2114 | static int snapshot_merge_map(struct dm_target *ti, struct bio *bio) |
3452c2a1 MP |
2115 | { |
2116 | struct dm_exception *e; | |
2117 | struct dm_snapshot *s = ti->private; | |
2118 | int r = DM_MAPIO_REMAPPED; | |
2119 | chunk_t chunk; | |
2120 | ||
ee18026a MP |
2121 | init_tracked_chunk(bio); |
2122 | ||
1eff9d32 | 2123 | if (bio->bi_opf & REQ_PREFLUSH) { |
55a62eef | 2124 | if (!dm_bio_get_target_bio_nr(bio)) |
74d46992 | 2125 | bio_set_dev(bio, s->origin->bdev); |
10b8106a | 2126 | else |
74d46992 | 2127 | bio_set_dev(bio, s->cow->bdev); |
10b8106a MS |
2128 | return DM_MAPIO_REMAPPED; |
2129 | } | |
2130 | ||
3ee25485 MS |
2131 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
2132 | /* Once merging, discards no longer effect change */ | |
2133 | bio_endio(bio); | |
2134 | return DM_MAPIO_SUBMITTED; | |
2135 | } | |
2136 | ||
4f024f37 | 2137 | chunk = sector_to_chunk(s->store, bio->bi_iter.bi_sector); |
3452c2a1 | 2138 | |
4ad8d880 | 2139 | down_write(&s->lock); |
3452c2a1 | 2140 | |
d2fdb776 MP |
2141 | /* Full merging snapshots are redirected to the origin */ |
2142 | if (!s->valid) | |
2143 | goto redirect_to_origin; | |
3452c2a1 MP |
2144 | |
2145 | /* If the block is already remapped - use that */ | |
2146 | e = dm_lookup_exception(&s->complete, chunk); | |
2147 | if (e) { | |
9fe86254 | 2148 | /* Queue writes overlapping with chunks being merged */ |
70246286 | 2149 | if (bio_data_dir(bio) == WRITE && |
9fe86254 MP |
2150 | chunk >= s->first_merging_chunk && |
2151 | chunk < (s->first_merging_chunk + | |
2152 | s->num_merging_chunks)) { | |
74d46992 | 2153 | bio_set_dev(bio, s->origin->bdev); |
9fe86254 MP |
2154 | bio_list_add(&s->bios_queued_during_merge, bio); |
2155 | r = DM_MAPIO_SUBMITTED; | |
2156 | goto out_unlock; | |
2157 | } | |
17aa0332 | 2158 | |
3452c2a1 | 2159 | remap_exception(s, e, bio, chunk); |
17aa0332 | 2160 | |
70246286 | 2161 | if (bio_data_dir(bio) == WRITE) |
ee18026a | 2162 | track_chunk(s, bio, chunk); |
3452c2a1 MP |
2163 | goto out_unlock; |
2164 | } | |
2165 | ||
d2fdb776 | 2166 | redirect_to_origin: |
74d46992 | 2167 | bio_set_dev(bio, s->origin->bdev); |
3452c2a1 | 2168 | |
70246286 | 2169 | if (bio_data_dir(bio) == WRITE) { |
4ad8d880 | 2170 | up_write(&s->lock); |
b2155578 | 2171 | return do_origin(s->origin, bio, false); |
3452c2a1 MP |
2172 | } |
2173 | ||
2174 | out_unlock: | |
4ad8d880 | 2175 | up_write(&s->lock); |
3452c2a1 MP |
2176 | |
2177 | return r; | |
2178 | } | |
2179 | ||
4e4cbee9 CH |
2180 | static int snapshot_end_io(struct dm_target *ti, struct bio *bio, |
2181 | blk_status_t *error) | |
cd45daff MP |
2182 | { |
2183 | struct dm_snapshot *s = ti->private; | |
cd45daff | 2184 | |
ee18026a MP |
2185 | if (is_bio_tracked(bio)) |
2186 | stop_tracking_chunk(s, bio); | |
cd45daff | 2187 | |
1be56909 | 2188 | return DM_ENDIO_DONE; |
cd45daff MP |
2189 | } |
2190 | ||
1e03f97e MP |
2191 | static void snapshot_merge_presuspend(struct dm_target *ti) |
2192 | { | |
2193 | struct dm_snapshot *s = ti->private; | |
2194 | ||
2195 | stop_merge(s); | |
2196 | } | |
2197 | ||
c1f0c183 MS |
2198 | static int snapshot_preresume(struct dm_target *ti) |
2199 | { | |
2200 | int r = 0; | |
2201 | struct dm_snapshot *s = ti->private; | |
2202 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
2203 | ||
2204 | down_read(&_origins_lock); | |
9d3b15c4 | 2205 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2206 | if (snap_src && snap_dest) { |
4ad8d880 | 2207 | down_read(&snap_src->lock); |
c1f0c183 | 2208 | if (s == snap_src) { |
2e84fecf | 2209 | DMERR("Unable to resume snapshot source until handover completes."); |
c1f0c183 | 2210 | r = -EINVAL; |
b83b2f29 | 2211 | } else if (!dm_suspended(snap_src->ti)) { |
2e84fecf | 2212 | DMERR("Unable to perform snapshot handover until source is suspended."); |
c1f0c183 MS |
2213 | r = -EINVAL; |
2214 | } | |
4ad8d880 | 2215 | up_read(&snap_src->lock); |
c1f0c183 MS |
2216 | } |
2217 | up_read(&_origins_lock); | |
2218 | ||
2219 | return r; | |
2220 | } | |
2221 | ||
1da177e4 LT |
2222 | static void snapshot_resume(struct dm_target *ti) |
2223 | { | |
028867ac | 2224 | struct dm_snapshot *s = ti->private; |
09ee96b2 | 2225 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL, *snap_merging = NULL; |
b735fede MP |
2226 | struct dm_origin *o; |
2227 | struct mapped_device *origin_md = NULL; | |
09ee96b2 | 2228 | bool must_restart_merging = false; |
c1f0c183 MS |
2229 | |
2230 | down_read(&_origins_lock); | |
b735fede MP |
2231 | |
2232 | o = __lookup_dm_origin(s->origin->bdev); | |
2233 | if (o) | |
2234 | origin_md = dm_table_get_md(o->ti->table); | |
09ee96b2 MP |
2235 | if (!origin_md) { |
2236 | (void) __find_snapshots_sharing_cow(s, NULL, NULL, &snap_merging); | |
2237 | if (snap_merging) | |
2238 | origin_md = dm_table_get_md(snap_merging->ti->table); | |
2239 | } | |
b735fede MP |
2240 | if (origin_md == dm_table_get_md(ti->table)) |
2241 | origin_md = NULL; | |
09ee96b2 MP |
2242 | if (origin_md) { |
2243 | if (dm_hold(origin_md)) | |
2244 | origin_md = NULL; | |
2245 | } | |
b735fede | 2246 | |
09ee96b2 MP |
2247 | up_read(&_origins_lock); |
2248 | ||
2249 | if (origin_md) { | |
b735fede | 2250 | dm_internal_suspend_fast(origin_md); |
09ee96b2 MP |
2251 | if (snap_merging && test_bit(RUNNING_MERGE, &snap_merging->state_bits)) { |
2252 | must_restart_merging = true; | |
2253 | stop_merge(snap_merging); | |
2254 | } | |
2255 | } | |
2256 | ||
2257 | down_read(&_origins_lock); | |
b735fede | 2258 | |
9d3b15c4 | 2259 | (void) __find_snapshots_sharing_cow(s, &snap_src, &snap_dest, NULL); |
c1f0c183 | 2260 | if (snap_src && snap_dest) { |
4ad8d880 NT |
2261 | down_write(&snap_src->lock); |
2262 | down_write_nested(&snap_dest->lock, SINGLE_DEPTH_NESTING); | |
c1f0c183 | 2263 | __handover_exceptions(snap_src, snap_dest); |
4ad8d880 NT |
2264 | up_write(&snap_dest->lock); |
2265 | up_write(&snap_src->lock); | |
c1f0c183 | 2266 | } |
b735fede | 2267 | |
c1f0c183 MS |
2268 | up_read(&_origins_lock); |
2269 | ||
09ee96b2 MP |
2270 | if (origin_md) { |
2271 | if (must_restart_merging) | |
2272 | start_merge(snap_merging); | |
2273 | dm_internal_resume_fast(origin_md); | |
2274 | dm_put(origin_md); | |
2275 | } | |
2276 | ||
c1f0c183 MS |
2277 | /* Now we have correct chunk size, reregister */ |
2278 | reregister_snapshot(s); | |
1da177e4 | 2279 | |
4ad8d880 | 2280 | down_write(&s->lock); |
aa14edeb | 2281 | s->active = 1; |
4ad8d880 | 2282 | up_write(&s->lock); |
1da177e4 LT |
2283 | } |
2284 | ||
542f9038 | 2285 | static uint32_t get_origin_minimum_chunksize(struct block_device *bdev) |
1e03f97e | 2286 | { |
542f9038 | 2287 | uint32_t min_chunksize; |
1e03f97e MP |
2288 | |
2289 | down_read(&_origins_lock); | |
2290 | min_chunksize = __minimum_chunk_size(__lookup_origin(bdev)); | |
2291 | up_read(&_origins_lock); | |
2292 | ||
2293 | return min_chunksize; | |
2294 | } | |
2295 | ||
2296 | static void snapshot_merge_resume(struct dm_target *ti) | |
2297 | { | |
2298 | struct dm_snapshot *s = ti->private; | |
2299 | ||
2300 | /* | |
2301 | * Handover exceptions from existing snapshot. | |
2302 | */ | |
2303 | snapshot_resume(ti); | |
2304 | ||
2305 | /* | |
542f9038 | 2306 | * snapshot-merge acts as an origin, so set ti->max_io_len |
1e03f97e | 2307 | */ |
542f9038 | 2308 | ti->max_io_len = get_origin_minimum_chunksize(s->origin->bdev); |
1e03f97e MP |
2309 | |
2310 | start_merge(s); | |
2311 | } | |
2312 | ||
fd7c092e | 2313 | static void snapshot_status(struct dm_target *ti, status_type_t type, |
86a3238c | 2314 | unsigned int status_flags, char *result, unsigned int maxlen) |
1da177e4 | 2315 | { |
86a3238c | 2316 | unsigned int sz = 0; |
028867ac | 2317 | struct dm_snapshot *snap = ti->private; |
86a3238c | 2318 | unsigned int num_features; |
1da177e4 LT |
2319 | |
2320 | switch (type) { | |
2321 | case STATUSTYPE_INFO: | |
94e76572 | 2322 | |
4ad8d880 | 2323 | down_write(&snap->lock); |
94e76572 | 2324 | |
1da177e4 | 2325 | if (!snap->valid) |
2e4a31df | 2326 | DMEMIT("Invalid"); |
d8ddb1cf MS |
2327 | else if (snap->merge_failed) |
2328 | DMEMIT("Merge failed"); | |
76c44f6d MP |
2329 | else if (snap->snapshot_overflowed) |
2330 | DMEMIT("Overflow"); | |
1da177e4 | 2331 | else { |
985903bb MS |
2332 | if (snap->store->type->usage) { |
2333 | sector_t total_sectors, sectors_allocated, | |
2334 | metadata_sectors; | |
2335 | snap->store->type->usage(snap->store, | |
2336 | &total_sectors, | |
2337 | §ors_allocated, | |
2338 | &metadata_sectors); | |
2339 | DMEMIT("%llu/%llu %llu", | |
2340 | (unsigned long long)sectors_allocated, | |
2341 | (unsigned long long)total_sectors, | |
2342 | (unsigned long long)metadata_sectors); | |
03b18887 | 2343 | } else |
2e4a31df | 2344 | DMEMIT("Unknown"); |
1da177e4 | 2345 | } |
94e76572 | 2346 | |
4ad8d880 | 2347 | up_write(&snap->lock); |
94e76572 | 2348 | |
1da177e4 LT |
2349 | break; |
2350 | ||
2351 | case STATUSTYPE_TABLE: | |
2352 | /* | |
2353 | * kdevname returns a static pointer so we need | |
2354 | * to make private copies if the output is to | |
2355 | * make sense. | |
2356 | */ | |
fc56f6fb | 2357 | DMEMIT("%s %s", snap->origin->name, snap->cow->name); |
2e602385 MS |
2358 | sz += snap->store->type->status(snap->store, type, result + sz, |
2359 | maxlen - sz); | |
2360 | num_features = snap->discard_zeroes_cow + snap->discard_passdown_origin; | |
2361 | if (num_features) { | |
2362 | DMEMIT(" %u", num_features); | |
2363 | if (snap->discard_zeroes_cow) | |
2364 | DMEMIT(" discard_zeroes_cow"); | |
2365 | if (snap->discard_passdown_origin) | |
2366 | DMEMIT(" discard_passdown_origin"); | |
2367 | } | |
1da177e4 | 2368 | break; |
8ec45662 TS |
2369 | |
2370 | case STATUSTYPE_IMA: | |
2371 | DMEMIT_TARGET_NAME_VERSION(ti->type); | |
2372 | DMEMIT(",snap_origin_name=%s", snap->origin->name); | |
2373 | DMEMIT(",snap_cow_name=%s", snap->cow->name); | |
2374 | DMEMIT(",snap_valid=%c", snap->valid ? 'y' : 'n'); | |
2375 | DMEMIT(",snap_merge_failed=%c", snap->merge_failed ? 'y' : 'n'); | |
2376 | DMEMIT(",snapshot_overflowed=%c", snap->snapshot_overflowed ? 'y' : 'n'); | |
2377 | DMEMIT(";"); | |
2378 | break; | |
1da177e4 | 2379 | } |
1da177e4 LT |
2380 | } |
2381 | ||
8811f46c MS |
2382 | static int snapshot_iterate_devices(struct dm_target *ti, |
2383 | iterate_devices_callout_fn fn, void *data) | |
2384 | { | |
2385 | struct dm_snapshot *snap = ti->private; | |
1e5554c8 MP |
2386 | int r; |
2387 | ||
2388 | r = fn(ti, snap->origin, 0, ti->len, data); | |
8811f46c | 2389 | |
1e5554c8 MP |
2390 | if (!r) |
2391 | r = fn(ti, snap->cow, 0, get_dev_size(snap->cow->bdev), data); | |
2392 | ||
2393 | return r; | |
8811f46c MS |
2394 | } |
2395 | ||
2e602385 MS |
2396 | static void snapshot_io_hints(struct dm_target *ti, struct queue_limits *limits) |
2397 | { | |
2398 | struct dm_snapshot *snap = ti->private; | |
2399 | ||
2400 | if (snap->discard_zeroes_cow) { | |
2401 | struct dm_snapshot *snap_src = NULL, *snap_dest = NULL; | |
2402 | ||
3ee25485 MS |
2403 | down_read(&_origins_lock); |
2404 | ||
2e602385 MS |
2405 | (void) __find_snapshots_sharing_cow(snap, &snap_src, &snap_dest, NULL); |
2406 | if (snap_src && snap_dest) | |
2407 | snap = snap_src; | |
2408 | ||
2409 | /* All discards are split on chunk_size boundary */ | |
2410 | limits->discard_granularity = snap->store->chunk_size; | |
2411 | limits->max_discard_sectors = snap->store->chunk_size; | |
3ee25485 MS |
2412 | |
2413 | up_read(&_origins_lock); | |
2e602385 MS |
2414 | } |
2415 | } | |
8811f46c | 2416 | |
a4a82ce3 HM |
2417 | /* |
2418 | *--------------------------------------------------------------- | |
1da177e4 | 2419 | * Origin methods |
a4a82ce3 HM |
2420 | *--------------------------------------------------------------- |
2421 | */ | |
9eaae8ff MP |
2422 | /* |
2423 | * If no exceptions need creating, DM_MAPIO_REMAPPED is returned and any | |
2424 | * supplied bio was ignored. The caller may submit it immediately. | |
2425 | * (No remapping actually occurs as the origin is always a direct linear | |
2426 | * map.) | |
2427 | * | |
2428 | * If further exceptions are required, DM_MAPIO_SUBMITTED is returned | |
2429 | * and any supplied bio is added to a list to be submitted once all | |
2430 | * the necessary exceptions exist. | |
2431 | */ | |
2432 | static int __origin_write(struct list_head *snapshots, sector_t sector, | |
2433 | struct bio *bio) | |
1da177e4 | 2434 | { |
515ad66c | 2435 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2436 | struct dm_snapshot *snap; |
1d4989c8 | 2437 | struct dm_exception *e; |
65fc7c37 | 2438 | struct dm_snap_pending_exception *pe, *pe2; |
515ad66c MP |
2439 | struct dm_snap_pending_exception *pe_to_start_now = NULL; |
2440 | struct dm_snap_pending_exception *pe_to_start_last = NULL; | |
f79ae415 | 2441 | struct dm_exception_table_lock lock; |
1da177e4 LT |
2442 | chunk_t chunk; |
2443 | ||
2444 | /* Do all the snapshots on this origin */ | |
43be9c74 | 2445 | list_for_each_entry(snap, snapshots, list) { |
3452c2a1 MP |
2446 | /* |
2447 | * Don't make new exceptions in a merging snapshot | |
2448 | * because it has effectively been deleted | |
2449 | */ | |
2450 | if (dm_target_is_snapshot_merge(snap->ti)) | |
2451 | continue; | |
2452 | ||
d5e404c1 | 2453 | /* Nothing to do if writing beyond end of snapshot */ |
9eaae8ff | 2454 | if (sector >= dm_table_get_size(snap->ti->table)) |
f79ae415 | 2455 | continue; |
1da177e4 LT |
2456 | |
2457 | /* | |
2458 | * Remember, different snapshots can have | |
2459 | * different chunk sizes. | |
2460 | */ | |
9eaae8ff | 2461 | chunk = sector_to_chunk(snap->store, sector); |
f79ae415 NT |
2462 | dm_exception_table_lock_init(snap, chunk, &lock); |
2463 | ||
3f1637f2 | 2464 | down_read(&snap->lock); |
f79ae415 NT |
2465 | dm_exception_table_lock(&lock); |
2466 | ||
2467 | /* Only deal with valid and active snapshots */ | |
2468 | if (!snap->valid || !snap->active) | |
2469 | goto next_snapshot; | |
1da177e4 | 2470 | |
2913808e | 2471 | pe = __lookup_pending_exception(snap, chunk); |
76df1c65 | 2472 | if (!pe) { |
65fc7c37 NT |
2473 | /* |
2474 | * Check exception table to see if block is already | |
2475 | * remapped in this snapshot and trigger an exception | |
2476 | * if not. | |
2477 | */ | |
2478 | e = dm_lookup_exception(&snap->complete, chunk); | |
2479 | if (e) | |
2480 | goto next_snapshot; | |
2481 | ||
f79ae415 | 2482 | dm_exception_table_unlock(&lock); |
c6621392 | 2483 | pe = alloc_pending_exception(snap); |
f79ae415 | 2484 | dm_exception_table_lock(&lock); |
c6621392 | 2485 | |
65fc7c37 NT |
2486 | pe2 = __lookup_pending_exception(snap, chunk); |
2487 | ||
2488 | if (!pe2) { | |
2489 | e = dm_lookup_exception(&snap->complete, chunk); | |
2490 | if (e) { | |
2491 | free_pending_exception(pe); | |
2492 | goto next_snapshot; | |
2493 | } | |
2494 | ||
2495 | pe = __insert_pending_exception(snap, pe, chunk); | |
2496 | if (!pe) { | |
f79ae415 | 2497 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2498 | up_read(&snap->lock); |
f79ae415 | 2499 | |
3f1637f2 | 2500 | invalidate_snapshot(snap, -ENOMEM); |
f79ae415 | 2501 | continue; |
65fc7c37 NT |
2502 | } |
2503 | } else { | |
35bf659b | 2504 | free_pending_exception(pe); |
65fc7c37 | 2505 | pe = pe2; |
2913808e | 2506 | } |
76df1c65 AK |
2507 | } |
2508 | ||
515ad66c | 2509 | r = DM_MAPIO_SUBMITTED; |
76df1c65 | 2510 | |
515ad66c MP |
2511 | /* |
2512 | * If an origin bio was supplied, queue it to wait for the | |
2513 | * completion of this exception, and start this one last, | |
2514 | * at the end of the function. | |
2515 | */ | |
2516 | if (bio) { | |
2517 | bio_list_add(&pe->origin_bios, bio); | |
2518 | bio = NULL; | |
76df1c65 | 2519 | |
515ad66c MP |
2520 | if (!pe->started) { |
2521 | pe->started = 1; | |
2522 | pe_to_start_last = pe; | |
2523 | } | |
76df1c65 AK |
2524 | } |
2525 | ||
2526 | if (!pe->started) { | |
2527 | pe->started = 1; | |
515ad66c | 2528 | pe_to_start_now = pe; |
1da177e4 LT |
2529 | } |
2530 | ||
a2d2b034 | 2531 | next_snapshot: |
f79ae415 | 2532 | dm_exception_table_unlock(&lock); |
3f1637f2 | 2533 | up_read(&snap->lock); |
1da177e4 | 2534 | |
515ad66c MP |
2535 | if (pe_to_start_now) { |
2536 | start_copy(pe_to_start_now); | |
2537 | pe_to_start_now = NULL; | |
2538 | } | |
b4b610f6 AK |
2539 | } |
2540 | ||
1da177e4 | 2541 | /* |
515ad66c MP |
2542 | * Submit the exception against which the bio is queued last, |
2543 | * to give the other exceptions a head start. | |
1da177e4 | 2544 | */ |
515ad66c MP |
2545 | if (pe_to_start_last) |
2546 | start_copy(pe_to_start_last); | |
1da177e4 LT |
2547 | |
2548 | return r; | |
2549 | } | |
2550 | ||
2551 | /* | |
2552 | * Called on a write from the origin driver. | |
2553 | */ | |
b2155578 | 2554 | static int do_origin(struct dm_dev *origin, struct bio *bio, bool limit) |
1da177e4 LT |
2555 | { |
2556 | struct origin *o; | |
d2a7ad29 | 2557 | int r = DM_MAPIO_REMAPPED; |
1da177e4 | 2558 | |
b2155578 | 2559 | again: |
1da177e4 LT |
2560 | down_read(&_origins_lock); |
2561 | o = __lookup_origin(origin->bdev); | |
b2155578 MP |
2562 | if (o) { |
2563 | if (limit) { | |
2564 | struct dm_snapshot *s; | |
0ef0b471 | 2565 | |
b2155578 MP |
2566 | list_for_each_entry(s, &o->snapshots, list) |
2567 | if (unlikely(!wait_for_in_progress(s, true))) | |
2568 | goto again; | |
2569 | } | |
2570 | ||
4f024f37 | 2571 | r = __origin_write(&o->snapshots, bio->bi_iter.bi_sector, bio); |
b2155578 | 2572 | } |
1da177e4 LT |
2573 | up_read(&_origins_lock); |
2574 | ||
2575 | return r; | |
2576 | } | |
2577 | ||
73dfd078 MP |
2578 | /* |
2579 | * Trigger exceptions in all non-merging snapshots. | |
2580 | * | |
2581 | * The chunk size of the merging snapshot may be larger than the chunk | |
2582 | * size of some other snapshot so we may need to reallocate multiple | |
2583 | * chunks in other snapshots. | |
2584 | * | |
2585 | * We scan all the overlapping exceptions in the other snapshots. | |
2586 | * Returns 1 if anything was reallocated and must be waited for, | |
2587 | * otherwise returns 0. | |
2588 | * | |
2589 | * size must be a multiple of merging_snap's chunk_size. | |
2590 | */ | |
2591 | static int origin_write_extent(struct dm_snapshot *merging_snap, | |
86a3238c | 2592 | sector_t sector, unsigned int size) |
73dfd078 MP |
2593 | { |
2594 | int must_wait = 0; | |
2595 | sector_t n; | |
2596 | struct origin *o; | |
2597 | ||
2598 | /* | |
542f9038 | 2599 | * The origin's __minimum_chunk_size() got stored in max_io_len |
73dfd078 MP |
2600 | * by snapshot_merge_resume(). |
2601 | */ | |
2602 | down_read(&_origins_lock); | |
2603 | o = __lookup_origin(merging_snap->origin->bdev); | |
542f9038 | 2604 | for (n = 0; n < size; n += merging_snap->ti->max_io_len) |
73dfd078 MP |
2605 | if (__origin_write(&o->snapshots, sector + n, NULL) == |
2606 | DM_MAPIO_SUBMITTED) | |
2607 | must_wait = 1; | |
2608 | up_read(&_origins_lock); | |
2609 | ||
2610 | return must_wait; | |
2611 | } | |
2612 | ||
1da177e4 LT |
2613 | /* |
2614 | * Origin: maps a linear range of a device, with hooks for snapshotting. | |
2615 | */ | |
2616 | ||
2617 | /* | |
2618 | * Construct an origin mapping: <dev_path> | |
2619 | * The context for an origin is merely a 'struct dm_dev *' | |
2620 | * pointing to the real device. | |
2621 | */ | |
2622 | static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
2623 | { | |
2624 | int r; | |
599cdf3b | 2625 | struct dm_origin *o; |
1da177e4 LT |
2626 | |
2627 | if (argc != 1) { | |
72d94861 | 2628 | ti->error = "origin: incorrect number of arguments"; |
1da177e4 LT |
2629 | return -EINVAL; |
2630 | } | |
2631 | ||
599cdf3b MP |
2632 | o = kmalloc(sizeof(struct dm_origin), GFP_KERNEL); |
2633 | if (!o) { | |
2634 | ti->error = "Cannot allocate private origin structure"; | |
2635 | r = -ENOMEM; | |
2636 | goto bad_alloc; | |
2637 | } | |
2638 | ||
2639 | r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &o->dev); | |
1da177e4 LT |
2640 | if (r) { |
2641 | ti->error = "Cannot get target device"; | |
599cdf3b | 2642 | goto bad_open; |
1da177e4 LT |
2643 | } |
2644 | ||
b735fede | 2645 | o->ti = ti; |
599cdf3b | 2646 | ti->private = o; |
55a62eef | 2647 | ti->num_flush_bios = 1; |
494b3ee7 | 2648 | |
1da177e4 | 2649 | return 0; |
599cdf3b MP |
2650 | |
2651 | bad_open: | |
2652 | kfree(o); | |
2653 | bad_alloc: | |
2654 | return r; | |
1da177e4 LT |
2655 | } |
2656 | ||
2657 | static void origin_dtr(struct dm_target *ti) | |
2658 | { | |
599cdf3b | 2659 | struct dm_origin *o = ti->private; |
b735fede | 2660 | |
599cdf3b MP |
2661 | dm_put_device(ti, o->dev); |
2662 | kfree(o); | |
1da177e4 LT |
2663 | } |
2664 | ||
7de3ee57 | 2665 | static int origin_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 2666 | { |
599cdf3b | 2667 | struct dm_origin *o = ti->private; |
86a3238c | 2668 | unsigned int available_sectors; |
1da177e4 | 2669 | |
74d46992 | 2670 | bio_set_dev(bio, o->dev->bdev); |
1da177e4 | 2671 | |
1eff9d32 | 2672 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) |
494b3ee7 MP |
2673 | return DM_MAPIO_REMAPPED; |
2674 | ||
70246286 | 2675 | if (bio_data_dir(bio) != WRITE) |
494b3ee7 MP |
2676 | return DM_MAPIO_REMAPPED; |
2677 | ||
298eaa89 | 2678 | available_sectors = o->split_boundary - |
86a3238c | 2679 | ((unsigned int)bio->bi_iter.bi_sector & (o->split_boundary - 1)); |
298eaa89 MP |
2680 | |
2681 | if (bio_sectors(bio) > available_sectors) | |
2682 | dm_accept_partial_bio(bio, available_sectors); | |
2683 | ||
1da177e4 | 2684 | /* Only tell snapshots if this is a write */ |
b2155578 | 2685 | return do_origin(o->dev, bio, true); |
1da177e4 LT |
2686 | } |
2687 | ||
1da177e4 | 2688 | /* |
542f9038 | 2689 | * Set the target "max_io_len" field to the minimum of all the snapshots' |
1da177e4 LT |
2690 | * chunk sizes. |
2691 | */ | |
2692 | static void origin_resume(struct dm_target *ti) | |
2693 | { | |
599cdf3b | 2694 | struct dm_origin *o = ti->private; |
1da177e4 | 2695 | |
298eaa89 | 2696 | o->split_boundary = get_origin_minimum_chunksize(o->dev->bdev); |
b735fede MP |
2697 | |
2698 | down_write(&_origins_lock); | |
2699 | __insert_dm_origin(o); | |
2700 | up_write(&_origins_lock); | |
2701 | } | |
2702 | ||
2703 | static void origin_postsuspend(struct dm_target *ti) | |
2704 | { | |
2705 | struct dm_origin *o = ti->private; | |
2706 | ||
2707 | down_write(&_origins_lock); | |
2708 | __remove_dm_origin(o); | |
2709 | up_write(&_origins_lock); | |
1da177e4 LT |
2710 | } |
2711 | ||
fd7c092e | 2712 | static void origin_status(struct dm_target *ti, status_type_t type, |
86a3238c | 2713 | unsigned int status_flags, char *result, unsigned int maxlen) |
1da177e4 | 2714 | { |
599cdf3b | 2715 | struct dm_origin *o = ti->private; |
1da177e4 LT |
2716 | |
2717 | switch (type) { | |
2718 | case STATUSTYPE_INFO: | |
2719 | result[0] = '\0'; | |
2720 | break; | |
2721 | ||
2722 | case STATUSTYPE_TABLE: | |
599cdf3b | 2723 | snprintf(result, maxlen, "%s", o->dev->name); |
1da177e4 | 2724 | break; |
8ec45662 TS |
2725 | case STATUSTYPE_IMA: |
2726 | result[0] = '\0'; | |
2727 | break; | |
1da177e4 | 2728 | } |
1da177e4 LT |
2729 | } |
2730 | ||
8811f46c MS |
2731 | static int origin_iterate_devices(struct dm_target *ti, |
2732 | iterate_devices_callout_fn fn, void *data) | |
2733 | { | |
599cdf3b | 2734 | struct dm_origin *o = ti->private; |
8811f46c | 2735 | |
599cdf3b | 2736 | return fn(ti, o->dev, 0, ti->len, data); |
8811f46c MS |
2737 | } |
2738 | ||
1da177e4 LT |
2739 | static struct target_type origin_target = { |
2740 | .name = "snapshot-origin", | |
b735fede | 2741 | .version = {1, 9, 0}, |
1da177e4 LT |
2742 | .module = THIS_MODULE, |
2743 | .ctr = origin_ctr, | |
2744 | .dtr = origin_dtr, | |
2745 | .map = origin_map, | |
2746 | .resume = origin_resume, | |
b735fede | 2747 | .postsuspend = origin_postsuspend, |
1da177e4 | 2748 | .status = origin_status, |
8811f46c | 2749 | .iterate_devices = origin_iterate_devices, |
1da177e4 LT |
2750 | }; |
2751 | ||
2752 | static struct target_type snapshot_target = { | |
2753 | .name = "snapshot", | |
2e602385 | 2754 | .version = {1, 16, 0}, |
1da177e4 LT |
2755 | .module = THIS_MODULE, |
2756 | .ctr = snapshot_ctr, | |
2757 | .dtr = snapshot_dtr, | |
2758 | .map = snapshot_map, | |
cd45daff | 2759 | .end_io = snapshot_end_io, |
c1f0c183 | 2760 | .preresume = snapshot_preresume, |
1da177e4 LT |
2761 | .resume = snapshot_resume, |
2762 | .status = snapshot_status, | |
8811f46c | 2763 | .iterate_devices = snapshot_iterate_devices, |
2e602385 | 2764 | .io_hints = snapshot_io_hints, |
1da177e4 LT |
2765 | }; |
2766 | ||
d698aa45 MP |
2767 | static struct target_type merge_target = { |
2768 | .name = dm_snapshot_merge_target_name, | |
2e602385 | 2769 | .version = {1, 5, 0}, |
d698aa45 MP |
2770 | .module = THIS_MODULE, |
2771 | .ctr = snapshot_ctr, | |
2772 | .dtr = snapshot_dtr, | |
3452c2a1 | 2773 | .map = snapshot_merge_map, |
d698aa45 | 2774 | .end_io = snapshot_end_io, |
1e03f97e | 2775 | .presuspend = snapshot_merge_presuspend, |
d698aa45 | 2776 | .preresume = snapshot_preresume, |
1e03f97e | 2777 | .resume = snapshot_merge_resume, |
d698aa45 MP |
2778 | .status = snapshot_status, |
2779 | .iterate_devices = snapshot_iterate_devices, | |
2e602385 | 2780 | .io_hints = snapshot_io_hints, |
d698aa45 MP |
2781 | }; |
2782 | ||
1da177e4 LT |
2783 | static int __init dm_snapshot_init(void) |
2784 | { | |
2785 | int r; | |
2786 | ||
4db6bfe0 AK |
2787 | r = dm_exception_store_init(); |
2788 | if (r) { | |
2789 | DMERR("Failed to initialize exception stores"); | |
2790 | return r; | |
2791 | } | |
2792 | ||
1da177e4 LT |
2793 | r = init_origin_hash(); |
2794 | if (r) { | |
2795 | DMERR("init_origin_hash failed."); | |
d698aa45 | 2796 | goto bad_origin_hash; |
1da177e4 LT |
2797 | } |
2798 | ||
1d4989c8 | 2799 | exception_cache = KMEM_CACHE(dm_exception, 0); |
1da177e4 LT |
2800 | if (!exception_cache) { |
2801 | DMERR("Couldn't create exception cache."); | |
2802 | r = -ENOMEM; | |
d698aa45 | 2803 | goto bad_exception_cache; |
1da177e4 LT |
2804 | } |
2805 | ||
028867ac | 2806 | pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0); |
1da177e4 LT |
2807 | if (!pending_cache) { |
2808 | DMERR("Couldn't create pending cache."); | |
2809 | r = -ENOMEM; | |
d698aa45 | 2810 | goto bad_pending_cache; |
1da177e4 LT |
2811 | } |
2812 | ||
7e6358d2 | 2813 | r = dm_register_target(&snapshot_target); |
b362c733 | 2814 | if (r < 0) |
7e6358d2 | 2815 | goto bad_register_snapshot_target; |
7e6358d2 | 2816 | |
2817 | r = dm_register_target(&origin_target); | |
b362c733 | 2818 | if (r < 0) |
7e6358d2 | 2819 | goto bad_register_origin_target; |
7e6358d2 | 2820 | |
2821 | r = dm_register_target(&merge_target); | |
b362c733 | 2822 | if (r < 0) |
7e6358d2 | 2823 | goto bad_register_merge_target; |
7e6358d2 | 2824 | |
1da177e4 LT |
2825 | return 0; |
2826 | ||
d698aa45 | 2827 | bad_register_merge_target: |
1da177e4 | 2828 | dm_unregister_target(&origin_target); |
d698aa45 | 2829 | bad_register_origin_target: |
1da177e4 | 2830 | dm_unregister_target(&snapshot_target); |
034a186d | 2831 | bad_register_snapshot_target: |
7e6358d2 | 2832 | kmem_cache_destroy(pending_cache); |
2833 | bad_pending_cache: | |
2834 | kmem_cache_destroy(exception_cache); | |
2835 | bad_exception_cache: | |
2836 | exit_origin_hash(); | |
2837 | bad_origin_hash: | |
034a186d | 2838 | dm_exception_store_exit(); |
d698aa45 | 2839 | |
1da177e4 LT |
2840 | return r; |
2841 | } | |
2842 | ||
2843 | static void __exit dm_snapshot_exit(void) | |
2844 | { | |
10d3bd09 MP |
2845 | dm_unregister_target(&snapshot_target); |
2846 | dm_unregister_target(&origin_target); | |
d698aa45 | 2847 | dm_unregister_target(&merge_target); |
1da177e4 LT |
2848 | |
2849 | exit_origin_hash(); | |
1da177e4 LT |
2850 | kmem_cache_destroy(pending_cache); |
2851 | kmem_cache_destroy(exception_cache); | |
4db6bfe0 AK |
2852 | |
2853 | dm_exception_store_exit(); | |
1da177e4 LT |
2854 | } |
2855 | ||
2856 | /* Module hooks */ | |
2857 | module_init(dm_snapshot_init); | |
2858 | module_exit(dm_snapshot_exit); | |
2859 | ||
2860 | MODULE_DESCRIPTION(DM_NAME " snapshot target"); | |
2861 | MODULE_AUTHOR("Joe Thornber"); | |
2862 | MODULE_LICENSE("GPL"); | |
23cb2109 MP |
2863 | MODULE_ALIAS("dm-snapshot-origin"); |
2864 | MODULE_ALIAS("dm-snapshot-merge"); |