OLD | NEW |
| (Empty) |
1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. | |
2 // Use of this source code is governed by a BSD-style license that can be | |
3 // found in the LICENSE file. | |
4 | |
5 #include "net/disk_cache/rankings.h" | |
6 | |
7 #include "base/metrics/histogram.h" | |
8 #include "net/disk_cache/backend_impl.h" | |
9 #include "net/disk_cache/disk_format.h" | |
10 #include "net/disk_cache/entry_impl.h" | |
11 #include "net/disk_cache/errors.h" | |
12 #include "net/disk_cache/histogram_macros.h" | |
13 #include "net/disk_cache/stress_support.h" | |
14 | |
15 using base::Time; | |
16 using base::TimeTicks; | |
17 | |
18 namespace disk_cache { | |
19 // This is used by crash_cache.exe to generate unit test files. | |
20 NET_EXPORT_PRIVATE RankCrashes g_rankings_crash = NO_CRASH; | |
21 } | |
22 | |
23 namespace { | |
24 | |
25 enum Operation { | |
26 INSERT = 1, | |
27 REMOVE | |
28 }; | |
29 | |
30 // This class provides a simple lock for the LRU list of rankings. Whenever an | |
31 // entry is to be inserted or removed from the list, a transaction object should | |
32 // be created to keep track of the operation. If the process crashes before | |
33 // finishing the operation, the transaction record (stored as part of the user | |
34 // data on the file header) can be used to finish the operation. | |
35 class Transaction { | |
36 public: | |
37 // addr is the cache addres of the node being inserted or removed. We want to | |
38 // avoid having the compiler doing optimizations on when to read or write | |
39 // from user_data because it is the basis of the crash detection. Maybe | |
40 // volatile is not enough for that, but it should be a good hint. | |
41 Transaction(volatile disk_cache::LruData* data, disk_cache::Addr addr, | |
42 Operation op, int list); | |
43 ~Transaction(); | |
44 private: | |
45 volatile disk_cache::LruData* data_; | |
46 DISALLOW_COPY_AND_ASSIGN(Transaction); | |
47 }; | |
48 | |
49 Transaction::Transaction(volatile disk_cache::LruData* data, | |
50 disk_cache::Addr addr, Operation op, int list) | |
51 : data_(data) { | |
52 DCHECK(!data_->transaction); | |
53 DCHECK(addr.is_initialized()); | |
54 data_->operation = op; | |
55 data_->operation_list = list; | |
56 data_->transaction = addr.value(); | |
57 } | |
58 | |
59 Transaction::~Transaction() { | |
60 DCHECK(data_->transaction); | |
61 data_->transaction = 0; | |
62 data_->operation = 0; | |
63 data_->operation_list = 0; | |
64 } | |
65 | |
66 // Code locations that can generate crashes. | |
67 enum CrashLocation { | |
68 ON_INSERT_1, ON_INSERT_2, ON_INSERT_3, ON_INSERT_4, ON_REMOVE_1, ON_REMOVE_2, | |
69 ON_REMOVE_3, ON_REMOVE_4, ON_REMOVE_5, ON_REMOVE_6, ON_REMOVE_7, ON_REMOVE_8 | |
70 }; | |
71 | |
72 #ifndef NDEBUG | |
73 void TerminateSelf() { | |
74 #if defined(OS_WIN) | |
75 // Windows does more work on _exit() than we would like, so we force exit. | |
76 TerminateProcess(GetCurrentProcess(), 0); | |
77 #elif defined(OS_POSIX) | |
78 // On POSIX, _exit() will terminate the process with minimal cleanup, | |
79 // and it is cleaner than killing. | |
80 _exit(0); | |
81 #endif | |
82 } | |
83 #endif // NDEBUG | |
84 | |
85 // Generates a crash on debug builds, acording to the value of g_rankings_crash. | |
86 // This used by crash_cache.exe to generate unit-test files. | |
87 void GenerateCrash(CrashLocation location) { | |
88 #ifndef NDEBUG | |
89 if (disk_cache::NO_CRASH == disk_cache::g_rankings_crash) | |
90 return; | |
91 switch (location) { | |
92 case ON_INSERT_1: | |
93 switch (disk_cache::g_rankings_crash) { | |
94 case disk_cache::INSERT_ONE_1: | |
95 case disk_cache::INSERT_LOAD_1: | |
96 TerminateSelf(); | |
97 default: | |
98 break; | |
99 } | |
100 break; | |
101 case ON_INSERT_2: | |
102 if (disk_cache::INSERT_EMPTY_1 == disk_cache::g_rankings_crash) | |
103 TerminateSelf(); | |
104 break; | |
105 case ON_INSERT_3: | |
106 switch (disk_cache::g_rankings_crash) { | |
107 case disk_cache::INSERT_EMPTY_2: | |
108 case disk_cache::INSERT_ONE_2: | |
109 case disk_cache::INSERT_LOAD_2: | |
110 TerminateSelf(); | |
111 default: | |
112 break; | |
113 } | |
114 break; | |
115 case ON_INSERT_4: | |
116 switch (disk_cache::g_rankings_crash) { | |
117 case disk_cache::INSERT_EMPTY_3: | |
118 case disk_cache::INSERT_ONE_3: | |
119 TerminateSelf(); | |
120 default: | |
121 break; | |
122 } | |
123 break; | |
124 case ON_REMOVE_1: | |
125 switch (disk_cache::g_rankings_crash) { | |
126 case disk_cache::REMOVE_ONE_1: | |
127 case disk_cache::REMOVE_HEAD_1: | |
128 case disk_cache::REMOVE_TAIL_1: | |
129 case disk_cache::REMOVE_LOAD_1: | |
130 TerminateSelf(); | |
131 default: | |
132 break; | |
133 } | |
134 break; | |
135 case ON_REMOVE_2: | |
136 if (disk_cache::REMOVE_ONE_2 == disk_cache::g_rankings_crash) | |
137 TerminateSelf(); | |
138 break; | |
139 case ON_REMOVE_3: | |
140 if (disk_cache::REMOVE_ONE_3 == disk_cache::g_rankings_crash) | |
141 TerminateSelf(); | |
142 break; | |
143 case ON_REMOVE_4: | |
144 if (disk_cache::REMOVE_HEAD_2 == disk_cache::g_rankings_crash) | |
145 TerminateSelf(); | |
146 break; | |
147 case ON_REMOVE_5: | |
148 if (disk_cache::REMOVE_TAIL_2 == disk_cache::g_rankings_crash) | |
149 TerminateSelf(); | |
150 break; | |
151 case ON_REMOVE_6: | |
152 if (disk_cache::REMOVE_TAIL_3 == disk_cache::g_rankings_crash) | |
153 TerminateSelf(); | |
154 break; | |
155 case ON_REMOVE_7: | |
156 switch (disk_cache::g_rankings_crash) { | |
157 case disk_cache::REMOVE_ONE_4: | |
158 case disk_cache::REMOVE_LOAD_2: | |
159 case disk_cache::REMOVE_HEAD_3: | |
160 TerminateSelf(); | |
161 default: | |
162 break; | |
163 } | |
164 break; | |
165 case ON_REMOVE_8: | |
166 switch (disk_cache::g_rankings_crash) { | |
167 case disk_cache::REMOVE_HEAD_4: | |
168 case disk_cache::REMOVE_LOAD_3: | |
169 TerminateSelf(); | |
170 default: | |
171 break; | |
172 } | |
173 break; | |
174 default: | |
175 NOTREACHED(); | |
176 return; | |
177 } | |
178 #endif // NDEBUG | |
179 } | |
180 | |
181 // Update the timestamp fields of |node|. | |
182 void UpdateTimes(disk_cache::CacheRankingsBlock* node, bool modified) { | |
183 base::Time now = base::Time::Now(); | |
184 node->Data()->last_used = now.ToInternalValue(); | |
185 if (modified) | |
186 node->Data()->last_modified = now.ToInternalValue(); | |
187 } | |
188 | |
189 } // namespace | |
190 | |
191 namespace disk_cache { | |
192 | |
193 Rankings::ScopedRankingsBlock::ScopedRankingsBlock() : rankings_(NULL) {} | |
194 | |
195 Rankings::ScopedRankingsBlock::ScopedRankingsBlock(Rankings* rankings) | |
196 : rankings_(rankings) {} | |
197 | |
198 Rankings::ScopedRankingsBlock::ScopedRankingsBlock( | |
199 Rankings* rankings, CacheRankingsBlock* node) | |
200 : scoped_ptr<CacheRankingsBlock>(node), rankings_(rankings) {} | |
201 | |
202 Rankings::Iterator::Iterator(Rankings* rankings) { | |
203 memset(this, 0, sizeof(Iterator)); | |
204 my_rankings = rankings; | |
205 } | |
206 | |
207 Rankings::Iterator::~Iterator() { | |
208 for (int i = 0; i < 3; i++) | |
209 ScopedRankingsBlock(my_rankings, nodes[i]); | |
210 } | |
211 | |
212 Rankings::Rankings() : init_(false) {} | |
213 | |
214 Rankings::~Rankings() {} | |
215 | |
216 bool Rankings::Init(BackendImpl* backend, bool count_lists) { | |
217 DCHECK(!init_); | |
218 if (init_) | |
219 return false; | |
220 | |
221 backend_ = backend; | |
222 control_data_ = backend_->GetLruData(); | |
223 count_lists_ = count_lists; | |
224 | |
225 ReadHeads(); | |
226 ReadTails(); | |
227 | |
228 if (control_data_->transaction) | |
229 CompleteTransaction(); | |
230 | |
231 init_ = true; | |
232 return true; | |
233 } | |
234 | |
235 void Rankings::Reset() { | |
236 init_ = false; | |
237 for (int i = 0; i < LAST_ELEMENT; i++) { | |
238 heads_[i].set_value(0); | |
239 tails_[i].set_value(0); | |
240 } | |
241 control_data_ = NULL; | |
242 } | |
243 | |
244 void Rankings::Insert(CacheRankingsBlock* node, bool modified, List list) { | |
245 Trace("Insert 0x%x l %d", node->address().value(), list); | |
246 DCHECK(node->HasData()); | |
247 Addr& my_head = heads_[list]; | |
248 Addr& my_tail = tails_[list]; | |
249 Transaction lock(control_data_, node->address(), INSERT, list); | |
250 CacheRankingsBlock head(backend_->File(my_head), my_head); | |
251 if (my_head.is_initialized()) { | |
252 if (!GetRanking(&head)) | |
253 return; | |
254 | |
255 if (head.Data()->prev != my_head.value() && // Normal path. | |
256 head.Data()->prev != node->address().value()) { // FinishInsert(). | |
257 backend_->CriticalError(ERR_INVALID_LINKS); | |
258 return; | |
259 } | |
260 | |
261 head.Data()->prev = node->address().value(); | |
262 head.Store(); | |
263 GenerateCrash(ON_INSERT_1); | |
264 UpdateIterators(&head); | |
265 } | |
266 | |
267 node->Data()->next = my_head.value(); | |
268 node->Data()->prev = node->address().value(); | |
269 my_head.set_value(node->address().value()); | |
270 | |
271 if (!my_tail.is_initialized() || my_tail.value() == node->address().value()) { | |
272 my_tail.set_value(node->address().value()); | |
273 node->Data()->next = my_tail.value(); | |
274 WriteTail(list); | |
275 GenerateCrash(ON_INSERT_2); | |
276 } | |
277 | |
278 UpdateTimes(node, modified); | |
279 node->Store(); | |
280 GenerateCrash(ON_INSERT_3); | |
281 | |
282 // The last thing to do is move our head to point to a node already stored. | |
283 WriteHead(list); | |
284 IncrementCounter(list); | |
285 GenerateCrash(ON_INSERT_4); | |
286 backend_->FlushIndex(); | |
287 } | |
288 | |
289 // If a, b and r are elements on the list, and we want to remove r, the possible | |
290 // states for the objects if a crash happens are (where y(x, z) means for object | |
291 // y, prev is x and next is z): | |
292 // A. One element: | |
293 // 1. r(r, r), head(r), tail(r) initial state | |
294 // 2. r(r, r), head(0), tail(r) WriteHead() | |
295 // 3. r(r, r), head(0), tail(0) WriteTail() | |
296 // 4. r(0, 0), head(0), tail(0) next.Store() | |
297 // | |
298 // B. Remove a random element: | |
299 // 1. a(x, r), r(a, b), b(r, y), head(x), tail(y) initial state | |
300 // 2. a(x, r), r(a, b), b(a, y), head(x), tail(y) next.Store() | |
301 // 3. a(x, b), r(a, b), b(a, y), head(x), tail(y) prev.Store() | |
302 // 4. a(x, b), r(0, 0), b(a, y), head(x), tail(y) node.Store() | |
303 // | |
304 // C. Remove head: | |
305 // 1. r(r, b), b(r, y), head(r), tail(y) initial state | |
306 // 2. r(r, b), b(r, y), head(b), tail(y) WriteHead() | |
307 // 3. r(r, b), b(b, y), head(b), tail(y) next.Store() | |
308 // 4. r(0, 0), b(b, y), head(b), tail(y) prev.Store() | |
309 // | |
310 // D. Remove tail: | |
311 // 1. a(x, r), r(a, r), head(x), tail(r) initial state | |
312 // 2. a(x, r), r(a, r), head(x), tail(a) WriteTail() | |
313 // 3. a(x, a), r(a, r), head(x), tail(a) prev.Store() | |
314 // 4. a(x, a), r(0, 0), head(x), tail(a) next.Store() | |
315 void Rankings::Remove(CacheRankingsBlock* node, List list, bool strict) { | |
316 Trace("Remove 0x%x (0x%x 0x%x) l %d", node->address().value(), | |
317 node->Data()->next, node->Data()->prev, list); | |
318 DCHECK(node->HasData()); | |
319 if (strict) | |
320 InvalidateIterators(node); | |
321 | |
322 Addr next_addr(node->Data()->next); | |
323 Addr prev_addr(node->Data()->prev); | |
324 if (!next_addr.is_initialized() || next_addr.is_separate_file() || | |
325 !prev_addr.is_initialized() || prev_addr.is_separate_file()) { | |
326 if (next_addr.is_initialized() || prev_addr.is_initialized()) { | |
327 LOG(ERROR) << "Invalid rankings info."; | |
328 STRESS_NOTREACHED(); | |
329 } | |
330 return; | |
331 } | |
332 | |
333 CacheRankingsBlock next(backend_->File(next_addr), next_addr); | |
334 CacheRankingsBlock prev(backend_->File(prev_addr), prev_addr); | |
335 if (!GetRanking(&next) || !GetRanking(&prev)) { | |
336 STRESS_NOTREACHED(); | |
337 return; | |
338 } | |
339 | |
340 if (!CheckLinks(node, &prev, &next, &list)) | |
341 return; | |
342 | |
343 Transaction lock(control_data_, node->address(), REMOVE, list); | |
344 prev.Data()->next = next.address().value(); | |
345 next.Data()->prev = prev.address().value(); | |
346 GenerateCrash(ON_REMOVE_1); | |
347 | |
348 CacheAddr node_value = node->address().value(); | |
349 Addr& my_head = heads_[list]; | |
350 Addr& my_tail = tails_[list]; | |
351 if (node_value == my_head.value() || node_value == my_tail.value()) { | |
352 if (my_head.value() == my_tail.value()) { | |
353 my_head.set_value(0); | |
354 my_tail.set_value(0); | |
355 | |
356 WriteHead(list); | |
357 GenerateCrash(ON_REMOVE_2); | |
358 WriteTail(list); | |
359 GenerateCrash(ON_REMOVE_3); | |
360 } else if (node_value == my_head.value()) { | |
361 my_head.set_value(next.address().value()); | |
362 next.Data()->prev = next.address().value(); | |
363 | |
364 WriteHead(list); | |
365 GenerateCrash(ON_REMOVE_4); | |
366 } else if (node_value == my_tail.value()) { | |
367 my_tail.set_value(prev.address().value()); | |
368 prev.Data()->next = prev.address().value(); | |
369 | |
370 WriteTail(list); | |
371 GenerateCrash(ON_REMOVE_5); | |
372 | |
373 // Store the new tail to make sure we can undo the operation if we crash. | |
374 prev.Store(); | |
375 GenerateCrash(ON_REMOVE_6); | |
376 } | |
377 } | |
378 | |
379 // Nodes out of the list can be identified by invalid pointers. | |
380 node->Data()->next = 0; | |
381 node->Data()->prev = 0; | |
382 | |
383 // The last thing to get to disk is the node itself, so before that there is | |
384 // enough info to recover. | |
385 next.Store(); | |
386 GenerateCrash(ON_REMOVE_7); | |
387 prev.Store(); | |
388 GenerateCrash(ON_REMOVE_8); | |
389 node->Store(); | |
390 DecrementCounter(list); | |
391 UpdateIterators(&next); | |
392 UpdateIterators(&prev); | |
393 backend_->FlushIndex(); | |
394 } | |
395 | |
396 // A crash in between Remove and Insert will lead to a dirty entry not on the | |
397 // list. We want to avoid that case as much as we can (as while waiting for IO), | |
398 // but the net effect is just an assert on debug when attempting to remove the | |
399 // entry. Otherwise we'll need reentrant transactions, which is an overkill. | |
400 void Rankings::UpdateRank(CacheRankingsBlock* node, bool modified, List list) { | |
401 Addr& my_head = heads_[list]; | |
402 if (my_head.value() == node->address().value()) { | |
403 UpdateTimes(node, modified); | |
404 node->set_modified(); | |
405 return; | |
406 } | |
407 | |
408 TimeTicks start = TimeTicks::Now(); | |
409 Remove(node, list, true); | |
410 Insert(node, modified, list); | |
411 CACHE_UMA(AGE_MS, "UpdateRank", 0, start); | |
412 } | |
413 | |
414 CacheRankingsBlock* Rankings::GetNext(CacheRankingsBlock* node, List list) { | |
415 ScopedRankingsBlock next(this); | |
416 if (!node) { | |
417 Addr& my_head = heads_[list]; | |
418 if (!my_head.is_initialized()) | |
419 return NULL; | |
420 next.reset(new CacheRankingsBlock(backend_->File(my_head), my_head)); | |
421 } else { | |
422 if (!node->HasData()) | |
423 node->Load(); | |
424 Addr& my_tail = tails_[list]; | |
425 if (!my_tail.is_initialized()) | |
426 return NULL; | |
427 if (my_tail.value() == node->address().value()) | |
428 return NULL; | |
429 Addr address(node->Data()->next); | |
430 if (address.value() == node->address().value()) | |
431 return NULL; // Another tail? fail it. | |
432 next.reset(new CacheRankingsBlock(backend_->File(address), address)); | |
433 } | |
434 | |
435 TrackRankingsBlock(next.get(), true); | |
436 | |
437 if (!GetRanking(next.get())) | |
438 return NULL; | |
439 | |
440 ConvertToLongLived(next.get()); | |
441 if (node && !CheckSingleLink(node, next.get())) | |
442 return NULL; | |
443 | |
444 return next.release(); | |
445 } | |
446 | |
447 CacheRankingsBlock* Rankings::GetPrev(CacheRankingsBlock* node, List list) { | |
448 ScopedRankingsBlock prev(this); | |
449 if (!node) { | |
450 Addr& my_tail = tails_[list]; | |
451 if (!my_tail.is_initialized()) | |
452 return NULL; | |
453 prev.reset(new CacheRankingsBlock(backend_->File(my_tail), my_tail)); | |
454 } else { | |
455 if (!node->HasData()) | |
456 node->Load(); | |
457 Addr& my_head = heads_[list]; | |
458 if (!my_head.is_initialized()) | |
459 return NULL; | |
460 if (my_head.value() == node->address().value()) | |
461 return NULL; | |
462 Addr address(node->Data()->prev); | |
463 if (address.value() == node->address().value()) | |
464 return NULL; // Another head? fail it. | |
465 prev.reset(new CacheRankingsBlock(backend_->File(address), address)); | |
466 } | |
467 | |
468 TrackRankingsBlock(prev.get(), true); | |
469 | |
470 if (!GetRanking(prev.get())) | |
471 return NULL; | |
472 | |
473 ConvertToLongLived(prev.get()); | |
474 if (node && !CheckSingleLink(prev.get(), node)) | |
475 return NULL; | |
476 | |
477 return prev.release(); | |
478 } | |
479 | |
480 void Rankings::FreeRankingsBlock(CacheRankingsBlock* node) { | |
481 TrackRankingsBlock(node, false); | |
482 } | |
483 | |
484 void Rankings::TrackRankingsBlock(CacheRankingsBlock* node, | |
485 bool start_tracking) { | |
486 if (!node) | |
487 return; | |
488 | |
489 IteratorPair current(node->address().value(), node); | |
490 | |
491 if (start_tracking) | |
492 iterators_.push_back(current); | |
493 else | |
494 iterators_.remove(current); | |
495 } | |
496 | |
497 int Rankings::SelfCheck() { | |
498 int total = 0; | |
499 int error = 0; | |
500 for (int i = 0; i < LAST_ELEMENT; i++) { | |
501 int partial = CheckList(static_cast<List>(i)); | |
502 if (partial < 0 && !error) | |
503 error = partial; | |
504 else if (partial > 0) | |
505 total += partial; | |
506 } | |
507 | |
508 return error ? error : total; | |
509 } | |
510 | |
511 bool Rankings::SanityCheck(CacheRankingsBlock* node, bool from_list) const { | |
512 if (!node->VerifyHash()) | |
513 return false; | |
514 | |
515 const RankingsNode* data = node->Data(); | |
516 | |
517 if ((!data->next && data->prev) || (data->next && !data->prev)) | |
518 return false; | |
519 | |
520 // Both pointers on zero is a node out of the list. | |
521 if (!data->next && !data->prev && from_list) | |
522 return false; | |
523 | |
524 List list = NO_USE; // Initialize it to something. | |
525 if ((node->address().value() == data->prev) && !IsHead(data->prev, &list)) | |
526 return false; | |
527 | |
528 if ((node->address().value() == data->next) && !IsTail(data->next, &list)) | |
529 return false; | |
530 | |
531 if (!data->next && !data->prev) | |
532 return true; | |
533 | |
534 Addr next_addr(data->next); | |
535 Addr prev_addr(data->prev); | |
536 if (!next_addr.SanityCheckV2() || next_addr.file_type() != RANKINGS || | |
537 !prev_addr.SanityCheckV2() || prev_addr.file_type() != RANKINGS) | |
538 return false; | |
539 | |
540 return true; | |
541 } | |
542 | |
543 bool Rankings::DataSanityCheck(CacheRankingsBlock* node, bool from_list) const { | |
544 const RankingsNode* data = node->Data(); | |
545 if (!data->contents) | |
546 return false; | |
547 | |
548 // It may have never been inserted. | |
549 if (from_list && (!data->last_used || !data->last_modified)) | |
550 return false; | |
551 | |
552 return true; | |
553 } | |
554 | |
555 void Rankings::SetContents(CacheRankingsBlock* node, CacheAddr address) { | |
556 node->Data()->contents = address; | |
557 node->Store(); | |
558 } | |
559 | |
560 void Rankings::ReadHeads() { | |
561 for (int i = 0; i < LAST_ELEMENT; i++) | |
562 heads_[i] = Addr(control_data_->heads[i]); | |
563 } | |
564 | |
565 void Rankings::ReadTails() { | |
566 for (int i = 0; i < LAST_ELEMENT; i++) | |
567 tails_[i] = Addr(control_data_->tails[i]); | |
568 } | |
569 | |
570 void Rankings::WriteHead(List list) { | |
571 control_data_->heads[list] = heads_[list].value(); | |
572 } | |
573 | |
574 void Rankings::WriteTail(List list) { | |
575 control_data_->tails[list] = tails_[list].value(); | |
576 } | |
577 | |
578 bool Rankings::GetRanking(CacheRankingsBlock* rankings) { | |
579 if (!rankings->address().is_initialized()) | |
580 return false; | |
581 | |
582 TimeTicks start = TimeTicks::Now(); | |
583 if (!rankings->Load()) | |
584 return false; | |
585 | |
586 if (!SanityCheck(rankings, true)) { | |
587 backend_->CriticalError(ERR_INVALID_LINKS); | |
588 return false; | |
589 } | |
590 | |
591 backend_->OnEvent(Stats::OPEN_RANKINGS); | |
592 | |
593 // Note that if the cache is in read_only mode, open entries are not marked | |
594 // as dirty, except when an entry is doomed. We have to look for open entries. | |
595 if (!backend_->read_only() && !rankings->Data()->dirty) | |
596 return true; | |
597 | |
598 EntryImpl* entry = backend_->GetOpenEntry(rankings); | |
599 if (!entry) { | |
600 if (backend_->read_only()) | |
601 return true; | |
602 | |
603 // We cannot trust this entry, but we cannot initiate a cleanup from this | |
604 // point (we may be in the middle of a cleanup already). The entry will be | |
605 // deleted when detected from a regular open/create path. | |
606 rankings->Data()->dirty = backend_->GetCurrentEntryId() - 1; | |
607 if (!rankings->Data()->dirty) | |
608 rankings->Data()->dirty--; | |
609 return true; | |
610 } | |
611 | |
612 // Note that we should not leave this module without deleting rankings first. | |
613 rankings->SetData(entry->rankings()->Data()); | |
614 | |
615 CACHE_UMA(AGE_MS, "GetRankings", 0, start); | |
616 return true; | |
617 } | |
618 | |
619 void Rankings::ConvertToLongLived(CacheRankingsBlock* rankings) { | |
620 if (rankings->own_data()) | |
621 return; | |
622 | |
623 // We cannot return a shared node because we are not keeping a reference | |
624 // to the entry that owns the buffer. Make this node a copy of the one that | |
625 // we have, and let the iterator logic update it when the entry changes. | |
626 CacheRankingsBlock temp(NULL, Addr(0)); | |
627 *temp.Data() = *rankings->Data(); | |
628 rankings->StopSharingData(); | |
629 *rankings->Data() = *temp.Data(); | |
630 } | |
631 | |
632 void Rankings::CompleteTransaction() { | |
633 Addr node_addr(static_cast<CacheAddr>(control_data_->transaction)); | |
634 if (!node_addr.is_initialized() || node_addr.is_separate_file()) { | |
635 NOTREACHED(); | |
636 LOG(ERROR) << "Invalid rankings info."; | |
637 return; | |
638 } | |
639 | |
640 Trace("CompleteTransaction 0x%x", node_addr.value()); | |
641 | |
642 CacheRankingsBlock node(backend_->File(node_addr), node_addr); | |
643 if (!node.Load()) | |
644 return; | |
645 | |
646 node.Store(); | |
647 | |
648 Addr& my_head = heads_[control_data_->operation_list]; | |
649 Addr& my_tail = tails_[control_data_->operation_list]; | |
650 | |
651 // We want to leave the node inside the list. The entry must me marked as | |
652 // dirty, and will be removed later. Otherwise, we'll get assertions when | |
653 // attempting to remove the dirty entry. | |
654 if (INSERT == control_data_->operation) { | |
655 Trace("FinishInsert h:0x%x t:0x%x", my_head.value(), my_tail.value()); | |
656 FinishInsert(&node); | |
657 } else if (REMOVE == control_data_->operation) { | |
658 Trace("RevertRemove h:0x%x t:0x%x", my_head.value(), my_tail.value()); | |
659 RevertRemove(&node); | |
660 } else { | |
661 NOTREACHED(); | |
662 LOG(ERROR) << "Invalid operation to recover."; | |
663 } | |
664 } | |
665 | |
666 void Rankings::FinishInsert(CacheRankingsBlock* node) { | |
667 control_data_->transaction = 0; | |
668 control_data_->operation = 0; | |
669 Addr& my_head = heads_[control_data_->operation_list]; | |
670 Addr& my_tail = tails_[control_data_->operation_list]; | |
671 if (my_head.value() != node->address().value()) { | |
672 if (my_tail.value() == node->address().value()) { | |
673 // This part will be skipped by the logic of Insert. | |
674 node->Data()->next = my_tail.value(); | |
675 } | |
676 | |
677 Insert(node, true, static_cast<List>(control_data_->operation_list)); | |
678 } | |
679 | |
680 // Tell the backend about this entry. | |
681 backend_->RecoveredEntry(node); | |
682 } | |
683 | |
684 void Rankings::RevertRemove(CacheRankingsBlock* node) { | |
685 Addr next_addr(node->Data()->next); | |
686 Addr prev_addr(node->Data()->prev); | |
687 if (!next_addr.is_initialized() || !prev_addr.is_initialized()) { | |
688 // The operation actually finished. Nothing to do. | |
689 control_data_->transaction = 0; | |
690 return; | |
691 } | |
692 if (next_addr.is_separate_file() || prev_addr.is_separate_file()) { | |
693 NOTREACHED(); | |
694 LOG(WARNING) << "Invalid rankings info."; | |
695 control_data_->transaction = 0; | |
696 return; | |
697 } | |
698 | |
699 CacheRankingsBlock next(backend_->File(next_addr), next_addr); | |
700 CacheRankingsBlock prev(backend_->File(prev_addr), prev_addr); | |
701 if (!next.Load() || !prev.Load()) | |
702 return; | |
703 | |
704 CacheAddr node_value = node->address().value(); | |
705 DCHECK(prev.Data()->next == node_value || | |
706 prev.Data()->next == prev_addr.value() || | |
707 prev.Data()->next == next.address().value()); | |
708 DCHECK(next.Data()->prev == node_value || | |
709 next.Data()->prev == next_addr.value() || | |
710 next.Data()->prev == prev.address().value()); | |
711 | |
712 if (node_value != prev_addr.value()) | |
713 prev.Data()->next = node_value; | |
714 if (node_value != next_addr.value()) | |
715 next.Data()->prev = node_value; | |
716 | |
717 List my_list = static_cast<List>(control_data_->operation_list); | |
718 Addr& my_head = heads_[my_list]; | |
719 Addr& my_tail = tails_[my_list]; | |
720 if (!my_head.is_initialized() || !my_tail.is_initialized()) { | |
721 my_head.set_value(node_value); | |
722 my_tail.set_value(node_value); | |
723 WriteHead(my_list); | |
724 WriteTail(my_list); | |
725 } else if (my_head.value() == next.address().value()) { | |
726 my_head.set_value(node_value); | |
727 prev.Data()->next = next.address().value(); | |
728 WriteHead(my_list); | |
729 } else if (my_tail.value() == prev.address().value()) { | |
730 my_tail.set_value(node_value); | |
731 next.Data()->prev = prev.address().value(); | |
732 WriteTail(my_list); | |
733 } | |
734 | |
735 next.Store(); | |
736 prev.Store(); | |
737 control_data_->transaction = 0; | |
738 control_data_->operation = 0; | |
739 backend_->FlushIndex(); | |
740 } | |
741 | |
742 bool Rankings::CheckLinks(CacheRankingsBlock* node, CacheRankingsBlock* prev, | |
743 CacheRankingsBlock* next, List* list) { | |
744 CacheAddr node_addr = node->address().value(); | |
745 if (prev->Data()->next == node_addr && | |
746 next->Data()->prev == node_addr) { | |
747 // A regular linked node. | |
748 return true; | |
749 } | |
750 | |
751 Trace("CheckLinks 0x%x (0x%x 0x%x)", node_addr, | |
752 prev->Data()->next, next->Data()->prev); | |
753 | |
754 if (node_addr != prev->address().value() && | |
755 node_addr != next->address().value() && | |
756 prev->Data()->next == next->address().value() && | |
757 next->Data()->prev == prev->address().value()) { | |
758 // The list is actually ok, node is wrong. | |
759 Trace("node 0x%x out of list %d", node_addr, list); | |
760 node->Data()->next = 0; | |
761 node->Data()->prev = 0; | |
762 node->Store(); | |
763 return false; | |
764 } | |
765 | |
766 if (prev->Data()->next == node_addr || | |
767 next->Data()->prev == node_addr) { | |
768 // Only one link is weird, lets double check. | |
769 if (prev->Data()->next != node_addr && IsHead(node_addr, list)) | |
770 return true; | |
771 | |
772 if (next->Data()->prev != node_addr && IsTail(node_addr, list)) | |
773 return true; | |
774 } | |
775 | |
776 LOG(ERROR) << "Inconsistent LRU."; | |
777 STRESS_NOTREACHED(); | |
778 | |
779 backend_->CriticalError(ERR_INVALID_LINKS); | |
780 return false; | |
781 } | |
782 | |
783 bool Rankings::CheckSingleLink(CacheRankingsBlock* prev, | |
784 CacheRankingsBlock* next) { | |
785 if (prev->Data()->next != next->address().value() || | |
786 next->Data()->prev != prev->address().value()) { | |
787 LOG(ERROR) << "Inconsistent LRU."; | |
788 | |
789 backend_->CriticalError(ERR_INVALID_LINKS); | |
790 return false; | |
791 } | |
792 | |
793 return true; | |
794 } | |
795 | |
796 int Rankings::CheckList(List list) { | |
797 Addr last1, last2; | |
798 int head_items; | |
799 int rv = CheckListSection(list, last1, last2, true, // Head to tail. | |
800 &last1, &last2, &head_items); | |
801 if (rv == ERR_NO_ERROR) | |
802 return head_items; | |
803 | |
804 return rv; | |
805 } | |
806 | |
807 // Note that the returned error codes assume a forward walk (from head to tail) | |
808 // so they have to be adjusted accordingly by the caller. We use two stop values | |
809 // to be able to detect a corrupt node at the end that is not linked going back. | |
810 int Rankings::CheckListSection(List list, Addr end1, Addr end2, bool forward, | |
811 Addr* last, Addr* second_last, int* num_items) { | |
812 Addr current = forward ? heads_[list] : tails_[list]; | |
813 *last = *second_last = current; | |
814 *num_items = 0; | |
815 if (!current.is_initialized()) | |
816 return ERR_NO_ERROR; | |
817 | |
818 if (!current.SanityCheckForRankings()) | |
819 return ERR_INVALID_HEAD; | |
820 | |
821 scoped_ptr<CacheRankingsBlock> node; | |
822 Addr prev_addr(current); | |
823 do { | |
824 node.reset(new CacheRankingsBlock(backend_->File(current), current)); | |
825 node->Load(); | |
826 if (!SanityCheck(node.get(), true)) | |
827 return ERR_INVALID_ENTRY; | |
828 | |
829 CacheAddr next = forward ? node->Data()->next : node->Data()->prev; | |
830 CacheAddr prev = forward ? node->Data()->prev : node->Data()->next; | |
831 | |
832 if (prev != prev_addr.value()) | |
833 return ERR_INVALID_PREV; | |
834 | |
835 Addr next_addr(next); | |
836 if (!next_addr.SanityCheckForRankings()) | |
837 return ERR_INVALID_NEXT; | |
838 | |
839 prev_addr = current; | |
840 current = next_addr; | |
841 *second_last = *last; | |
842 *last = current; | |
843 (*num_items)++; | |
844 | |
845 if (next_addr == prev_addr) { | |
846 Addr last = forward ? tails_[list] : heads_[list]; | |
847 if (next_addr == last) | |
848 return ERR_NO_ERROR; | |
849 return ERR_INVALID_TAIL; | |
850 } | |
851 } while (current != end1 && current != end2); | |
852 return ERR_NO_ERROR; | |
853 } | |
854 | |
855 bool Rankings::IsHead(CacheAddr addr, List* list) const { | |
856 for (int i = 0; i < LAST_ELEMENT; i++) { | |
857 if (addr == heads_[i].value()) { | |
858 if (*list != i) | |
859 Trace("Changing list %d to %d", *list, i); | |
860 *list = static_cast<List>(i); | |
861 return true; | |
862 } | |
863 } | |
864 return false; | |
865 } | |
866 | |
867 bool Rankings::IsTail(CacheAddr addr, List* list) const { | |
868 for (int i = 0; i < LAST_ELEMENT; i++) { | |
869 if (addr == tails_[i].value()) { | |
870 if (*list != i) | |
871 Trace("Changing list %d to %d", *list, i); | |
872 *list = static_cast<List>(i); | |
873 return true; | |
874 } | |
875 } | |
876 return false; | |
877 } | |
878 | |
879 // We expect to have just a few iterators at any given time, maybe two or three, | |
880 // But we could have more than one pointing at the same mode. We walk the list | |
881 // of cache iterators and update all that are pointing to the given node. | |
882 void Rankings::UpdateIterators(CacheRankingsBlock* node) { | |
883 CacheAddr address = node->address().value(); | |
884 for (IteratorList::iterator it = iterators_.begin(); it != iterators_.end(); | |
885 ++it) { | |
886 if (it->first == address && it->second->HasData()) { | |
887 CacheRankingsBlock* other = it->second; | |
888 *other->Data() = *node->Data(); | |
889 } | |
890 } | |
891 } | |
892 | |
893 void Rankings::InvalidateIterators(CacheRankingsBlock* node) { | |
894 CacheAddr address = node->address().value(); | |
895 for (IteratorList::iterator it = iterators_.begin(); it != iterators_.end(); | |
896 ++it) { | |
897 if (it->first == address) { | |
898 DLOG(INFO) << "Invalidating iterator at 0x" << std::hex << address; | |
899 it->second->Discard(); | |
900 } | |
901 } | |
902 } | |
903 | |
904 void Rankings::IncrementCounter(List list) { | |
905 if (!count_lists_) | |
906 return; | |
907 | |
908 DCHECK(control_data_->sizes[list] < kint32max); | |
909 if (control_data_->sizes[list] < kint32max) | |
910 control_data_->sizes[list]++; | |
911 } | |
912 | |
913 void Rankings::DecrementCounter(List list) { | |
914 if (!count_lists_) | |
915 return; | |
916 | |
917 DCHECK(control_data_->sizes[list] > 0); | |
918 if (control_data_->sizes[list] > 0) | |
919 control_data_->sizes[list]--; | |
920 } | |
921 | |
922 } // namespace disk_cache | |
OLD | NEW |