Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(590)

Side by Side Diff: net/disk_cache/v3/index_table.cc

Issue 15203004: Disk cache: Reference CL for the implementation of file format version 3. (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/src/
Patch Set: IndexTable review Created 7 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
« no previous file with comments | « net/disk_cache/v3/index_table.h ('k') | net/disk_cache/v3/index_table_unittest.cc » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Property Changes:
Added: svn:eol-style
+ LF
OLDNEW
(Empty)
1 // Copyright (c) 2013 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/v3/index_table.h"
6
7 #include <algorithm>
8 #include <set>
9 #include <utility>
10
11 #include "base/bits.h"
12 #include "net/base/io_buffer.h"
13 #include "net/base/net_errors.h"
14 #include "net/disk_cache/disk_cache.h"
15
16 using base::Time;
17 using base::TimeDelta;
18 using disk_cache::CellInfo;
19 using disk_cache::CellList;
20 using disk_cache::IndexCell;
21 using disk_cache::IndexIterator;
22
23 namespace {
24
25 const int kCellHashOffset = 22;
26 const int kCellSmallTableHashOffset = 16;
27 const int kCellTimestampOffset = 40;
28 const int kCellReuseOffset = 60;
29 const int kCellGroupOffset = 3;
30 const int kCellSumOffset = 6;
31
32 const uint64 kCellAddressMask = 0x3FFFFF;
33 const uint64 kCellSmallTableAddressMask = 0xFFFF;
34 const uint64 kCellHashMask = 0x3FFFF;
35 const uint64 kCellSmallTableHashMask = 0xFFFFFF;
36 const uint64 kCellTimestampMask = 0xFFFFF;
37 const uint64 kCellReuseMask = 0xF;
38 const uint8 kCellStateMask = 0x7;
39 const uint8 kCellGroupMask = 0x7;
40 const uint8 kCellSumMask = 0x3;
41
42 const int kHashShift = 14;
43 const int kHashSmallTableShift = 8;
44
45 // Unfortunately we have to break the abstaction a little here: the file number
46 // where entries are stored is outside of the control of this code, and it is
47 // usually part of the stored address. However, for small tables we only store
48 // 16 bits of the address so the file number is never stored on a cell. We have
49 // to infere the file number from the type of entry (normal vs evicted), and
50 // the knowledge that given that the table will not keep more than 64k entries,
51 // a single file of each type is enough.
52 const int kEntriesFile = disk_cache::BLOCK_ENTRIES - 1;
53 const int kEvictedEntriesFile = disk_cache::BLOCK_EVICTED - 1;
54 const int kMaxAddress = 1 << 22;
55 const int kMinFileNumber = 1 << 16;
56
57 uint32 GetCellAddress(const IndexCell& cell) {
58 return cell.first_part & kCellAddressMask;
59 }
60
61 uint32 GetCellSmallTableAddress(const IndexCell& cell) {
62 return cell.first_part & kCellSmallTableAddressMask;
63 }
64
65 uint32 GetCellHash(const IndexCell& cell) {
66 return (cell.first_part >> kCellHashOffset) & kCellHashMask;
67 }
68
69 uint32 GetCellSmallTableHash(const IndexCell& cell) {
70 return (cell.first_part >> kCellSmallTableHashOffset) &
71 kCellSmallTableHashMask;
72 }
73
74 int GetCellTimestamp(const IndexCell& cell) {
75 return (cell.first_part >> kCellTimestampOffset) & kCellTimestampMask;
76 }
77
78 int GetCellReuse(const IndexCell& cell) {
79 return (cell.first_part >> kCellReuseOffset) & kCellReuseMask;
80 }
81
82 int GetCellState(const IndexCell& cell) {
83 return cell.last_part & kCellStateMask;
84 }
85
86 int GetCellGroup(const IndexCell& cell) {
87 return (cell.last_part >> kCellGroupOffset) & kCellGroupMask;
88 }
89
90 int GetCellSum(const IndexCell& cell) {
91 return (cell.last_part >> kCellSumOffset) & kCellSumMask;
92 }
93
94 void SetCellAddress(IndexCell* cell, uint32 address) {
95 DCHECK_LE(address, static_cast<uint32>(kCellAddressMask));
96 cell->first_part &= ~kCellAddressMask;
97 cell->first_part |= address;
98 }
99
100 void SetCellSmallTableAddress(IndexCell* cell, uint32 address) {
101 DCHECK_LE(address, static_cast<uint32>(kCellSmallTableAddressMask));
102 cell->first_part &= ~kCellSmallTableAddressMask;
103 cell->first_part |= address;
104 }
105
106 void SetCellHash(IndexCell* cell, uint32 hash) {
107 DCHECK_LE(hash, static_cast<uint32>(kCellHashMask));
108 cell->first_part &= ~(kCellHashMask << kCellHashOffset);
109 cell->first_part |= static_cast<int64>(hash) << kCellHashOffset;
110 }
111
112 void SetCellSmallTableHash(IndexCell* cell, uint32 hash) {
113 DCHECK_LE(hash, static_cast<uint32>(kCellSmallTableHashMask));
114 cell->first_part &= ~(kCellSmallTableHashMask << kCellSmallTableHashOffset);
115 cell->first_part |= static_cast<int64>(hash) << kCellSmallTableHashOffset;
116 }
117
118 void SetCellTimestamp(IndexCell* cell, int timestamp) {
119 DCHECK_LT(timestamp, 1 << 20);
120 DCHECK_GE(timestamp, 0);
121 cell->first_part &= ~(kCellTimestampMask << kCellTimestampOffset);
122 cell->first_part |= static_cast<int64>(timestamp) << kCellTimestampOffset;
123 }
124
125 void SetCellReuse(IndexCell* cell, int count) {
126 DCHECK_LT(count, 16);
127 DCHECK_GE(count, 0);
128 cell->first_part &= ~(kCellReuseMask << kCellReuseOffset);
129 cell->first_part |= static_cast<int64>(count) << kCellReuseOffset;
130 }
131
132 void SetCellState(IndexCell* cell, disk_cache::EntryState state) {
133 cell->last_part &= ~kCellStateMask;
134 cell->last_part |= state;
135 }
136
137 void SetCellGroup(IndexCell* cell, disk_cache::EntryGroup group) {
138 cell->last_part &= ~(kCellGroupMask << kCellGroupOffset);
139 cell->last_part |= group << kCellGroupOffset;
140 }
141
142 void SetCellSum(IndexCell* cell, int sum) {
143 DCHECK_LT(sum, 4);
144 DCHECK_GE(sum, 0);
145 cell->last_part &= ~(kCellSumMask << kCellSumOffset);
146 cell->last_part |= sum << kCellSumOffset;
147 }
148
149 // This is a very particular way to calculate the sum, so it will not match if
150 // compared a gainst a pure 2 bit, modulo 2 sum.
151 int CalculateCellSum(const IndexCell& cell) {
152 uint32* words = bit_cast<uint32*>(&cell);
153 uint8* bytes = bit_cast<uint8*>(&cell);
154 uint32 result = words[0] + words[1];
155 result += result >> 16;
156 result += (result >> 8) + (bytes[8] & 0x3f);
157 result += result >> 4;
158 result += result >> 2;
159 return result & 3;
160 }
161
162 bool SanityCheck(const IndexCell& cell) {
163 if (GetCellSum(cell) != CalculateCellSum(cell))
164 return false;
165
166 if (GetCellState(cell) > disk_cache::ENTRY_USED ||
167 GetCellGroup(cell) == disk_cache::ENTRY_RESERVED ||
168 GetCellGroup(cell) > disk_cache::ENTRY_EVICTED) {
169 return false;
170 }
171
172 return true;
173 }
174
175 int FileNumberFromAddress(int index_address) {
176 return index_address / kMinFileNumber;
177 }
178
179 int StartBlockFromAddress(int index_address) {
180 return index_address % kMinFileNumber;
181 }
182
183 bool IsValidAddress(disk_cache::Addr address) {
184 if (!address.is_initialized() ||
185 (address.file_type() != disk_cache::BLOCK_EVICTED &&
186 address.file_type() != disk_cache::BLOCK_ENTRIES)) {
187 return false;
188 }
189
190 return address.FileNumber() < FileNumberFromAddress(kMaxAddress);
191 }
192
193 bool IsNormalState(const IndexCell& cell) {
194 disk_cache::EntryState state =
195 static_cast<disk_cache::EntryState>(GetCellState(cell));
196 DCHECK_NE(state, disk_cache::ENTRY_FREE);
197 return state != disk_cache::ENTRY_DELETED &&
198 state != disk_cache::ENTRY_FIXING;
199 }
200
201 inline int GetNextBucket(int min_bucket_id, int max_bucket_id,
202 disk_cache::IndexBucket* table,
203 disk_cache::IndexBucket** bucket) {
204 if (!(*bucket)->next)
205 return 0;
206
207 int bucket_id = (*bucket)->next / disk_cache::kCellsPerBucket;
208 if (bucket_id < min_bucket_id || bucket_id > max_bucket_id) {
209 (*bucket)->next = 0;
210 return 0;
211 }
212 *bucket = &table[bucket_id - min_bucket_id];
213 return bucket_id;
214 }
215
216 // Updates the |iterator| with the current |cell|. This cell may cause all
217 // previous cells to be deleted (when a new target timestamp is found), the cell
218 // may be added to the list (if it matches the target timestamp), or may it be
219 // ignored.
220 void UpdateIterator(const disk_cache::EntryCell& cell,
221 int limit_time,
222 IndexIterator* iterator) {
223 int time = cell.GetTimestamp();
224 // Look for not interesting times.
225 if (iterator->forward && time <= limit_time)
226 return;
227 if (!iterator->forward && time >= limit_time)
228 return;
229
230 if ((iterator->forward && time < iterator->timestamp) ||
231 (!iterator->forward && time > iterator->timestamp)) {
232 // This timestamp is better than the one we had.
233 iterator->timestamp = time;
234 iterator->cells.clear();
235 }
236 if (time == iterator->timestamp) {
237 CellInfo cell_info = { cell.hash(), cell.GetAddress() };
238 iterator->cells.push_back(cell_info);
239 }
240 }
241
242 void InitIterator(IndexIterator* iterator) {
243 iterator->cells.clear();
244 iterator->timestamp = iterator->forward ? kint32max : 0;
245 }
246
247 } // namespace
248
249 namespace disk_cache {
250
251 EntryCell::~EntryCell() {
252 }
253
254 bool EntryCell::IsValid() const {
255 return GetCellAddress(cell_) != 0;
256 }
257
258 // This code has to map the cell address (up to 22 bits) to a general cache Addr
259 // (up to 24 bits of general addressing). It also set the implied file_number
260 // in the case of small tables. See also the comment by the definition of
261 // kEntriesFile.
262 Addr EntryCell::GetAddress() const {
263 uint32 address_value = GetAddressValue();
264 int file_number = FileNumberFromAddress(address_value);
265 if (small_table_) {
266 DCHECK_EQ(0, file_number);
267 file_number = (GetGroup() == ENTRY_EVICTED) ? kEvictedEntriesFile :
268 kEntriesFile;
269 }
270 DCHECK_NE(0, file_number);
271 FileType file_type = (GetGroup() == ENTRY_EVICTED) ? BLOCK_EVICTED :
272 BLOCK_ENTRIES;
273 return Addr(file_type, 1, file_number, StartBlockFromAddress(address_value));
274 }
275
276 EntryState EntryCell::GetState() const {
277 return static_cast<EntryState>(cell_.last_part & kCellStateMask);
278 }
279
280 EntryGroup EntryCell::GetGroup() const {
281 return static_cast<EntryGroup>((cell_.last_part >> kCellGroupOffset) &
282 kCellGroupMask);
283 }
284
285 int EntryCell::GetReuse() const {
286 return (cell_.first_part >> kCellReuseOffset) & kCellReuseMask;
287 }
288
289 int EntryCell::GetTimestamp() const {
290 return GetCellTimestamp(cell_);
291 }
292
293 void EntryCell::SetState(EntryState state) {
294 SetCellState(&cell_, state);
295 }
296
297 void EntryCell::SetGroup(EntryGroup group) {
298 SetCellGroup(&cell_, group);
299 }
300
301 void EntryCell::SetReuse(int count) {
302 SetCellReuse(&cell_, count);
303 }
304
305 void EntryCell::SetTimestamp(int timestamp) {
306 SetCellTimestamp(&cell_, timestamp);
307 }
308
309 // Static.
310 EntryCell EntryCell::GetEntryCellForTest(int32 cell_id,
311 uint32 hash,
312 Addr address,
313 IndexCell* cell,
314 bool small_table) {
315 if (cell) {
316 EntryCell entry_cell(cell_id, hash, *cell, small_table);
317 return entry_cell;
318 }
319
320 return EntryCell(cell_id, hash, address, small_table);
321 }
322
323 void EntryCell::SerializaForTest(IndexCell* destination) {
324 FixSum();
325 Serialize(destination);
326 }
327
328 EntryCell::EntryCell() : cell_id_(0), hash_(0), small_table_(false) {
329 cell_.Clear();
330 }
331
332 EntryCell::EntryCell(int32 cell_id, uint32 hash, Addr address, bool small_table)
333 : cell_id_(cell_id),
334 hash_(hash),
335 small_table_(small_table) {
336 DCHECK(IsValidAddress(address) || !address.value());
337
338 cell_.Clear();
339 SetCellState(&cell_, ENTRY_NEW);
340 SetCellGroup(&cell_, ENTRY_NO_USE);
341 if (small_table) {
342 DCHECK(address.FileNumber() == kEntriesFile ||
343 address.FileNumber() == kEvictedEntriesFile);
344 SetCellSmallTableAddress(&cell_, address.start_block());
345 SetCellSmallTableHash(&cell_, hash >> kHashSmallTableShift);
346 } else {
347 uint32 cell_address = address.FileNumber() << 16 | address.start_block();
348 SetCellAddress(&cell_, cell_address);
349 SetCellHash(&cell_, hash >> kHashShift);
350 }
351 }
352
353 EntryCell::EntryCell(int32 cell_id,
354 uint32 hash,
355 const IndexCell& cell,
356 bool small_table)
357 : cell_id_(cell_id),
358 hash_(hash),
359 cell_(cell),
360 small_table_(small_table) {
361 }
362
363 void EntryCell::FixSum() {
364 SetCellSum(&cell_, CalculateCellSum(cell_));
365 }
366
367 uint32 EntryCell::GetAddressValue() const {
368 if (small_table_)
369 return GetCellSmallTableAddress(cell_);
370
371 return GetCellAddress(cell_);
372 }
373
374 uint32 EntryCell::RecomputeHash() {
375 if (small_table_) {
376 hash_ &= (1 << kHashSmallTableShift) - 1;
377 hash_ |= GetCellSmallTableHash(cell_) << kHashSmallTableShift;
378 return hash_;
379 }
380
381 hash_ &= (1 << kHashShift) - 1;
382 hash_ |= GetCellHash(cell_) << kHashShift;
383 return hash_;
384 }
385
386 void EntryCell::Serialize(IndexCell* destination) const {
387 *destination = cell_;
388 }
389
390 EntrySet::EntrySet() : evicted_count(0), current(0) {
391 }
392
393 EntrySet::~EntrySet() {
394 }
395
396 IndexIterator::IndexIterator() {
397 }
398
399 IndexIterator::~IndexIterator() {
400 }
401
402 IndexTableInitData::IndexTableInitData() {
403 }
404
405 IndexTableInitData::~IndexTableInitData() {
406 }
407
408 // -----------------------------------------------------------------------
409
410 IndexTable::IndexTable(IndexTableBackend* backend)
411 : backend_(backend),
412 header_(NULL),
413 main_table_(NULL),
414 extra_table_(NULL),
415 modified_(false),
416 small_table_(false) {
417 }
418
419 IndexTable::~IndexTable() {
420 }
421
422 // For a general description of the index tables see:
423 // http://www.chromium.org/developers/design-documents/network-stack/disk-cache/ disk-cache-v3#TOC-Index
424 //
425 // The index is split between two tables: the main_table_ and the extra_table_.
426 // The main table can grow only by doubling its number of cells, while the
427 // extra table can grow slowly, because it only contain cells that overflow
428 // from the main table. In order to locate a given cell, part of the hash is
429 // used directly as an index into the main table; once that bucket is located,
430 // all cells with that partial hash (i.e., belonging to that bucket) are
431 // inspected, and if present, the next bucket (located on the extra table) is
432 // then located. For more information on bucket chaining see:
433 // http://www.chromium.org/developers/design-documents/network-stack/disk-cache/ disk-cache-v3#TOC-Buckets
434 //
435 // There are two cases when increasing the size:
436 // - Doubling the size of the main table
437 // - Adding more entries to the extra table
438 //
439 // For example, consider a 64k main table with 8k cells on the extra table (for
440 // a total of 72k cells). Init can be called to add another 8k cells at the end
441 // (grow to 80k cells). When the size of the extra table approaches 64k, Init
442 // can be called to double the main table (to 128k) and go back to a small extra
443 // table.
444 void IndexTable::Init(IndexTableInitData* params) {
445 bool growing = header_ != NULL;
446 scoped_ptr<IndexBucket[]> old_extra_table;
447 header_ = &params->index_bitmap->header;
448
449 if (params->main_table) {
450 if (main_table_) {
451 // This is doubling the size of main table.
452 DCHECK_EQ(base::bits::Log2Floor(header_->table_len),
453 base::bits::Log2Floor(backup_header_->table_len) + 1);
454 int extra_size = (header()->max_bucket - mask_) * kCellsPerBucket;
455 DCHECK_GE(extra_size, 0);
456
457 // Doubling the size implies deleting the extra table and moving as many
458 // cells as we can to the main table, so we first copy the old one. This
459 // is not required when just growing the extra table because we don't
460 // move any cell in that case.
461 old_extra_table.reset(new IndexBucket[extra_size]);
462 memcpy(old_extra_table.get(), extra_table_,
463 extra_size * sizeof(IndexBucket));
464 memset(params->extra_table, 0, extra_size * sizeof(IndexBucket));
465 }
466 main_table_ = params->main_table;
467 }
468 DCHECK(main_table_);
469 extra_table_ = params->extra_table;
470
471 // extra_bits_ is really measured against table-size specific values.
472 const int kMaxAbsoluteExtraBits = 11; // From smaller to largest table.
473 const int kMaxExtraBitsSmallTable = 6;
474
475 extra_bits_ = base::bits::Log2Floor(header_->table_len) -
476 base::bits::Log2Floor(kBaseTableLen);
477 DCHECK_GE(extra_bits_, 0);
478 DCHECK_LE(extra_bits_, kMaxAbsoluteExtraBits);
479 mask_ = ((kBaseTableLen / kCellsPerBucket) << extra_bits_) - 1;
480 small_table_ = extra_bits_ < kMaxExtraBitsSmallTable;
481 if (!small_table_)
482 extra_bits_ -= kMaxExtraBitsSmallTable;
483
484 // table_len keeps the max number of cells stored by the index. We need a
485 // bitmap with 1 bit per cell, and that bitmap has num_words 32-bit words.
486 int num_words = (header_->table_len + 31) / 32;
487
488 if (old_extra_table) {
489 // All the cells from the extra table are moving to the new tables so before
490 // creating the bitmaps, clear the part of the bitmap referring to the extra
491 // table.
492 int main_table_bit_words = ((mask_ >> 1) + 1) * kCellsPerBucket / 32;
493 DCHECK_GT(num_words, main_table_bit_words);
494 memset(params->index_bitmap->bitmap + main_table_bit_words, 0,
495 (num_words - main_table_bit_words) * sizeof(int32));
496
497 DCHECK(growing);
498 int old_num_words = (backup_header_.get()->table_len + 31) / 32;
499 DCHECK_GT(old_num_words, main_table_bit_words);
500 memset(backup_bitmap_storage_.get() + main_table_bit_words, 0,
501 (old_num_words - main_table_bit_words) * sizeof(int32));
502 }
503 bitmap_.reset(new Bitmap(params->index_bitmap->bitmap, header_->table_len,
504 num_words));
505
506 if (growing) {
507 int old_num_words = (backup_header_.get()->table_len + 31) / 32;
508 DCHECK_GE(num_words, old_num_words);
509 scoped_ptr<uint32[]> storage(new uint32[num_words]);
510 memcpy(storage.get(), backup_bitmap_storage_.get(),
511 old_num_words * sizeof(int32));
512 memset(storage.get() + old_num_words, 0,
513 (num_words - old_num_words) * sizeof(int32));
514
515 backup_bitmap_storage_.swap(storage);
516 backup_header_->table_len = header_->table_len;
517 } else {
518 backup_bitmap_storage_.reset(params->backup_bitmap.release());
519 backup_header_.reset(params->backup_header.release());
520 }
521
522 num_words = (backup_header_->table_len + 31) / 32;
523 backup_bitmap_.reset(new Bitmap(backup_bitmap_storage_.get(),
524 backup_header_->table_len, num_words));
525 if (old_extra_table)
526 MoveCells(old_extra_table.get());
527
528 if (small_table_)
529 DCHECK(header_->flags & SMALL_CACHE);
530 }
531
532 void IndexTable::Reset() {
533 header_ = NULL;
534 main_table_ = NULL;
535 extra_table_ = NULL;
536 bitmap_.reset();
537 backup_bitmap_.reset();
538 backup_header_.reset();
539 backup_bitmap_storage_.reset();
540 modified_ = false;
541 }
542
543 // The general method for locating cells is to:
544 // 1. Get the first bucket. This usually means directly indexing the table (as
545 // this method does), or iterating through all possible buckets.
546 // 2. Iterate through all the cells in that first bucket.
547 // 3. If there is a linked bucket, locate it directly in the extra table.
548 // 4. Go back to 2, as needed.
549 //
550 // One consequence of this pattern is that we never start looking at buckets in
551 // the extra table, unless we are following a link from the main table.
552 EntrySet IndexTable::LookupEntries(uint32 hash) {
553 EntrySet entries;
554 int bucket_id = static_cast<int>(hash & mask_);
555 IndexBucket* bucket = &main_table_[bucket_id];
556 for (;;) {
557 for (int i = 0; i < kCellsPerBucket; i++) {
558 IndexCell* current_cell = &bucket->cells[i];
559 if (!GetAddressValue(*current_cell))
560 continue;
561 if (!SanityCheck(*current_cell)) {
562 NOTREACHED();
563 int cell_id = bucket_id * kCellsPerBucket + i;
564 current_cell->Clear();
565 bitmap_->Set(cell_id, false);
566 backup_bitmap_->Set(cell_id, false);
567 modified_ = true;
568 continue;
569 }
570 int cell_id = bucket_id * kCellsPerBucket + i;
571 if (MisplacedHash(*current_cell, hash)) {
572 HandleMisplacedCell(current_cell, cell_id, hash & mask_);
573 } else if (IsHashMatch(*current_cell, hash)) {
574 EntryCell entry_cell(cell_id, hash, *current_cell, small_table_);
575 CheckState(entry_cell);
576 if (entry_cell.GetState() != ENTRY_DELETED) {
577 entries.cells.push_back(entry_cell);
578 if (entry_cell.GetGroup() == ENTRY_EVICTED)
579 entries.evicted_count++;
580 }
581 }
582 }
583 bucket_id = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
584 &bucket);
585 if (!bucket_id)
586 break;
587 }
588 return entries;
589 }
590
591 EntryCell IndexTable::CreateEntryCell(uint32 hash, Addr address) {
592 DCHECK(IsValidAddress(address));
593 DCHECK(address.FileNumber() || address.start_block());
594
595 int bucket_id = static_cast<int>(hash & mask_);
596 int cell_id = 0;
597 IndexBucket* bucket = &main_table_[bucket_id];
598 IndexCell* current_cell = NULL;
599 bool found = false;
600 for (; !found;) {
601 for (int i = 0; i < kCellsPerBucket && !found; i++) {
602 current_cell = &bucket->cells[i];
603 if (!GetAddressValue(*current_cell)) {
604 cell_id = bucket_id * kCellsPerBucket + i;
605 found = true;
606 }
607 }
608 if (found)
609 break;
610 bucket_id = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
611 &bucket);
612 if (!bucket_id)
613 break;
614 }
615
616 if (!found) {
617 bucket_id = NewExtraBucket();
618 if (bucket_id) {
619 cell_id = bucket_id * kCellsPerBucket;
620 bucket->next = cell_id;
621 bucket = &extra_table_[bucket_id - (mask_ + 1)];
622 bucket->hash = hash & mask_;
623 found = true;
624 } else {
625 // address 0 is a reserved value, and the caller interprets it as invalid.
626 address.set_value(0);
627 }
628 }
629
630 EntryCell entry_cell(cell_id, hash, address, small_table_);
631 if (address.file_type() == BLOCK_EVICTED)
632 entry_cell.SetGroup(ENTRY_EVICTED);
633 else
634 entry_cell.SetGroup(ENTRY_NO_USE);
635 Save(&entry_cell);
636
637 if (found) {
638 bitmap_->Set(cell_id, true);
639 backup_bitmap_->Set(cell_id, true);
640 header()->used_cells++;
641 modified_ = true;
642 }
643
644 return entry_cell;
645 }
646
647 EntryCell IndexTable::FindEntryCell(uint32 hash, Addr address) {
648 return FindEntryCellImpl(hash, address, false);
649 }
650
651 int IndexTable::CalculateTimestamp(Time time) {
652 TimeDelta delta = time - Time::FromInternalValue(header_->base_time);
653 return std::max(delta.InMinutes(), 0);
654 }
655
656 base::Time IndexTable::TimeFromTimestamp(int timestamp) {
657 return Time::FromInternalValue(header_->base_time) +
658 TimeDelta::FromMinutes(timestamp);
659 }
660
661 void IndexTable::SetSate(uint32 hash, Addr address, EntryState state) {
662 EntryCell cell = FindEntryCellImpl(hash, address, state == ENTRY_FREE);
663 if (!cell.IsValid()) {
664 NOTREACHED();
665 return;
666 }
667
668 EntryState old_state = cell.GetState();
669 if (state == ENTRY_FREE) {
670 DCHECK_EQ(old_state, ENTRY_DELETED);
671 } else if (state == ENTRY_NEW) {
672 DCHECK_EQ(old_state, ENTRY_FREE);
673 } else if (state == ENTRY_OPEN) {
674 DCHECK_EQ(old_state, ENTRY_USED);
675 } else if (state == ENTRY_MODIFIED) {
676 DCHECK_EQ(old_state, ENTRY_OPEN);
677 } else if (state == ENTRY_DELETED) {
678 DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN ||
679 old_state == ENTRY_MODIFIED);
680 } else if (state == ENTRY_USED) {
681 DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN ||
682 old_state == ENTRY_MODIFIED);
683 }
684
685 modified_ = true;
686 if (state == ENTRY_DELETED) {
687 bitmap_->Set(cell.cell_id(), false);
688 backup_bitmap_->Set(cell.cell_id(), false);
689 } else if (state == ENTRY_FREE) {
690 cell.Clear();
691 Write(cell);
692 header()->used_cells--;
693 return;
694 }
695 cell.SetState(state);
696
697 Save(&cell);
698 }
699
700 void IndexTable::UpdateTime(uint32 hash, Addr address, base::Time current) {
701 EntryCell cell = FindEntryCell(hash, address);
702 if (!cell.IsValid())
703 return;
704
705 int minutes = CalculateTimestamp(current);
706
707 // Keep about 3 months of headroom.
708 const int kMaxTimestamp = (1 << 20) - 60 * 24 * 90;
709 if (minutes > kMaxTimestamp) {
710 // TODO(rvargas):
711 // Update header->old_time and trigger a timer
712 // Rebaseline timestamps and don't update sums
713 // Start a timer (about 2 backups)
714 // fix all ckecksums and trigger another timer
715 // update header->old_time because rebaseline is done.
716 minutes = std::min(minutes, (1 << 20) - 1);
717 }
718
719 cell.SetTimestamp(minutes);
720 Save(&cell);
721 }
722
723 void IndexTable::Save(EntryCell* cell) {
724 cell->FixSum();
725 Write(*cell);
726 }
727
728 void IndexTable::GetOldest(IndexIterator* no_use,
729 IndexIterator* low_use,
730 IndexIterator* high_use) {
731 no_use->forward = true;
732 low_use->forward = true;
733 high_use->forward = true;
734 InitIterator(no_use);
735 InitIterator(low_use);
736 InitIterator(high_use);
737
738 WalkTables(-1, no_use, low_use, high_use);
739 }
740
741 bool IndexTable::GetNextCells(IndexIterator* iterator) {
742 int current_time = iterator->timestamp;
743 InitIterator(iterator);
744
745 WalkTables(current_time, iterator, iterator, iterator);
746 return !iterator->cells.empty();
747 }
748
749 void IndexTable::OnBackupTimer() {
750 if (!modified_)
751 return;
752
753 int num_words = (header_->table_len + 31) / 32;
754 int num_bytes = num_words * 4 + static_cast<int>(sizeof(*header_));
755 scoped_refptr<net::IOBuffer> buffer(new net::IOBuffer(num_bytes));
756 memcpy(buffer->data(), header_, sizeof(*header_));
757 memcpy(buffer->data() + sizeof(*header_), backup_bitmap_storage_.get(),
758 num_words * 4);
759 backend_->SaveIndex(buffer, num_bytes);
760 modified_ = false;
761 }
762
763 // -----------------------------------------------------------------------
764
765 EntryCell IndexTable::FindEntryCellImpl(uint32 hash, Addr address,
766 bool allow_deleted) {
767 int bucket_id = static_cast<int>(hash & mask_);
768 IndexBucket* bucket = &main_table_[bucket_id];
769 for (;;) {
770 for (int i = 0; i < kCellsPerBucket; i++) {
771 IndexCell* current_cell = &bucket->cells[i];
772 if (!GetAddressValue(*current_cell))
773 continue;
774 DCHECK(SanityCheck(*current_cell));
775 if (IsHashMatch(*current_cell, hash)) {
776 // We have a match.
777 int cell_id = bucket_id * kCellsPerBucket + i;
778 EntryCell entry_cell(cell_id, hash, *current_cell, small_table_);
779 if (entry_cell.GetAddress() != address)
780 continue;
781
782 if (!allow_deleted && entry_cell.GetState() == ENTRY_DELETED)
783 continue;
784
785 return entry_cell;
786 }
787 }
788 bucket_id = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
789 &bucket);
790 if (!bucket_id)
791 break;
792 }
793 return EntryCell();
794 }
795
796 void IndexTable::CheckState(const EntryCell& cell) {
797 int current_state = cell.GetState();
798 if (current_state != ENTRY_FIXING) {
799 bool present = ((current_state & 3) != 0); // Look at the last two bits.
800 if (present != bitmap_->Get(cell.cell_id()) ||
801 present != backup_bitmap_->Get(cell.cell_id())) {
802 // There's a mismatch.
803 if (current_state == ENTRY_DELETED) {
804 // We were in the process of deleting this entry. Finish now.
805 backend_->DeleteCell(cell);
806 } else {
807 current_state = ENTRY_FIXING;
808 EntryCell bad_cell(cell);
809 bad_cell.SetState(ENTRY_FIXING);
810 Save(&bad_cell);
811 }
812 }
813 }
814
815 if (current_state == ENTRY_FIXING)
816 backend_->FixCell(cell);
817 }
818
819 void IndexTable::Write(const EntryCell& cell) {
820 IndexBucket* bucket = NULL;
821 int bucket_id = cell.cell_id() / kCellsPerBucket;
822 if (bucket_id < static_cast<int32>(mask_ + 1)) {
823 bucket = &main_table_[bucket_id];
824 } else {
825 DCHECK_LE(bucket_id, header()->max_bucket);
826 bucket = &extra_table_[bucket_id - (mask_ + 1)];
827 }
828
829 int cell_number = cell.cell_id() % kCellsPerBucket;
830 if (GetAddressValue(bucket->cells[cell_number]) && cell.GetAddressValue()) {
831 DCHECK_EQ(cell.GetAddressValue(),
832 GetAddressValue(bucket->cells[cell_number]));
833 }
834 cell.Serialize(&bucket->cells[cell_number]);
835 }
836
837 int IndexTable::NewExtraBucket() {
838 int safe_window = (header()->table_len < kNumExtraBlocks * 2) ?
839 kNumExtraBlocks / 4 : kNumExtraBlocks;
840 if (header()->table_len - header()->max_bucket * kCellsPerBucket <
841 safe_window) {
842 backend_->GrowIndex();
843 }
844
845 if (header()->max_bucket * kCellsPerBucket ==
846 header()->table_len - kCellsPerBucket) {
847 return 0;
848 }
849
850 header()->max_bucket++;
851 return header()->max_bucket;
852 }
853
854 void IndexTable::WalkTables(int limit_time,
855 IndexIterator* no_use,
856 IndexIterator* low_use,
857 IndexIterator* high_use) {
858 header_->num_no_use_entries = 0;
859 header_->num_low_use_entries = 0;
860 header_->num_high_use_entries = 0;
861 header_->num_evicted_entries = 0;
862
863 for (int i = 0; i < static_cast<int32>(mask_ + 1); i++) {
864 int bucket_id = i;
865 IndexBucket* bucket = &main_table_[i];
866 for (;;) {
867 UpdateFromBucket(bucket, i, limit_time, no_use, low_use, high_use);
868
869 bucket_id = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
870 &bucket);
871 if (!bucket_id)
872 break;
873 }
874 }
875 header_->num_entries = header_->num_no_use_entries +
876 header_->num_low_use_entries +
877 header_->num_high_use_entries +
878 header_->num_evicted_entries;
879 modified_ = true;
880 }
881
882 void IndexTable::UpdateFromBucket(IndexBucket* bucket, int bucket_hash,
883 int limit_time,
884 IndexIterator* no_use,
885 IndexIterator* low_use,
886 IndexIterator* high_use) {
887 for (int i = 0; i < kCellsPerBucket; i++) {
888 IndexCell& current_cell = bucket->cells[i];
889 if (!GetAddressValue(current_cell))
890 continue;
891 DCHECK(SanityCheck(current_cell));
892 if (!IsNormalState(current_cell))
893 continue;
894
895 EntryCell entry_cell(0, GetFullHash(current_cell, bucket_hash),
896 current_cell, small_table_);
897 switch (GetCellGroup(current_cell)) {
898 case ENTRY_NO_USE:
899 UpdateIterator(entry_cell, limit_time, no_use);
900 header_->num_no_use_entries++;
901 break;
902 case ENTRY_LOW_USE:
903 UpdateIterator(entry_cell, limit_time, low_use);
904 header_->num_low_use_entries++;
905 break;
906 case ENTRY_HIGH_USE:
907 UpdateIterator(entry_cell, limit_time, high_use);
908 header_->num_high_use_entries++;
909 break;
910 case ENTRY_EVICTED:
911 header_->num_evicted_entries++;
912 break;
913 default:
914 NOTREACHED();
915 }
916 }
917 }
918
919 void IndexTable::MoveCells(IndexBucket* old_extra_table) {
920 int max_hash = (mask_ + 1) / 2;
921 int max_bucket = header()->max_bucket;
922 header()->max_bucket = mask_;
923 int used_cells = header()->used_cells;
924
925 // Consider a large cache: a cell stores the upper 18 bits of the hash
926 // (h >> 14). If the table is say 8 times the original size (growing from 4x),
927 // the bit that we are interested in would be the 3rd bit of the stored value,
928 // in other words 'multiplier' >> 1.
929 uint32 new_bit = (1 << extra_bits_) >> 1;
930
931 scoped_ptr<IndexBucket[]> old_main_table;
932 IndexBucket* source_table = main_table_;
933 bool upgrade_format = !extra_bits_;
934 if (upgrade_format) {
935 // This method should deal with migrating a small table to a big one. Given
936 // that the first thing to do is read the old table, set small_table_ for
937 // the size of the old table. Now, when moving a cell, the result cannot be
938 // placed in the old table or we will end up reading it again and attempting
939 // to move it, so we have to copy the whole table at once.
940 DCHECK(!small_table_);
941 small_table_ = true;
942 old_main_table.reset(new IndexBucket[max_hash]);
943 memcpy(old_main_table.get(), main_table_, max_hash * sizeof(IndexBucket));
944 memset(main_table_, 0, max_hash * sizeof(IndexBucket));
945 source_table = old_main_table.get();
946 }
947
948 for (int i = 0; i < max_hash; i++) {
949 int bucket_id = i;
950 IndexBucket* bucket = &source_table[i];
951 for (;;) {
952 for (int j = 0; j < kCellsPerBucket; j++) {
953 IndexCell& current_cell = bucket->cells[j];
954 if (!GetAddressValue(current_cell))
955 continue;
956 DCHECK(SanityCheck(current_cell));
957 if (bucket_id == i) {
958 if (upgrade_format || (GetHashValue(current_cell) & new_bit)) {
959 // Move this cell to the upper half of the table.
960 MoveSingleCell(&current_cell, bucket_id * kCellsPerBucket + j, i,
961 true);
962 }
963 } else {
964 // All cells on extra buckets have to move.
965 MoveSingleCell(&current_cell, bucket_id * kCellsPerBucket + j, i,
966 true);
967 }
968 }
969
970 bucket_id = GetNextBucket(max_hash, max_bucket, old_extra_table, &bucket);
971 if (!bucket_id)
972 break;
973 }
974 }
975
976 DCHECK_EQ(header()->used_cells, used_cells);
977
978 if (upgrade_format) {
979 small_table_ = false;
980 header()->flags &= ~SMALL_CACHE;
981 }
982 }
983
984 void IndexTable::MoveSingleCell(IndexCell* current_cell, int cell_id,
985 int main_table_index, bool growing) {
986 uint32 hash = GetFullHash(*current_cell, main_table_index);
987 EntryCell old_cell(cell_id, hash, *current_cell, small_table_);
988
989 bool upgrade_format = !extra_bits_ && growing;
990 if (upgrade_format)
991 small_table_ = false;
992 EntryCell new_cell = CreateEntryCell(hash, old_cell.GetAddress());
993
994 if (!new_cell.IsValid()) {
995 // We'll deal with this entry later.
996 if (upgrade_format)
997 small_table_ = true;
998 return;
999 }
1000
1001 new_cell.SetState(old_cell.GetState());
1002 new_cell.SetGroup(old_cell.GetGroup());
1003 new_cell.SetReuse(old_cell.GetReuse());
1004 new_cell.SetTimestamp(old_cell.GetTimestamp());
1005 Save(&new_cell);
1006 modified_ = true;
1007 if (upgrade_format)
1008 small_table_ = true;
1009
1010 if (old_cell.GetState() == ENTRY_DELETED) {
1011 bitmap_->Set(new_cell.cell_id(), false);
1012 backup_bitmap_->Set(new_cell.cell_id(), false);
1013 }
1014
1015 if (!growing || cell_id / kCellsPerBucket == main_table_index) {
1016 // Only delete entries that live on the main table.
1017 if (!upgrade_format) {
1018 old_cell.Clear();
1019 Write(old_cell);
1020 }
1021
1022 if (cell_id != new_cell.cell_id()) {
1023 bitmap_->Set(old_cell.cell_id(), false);
1024 backup_bitmap_->Set(old_cell.cell_id(), false);
1025 }
1026 }
1027 header()->used_cells--;
1028 }
1029
1030 void IndexTable::HandleMisplacedCell(IndexCell* current_cell, int cell_id,
1031 int main_table_index) {
1032 // The cell may be misplaced, or a duplicate cell exists with this data.
1033 uint32 hash = GetFullHash(*current_cell, main_table_index);
1034 MoveSingleCell(current_cell, cell_id, main_table_index, false);
1035
1036 // Now look for a duplicate cell.
1037 CheckBucketList(hash & mask_);
1038 }
1039
1040 void IndexTable::CheckBucketList(int bucket_id) {
1041 typedef std::pair<int, EntryGroup> AddressAndGroup;
1042 std::set<AddressAndGroup> entries;
1043 IndexBucket* bucket = &main_table_[bucket_id];
1044 int bucket_hash = bucket_id;
1045 for (;;) {
1046 for (int i = 0; i < kCellsPerBucket; i++) {
1047 IndexCell* current_cell = &bucket->cells[i];
1048 if (!GetAddressValue(*current_cell))
1049 continue;
1050 if (!SanityCheck(*current_cell)) {
1051 NOTREACHED();
1052 current_cell->Clear();
1053 continue;
1054 }
1055 int cell_id = bucket_id * kCellsPerBucket + i;
1056 EntryCell cell(cell_id, GetFullHash(*current_cell, bucket_hash),
1057 *current_cell, small_table_);
1058 if (!entries.insert(std::make_pair(cell.GetAddress().value(),
1059 cell.GetGroup())).second) {
1060 current_cell->Clear();
1061 continue;
1062 }
1063 CheckState(cell);
1064 }
1065
1066 bucket_id = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
1067 &bucket);
1068 if (!bucket_id)
1069 break;
1070 }
1071 }
1072
1073 uint32 IndexTable::GetAddressValue(const IndexCell& cell) {
1074 if (small_table_)
1075 return GetCellSmallTableAddress(cell);
1076
1077 return GetCellAddress(cell);
1078 }
1079
1080 uint32 IndexTable::GetHashValue(const IndexCell& cell) {
1081 if (small_table_)
1082 return GetCellSmallTableHash(cell);
1083
1084 return GetCellHash(cell);
1085 }
1086
1087 uint32 IndexTable::GetFullHash(const IndexCell& cell, uint32 lower_part) {
1088 // It is OK for the high order bits of lower_part to overlap with the stored
1089 // part of the hash.
1090 if (small_table_)
1091 return (GetCellSmallTableHash(cell) << kHashSmallTableShift) | lower_part;
1092
1093 return (GetCellHash(cell) << kHashShift) | lower_part;
1094 }
1095
1096 // All the bits stored in the cell should match the provided hash.
1097 bool IndexTable::IsHashMatch(const IndexCell& cell, uint32 hash) {
1098 hash = small_table_ ? hash >> kHashSmallTableShift : hash >> kHashShift;
1099 return GetHashValue(cell) == hash;
1100 }
1101
1102 // A partial match ignores the redundant bits stored in the cell. In other
1103 // words, returns true for misplaced cells as well as cells that have a regular
1104 // hash match.
1105 bool IndexTable::IsPartialHashMatch(const IndexCell& cell, uint32 hash) {
1106 hash = small_table_ ? hash >> kHashSmallTableShift : hash >> kHashShift;
1107 return (GetHashValue(cell) >> extra_bits_) == (hash >> extra_bits_);
1108 }
1109
1110 bool IndexTable::MisplacedHash(const IndexCell& cell, uint32 hash) {
1111 if (!extra_bits_)
1112 return false;
1113
1114 uint32 mask = (1 << extra_bits_) - 1;
1115 hash = small_table_ ? hash >> kHashSmallTableShift : hash >> kHashShift;
1116 return (GetHashValue(cell) & mask) != (hash & mask);
1117 }
1118
1119 // Things we can be doing:
1120 //
1121 // - Updating timestamps
1122 // Fast, but we go through a few backup cycles to make sure it sticks
1123 // - Growing the extra table -> increase bitmap size, remap
1124 // Just a matter of tripping to the cache thread. Everything keeps moving
1125 // - Growing the whole table -> relocating all entries, may take a while
1126 // - Evictions... only when we are NOT doing something else? just find the
1127 // entries again so that there's no invalidation of lists
1128
1129 } // namespace disk_cache
OLDNEW
« no previous file with comments | « net/disk_cache/v3/index_table.h ('k') | net/disk_cache/v3/index_table_unittest.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698