• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2014 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/blockfile/index_table_v3.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 // The following constants describe the bitfields of an IndexCell so they are
26 // implicitly synchronized with the descrption of IndexCell on file_format_v3.h.
27 const uint64 kCellLocationMask = (1 << 22) - 1;
28 const uint64 kCellIdMask = (1 << 18) - 1;
29 const uint64 kCellTimestampMask = (1 << 20) - 1;
30 const uint64 kCellReuseMask = (1 << 4) - 1;
31 const uint8 kCellStateMask = (1 << 3) - 1;
32 const uint8 kCellGroupMask = (1 << 3) - 1;
33 const uint8 kCellSumMask = (1 << 2) - 1;
34 
35 const uint64 kCellSmallTableLocationMask = (1 << 16) - 1;
36 const uint64 kCellSmallTableIdMask = (1 << 24) - 1;
37 
38 const int kCellIdOffset = 22;
39 const int kCellTimestampOffset = 40;
40 const int kCellReuseOffset = 60;
41 const int kCellGroupOffset = 3;
42 const int kCellSumOffset = 6;
43 
44 const int kCellSmallTableIdOffset = 16;
45 
46 // The number of bits that a hash has to be shifted to grab the part that
47 // defines the cell id.
48 const int kHashShift = 14;
49 const int kSmallTableHashShift = 8;
50 
51 // Unfortunately we have to break the abstaction a little here: the file number
52 // where entries are stored is outside of the control of this code, and it is
53 // usually part of the stored address. However, for small tables we only store
54 // 16 bits of the address so the file number is never stored on a cell. We have
55 // to infere the file number from the type of entry (normal vs evicted), and
56 // the knowledge that given that the table will not keep more than 64k entries,
57 // a single file of each type is enough.
58 const int kEntriesFile = disk_cache::BLOCK_ENTRIES - 1;
59 const int kEvictedEntriesFile = disk_cache::BLOCK_EVICTED - 1;
60 const int kMaxLocation = 1 << 22;
61 const int kMinFileNumber = 1 << 16;
62 
GetCellLocation(const IndexCell & cell)63 uint32 GetCellLocation(const IndexCell& cell) {
64   return cell.first_part & kCellLocationMask;
65 }
66 
GetCellSmallTableLocation(const IndexCell & cell)67 uint32 GetCellSmallTableLocation(const IndexCell& cell) {
68   return cell.first_part & kCellSmallTableLocationMask;
69 }
70 
GetCellId(const IndexCell & cell)71 uint32 GetCellId(const IndexCell& cell) {
72   return (cell.first_part >> kCellIdOffset) & kCellIdMask;
73 }
74 
GetCellSmallTableId(const IndexCell & cell)75 uint32 GetCellSmallTableId(const IndexCell& cell) {
76   return (cell.first_part >> kCellSmallTableIdOffset) &
77          kCellSmallTableIdMask;
78 }
79 
GetCellTimestamp(const IndexCell & cell)80 int GetCellTimestamp(const IndexCell& cell) {
81   return (cell.first_part >> kCellTimestampOffset) & kCellTimestampMask;
82 }
83 
GetCellReuse(const IndexCell & cell)84 int GetCellReuse(const IndexCell& cell) {
85   return (cell.first_part >> kCellReuseOffset) & kCellReuseMask;
86 }
87 
GetCellState(const IndexCell & cell)88 int GetCellState(const IndexCell& cell) {
89   return cell.last_part & kCellStateMask;
90 }
91 
GetCellGroup(const IndexCell & cell)92 int GetCellGroup(const IndexCell& cell) {
93   return (cell.last_part >> kCellGroupOffset) & kCellGroupMask;
94 }
95 
GetCellSum(const IndexCell & cell)96 int GetCellSum(const IndexCell& cell) {
97   return (cell.last_part >> kCellSumOffset) & kCellSumMask;
98 }
99 
SetCellLocation(IndexCell * cell,uint32 address)100 void SetCellLocation(IndexCell* cell, uint32 address) {
101   DCHECK_LE(address, static_cast<uint32>(kCellLocationMask));
102   cell->first_part &= ~kCellLocationMask;
103   cell->first_part |= address;
104 }
105 
SetCellSmallTableLocation(IndexCell * cell,uint32 address)106 void SetCellSmallTableLocation(IndexCell* cell, uint32 address) {
107   DCHECK_LE(address, static_cast<uint32>(kCellSmallTableLocationMask));
108   cell->first_part &= ~kCellSmallTableLocationMask;
109   cell->first_part |= address;
110 }
111 
SetCellId(IndexCell * cell,uint32 hash)112 void SetCellId(IndexCell* cell, uint32 hash) {
113   DCHECK_LE(hash, static_cast<uint32>(kCellIdMask));
114   cell->first_part &= ~(kCellIdMask << kCellIdOffset);
115   cell->first_part |= static_cast<int64>(hash) << kCellIdOffset;
116 }
117 
SetCellSmallTableId(IndexCell * cell,uint32 hash)118 void SetCellSmallTableId(IndexCell* cell, uint32 hash) {
119   DCHECK_LE(hash, static_cast<uint32>(kCellSmallTableIdMask));
120   cell->first_part &= ~(kCellSmallTableIdMask << kCellSmallTableIdOffset);
121   cell->first_part |= static_cast<int64>(hash) << kCellSmallTableIdOffset;
122 }
123 
SetCellTimestamp(IndexCell * cell,int timestamp)124 void SetCellTimestamp(IndexCell* cell, int timestamp) {
125   DCHECK_LT(timestamp, 1 << 20);
126   DCHECK_GE(timestamp, 0);
127   cell->first_part &= ~(kCellTimestampMask << kCellTimestampOffset);
128   cell->first_part |= static_cast<int64>(timestamp) << kCellTimestampOffset;
129 }
130 
SetCellReuse(IndexCell * cell,int count)131 void SetCellReuse(IndexCell* cell, int count) {
132   DCHECK_LT(count, 16);
133   DCHECK_GE(count, 0);
134   cell->first_part &= ~(kCellReuseMask << kCellReuseOffset);
135   cell->first_part |= static_cast<int64>(count) << kCellReuseOffset;
136 }
137 
SetCellState(IndexCell * cell,disk_cache::EntryState state)138 void SetCellState(IndexCell* cell, disk_cache::EntryState state) {
139   cell->last_part &= ~kCellStateMask;
140   cell->last_part |= state;
141 }
142 
SetCellGroup(IndexCell * cell,disk_cache::EntryGroup group)143 void SetCellGroup(IndexCell* cell, disk_cache::EntryGroup group) {
144   cell->last_part &= ~(kCellGroupMask << kCellGroupOffset);
145   cell->last_part |= group << kCellGroupOffset;
146 }
147 
SetCellSum(IndexCell * cell,int sum)148 void SetCellSum(IndexCell* cell, int sum) {
149   DCHECK_LT(sum, 4);
150   DCHECK_GE(sum, 0);
151   cell->last_part &= ~(kCellSumMask << kCellSumOffset);
152   cell->last_part |= sum << kCellSumOffset;
153 }
154 
155 // This is a very particular way to calculate the sum, so it will not match if
156 // compared a gainst a pure 2 bit, modulo 2 sum.
CalculateCellSum(const IndexCell & cell)157 int CalculateCellSum(const IndexCell& cell) {
158   uint32* words = bit_cast<uint32*>(&cell);
159   uint8* bytes = bit_cast<uint8*>(&cell);
160   uint32 result = words[0] + words[1];
161   result += result >> 16;
162   result += (result >> 8) + (bytes[8] & 0x3f);
163   result += result >> 4;
164   result += result >> 2;
165   return result & 3;
166 }
167 
SanityCheck(const IndexCell & cell)168 bool SanityCheck(const IndexCell& cell) {
169   if (GetCellSum(cell) != CalculateCellSum(cell))
170     return false;
171 
172   if (GetCellState(cell) > disk_cache::ENTRY_USED ||
173       GetCellGroup(cell) == disk_cache::ENTRY_RESERVED ||
174       GetCellGroup(cell) > disk_cache::ENTRY_EVICTED) {
175     return false;
176   }
177 
178   return true;
179 }
180 
FileNumberFromLocation(int location)181 int FileNumberFromLocation(int location) {
182   return location / kMinFileNumber;
183 }
184 
StartBlockFromLocation(int location)185 int StartBlockFromLocation(int location) {
186   return location % kMinFileNumber;
187 }
188 
IsValidAddress(disk_cache::Addr address)189 bool IsValidAddress(disk_cache::Addr address) {
190   if (!address.is_initialized() ||
191       (address.file_type() != disk_cache::BLOCK_EVICTED &&
192        address.file_type() != disk_cache::BLOCK_ENTRIES)) {
193     return false;
194   }
195 
196   return address.FileNumber() < FileNumberFromLocation(kMaxLocation);
197 }
198 
IsNormalState(const IndexCell & cell)199 bool IsNormalState(const IndexCell& cell) {
200   disk_cache::EntryState state =
201       static_cast<disk_cache::EntryState>(GetCellState(cell));
202   DCHECK_NE(state, disk_cache::ENTRY_FREE);
203   return state != disk_cache::ENTRY_DELETED &&
204          state != disk_cache::ENTRY_FIXING;
205 }
206 
GetNextBucket(int min_bucket_num,int max_bucket_num,disk_cache::IndexBucket * table,disk_cache::IndexBucket ** bucket)207 inline int GetNextBucket(int min_bucket_num, int max_bucket_num,
208                          disk_cache::IndexBucket* table,
209                          disk_cache::IndexBucket** bucket) {
210   if (!(*bucket)->next)
211     return 0;
212 
213   int bucket_num = (*bucket)->next / disk_cache::kCellsPerBucket;
214   if (bucket_num < min_bucket_num || bucket_num > max_bucket_num) {
215     // The next bucket must fall within the extra table. Note that this is not
216     // an uncommon path as growing the table may not cleanup the link from the
217     // main table to the extra table, and that cleanup is performed here when
218     // accessing that bucket for the first time. This behavior has to change if
219     // the tables are ever shrinked.
220     (*bucket)->next = 0;
221     return 0;
222   }
223   *bucket = &table[bucket_num - min_bucket_num];
224   return bucket_num;
225 }
226 
227 // Updates the |iterator| with the current |cell|. This cell may cause all
228 // previous cells to be deleted (when a new target timestamp is found), the cell
229 // may be added to the list (if it matches the target timestamp), or may it be
230 // ignored.
UpdateIterator(const disk_cache::EntryCell & cell,int limit_time,IndexIterator * iterator)231 void UpdateIterator(const disk_cache::EntryCell& cell,
232                     int limit_time,
233                     IndexIterator* iterator) {
234   int time = cell.GetTimestamp();
235   // Look for not interesting times.
236   if (iterator->forward && time <= limit_time)
237     return;
238   if (!iterator->forward && time >= limit_time)
239     return;
240 
241   if ((iterator->forward && time < iterator->timestamp) ||
242       (!iterator->forward && time > iterator->timestamp)) {
243     // This timestamp is better than the one we had.
244     iterator->timestamp = time;
245     iterator->cells.clear();
246   }
247   if (time == iterator->timestamp) {
248     CellInfo cell_info = { cell.hash(), cell.GetAddress() };
249     iterator->cells.push_back(cell_info);
250   }
251 }
252 
InitIterator(IndexIterator * iterator)253 void InitIterator(IndexIterator* iterator) {
254   iterator->cells.clear();
255   iterator->timestamp = iterator->forward ? kint32max : 0;
256 }
257 
258 }  // namespace
259 
260 namespace disk_cache {
261 
~EntryCell()262 EntryCell::~EntryCell() {
263 }
264 
IsValid() const265 bool EntryCell::IsValid() const {
266   return GetCellLocation(cell_) != 0;
267 }
268 
269 // This code has to map the cell address (up to 22 bits) to a general cache Addr
270 // (up to 24 bits of general addressing). It also set the implied file_number
271 // in the case of small tables. See also the comment by the definition of
272 // kEntriesFile.
GetAddress() const273 Addr EntryCell::GetAddress() const {
274   uint32 location = GetLocation();
275   int file_number = FileNumberFromLocation(location);
276   if (small_table_) {
277     DCHECK_EQ(0, file_number);
278     file_number = (GetGroup() == ENTRY_EVICTED) ? kEvictedEntriesFile :
279                                                   kEntriesFile;
280   }
281   DCHECK_NE(0, file_number);
282   FileType file_type = (GetGroup() == ENTRY_EVICTED) ? BLOCK_EVICTED :
283                                                        BLOCK_ENTRIES;
284   return Addr(file_type, 1, file_number, StartBlockFromLocation(location));
285 }
286 
GetState() const287 EntryState EntryCell::GetState() const {
288   return static_cast<EntryState>(GetCellState(cell_));
289 }
290 
GetGroup() const291 EntryGroup EntryCell::GetGroup() const {
292   return static_cast<EntryGroup>(GetCellGroup(cell_));
293 }
294 
GetReuse() const295 int EntryCell::GetReuse() const {
296   return GetCellReuse(cell_);
297 }
298 
GetTimestamp() const299 int EntryCell::GetTimestamp() const {
300   return GetCellTimestamp(cell_);
301 }
302 
SetState(EntryState state)303 void EntryCell::SetState(EntryState state) {
304   SetCellState(&cell_, state);
305 }
306 
SetGroup(EntryGroup group)307 void EntryCell::SetGroup(EntryGroup group) {
308   SetCellGroup(&cell_, group);
309 }
310 
SetReuse(int count)311 void EntryCell::SetReuse(int count) {
312   SetCellReuse(&cell_, count);
313 }
314 
SetTimestamp(int timestamp)315 void EntryCell::SetTimestamp(int timestamp) {
316   SetCellTimestamp(&cell_, timestamp);
317 }
318 
319 // Static.
GetEntryCellForTest(int32 cell_num,uint32 hash,Addr address,IndexCell * cell,bool small_table)320 EntryCell EntryCell::GetEntryCellForTest(int32 cell_num,
321                                          uint32 hash,
322                                          Addr address,
323                                          IndexCell* cell,
324                                          bool small_table) {
325   if (cell) {
326     EntryCell entry_cell(cell_num, hash, *cell, small_table);
327     return entry_cell;
328   }
329 
330   return EntryCell(cell_num, hash, address, small_table);
331 }
332 
SerializaForTest(IndexCell * destination)333 void EntryCell::SerializaForTest(IndexCell* destination) {
334   FixSum();
335   Serialize(destination);
336 }
337 
EntryCell()338 EntryCell::EntryCell() : cell_num_(0), hash_(0), small_table_(false) {
339   cell_.Clear();
340 }
341 
EntryCell(int32 cell_num,uint32 hash,Addr address,bool small_table)342 EntryCell::EntryCell(int32 cell_num,
343                      uint32 hash,
344                      Addr address,
345                      bool small_table)
346     : cell_num_(cell_num),
347       hash_(hash),
348       small_table_(small_table) {
349   DCHECK(IsValidAddress(address) || !address.value());
350 
351   cell_.Clear();
352   SetCellState(&cell_, ENTRY_NEW);
353   SetCellGroup(&cell_, ENTRY_NO_USE);
354   if (small_table) {
355     DCHECK(address.FileNumber() == kEntriesFile ||
356            address.FileNumber() == kEvictedEntriesFile);
357     SetCellSmallTableLocation(&cell_, address.start_block());
358     SetCellSmallTableId(&cell_, hash >> kSmallTableHashShift);
359   } else {
360     uint32 location = address.FileNumber() << 16 | address.start_block();
361     SetCellLocation(&cell_, location);
362     SetCellId(&cell_, hash >> kHashShift);
363   }
364 }
365 
EntryCell(int32 cell_num,uint32 hash,const IndexCell & cell,bool small_table)366 EntryCell::EntryCell(int32 cell_num,
367                      uint32 hash,
368                      const IndexCell& cell,
369                      bool small_table)
370     : cell_num_(cell_num),
371       hash_(hash),
372       cell_(cell),
373       small_table_(small_table) {
374 }
375 
FixSum()376 void EntryCell::FixSum() {
377   SetCellSum(&cell_, CalculateCellSum(cell_));
378 }
379 
GetLocation() const380 uint32 EntryCell::GetLocation() const {
381   if (small_table_)
382     return GetCellSmallTableLocation(cell_);
383 
384   return GetCellLocation(cell_);
385 }
386 
RecomputeHash()387 uint32 EntryCell::RecomputeHash() {
388   if (small_table_) {
389     hash_ &= (1 << kSmallTableHashShift) - 1;
390     hash_ |= GetCellSmallTableId(cell_) << kSmallTableHashShift;
391     return hash_;
392   }
393 
394   hash_ &= (1 << kHashShift) - 1;
395   hash_ |= GetCellId(cell_) << kHashShift;
396   return hash_;
397 }
398 
Serialize(IndexCell * destination) const399 void EntryCell::Serialize(IndexCell* destination) const {
400   *destination = cell_;
401 }
402 
EntrySet()403 EntrySet::EntrySet() : evicted_count(0), current(0) {
404 }
405 
~EntrySet()406 EntrySet::~EntrySet() {
407 }
408 
IndexIterator()409 IndexIterator::IndexIterator() {
410 }
411 
~IndexIterator()412 IndexIterator::~IndexIterator() {
413 }
414 
IndexTableInitData()415 IndexTableInitData::IndexTableInitData() {
416 }
417 
~IndexTableInitData()418 IndexTableInitData::~IndexTableInitData() {
419 }
420 
421 // -----------------------------------------------------------------------
422 
IndexTable(IndexTableBackend * backend)423 IndexTable::IndexTable(IndexTableBackend* backend)
424     : backend_(backend),
425       header_(NULL),
426       main_table_(NULL),
427       extra_table_(NULL),
428       modified_(false),
429       small_table_(false) {
430 }
431 
~IndexTable()432 IndexTable::~IndexTable() {
433 }
434 
435 // For a general description of the index tables see:
436 // http://www.chromium.org/developers/design-documents/network-stack/disk-cache/disk-cache-v3#TOC-Index
437 //
438 // The index is split between two tables: the main_table_ and the extra_table_.
439 // The main table can grow only by doubling its number of cells, while the
440 // extra table can grow slowly, because it only contain cells that overflow
441 // from the main table. In order to locate a given cell, part of the hash is
442 // used directly as an index into the main table; once that bucket is located,
443 // all cells with that partial hash (i.e., belonging to that bucket) are
444 // inspected, and if present, the next bucket (located on the extra table) is
445 // then located. For more information on bucket chaining see:
446 // http://www.chromium.org/developers/design-documents/network-stack/disk-cache/disk-cache-v3#TOC-Buckets
447 //
448 // There are two cases when increasing the size:
449 //  - Doubling the size of the main table
450 //  - Adding more entries to the extra table
451 //
452 // For example, consider a 64k main table with 8k cells on the extra table (for
453 // a total of 72k cells). Init can be called to add another 8k cells at the end
454 // (grow to 80k cells). When the size of the extra table approaches 64k, Init
455 // can be called to double the main table (to 128k) and go back to a small extra
456 // table.
Init(IndexTableInitData * params)457 void IndexTable::Init(IndexTableInitData* params) {
458   bool growing = header_ != NULL;
459   scoped_ptr<IndexBucket[]> old_extra_table;
460   header_ = &params->index_bitmap->header;
461 
462   if (params->main_table) {
463     if (main_table_) {
464       // This is doubling the size of main table.
465       DCHECK_EQ(base::bits::Log2Floor(header_->table_len),
466                 base::bits::Log2Floor(backup_header_->table_len) + 1);
467       int extra_size = (header()->max_bucket - mask_) * kCellsPerBucket;
468       DCHECK_GE(extra_size, 0);
469 
470       // Doubling the size implies deleting the extra table and moving as many
471       // cells as we can to the main table, so we first copy the old one. This
472       // is not required when just growing the extra table because we don't
473       // move any cell in that case.
474       old_extra_table.reset(new IndexBucket[extra_size]);
475       memcpy(old_extra_table.get(), extra_table_,
476              extra_size * sizeof(IndexBucket));
477       memset(params->extra_table, 0, extra_size * sizeof(IndexBucket));
478     }
479     main_table_ = params->main_table;
480   }
481   DCHECK(main_table_);
482   extra_table_ = params->extra_table;
483 
484   // extra_bits_ is really measured against table-size specific values.
485   const int kMaxAbsoluteExtraBits = 12;  // From smallest to largest table.
486   const int kMaxExtraBitsSmallTable = 6;  // From smallest to 64K table.
487 
488   extra_bits_ = base::bits::Log2Floor(header_->table_len) -
489                 base::bits::Log2Floor(kBaseTableLen);
490   DCHECK_GE(extra_bits_, 0);
491   DCHECK_LT(extra_bits_, kMaxAbsoluteExtraBits);
492 
493   // Note that following the previous code the constants could be derived as
494   // kMaxAbsoluteExtraBits = base::bits::Log2Floor(max table len) -
495   //                         base::bits::Log2Floor(kBaseTableLen);
496   //                       = 22 - base::bits::Log2Floor(1024) = 22 - 10;
497   // kMaxExtraBitsSmallTable = base::bits::Log2Floor(max 16 bit table) - 10.
498 
499   mask_ = ((kBaseTableLen / kCellsPerBucket) << extra_bits_) - 1;
500   small_table_ = extra_bits_ < kMaxExtraBitsSmallTable;
501   if (!small_table_)
502     extra_bits_ -= kMaxExtraBitsSmallTable;
503 
504   // table_len keeps the max number of cells stored by the index. We need a
505   // bitmap with 1 bit per cell, and that bitmap has num_words 32-bit words.
506   int num_words = (header_->table_len + 31) / 32;
507 
508   if (old_extra_table) {
509     // All the cells from the extra table are moving to the new tables so before
510     // creating the bitmaps, clear the part of the bitmap referring to the extra
511     // table.
512     int old_main_table_bit_words = ((mask_ >> 1) + 1) * kCellsPerBucket / 32;
513     DCHECK_GT(num_words, old_main_table_bit_words);
514     memset(params->index_bitmap->bitmap + old_main_table_bit_words, 0,
515            (num_words - old_main_table_bit_words) * sizeof(int32));
516 
517     DCHECK(growing);
518     int old_num_words = (backup_header_.get()->table_len + 31) / 32;
519     DCHECK_GT(old_num_words, old_main_table_bit_words);
520     memset(backup_bitmap_storage_.get() + old_main_table_bit_words, 0,
521            (old_num_words - old_main_table_bit_words) * sizeof(int32));
522   }
523   bitmap_.reset(new Bitmap(params->index_bitmap->bitmap, header_->table_len,
524                            num_words));
525 
526   if (growing) {
527     int old_num_words = (backup_header_.get()->table_len + 31) / 32;
528     DCHECK_GE(num_words, old_num_words);
529     scoped_ptr<uint32[]> storage(new uint32[num_words]);
530     memcpy(storage.get(), backup_bitmap_storage_.get(),
531            old_num_words * sizeof(int32));
532     memset(storage.get() + old_num_words, 0,
533            (num_words - old_num_words) * sizeof(int32));
534 
535     backup_bitmap_storage_.swap(storage);
536     backup_header_->table_len = header_->table_len;
537   } else {
538     backup_bitmap_storage_.reset(params->backup_bitmap.release());
539     backup_header_.reset(params->backup_header.release());
540   }
541 
542   num_words = (backup_header_->table_len + 31) / 32;
543   backup_bitmap_.reset(new Bitmap(backup_bitmap_storage_.get(),
544                                   backup_header_->table_len, num_words));
545   if (old_extra_table)
546     MoveCells(old_extra_table.get());
547 
548   if (small_table_)
549     DCHECK(header_->flags & SMALL_CACHE);
550 
551   // All tables and backups are needed for operation.
552   DCHECK(main_table_);
553   DCHECK(extra_table_);
554   DCHECK(bitmap_.get());
555 }
556 
Shutdown()557 void IndexTable::Shutdown() {
558   header_ = NULL;
559   main_table_ = NULL;
560   extra_table_ = NULL;
561   bitmap_.reset();
562   backup_bitmap_.reset();
563   backup_header_.reset();
564   backup_bitmap_storage_.reset();
565   modified_ = false;
566 }
567 
568 // The general method for locating cells is to:
569 // 1. Get the first bucket. This usually means directly indexing the table (as
570 //     this method does), or iterating through all possible buckets.
571 // 2. Iterate through all the cells in that first bucket.
572 // 3. If there is a linked bucket, locate it directly in the extra table.
573 // 4. Go back to 2, as needed.
574 //
575 // One consequence of this pattern is that we never start looking at buckets in
576 // the extra table, unless we are following a link from the main table.
LookupEntries(uint32 hash)577 EntrySet IndexTable::LookupEntries(uint32 hash) {
578   EntrySet entries;
579   int bucket_num = static_cast<int>(hash & mask_);
580   IndexBucket* bucket = &main_table_[bucket_num];
581   do {
582     for (int i = 0; i < kCellsPerBucket; i++) {
583       IndexCell* current_cell = &bucket->cells[i];
584       if (!GetLocation(*current_cell))
585         continue;
586       if (!SanityCheck(*current_cell)) {
587         NOTREACHED();
588         int cell_num = bucket_num * kCellsPerBucket + i;
589         current_cell->Clear();
590         bitmap_->Set(cell_num, false);
591         backup_bitmap_->Set(cell_num, false);
592         modified_ = true;
593         continue;
594       }
595       int cell_num = bucket_num * kCellsPerBucket + i;
596       if (MisplacedHash(*current_cell, hash)) {
597         HandleMisplacedCell(current_cell, cell_num, hash & mask_);
598       } else if (IsHashMatch(*current_cell, hash)) {
599         EntryCell entry_cell(cell_num, hash, *current_cell, small_table_);
600         CheckState(entry_cell);
601         if (entry_cell.GetState() != ENTRY_DELETED) {
602           entries.cells.push_back(entry_cell);
603           if (entry_cell.GetGroup() == ENTRY_EVICTED)
604             entries.evicted_count++;
605         }
606       }
607     }
608     bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
609                                &bucket);
610   } while (bucket_num);
611   return entries;
612 }
613 
CreateEntryCell(uint32 hash,Addr address)614 EntryCell IndexTable::CreateEntryCell(uint32 hash, Addr address) {
615   DCHECK(IsValidAddress(address));
616   DCHECK(address.FileNumber() || address.start_block());
617 
618   int bucket_num = static_cast<int>(hash & mask_);
619   int cell_num = 0;
620   IndexBucket* bucket = &main_table_[bucket_num];
621   IndexCell* current_cell = NULL;
622   bool found = false;
623   do {
624     for (int i = 0; i < kCellsPerBucket && !found; i++) {
625       current_cell = &bucket->cells[i];
626       if (!GetLocation(*current_cell)) {
627         cell_num = bucket_num * kCellsPerBucket + i;
628         found = true;
629       }
630     }
631     if (found)
632       break;
633     bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
634                                &bucket);
635   } while (bucket_num);
636 
637   if (!found) {
638     bucket_num = NewExtraBucket();
639     if (bucket_num) {
640       cell_num = bucket_num * kCellsPerBucket;
641       bucket->next = cell_num;
642       bucket = &extra_table_[bucket_num - (mask_ + 1)];
643       bucket->hash = hash & mask_;
644       found = true;
645     } else {
646       // address 0 is a reserved value, and the caller interprets it as invalid.
647       address.set_value(0);
648     }
649   }
650 
651   EntryCell entry_cell(cell_num, hash, address, small_table_);
652   if (address.file_type() == BLOCK_EVICTED)
653     entry_cell.SetGroup(ENTRY_EVICTED);
654   else
655     entry_cell.SetGroup(ENTRY_NO_USE);
656   Save(&entry_cell);
657 
658   if (found) {
659     bitmap_->Set(cell_num, true);
660     backup_bitmap_->Set(cell_num, true);
661     header()->used_cells++;
662     modified_ = true;
663   }
664 
665   return entry_cell;
666 }
667 
FindEntryCell(uint32 hash,Addr address)668 EntryCell IndexTable::FindEntryCell(uint32 hash, Addr address) {
669   return FindEntryCellImpl(hash, address, false);
670 }
671 
CalculateTimestamp(Time time)672 int IndexTable::CalculateTimestamp(Time time) {
673   TimeDelta delta = time - Time::FromInternalValue(header_->base_time);
674   return std::max(delta.InMinutes(), 0);
675 }
676 
TimeFromTimestamp(int timestamp)677 base::Time IndexTable::TimeFromTimestamp(int timestamp) {
678   return Time::FromInternalValue(header_->base_time) +
679          TimeDelta::FromMinutes(timestamp);
680 }
681 
SetSate(uint32 hash,Addr address,EntryState state)682 void IndexTable::SetSate(uint32 hash, Addr address, EntryState state) {
683   EntryCell cell = FindEntryCellImpl(hash, address, state == ENTRY_FREE);
684   if (!cell.IsValid()) {
685     NOTREACHED();
686     return;
687   }
688 
689   EntryState old_state = cell.GetState();
690   switch (state) {
691     case ENTRY_FREE:
692       DCHECK_EQ(old_state, ENTRY_DELETED);
693       break;
694     case ENTRY_NEW:
695       DCHECK_EQ(old_state, ENTRY_FREE);
696       break;
697     case ENTRY_OPEN:
698       DCHECK_EQ(old_state, ENTRY_USED);
699       break;
700     case ENTRY_MODIFIED:
701       DCHECK_EQ(old_state, ENTRY_OPEN);
702       break;
703     case ENTRY_DELETED:
704       DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN ||
705              old_state == ENTRY_MODIFIED);
706       break;
707     case ENTRY_USED:
708       DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN ||
709              old_state == ENTRY_MODIFIED);
710       break;
711     case ENTRY_FIXING:
712       break;
713   };
714 
715   modified_ = true;
716   if (state == ENTRY_DELETED) {
717     bitmap_->Set(cell.cell_num(), false);
718     backup_bitmap_->Set(cell.cell_num(), false);
719   } else if (state == ENTRY_FREE) {
720     cell.Clear();
721     Write(cell);
722     header()->used_cells--;
723     return;
724   }
725   cell.SetState(state);
726 
727   Save(&cell);
728 }
729 
UpdateTime(uint32 hash,Addr address,base::Time current)730 void IndexTable::UpdateTime(uint32 hash, Addr address, base::Time current) {
731   EntryCell cell = FindEntryCell(hash, address);
732   if (!cell.IsValid())
733     return;
734 
735   int minutes = CalculateTimestamp(current);
736 
737   // Keep about 3 months of headroom.
738   const int kMaxTimestamp = (1 << 20) - 60 * 24 * 90;
739   if (minutes > kMaxTimestamp) {
740     // TODO(rvargas):
741     // Update header->old_time and trigger a timer
742     // Rebaseline timestamps and don't update sums
743     // Start a timer (about 2 backups)
744     // fix all ckecksums and trigger another timer
745     // update header->old_time because rebaseline is done.
746     minutes = std::min(minutes, (1 << 20) - 1);
747   }
748 
749   cell.SetTimestamp(minutes);
750   Save(&cell);
751 }
752 
Save(EntryCell * cell)753 void IndexTable::Save(EntryCell* cell) {
754   cell->FixSum();
755   Write(*cell);
756 }
757 
GetOldest(IndexIterator * no_use,IndexIterator * low_use,IndexIterator * high_use)758 void IndexTable::GetOldest(IndexIterator* no_use,
759                            IndexIterator* low_use,
760                            IndexIterator* high_use) {
761   no_use->forward = true;
762   low_use->forward = true;
763   high_use->forward = true;
764   InitIterator(no_use);
765   InitIterator(low_use);
766   InitIterator(high_use);
767 
768   WalkTables(-1, no_use, low_use, high_use);
769 }
770 
GetNextCells(IndexIterator * iterator)771 bool IndexTable::GetNextCells(IndexIterator* iterator) {
772   int current_time = iterator->timestamp;
773   InitIterator(iterator);
774 
775   WalkTables(current_time, iterator, iterator, iterator);
776   return !iterator->cells.empty();
777 }
778 
OnBackupTimer()779 void IndexTable::OnBackupTimer() {
780   if (!modified_)
781     return;
782 
783   int num_words = (header_->table_len + 31) / 32;
784   int num_bytes = num_words * 4 + static_cast<int>(sizeof(*header_));
785   scoped_refptr<net::IOBuffer> buffer(new net::IOBuffer(num_bytes));
786   memcpy(buffer->data(), header_, sizeof(*header_));
787   memcpy(buffer->data() + sizeof(*header_), backup_bitmap_storage_.get(),
788          num_words * 4);
789   backend_->SaveIndex(buffer, num_bytes);
790   modified_ = false;
791 }
792 
793 // -----------------------------------------------------------------------
794 
FindEntryCellImpl(uint32 hash,Addr address,bool allow_deleted)795 EntryCell IndexTable::FindEntryCellImpl(uint32 hash, Addr address,
796                                         bool allow_deleted) {
797   int bucket_num = static_cast<int>(hash & mask_);
798   IndexBucket* bucket = &main_table_[bucket_num];
799   do {
800     for (int i = 0; i < kCellsPerBucket; i++) {
801       IndexCell* current_cell = &bucket->cells[i];
802       if (!GetLocation(*current_cell))
803         continue;
804       DCHECK(SanityCheck(*current_cell));
805       if (IsHashMatch(*current_cell, hash)) {
806         // We have a match.
807         int cell_num = bucket_num * kCellsPerBucket + i;
808         EntryCell entry_cell(cell_num, hash, *current_cell, small_table_);
809         if (entry_cell.GetAddress() != address)
810           continue;
811 
812         if (!allow_deleted && entry_cell.GetState() == ENTRY_DELETED)
813           continue;
814 
815         return entry_cell;
816       }
817     }
818     bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
819                                &bucket);
820   } while (bucket_num);
821   return EntryCell();
822 }
823 
CheckState(const EntryCell & cell)824 void IndexTable::CheckState(const EntryCell& cell) {
825   int current_state = cell.GetState();
826   if (current_state != ENTRY_FIXING) {
827     bool present = ((current_state & 3) != 0);  // Look at the last two bits.
828     if (present != bitmap_->Get(cell.cell_num()) ||
829         present != backup_bitmap_->Get(cell.cell_num())) {
830       // There's a mismatch.
831       if (current_state == ENTRY_DELETED) {
832         // We were in the process of deleting this entry. Finish now.
833         backend_->DeleteCell(cell);
834       } else {
835         current_state = ENTRY_FIXING;
836         EntryCell bad_cell(cell);
837         bad_cell.SetState(ENTRY_FIXING);
838         Save(&bad_cell);
839       }
840     }
841   }
842 
843   if (current_state == ENTRY_FIXING)
844     backend_->FixCell(cell);
845 }
846 
Write(const EntryCell & cell)847 void IndexTable::Write(const EntryCell& cell) {
848   IndexBucket* bucket = NULL;
849   int bucket_num = cell.cell_num() / kCellsPerBucket;
850   if (bucket_num < static_cast<int32>(mask_ + 1)) {
851     bucket = &main_table_[bucket_num];
852   } else {
853     DCHECK_LE(bucket_num, header()->max_bucket);
854     bucket = &extra_table_[bucket_num - (mask_ + 1)];
855   }
856 
857   int cell_number = cell.cell_num() % kCellsPerBucket;
858   if (GetLocation(bucket->cells[cell_number]) && cell.GetLocation()) {
859     DCHECK_EQ(cell.GetLocation(),
860               GetLocation(bucket->cells[cell_number]));
861   }
862   cell.Serialize(&bucket->cells[cell_number]);
863 }
864 
NewExtraBucket()865 int IndexTable::NewExtraBucket() {
866   int safe_window = (header()->table_len < kNumExtraBlocks * 2) ?
867                     kNumExtraBlocks / 4 : kNumExtraBlocks;
868   if (header()->table_len - header()->max_bucket * kCellsPerBucket <
869       safe_window) {
870     backend_->GrowIndex();
871   }
872 
873   if (header()->max_bucket * kCellsPerBucket ==
874       header()->table_len - kCellsPerBucket) {
875     return 0;
876   }
877 
878   header()->max_bucket++;
879   return header()->max_bucket;
880 }
881 
WalkTables(int limit_time,IndexIterator * no_use,IndexIterator * low_use,IndexIterator * high_use)882 void IndexTable::WalkTables(int limit_time,
883                             IndexIterator* no_use,
884                             IndexIterator* low_use,
885                             IndexIterator* high_use) {
886   header_->num_no_use_entries = 0;
887   header_->num_low_use_entries = 0;
888   header_->num_high_use_entries = 0;
889   header_->num_evicted_entries = 0;
890 
891   for (int i = 0; i < static_cast<int32>(mask_ + 1); i++) {
892     int bucket_num = i;
893     IndexBucket* bucket = &main_table_[i];
894     do {
895       UpdateFromBucket(bucket, i, limit_time, no_use, low_use, high_use);
896 
897       bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
898                                  &bucket);
899     } while (bucket_num);
900   }
901   header_->num_entries = header_->num_no_use_entries +
902                          header_->num_low_use_entries +
903                          header_->num_high_use_entries +
904                          header_->num_evicted_entries;
905   modified_ = true;
906 }
907 
UpdateFromBucket(IndexBucket * bucket,int bucket_hash,int limit_time,IndexIterator * no_use,IndexIterator * low_use,IndexIterator * high_use)908 void IndexTable::UpdateFromBucket(IndexBucket* bucket, int bucket_hash,
909                                   int limit_time,
910                                   IndexIterator* no_use,
911                                   IndexIterator* low_use,
912                                   IndexIterator* high_use) {
913   for (int i = 0; i < kCellsPerBucket; i++) {
914     IndexCell& current_cell = bucket->cells[i];
915     if (!GetLocation(current_cell))
916       continue;
917     DCHECK(SanityCheck(current_cell));
918     if (!IsNormalState(current_cell))
919       continue;
920 
921     EntryCell entry_cell(0, GetFullHash(current_cell, bucket_hash),
922                          current_cell, small_table_);
923     switch (GetCellGroup(current_cell)) {
924       case ENTRY_NO_USE:
925         UpdateIterator(entry_cell, limit_time, no_use);
926         header_->num_no_use_entries++;
927         break;
928       case ENTRY_LOW_USE:
929         UpdateIterator(entry_cell, limit_time, low_use);
930         header_->num_low_use_entries++;
931         break;
932       case ENTRY_HIGH_USE:
933         UpdateIterator(entry_cell, limit_time, high_use);
934         header_->num_high_use_entries++;
935         break;
936       case ENTRY_EVICTED:
937         header_->num_evicted_entries++;
938         break;
939       default:
940         NOTREACHED();
941     }
942   }
943 }
944 
945 // This code is only called from Init() so the internal state of this object is
946 // in flux (this method is performing the last steps of re-initialization). As
947 // such, random methods are not supposed to work at this point, so whatever this
948 // method calls should be relatively well controlled and it may require some
949 // degree of "stable state faking".
MoveCells(IndexBucket * old_extra_table)950 void IndexTable::MoveCells(IndexBucket* old_extra_table) {
951   int max_hash = (mask_ + 1) / 2;
952   int max_bucket = header()->max_bucket;
953   header()->max_bucket = mask_;
954   int used_cells = header()->used_cells;
955 
956   // Consider a large cache: a cell stores the upper 18 bits of the hash
957   // (h >> 14). If the table is say 8 times the original size (growing from 4x),
958   // the bit that we are interested in would be the 3rd bit of the stored value,
959   // in other words 'multiplier' >> 1.
960   uint32 new_bit = (1 << extra_bits_) >> 1;
961 
962   scoped_ptr<IndexBucket[]> old_main_table;
963   IndexBucket* source_table = main_table_;
964   bool upgrade_format = !extra_bits_;
965   if (upgrade_format) {
966     // This method should deal with migrating a small table to a big one. Given
967     // that the first thing to do is read the old table, set small_table_ for
968     // the size of the old table. Now, when moving a cell, the result cannot be
969     // placed in the old table or we will end up reading it again and attempting
970     // to move it, so we have to copy the whole table at once.
971     DCHECK(!small_table_);
972     small_table_ = true;
973     old_main_table.reset(new IndexBucket[max_hash]);
974     memcpy(old_main_table.get(), main_table_, max_hash * sizeof(IndexBucket));
975     memset(main_table_, 0, max_hash * sizeof(IndexBucket));
976     source_table = old_main_table.get();
977   }
978 
979   for (int i = 0; i < max_hash; i++) {
980     int bucket_num = i;
981     IndexBucket* bucket = &source_table[i];
982     do {
983       for (int j = 0; j < kCellsPerBucket; j++) {
984         IndexCell& current_cell = bucket->cells[j];
985         if (!GetLocation(current_cell))
986           continue;
987         DCHECK(SanityCheck(current_cell));
988         if (bucket_num == i) {
989           if (upgrade_format || (GetHashValue(current_cell) & new_bit)) {
990             // Move this cell to the upper half of the table.
991             MoveSingleCell(&current_cell, bucket_num * kCellsPerBucket + j, i,
992                            true);
993           }
994         } else {
995           // All cells on extra buckets have to move.
996           MoveSingleCell(&current_cell, bucket_num * kCellsPerBucket + j, i,
997                          true);
998         }
999       }
1000 
1001       // There is no need to clear the old bucket->next value because if falls
1002       // within the main table so it will be fixed when attempting to follow
1003       // the link.
1004       bucket_num = GetNextBucket(max_hash, max_bucket, old_extra_table,
1005                                  &bucket);
1006     } while (bucket_num);
1007   }
1008 
1009   DCHECK_EQ(header()->used_cells, used_cells);
1010 
1011   if (upgrade_format) {
1012     small_table_ = false;
1013     header()->flags &= ~SMALL_CACHE;
1014   }
1015 }
1016 
MoveSingleCell(IndexCell * current_cell,int cell_num,int main_table_index,bool growing)1017 void IndexTable::MoveSingleCell(IndexCell* current_cell, int cell_num,
1018                                 int main_table_index, bool growing) {
1019   uint32 hash = GetFullHash(*current_cell, main_table_index);
1020   EntryCell old_cell(cell_num, hash, *current_cell, small_table_);
1021 
1022   // This method may be called when moving entries from a small table to a
1023   // normal table. In that case, the caller (MoveCells) has to read the old
1024   // table, so it needs small_table_ set to true, but this method needs to
1025   // write to the new table so small_table_ has to be set to false, and the
1026   // value restored to true before returning.
1027   bool upgrade_format = !extra_bits_ && growing;
1028   if (upgrade_format)
1029     small_table_ = false;
1030   EntryCell new_cell = CreateEntryCell(hash, old_cell.GetAddress());
1031 
1032   if (!new_cell.IsValid()) {
1033     // We'll deal with this entry later.
1034     if (upgrade_format)
1035       small_table_ = true;
1036     return;
1037   }
1038 
1039   new_cell.SetState(old_cell.GetState());
1040   new_cell.SetGroup(old_cell.GetGroup());
1041   new_cell.SetReuse(old_cell.GetReuse());
1042   new_cell.SetTimestamp(old_cell.GetTimestamp());
1043   Save(&new_cell);
1044   modified_ = true;
1045   if (upgrade_format)
1046     small_table_ = true;
1047 
1048   if (old_cell.GetState() == ENTRY_DELETED) {
1049     bitmap_->Set(new_cell.cell_num(), false);
1050     backup_bitmap_->Set(new_cell.cell_num(), false);
1051   }
1052 
1053   if (!growing || cell_num / kCellsPerBucket == main_table_index) {
1054     // Only delete entries that live on the main table.
1055     if (!upgrade_format) {
1056       old_cell.Clear();
1057       Write(old_cell);
1058     }
1059 
1060     if (cell_num != new_cell.cell_num()) {
1061       bitmap_->Set(old_cell.cell_num(), false);
1062       backup_bitmap_->Set(old_cell.cell_num(), false);
1063     }
1064   }
1065   header()->used_cells--;
1066 }
1067 
HandleMisplacedCell(IndexCell * current_cell,int cell_num,int main_table_index)1068 void IndexTable::HandleMisplacedCell(IndexCell* current_cell, int cell_num,
1069                                      int main_table_index) {
1070   NOTREACHED();  // No unit tests yet.
1071 
1072   // The cell may be misplaced, or a duplicate cell exists with this data.
1073   uint32 hash = GetFullHash(*current_cell, main_table_index);
1074   MoveSingleCell(current_cell, cell_num, main_table_index, false);
1075 
1076   // Now look for a duplicate cell.
1077   CheckBucketList(hash & mask_);
1078 }
1079 
CheckBucketList(int bucket_num)1080 void IndexTable::CheckBucketList(int bucket_num) {
1081   typedef std::pair<int, EntryGroup> AddressAndGroup;
1082   std::set<AddressAndGroup> entries;
1083   IndexBucket* bucket = &main_table_[bucket_num];
1084   int bucket_hash = bucket_num;
1085   do {
1086     for (int i = 0; i < kCellsPerBucket; i++) {
1087       IndexCell* current_cell = &bucket->cells[i];
1088       if (!GetLocation(*current_cell))
1089         continue;
1090       if (!SanityCheck(*current_cell)) {
1091         NOTREACHED();
1092         current_cell->Clear();
1093         continue;
1094       }
1095       int cell_num = bucket_num * kCellsPerBucket + i;
1096       EntryCell cell(cell_num, GetFullHash(*current_cell, bucket_hash),
1097                      *current_cell, small_table_);
1098       if (!entries.insert(std::make_pair(cell.GetAddress().value(),
1099                                          cell.GetGroup())).second) {
1100         current_cell->Clear();
1101         continue;
1102       }
1103       CheckState(cell);
1104     }
1105 
1106     bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_,
1107                               &bucket);
1108   } while (bucket_num);
1109 }
1110 
GetLocation(const IndexCell & cell)1111 uint32 IndexTable::GetLocation(const IndexCell& cell) {
1112   if (small_table_)
1113     return GetCellSmallTableLocation(cell);
1114 
1115   return GetCellLocation(cell);
1116 }
1117 
GetHashValue(const IndexCell & cell)1118 uint32 IndexTable::GetHashValue(const IndexCell& cell) {
1119   if (small_table_)
1120     return GetCellSmallTableId(cell);
1121 
1122   return GetCellId(cell);
1123 }
1124 
GetFullHash(const IndexCell & cell,uint32 lower_part)1125 uint32 IndexTable::GetFullHash(const IndexCell& cell, uint32 lower_part) {
1126   // It is OK for the high order bits of lower_part to overlap with the stored
1127   // part of the hash.
1128   if (small_table_)
1129     return (GetCellSmallTableId(cell) << kSmallTableHashShift) | lower_part;
1130 
1131   return (GetCellId(cell) << kHashShift) | lower_part;
1132 }
1133 
1134 // All the bits stored in the cell should match the provided hash.
IsHashMatch(const IndexCell & cell,uint32 hash)1135 bool IndexTable::IsHashMatch(const IndexCell& cell, uint32 hash) {
1136   hash = small_table_ ? hash >> kSmallTableHashShift : hash >> kHashShift;
1137   return GetHashValue(cell) == hash;
1138 }
1139 
MisplacedHash(const IndexCell & cell,uint32 hash)1140 bool IndexTable::MisplacedHash(const IndexCell& cell, uint32 hash) {
1141   if (!extra_bits_)
1142     return false;
1143 
1144   uint32 mask = (1 << extra_bits_) - 1;
1145   hash = small_table_ ? hash >> kSmallTableHashShift : hash >> kHashShift;
1146   return (GetHashValue(cell) & mask) != (hash & mask);
1147 }
1148 
1149 }  // namespace disk_cache
1150