diff options
author | rvargas@chromium.org <rvargas@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2014-01-10 00:25:15 +0000 |
---|---|---|
committer | rvargas@chromium.org <rvargas@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | 2014-01-10 00:25:15 +0000 |
commit | b46767a36914907cbe36d7953980a5a399041806 (patch) | |
tree | 9870b5ac0004c21730d372c8fe7da1bf5bbf297b /net | |
parent | 86ba1fe0d2582adf113f68a32532277017c4449f (diff) | |
download | chromium_src-b46767a36914907cbe36d7953980a5a399041806.zip chromium_src-b46767a36914907cbe36d7953980a5a399041806.tar.gz chromium_src-b46767a36914907cbe36d7953980a5a399041806.tar.bz2 |
Disk cache v3: The main index table.
The IndexTable controls what entries are stored in the cache.
This class provides a memory-only view of the underlying structures
that store "the index"
BUG=241277
TEST=net_unittests
Review URL: https://codereview.chromium.org/53313004
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@244027 0039d316-1c4b-4281-b951-d872f2087c98
Diffstat (limited to 'net')
-rw-r--r-- | net/disk_cache/addr.h | 20 | ||||
-rw-r--r-- | net/disk_cache/disk_format_base.h | 1 | ||||
-rw-r--r-- | net/disk_cache/v3/disk_format_v3.h | 86 | ||||
-rw-r--r-- | net/disk_cache/v3/index_table.cc | 1149 | ||||
-rw-r--r-- | net/disk_cache/v3/index_table.h | 279 | ||||
-rw-r--r-- | net/disk_cache/v3/index_table_unittest.cc | 706 | ||||
-rw-r--r-- | net/net.gyp | 3 |
7 files changed, 2221 insertions, 23 deletions
diff --git a/net/disk_cache/addr.h b/net/disk_cache/addr.h index f0fb1ca..664e462 100644 --- a/net/disk_cache/addr.h +++ b/net/disk_cache/addr.h @@ -54,6 +54,18 @@ const int kFirstAdditionalBlockFileV3 = 7; // 0000 0011 0000 0000 0000 0000 0000 0000 : number of contiguous blocks 1-4 // 0000 0000 1111 1111 0000 0000 0000 0000 : file selector 0 - 255 // 0000 0000 0000 0000 1111 1111 1111 1111 : block# 0 - 65,535 (2^16) +// +// Note that an Addr can be used to "point" to a variety of different objects, +// from a given type of entry to random blobs of data. Conceptually, an Addr is +// just a number that someone can inspect to find out how to locate the desired +// record. Most users will not care about the specific bits inside Addr, for +// example, what parts of it point to a file number; only the code that has to +// select a specific file would care about those specific bits. +// +// From a general point of view, an Addr has a total capacity of 2^24 entities, +// in that it has 24 bits that can identify individual records. Note that the +// address space is bigger for independent files (2^28), but that would not be +// the general case. class NET_EXPORT_PRIVATE Addr { public: Addr() : value_(0) {} @@ -108,14 +120,6 @@ class NET_EXPORT_PRIVATE Addr { return value_ != other.value_; } - static Addr FromEntryAddress(uint32 value) { - return Addr(kInitializedMask + (BLOCK_ENTRIES << kFileTypeOffset) + value); - } - - static Addr FromEvictedAddress(uint32 value) { - return Addr(kInitializedMask + (BLOCK_EVICTED << kFileTypeOffset) + value); - } - static int BlockSizeForFileType(FileType file_type) { switch (file_type) { case RANKINGS: diff --git a/net/disk_cache/disk_format_base.h b/net/disk_cache/disk_format_base.h index 3198381..419f41b 100644 --- a/net/disk_cache/disk_format_base.h +++ b/net/disk_cache/disk_format_base.h @@ -33,6 +33,7 @@ const uint32 kBlockCurrentVersion = 0x30000; // Version 3.0. const uint32 kBlockMagic = 0xC104CAC3; const int kBlockHeaderSize = 8192; // Two pages: almost 64k entries const int kMaxBlocks = (kBlockHeaderSize - 80) * 8; +const int kNumExtraBlocks = 1024; // How fast files grow. // Bitmap to track used blocks on a block-file. typedef uint32 AllocBitmap[kMaxBlocks / 32]; diff --git a/net/disk_cache/v3/disk_format_v3.h b/net/disk_cache/v3/disk_format_v3.h index 5616377..7777c783 100644 --- a/net/disk_cache/v3/disk_format_v3.h +++ b/net/disk_cache/v3/disk_format_v3.h @@ -37,6 +37,10 @@ // internal structures are modified, so it is possible to detect (most of the // time) when the process dies in the middle of an update. There are dedicated // backup files for cache bitmaps, used to detect entries out of date. +// +// Although cache files are to be consumed on the same machine that creates +// them, if files are to be moved accross machines, little endian storage is +// assumed. #ifndef NET_DISK_CACHE_V3_DISK_FORMAT_V3_H_ #define NET_DISK_CACHE_V3_DISK_FORMAT_V3_H_ @@ -46,14 +50,15 @@ namespace disk_cache { -const int kBaseTableLen = 0x10000; +const int kBaseTableLen = 0x400; const uint32 kIndexMagicV3 = 0xC103CAC3; const uint32 kVersion3 = 0x30000; // Version 3.0. // Flags for a given cache. enum CacheFlags { - CACHE_EVICTION_2 = 1, // Keep multiple lists for eviction. - CACHE_EVICTED = 1 << 1 // Already evicted at least one entry. + SMALL_CACHE = 1 << 0, // See IndexCell. + CACHE_EVICTION_2 = 1 << 1, // Keep multiple lists for eviction. + CACHE_EVICTED = 1 << 2 // Already evicted at least one entry. }; // Header for the master index file. @@ -119,30 +124,81 @@ COMPILE_ASSERT(ENTRY_USED <= 7, group_uses_3_bits); struct IndexCell { void Clear() { memset(this, 0, sizeof(*this)); } - uint64 address : 22; - uint64 hash : 18; - uint64 timestamp : 20; - uint64 reuse : 4; - uint8 state : 3; - uint8 group : 3; - uint8 sum : 2; + // A cell is a 9 byte bit-field that stores 7 values: + // location : 22 bits + // id : 18 bits + // timestamp : 20 bits + // reuse : 4 bits + // state : 3 bits + // group : 3 bits + // sum : 2 bits + // The id is derived from the full hash of the entry. + // + // The actual layout is as follows: + // + // first_part (low order 32 bits): + // 0000 0000 0011 1111 1111 1111 1111 1111 : location + // 1111 1111 1100 0000 0000 0000 0000 0000 : id + // + // first_part (high order 32 bits): + // 0000 0000 0000 0000 0000 0000 1111 1111 : id + // 0000 1111 1111 1111 1111 1111 0000 0000 : timestamp + // 1111 0000 0000 0000 0000 0000 0000 0000 : reuse + // + // last_part: + // 0000 0111 : state + // 0011 1000 : group + // 1100 0000 : sum + // + // The small-cache version of the format moves some bits from the location to + // the id fileds, like so: + // location : 16 bits + // id : 24 bits + // + // first_part (low order 32 bits): + // 0000 0000 0000 0000 1111 1111 1111 1111 : location + // 1111 1111 1111 1111 0000 0000 0000 0000 : id + // + // The actual bit distribution between location and id is determined by the + // table size (IndexHeaderV3.table_len). Tables smaller than 65536 entries + // use the small-cache version; after that size, caches should have the + // SMALL_CACHE flag cleared. + // + // To locate a given entry after recovering the location from the cell, the + // file type and file number are appended (see disk_cache/addr.h). For a large + // table only the file type is implied; for a small table, the file number + // is also implied, and it should be the first file for that type of entry, + // as determined by the EntryGroup (two files in total, one for active entries + // and another one for evicted entries). + // + // For example, a small table may store something like 0x1234 as the location + // field. That means it stores the entry number 0x1234. If that record belongs + // to a deleted entry, the regular cache address may look something like + // BLOCK_EVICTED + 1 block + file number 6 + entry number 0x1234 + // so Addr = 0xf0061234 + // + // If that same Addr is stored on a large table, the location field would be + // 0x61234 + + uint64 first_part; + uint8 last_part; }; COMPILE_ASSERT(sizeof(IndexCell) == 9, bad_IndexCell); +const int kCellsPerBucket = 4; struct IndexBucket { - IndexCell cells[4]; + IndexCell cells[kCellsPerBucket]; int32 next; - uint32 hash : 24; // The last byte is only defined for buckets of - uint32 reserved : 8; // the extra table. + uint32 hash; // The high order byte is reserved (should be zero). }; COMPILE_ASSERT(sizeof(IndexBucket) == 44, bad_IndexBucket); -const int kBytesPerCell = 44 / 4; +const int kBytesPerCell = 44 / kCellsPerBucket; // The main cache index. Backed by a file named index_tb1. // The extra table (index_tb2) has a similar format, but different size. struct Index { // Default size. Actual size controlled by header.table_len. - IndexBucket table[kBaseTableLen / 4]; + IndexBucket table[kBaseTableLen / kCellsPerBucket]; }; #pragma pack(pop) diff --git a/net/disk_cache/v3/index_table.cc b/net/disk_cache/v3/index_table.cc new file mode 100644 index 0000000..3bbb130 --- /dev/null +++ b/net/disk_cache/v3/index_table.cc @@ -0,0 +1,1149 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "net/disk_cache/v3/index_table.h" + +#include <algorithm> +#include <set> +#include <utility> + +#include "base/bits.h" +#include "net/base/io_buffer.h" +#include "net/base/net_errors.h" +#include "net/disk_cache/disk_cache.h" + +using base::Time; +using base::TimeDelta; +using disk_cache::CellInfo; +using disk_cache::CellList; +using disk_cache::IndexCell; +using disk_cache::IndexIterator; + +namespace { + +// The following constants describe the bitfields of an IndexCell so they are +// implicitly synchronized with the descrption of IndexCell on file_format_v3.h. +const uint64 kCellLocationMask = (1 << 22) - 1; +const uint64 kCellIdMask = (1 << 18) - 1; +const uint64 kCellTimestampMask = (1 << 20) - 1; +const uint64 kCellReuseMask = (1 << 4) - 1; +const uint8 kCellStateMask = (1 << 3) - 1; +const uint8 kCellGroupMask = (1 << 3) - 1; +const uint8 kCellSumMask = (1 << 2) - 1; + +const uint64 kCellSmallTableLocationMask = (1 << 16) - 1; +const uint64 kCellSmallTableIdMask = (1 << 24) - 1; + +const int kCellIdOffset = 22; +const int kCellTimestampOffset = 40; +const int kCellReuseOffset = 60; +const int kCellGroupOffset = 3; +const int kCellSumOffset = 6; + +const int kCellSmallTableIdOffset = 16; + +// The number of bits that a hash has to be shifted to grab the part that +// defines the cell id. +const int kHashShift = 14; +const int kSmallTableHashShift = 8; + +// Unfortunately we have to break the abstaction a little here: the file number +// where entries are stored is outside of the control of this code, and it is +// usually part of the stored address. However, for small tables we only store +// 16 bits of the address so the file number is never stored on a cell. We have +// to infere the file number from the type of entry (normal vs evicted), and +// the knowledge that given that the table will not keep more than 64k entries, +// a single file of each type is enough. +const int kEntriesFile = disk_cache::BLOCK_ENTRIES - 1; +const int kEvictedEntriesFile = disk_cache::BLOCK_EVICTED - 1; +const int kMaxLocation = 1 << 22; +const int kMinFileNumber = 1 << 16; + +uint32 GetCellLocation(const IndexCell& cell) { + return cell.first_part & kCellLocationMask; +} + +uint32 GetCellSmallTableLocation(const IndexCell& cell) { + return cell.first_part & kCellSmallTableLocationMask; +} + +uint32 GetCellId(const IndexCell& cell) { + return (cell.first_part >> kCellIdOffset) & kCellIdMask; +} + +uint32 GetCellSmallTableId(const IndexCell& cell) { + return (cell.first_part >> kCellSmallTableIdOffset) & + kCellSmallTableIdMask; +} + +int GetCellTimestamp(const IndexCell& cell) { + return (cell.first_part >> kCellTimestampOffset) & kCellTimestampMask; +} + +int GetCellReuse(const IndexCell& cell) { + return (cell.first_part >> kCellReuseOffset) & kCellReuseMask; +} + +int GetCellState(const IndexCell& cell) { + return cell.last_part & kCellStateMask; +} + +int GetCellGroup(const IndexCell& cell) { + return (cell.last_part >> kCellGroupOffset) & kCellGroupMask; +} + +int GetCellSum(const IndexCell& cell) { + return (cell.last_part >> kCellSumOffset) & kCellSumMask; +} + +void SetCellLocation(IndexCell* cell, uint32 address) { + DCHECK_LE(address, static_cast<uint32>(kCellLocationMask)); + cell->first_part &= ~kCellLocationMask; + cell->first_part |= address; +} + +void SetCellSmallTableLocation(IndexCell* cell, uint32 address) { + DCHECK_LE(address, static_cast<uint32>(kCellSmallTableLocationMask)); + cell->first_part &= ~kCellSmallTableLocationMask; + cell->first_part |= address; +} + +void SetCellId(IndexCell* cell, uint32 hash) { + DCHECK_LE(hash, static_cast<uint32>(kCellIdMask)); + cell->first_part &= ~(kCellIdMask << kCellIdOffset); + cell->first_part |= static_cast<int64>(hash) << kCellIdOffset; +} + +void SetCellSmallTableId(IndexCell* cell, uint32 hash) { + DCHECK_LE(hash, static_cast<uint32>(kCellSmallTableIdMask)); + cell->first_part &= ~(kCellSmallTableIdMask << kCellSmallTableIdOffset); + cell->first_part |= static_cast<int64>(hash) << kCellSmallTableIdOffset; +} + +void SetCellTimestamp(IndexCell* cell, int timestamp) { + DCHECK_LT(timestamp, 1 << 20); + DCHECK_GE(timestamp, 0); + cell->first_part &= ~(kCellTimestampMask << kCellTimestampOffset); + cell->first_part |= static_cast<int64>(timestamp) << kCellTimestampOffset; +} + +void SetCellReuse(IndexCell* cell, int count) { + DCHECK_LT(count, 16); + DCHECK_GE(count, 0); + cell->first_part &= ~(kCellReuseMask << kCellReuseOffset); + cell->first_part |= static_cast<int64>(count) << kCellReuseOffset; +} + +void SetCellState(IndexCell* cell, disk_cache::EntryState state) { + cell->last_part &= ~kCellStateMask; + cell->last_part |= state; +} + +void SetCellGroup(IndexCell* cell, disk_cache::EntryGroup group) { + cell->last_part &= ~(kCellGroupMask << kCellGroupOffset); + cell->last_part |= group << kCellGroupOffset; +} + +void SetCellSum(IndexCell* cell, int sum) { + DCHECK_LT(sum, 4); + DCHECK_GE(sum, 0); + cell->last_part &= ~(kCellSumMask << kCellSumOffset); + cell->last_part |= sum << kCellSumOffset; +} + +// This is a very particular way to calculate the sum, so it will not match if +// compared a gainst a pure 2 bit, modulo 2 sum. +int CalculateCellSum(const IndexCell& cell) { + uint32* words = bit_cast<uint32*>(&cell); + uint8* bytes = bit_cast<uint8*>(&cell); + uint32 result = words[0] + words[1]; + result += result >> 16; + result += (result >> 8) + (bytes[8] & 0x3f); + result += result >> 4; + result += result >> 2; + return result & 3; +} + +bool SanityCheck(const IndexCell& cell) { + if (GetCellSum(cell) != CalculateCellSum(cell)) + return false; + + if (GetCellState(cell) > disk_cache::ENTRY_USED || + GetCellGroup(cell) == disk_cache::ENTRY_RESERVED || + GetCellGroup(cell) > disk_cache::ENTRY_EVICTED) { + return false; + } + + return true; +} + +int FileNumberFromLocation(int location) { + return location / kMinFileNumber; +} + +int StartBlockFromLocation(int location) { + return location % kMinFileNumber; +} + +bool IsValidAddress(disk_cache::Addr address) { + if (!address.is_initialized() || + (address.file_type() != disk_cache::BLOCK_EVICTED && + address.file_type() != disk_cache::BLOCK_ENTRIES)) { + return false; + } + + return address.FileNumber() < FileNumberFromLocation(kMaxLocation); +} + +bool IsNormalState(const IndexCell& cell) { + disk_cache::EntryState state = + static_cast<disk_cache::EntryState>(GetCellState(cell)); + DCHECK_NE(state, disk_cache::ENTRY_FREE); + return state != disk_cache::ENTRY_DELETED && + state != disk_cache::ENTRY_FIXING; +} + +inline int GetNextBucket(int min_bucket_num, int max_bucket_num, + disk_cache::IndexBucket* table, + disk_cache::IndexBucket** bucket) { + if (!(*bucket)->next) + return 0; + + int bucket_num = (*bucket)->next / disk_cache::kCellsPerBucket; + if (bucket_num < min_bucket_num || bucket_num > max_bucket_num) { + // The next bucket must fall within the extra table. Note that this is not + // an uncommon path as growing the table may not cleanup the link from the + // main table to the extra table, and that cleanup is performed here when + // accessing that bucket for the first time. This behavior has to change if + // the tables are ever shrinked. + (*bucket)->next = 0; + return 0; + } + *bucket = &table[bucket_num - min_bucket_num]; + return bucket_num; +} + +// Updates the |iterator| with the current |cell|. This cell may cause all +// previous cells to be deleted (when a new target timestamp is found), the cell +// may be added to the list (if it matches the target timestamp), or may it be +// ignored. +void UpdateIterator(const disk_cache::EntryCell& cell, + int limit_time, + IndexIterator* iterator) { + int time = cell.GetTimestamp(); + // Look for not interesting times. + if (iterator->forward && time <= limit_time) + return; + if (!iterator->forward && time >= limit_time) + return; + + if ((iterator->forward && time < iterator->timestamp) || + (!iterator->forward && time > iterator->timestamp)) { + // This timestamp is better than the one we had. + iterator->timestamp = time; + iterator->cells.clear(); + } + if (time == iterator->timestamp) { + CellInfo cell_info = { cell.hash(), cell.GetAddress() }; + iterator->cells.push_back(cell_info); + } +} + +void InitIterator(IndexIterator* iterator) { + iterator->cells.clear(); + iterator->timestamp = iterator->forward ? kint32max : 0; +} + +} // namespace + +namespace disk_cache { + +EntryCell::~EntryCell() { +} + +bool EntryCell::IsValid() const { + return GetCellLocation(cell_) != 0; +} + +// This code has to map the cell address (up to 22 bits) to a general cache Addr +// (up to 24 bits of general addressing). It also set the implied file_number +// in the case of small tables. See also the comment by the definition of +// kEntriesFile. +Addr EntryCell::GetAddress() const { + uint32 location = GetLocation(); + int file_number = FileNumberFromLocation(location); + if (small_table_) { + DCHECK_EQ(0, file_number); + file_number = (GetGroup() == ENTRY_EVICTED) ? kEvictedEntriesFile : + kEntriesFile; + } + DCHECK_NE(0, file_number); + FileType file_type = (GetGroup() == ENTRY_EVICTED) ? BLOCK_EVICTED : + BLOCK_ENTRIES; + return Addr(file_type, 1, file_number, StartBlockFromLocation(location)); +} + +EntryState EntryCell::GetState() const { + return static_cast<EntryState>(GetCellState(cell_)); +} + +EntryGroup EntryCell::GetGroup() const { + return static_cast<EntryGroup>(GetCellGroup(cell_)); +} + +int EntryCell::GetReuse() const { + return GetCellReuse(cell_); +} + +int EntryCell::GetTimestamp() const { + return GetCellTimestamp(cell_); +} + +void EntryCell::SetState(EntryState state) { + SetCellState(&cell_, state); +} + +void EntryCell::SetGroup(EntryGroup group) { + SetCellGroup(&cell_, group); +} + +void EntryCell::SetReuse(int count) { + SetCellReuse(&cell_, count); +} + +void EntryCell::SetTimestamp(int timestamp) { + SetCellTimestamp(&cell_, timestamp); +} + +// Static. +EntryCell EntryCell::GetEntryCellForTest(int32 cell_num, + uint32 hash, + Addr address, + IndexCell* cell, + bool small_table) { + if (cell) { + EntryCell entry_cell(cell_num, hash, *cell, small_table); + return entry_cell; + } + + return EntryCell(cell_num, hash, address, small_table); +} + +void EntryCell::SerializaForTest(IndexCell* destination) { + FixSum(); + Serialize(destination); +} + +EntryCell::EntryCell() : cell_num_(0), hash_(0), small_table_(false) { + cell_.Clear(); +} + +EntryCell::EntryCell(int32 cell_num, + uint32 hash, + Addr address, + bool small_table) + : cell_num_(cell_num), + hash_(hash), + small_table_(small_table) { + DCHECK(IsValidAddress(address) || !address.value()); + + cell_.Clear(); + SetCellState(&cell_, ENTRY_NEW); + SetCellGroup(&cell_, ENTRY_NO_USE); + if (small_table) { + DCHECK(address.FileNumber() == kEntriesFile || + address.FileNumber() == kEvictedEntriesFile); + SetCellSmallTableLocation(&cell_, address.start_block()); + SetCellSmallTableId(&cell_, hash >> kSmallTableHashShift); + } else { + uint32 location = address.FileNumber() << 16 | address.start_block(); + SetCellLocation(&cell_, location); + SetCellId(&cell_, hash >> kHashShift); + } +} + +EntryCell::EntryCell(int32 cell_num, + uint32 hash, + const IndexCell& cell, + bool small_table) + : cell_num_(cell_num), + hash_(hash), + cell_(cell), + small_table_(small_table) { +} + +void EntryCell::FixSum() { + SetCellSum(&cell_, CalculateCellSum(cell_)); +} + +uint32 EntryCell::GetLocation() const { + if (small_table_) + return GetCellSmallTableLocation(cell_); + + return GetCellLocation(cell_); +} + +uint32 EntryCell::RecomputeHash() { + if (small_table_) { + hash_ &= (1 << kSmallTableHashShift) - 1; + hash_ |= GetCellSmallTableId(cell_) << kSmallTableHashShift; + return hash_; + } + + hash_ &= (1 << kHashShift) - 1; + hash_ |= GetCellId(cell_) << kHashShift; + return hash_; +} + +void EntryCell::Serialize(IndexCell* destination) const { + *destination = cell_; +} + +EntrySet::EntrySet() : evicted_count(0), current(0) { +} + +EntrySet::~EntrySet() { +} + +IndexIterator::IndexIterator() { +} + +IndexIterator::~IndexIterator() { +} + +IndexTableInitData::IndexTableInitData() { +} + +IndexTableInitData::~IndexTableInitData() { +} + +// ----------------------------------------------------------------------- + +IndexTable::IndexTable(IndexTableBackend* backend) + : backend_(backend), + header_(NULL), + main_table_(NULL), + extra_table_(NULL), + modified_(false), + small_table_(false) { +} + +IndexTable::~IndexTable() { +} + +// For a general description of the index tables see: +// http://www.chromium.org/developers/design-documents/network-stack/disk-cache/disk-cache-v3#TOC-Index +// +// The index is split between two tables: the main_table_ and the extra_table_. +// The main table can grow only by doubling its number of cells, while the +// extra table can grow slowly, because it only contain cells that overflow +// from the main table. In order to locate a given cell, part of the hash is +// used directly as an index into the main table; once that bucket is located, +// all cells with that partial hash (i.e., belonging to that bucket) are +// inspected, and if present, the next bucket (located on the extra table) is +// then located. For more information on bucket chaining see: +// http://www.chromium.org/developers/design-documents/network-stack/disk-cache/disk-cache-v3#TOC-Buckets +// +// There are two cases when increasing the size: +// - Doubling the size of the main table +// - Adding more entries to the extra table +// +// For example, consider a 64k main table with 8k cells on the extra table (for +// a total of 72k cells). Init can be called to add another 8k cells at the end +// (grow to 80k cells). When the size of the extra table approaches 64k, Init +// can be called to double the main table (to 128k) and go back to a small extra +// table. +void IndexTable::Init(IndexTableInitData* params) { + bool growing = header_ != NULL; + scoped_ptr<IndexBucket[]> old_extra_table; + header_ = ¶ms->index_bitmap->header; + + if (params->main_table) { + if (main_table_) { + // This is doubling the size of main table. + DCHECK_EQ(base::bits::Log2Floor(header_->table_len), + base::bits::Log2Floor(backup_header_->table_len) + 1); + int extra_size = (header()->max_bucket - mask_) * kCellsPerBucket; + DCHECK_GE(extra_size, 0); + + // Doubling the size implies deleting the extra table and moving as many + // cells as we can to the main table, so we first copy the old one. This + // is not required when just growing the extra table because we don't + // move any cell in that case. + old_extra_table.reset(new IndexBucket[extra_size]); + memcpy(old_extra_table.get(), extra_table_, + extra_size * sizeof(IndexBucket)); + memset(params->extra_table, 0, extra_size * sizeof(IndexBucket)); + } + main_table_ = params->main_table; + } + DCHECK(main_table_); + extra_table_ = params->extra_table; + + // extra_bits_ is really measured against table-size specific values. + const int kMaxAbsoluteExtraBits = 12; // From smallest to largest table. + const int kMaxExtraBitsSmallTable = 6; // From smallest to 64K table. + + extra_bits_ = base::bits::Log2Floor(header_->table_len) - + base::bits::Log2Floor(kBaseTableLen); + DCHECK_GE(extra_bits_, 0); + DCHECK_LT(extra_bits_, kMaxAbsoluteExtraBits); + + // Note that following the previous code the constants could be derived as + // kMaxAbsoluteExtraBits = base::bits::Log2Floor(max table len) - + // base::bits::Log2Floor(kBaseTableLen); + // = 22 - base::bits::Log2Floor(1024) = 22 - 10; + // kMaxExtraBitsSmallTable = base::bits::Log2Floor(max 16 bit table) - 10. + + mask_ = ((kBaseTableLen / kCellsPerBucket) << extra_bits_) - 1; + small_table_ = extra_bits_ < kMaxExtraBitsSmallTable; + if (!small_table_) + extra_bits_ -= kMaxExtraBitsSmallTable; + + // table_len keeps the max number of cells stored by the index. We need a + // bitmap with 1 bit per cell, and that bitmap has num_words 32-bit words. + int num_words = (header_->table_len + 31) / 32; + + if (old_extra_table) { + // All the cells from the extra table are moving to the new tables so before + // creating the bitmaps, clear the part of the bitmap referring to the extra + // table. + int old_main_table_bit_words = ((mask_ >> 1) + 1) * kCellsPerBucket / 32; + DCHECK_GT(num_words, old_main_table_bit_words); + memset(params->index_bitmap->bitmap + old_main_table_bit_words, 0, + (num_words - old_main_table_bit_words) * sizeof(int32)); + + DCHECK(growing); + int old_num_words = (backup_header_.get()->table_len + 31) / 32; + DCHECK_GT(old_num_words, old_main_table_bit_words); + memset(backup_bitmap_storage_.get() + old_main_table_bit_words, 0, + (old_num_words - old_main_table_bit_words) * sizeof(int32)); + } + bitmap_.reset(new Bitmap(params->index_bitmap->bitmap, header_->table_len, + num_words)); + + if (growing) { + int old_num_words = (backup_header_.get()->table_len + 31) / 32; + DCHECK_GE(num_words, old_num_words); + scoped_ptr<uint32[]> storage(new uint32[num_words]); + memcpy(storage.get(), backup_bitmap_storage_.get(), + old_num_words * sizeof(int32)); + memset(storage.get() + old_num_words, 0, + (num_words - old_num_words) * sizeof(int32)); + + backup_bitmap_storage_.swap(storage); + backup_header_->table_len = header_->table_len; + } else { + backup_bitmap_storage_.reset(params->backup_bitmap.release()); + backup_header_.reset(params->backup_header.release()); + } + + num_words = (backup_header_->table_len + 31) / 32; + backup_bitmap_.reset(new Bitmap(backup_bitmap_storage_.get(), + backup_header_->table_len, num_words)); + if (old_extra_table) + MoveCells(old_extra_table.get()); + + if (small_table_) + DCHECK(header_->flags & SMALL_CACHE); + + // All tables and backups are needed for operation. + DCHECK(main_table_); + DCHECK(extra_table_); + DCHECK(bitmap_.get()); +} + +void IndexTable::Shutdown() { + header_ = NULL; + main_table_ = NULL; + extra_table_ = NULL; + bitmap_.reset(); + backup_bitmap_.reset(); + backup_header_.reset(); + backup_bitmap_storage_.reset(); + modified_ = false; +} + +// The general method for locating cells is to: +// 1. Get the first bucket. This usually means directly indexing the table (as +// this method does), or iterating through all possible buckets. +// 2. Iterate through all the cells in that first bucket. +// 3. If there is a linked bucket, locate it directly in the extra table. +// 4. Go back to 2, as needed. +// +// One consequence of this pattern is that we never start looking at buckets in +// the extra table, unless we are following a link from the main table. +EntrySet IndexTable::LookupEntries(uint32 hash) { + EntrySet entries; + int bucket_num = static_cast<int>(hash & mask_); + IndexBucket* bucket = &main_table_[bucket_num]; + do { + for (int i = 0; i < kCellsPerBucket; i++) { + IndexCell* current_cell = &bucket->cells[i]; + if (!GetLocation(*current_cell)) + continue; + if (!SanityCheck(*current_cell)) { + NOTREACHED(); + int cell_num = bucket_num * kCellsPerBucket + i; + current_cell->Clear(); + bitmap_->Set(cell_num, false); + backup_bitmap_->Set(cell_num, false); + modified_ = true; + continue; + } + int cell_num = bucket_num * kCellsPerBucket + i; + if (MisplacedHash(*current_cell, hash)) { + HandleMisplacedCell(current_cell, cell_num, hash & mask_); + } else if (IsHashMatch(*current_cell, hash)) { + EntryCell entry_cell(cell_num, hash, *current_cell, small_table_); + CheckState(entry_cell); + if (entry_cell.GetState() != ENTRY_DELETED) { + entries.cells.push_back(entry_cell); + if (entry_cell.GetGroup() == ENTRY_EVICTED) + entries.evicted_count++; + } + } + } + bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_, + &bucket); + } while (bucket_num); + return entries; +} + +EntryCell IndexTable::CreateEntryCell(uint32 hash, Addr address) { + DCHECK(IsValidAddress(address)); + DCHECK(address.FileNumber() || address.start_block()); + + int bucket_num = static_cast<int>(hash & mask_); + int cell_num = 0; + IndexBucket* bucket = &main_table_[bucket_num]; + IndexCell* current_cell = NULL; + bool found = false; + do { + for (int i = 0; i < kCellsPerBucket && !found; i++) { + current_cell = &bucket->cells[i]; + if (!GetLocation(*current_cell)) { + cell_num = bucket_num * kCellsPerBucket + i; + found = true; + } + } + if (found) + break; + bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_, + &bucket); + } while (bucket_num); + + if (!found) { + bucket_num = NewExtraBucket(); + if (bucket_num) { + cell_num = bucket_num * kCellsPerBucket; + bucket->next = cell_num; + bucket = &extra_table_[bucket_num - (mask_ + 1)]; + bucket->hash = hash & mask_; + found = true; + } else { + // address 0 is a reserved value, and the caller interprets it as invalid. + address.set_value(0); + } + } + + EntryCell entry_cell(cell_num, hash, address, small_table_); + if (address.file_type() == BLOCK_EVICTED) + entry_cell.SetGroup(ENTRY_EVICTED); + else + entry_cell.SetGroup(ENTRY_NO_USE); + Save(&entry_cell); + + if (found) { + bitmap_->Set(cell_num, true); + backup_bitmap_->Set(cell_num, true); + header()->used_cells++; + modified_ = true; + } + + return entry_cell; +} + +EntryCell IndexTable::FindEntryCell(uint32 hash, Addr address) { + return FindEntryCellImpl(hash, address, false); +} + +int IndexTable::CalculateTimestamp(Time time) { + TimeDelta delta = time - Time::FromInternalValue(header_->base_time); + return std::max(delta.InMinutes(), 0); +} + +base::Time IndexTable::TimeFromTimestamp(int timestamp) { + return Time::FromInternalValue(header_->base_time) + + TimeDelta::FromMinutes(timestamp); +} + +void IndexTable::SetSate(uint32 hash, Addr address, EntryState state) { + EntryCell cell = FindEntryCellImpl(hash, address, state == ENTRY_FREE); + if (!cell.IsValid()) { + NOTREACHED(); + return; + } + + EntryState old_state = cell.GetState(); + switch (state) { + case ENTRY_FREE: + DCHECK_EQ(old_state, ENTRY_DELETED); + break; + case ENTRY_NEW: + DCHECK_EQ(old_state, ENTRY_FREE); + break; + case ENTRY_OPEN: + DCHECK_EQ(old_state, ENTRY_USED); + break; + case ENTRY_MODIFIED: + DCHECK_EQ(old_state, ENTRY_OPEN); + break; + case ENTRY_DELETED: + DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN || + old_state == ENTRY_MODIFIED); + break; + case ENTRY_USED: + DCHECK(old_state == ENTRY_NEW || old_state == ENTRY_OPEN || + old_state == ENTRY_MODIFIED); + break; + case ENTRY_FIXING: + break; + }; + + modified_ = true; + if (state == ENTRY_DELETED) { + bitmap_->Set(cell.cell_num(), false); + backup_bitmap_->Set(cell.cell_num(), false); + } else if (state == ENTRY_FREE) { + cell.Clear(); + Write(cell); + header()->used_cells--; + return; + } + cell.SetState(state); + + Save(&cell); +} + +void IndexTable::UpdateTime(uint32 hash, Addr address, base::Time current) { + EntryCell cell = FindEntryCell(hash, address); + if (!cell.IsValid()) + return; + + int minutes = CalculateTimestamp(current); + + // Keep about 3 months of headroom. + const int kMaxTimestamp = (1 << 20) - 60 * 24 * 90; + if (minutes > kMaxTimestamp) { + // TODO(rvargas): + // Update header->old_time and trigger a timer + // Rebaseline timestamps and don't update sums + // Start a timer (about 2 backups) + // fix all ckecksums and trigger another timer + // update header->old_time because rebaseline is done. + minutes = std::min(minutes, (1 << 20) - 1); + } + + cell.SetTimestamp(minutes); + Save(&cell); +} + +void IndexTable::Save(EntryCell* cell) { + cell->FixSum(); + Write(*cell); +} + +void IndexTable::GetOldest(IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use) { + no_use->forward = true; + low_use->forward = true; + high_use->forward = true; + InitIterator(no_use); + InitIterator(low_use); + InitIterator(high_use); + + WalkTables(-1, no_use, low_use, high_use); +} + +bool IndexTable::GetNextCells(IndexIterator* iterator) { + int current_time = iterator->timestamp; + InitIterator(iterator); + + WalkTables(current_time, iterator, iterator, iterator); + return !iterator->cells.empty(); +} + +void IndexTable::OnBackupTimer() { + if (!modified_) + return; + + int num_words = (header_->table_len + 31) / 32; + int num_bytes = num_words * 4 + static_cast<int>(sizeof(*header_)); + scoped_refptr<net::IOBuffer> buffer(new net::IOBuffer(num_bytes)); + memcpy(buffer->data(), header_, sizeof(*header_)); + memcpy(buffer->data() + sizeof(*header_), backup_bitmap_storage_.get(), + num_words * 4); + backend_->SaveIndex(buffer, num_bytes); + modified_ = false; +} + +// ----------------------------------------------------------------------- + +EntryCell IndexTable::FindEntryCellImpl(uint32 hash, Addr address, + bool allow_deleted) { + int bucket_num = static_cast<int>(hash & mask_); + IndexBucket* bucket = &main_table_[bucket_num]; + do { + for (int i = 0; i < kCellsPerBucket; i++) { + IndexCell* current_cell = &bucket->cells[i]; + if (!GetLocation(*current_cell)) + continue; + DCHECK(SanityCheck(*current_cell)); + if (IsHashMatch(*current_cell, hash)) { + // We have a match. + int cell_num = bucket_num * kCellsPerBucket + i; + EntryCell entry_cell(cell_num, hash, *current_cell, small_table_); + if (entry_cell.GetAddress() != address) + continue; + + if (!allow_deleted && entry_cell.GetState() == ENTRY_DELETED) + continue; + + return entry_cell; + } + } + bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_, + &bucket); + } while (bucket_num); + return EntryCell(); +} + +void IndexTable::CheckState(const EntryCell& cell) { + int current_state = cell.GetState(); + if (current_state != ENTRY_FIXING) { + bool present = ((current_state & 3) != 0); // Look at the last two bits. + if (present != bitmap_->Get(cell.cell_num()) || + present != backup_bitmap_->Get(cell.cell_num())) { + // There's a mismatch. + if (current_state == ENTRY_DELETED) { + // We were in the process of deleting this entry. Finish now. + backend_->DeleteCell(cell); + } else { + current_state = ENTRY_FIXING; + EntryCell bad_cell(cell); + bad_cell.SetState(ENTRY_FIXING); + Save(&bad_cell); + } + } + } + + if (current_state == ENTRY_FIXING) + backend_->FixCell(cell); +} + +void IndexTable::Write(const EntryCell& cell) { + IndexBucket* bucket = NULL; + int bucket_num = cell.cell_num() / kCellsPerBucket; + if (bucket_num < static_cast<int32>(mask_ + 1)) { + bucket = &main_table_[bucket_num]; + } else { + DCHECK_LE(bucket_num, header()->max_bucket); + bucket = &extra_table_[bucket_num - (mask_ + 1)]; + } + + int cell_number = cell.cell_num() % kCellsPerBucket; + if (GetLocation(bucket->cells[cell_number]) && cell.GetLocation()) { + DCHECK_EQ(cell.GetLocation(), + GetLocation(bucket->cells[cell_number])); + } + cell.Serialize(&bucket->cells[cell_number]); +} + +int IndexTable::NewExtraBucket() { + int safe_window = (header()->table_len < kNumExtraBlocks * 2) ? + kNumExtraBlocks / 4 : kNumExtraBlocks; + if (header()->table_len - header()->max_bucket * kCellsPerBucket < + safe_window) { + backend_->GrowIndex(); + } + + if (header()->max_bucket * kCellsPerBucket == + header()->table_len - kCellsPerBucket) { + return 0; + } + + header()->max_bucket++; + return header()->max_bucket; +} + +void IndexTable::WalkTables(int limit_time, + IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use) { + header_->num_no_use_entries = 0; + header_->num_low_use_entries = 0; + header_->num_high_use_entries = 0; + header_->num_evicted_entries = 0; + + for (int i = 0; i < static_cast<int32>(mask_ + 1); i++) { + int bucket_num = i; + IndexBucket* bucket = &main_table_[i]; + do { + UpdateFromBucket(bucket, i, limit_time, no_use, low_use, high_use); + + bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_, + &bucket); + } while (bucket_num); + } + header_->num_entries = header_->num_no_use_entries + + header_->num_low_use_entries + + header_->num_high_use_entries + + header_->num_evicted_entries; + modified_ = true; +} + +void IndexTable::UpdateFromBucket(IndexBucket* bucket, int bucket_hash, + int limit_time, + IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use) { + for (int i = 0; i < kCellsPerBucket; i++) { + IndexCell& current_cell = bucket->cells[i]; + if (!GetLocation(current_cell)) + continue; + DCHECK(SanityCheck(current_cell)); + if (!IsNormalState(current_cell)) + continue; + + EntryCell entry_cell(0, GetFullHash(current_cell, bucket_hash), + current_cell, small_table_); + switch (GetCellGroup(current_cell)) { + case ENTRY_NO_USE: + UpdateIterator(entry_cell, limit_time, no_use); + header_->num_no_use_entries++; + break; + case ENTRY_LOW_USE: + UpdateIterator(entry_cell, limit_time, low_use); + header_->num_low_use_entries++; + break; + case ENTRY_HIGH_USE: + UpdateIterator(entry_cell, limit_time, high_use); + header_->num_high_use_entries++; + break; + case ENTRY_EVICTED: + header_->num_evicted_entries++; + break; + default: + NOTREACHED(); + } + } +} + +// This code is only called from Init() so the internal state of this object is +// in flux (this method is performing the last steps of re-initialization). As +// such, random methods are not supposed to work at this point, so whatever this +// method calls should be relatively well controlled and it may require some +// degree of "stable state faking". +void IndexTable::MoveCells(IndexBucket* old_extra_table) { + int max_hash = (mask_ + 1) / 2; + int max_bucket = header()->max_bucket; + header()->max_bucket = mask_; + int used_cells = header()->used_cells; + + // Consider a large cache: a cell stores the upper 18 bits of the hash + // (h >> 14). If the table is say 8 times the original size (growing from 4x), + // the bit that we are interested in would be the 3rd bit of the stored value, + // in other words 'multiplier' >> 1. + uint32 new_bit = (1 << extra_bits_) >> 1; + + scoped_ptr<IndexBucket[]> old_main_table; + IndexBucket* source_table = main_table_; + bool upgrade_format = !extra_bits_; + if (upgrade_format) { + // This method should deal with migrating a small table to a big one. Given + // that the first thing to do is read the old table, set small_table_ for + // the size of the old table. Now, when moving a cell, the result cannot be + // placed in the old table or we will end up reading it again and attempting + // to move it, so we have to copy the whole table at once. + DCHECK(!small_table_); + small_table_ = true; + old_main_table.reset(new IndexBucket[max_hash]); + memcpy(old_main_table.get(), main_table_, max_hash * sizeof(IndexBucket)); + memset(main_table_, 0, max_hash * sizeof(IndexBucket)); + source_table = old_main_table.get(); + } + + for (int i = 0; i < max_hash; i++) { + int bucket_num = i; + IndexBucket* bucket = &source_table[i]; + do { + for (int j = 0; j < kCellsPerBucket; j++) { + IndexCell& current_cell = bucket->cells[j]; + if (!GetLocation(current_cell)) + continue; + DCHECK(SanityCheck(current_cell)); + if (bucket_num == i) { + if (upgrade_format || (GetHashValue(current_cell) & new_bit)) { + // Move this cell to the upper half of the table. + MoveSingleCell(¤t_cell, bucket_num * kCellsPerBucket + j, i, + true); + } + } else { + // All cells on extra buckets have to move. + MoveSingleCell(¤t_cell, bucket_num * kCellsPerBucket + j, i, + true); + } + } + + // There is no need to clear the old bucket->next value because if falls + // within the main table so it will be fixed when attempting to follow + // the link. + bucket_num = GetNextBucket(max_hash, max_bucket, old_extra_table, + &bucket); + } while (bucket_num); + } + + DCHECK_EQ(header()->used_cells, used_cells); + + if (upgrade_format) { + small_table_ = false; + header()->flags &= ~SMALL_CACHE; + } +} + +void IndexTable::MoveSingleCell(IndexCell* current_cell, int cell_num, + int main_table_index, bool growing) { + uint32 hash = GetFullHash(*current_cell, main_table_index); + EntryCell old_cell(cell_num, hash, *current_cell, small_table_); + + // This method may be called when moving entries from a small table to a + // normal table. In that case, the caller (MoveCells) has to read the old + // table, so it needs small_table_ set to true, but this method needs to + // write to the new table so small_table_ has to be set to false, and the + // value restored to true before returning. + bool upgrade_format = !extra_bits_ && growing; + if (upgrade_format) + small_table_ = false; + EntryCell new_cell = CreateEntryCell(hash, old_cell.GetAddress()); + + if (!new_cell.IsValid()) { + // We'll deal with this entry later. + if (upgrade_format) + small_table_ = true; + return; + } + + new_cell.SetState(old_cell.GetState()); + new_cell.SetGroup(old_cell.GetGroup()); + new_cell.SetReuse(old_cell.GetReuse()); + new_cell.SetTimestamp(old_cell.GetTimestamp()); + Save(&new_cell); + modified_ = true; + if (upgrade_format) + small_table_ = true; + + if (old_cell.GetState() == ENTRY_DELETED) { + bitmap_->Set(new_cell.cell_num(), false); + backup_bitmap_->Set(new_cell.cell_num(), false); + } + + if (!growing || cell_num / kCellsPerBucket == main_table_index) { + // Only delete entries that live on the main table. + if (!upgrade_format) { + old_cell.Clear(); + Write(old_cell); + } + + if (cell_num != new_cell.cell_num()) { + bitmap_->Set(old_cell.cell_num(), false); + backup_bitmap_->Set(old_cell.cell_num(), false); + } + } + header()->used_cells--; +} + +void IndexTable::HandleMisplacedCell(IndexCell* current_cell, int cell_num, + int main_table_index) { + NOTREACHED(); // No unit tests yet. + + // The cell may be misplaced, or a duplicate cell exists with this data. + uint32 hash = GetFullHash(*current_cell, main_table_index); + MoveSingleCell(current_cell, cell_num, main_table_index, false); + + // Now look for a duplicate cell. + CheckBucketList(hash & mask_); +} + +void IndexTable::CheckBucketList(int bucket_num) { + typedef std::pair<int, EntryGroup> AddressAndGroup; + std::set<AddressAndGroup> entries; + IndexBucket* bucket = &main_table_[bucket_num]; + int bucket_hash = bucket_num; + do { + for (int i = 0; i < kCellsPerBucket; i++) { + IndexCell* current_cell = &bucket->cells[i]; + if (!GetLocation(*current_cell)) + continue; + if (!SanityCheck(*current_cell)) { + NOTREACHED(); + current_cell->Clear(); + continue; + } + int cell_num = bucket_num * kCellsPerBucket + i; + EntryCell cell(cell_num, GetFullHash(*current_cell, bucket_hash), + *current_cell, small_table_); + if (!entries.insert(std::make_pair(cell.GetAddress().value(), + cell.GetGroup())).second) { + current_cell->Clear(); + continue; + } + CheckState(cell); + } + + bucket_num = GetNextBucket(mask_ + 1, header()->max_bucket, extra_table_, + &bucket); + } while (bucket_num); +} + +uint32 IndexTable::GetLocation(const IndexCell& cell) { + if (small_table_) + return GetCellSmallTableLocation(cell); + + return GetCellLocation(cell); +} + +uint32 IndexTable::GetHashValue(const IndexCell& cell) { + if (small_table_) + return GetCellSmallTableId(cell); + + return GetCellId(cell); +} + +uint32 IndexTable::GetFullHash(const IndexCell& cell, uint32 lower_part) { + // It is OK for the high order bits of lower_part to overlap with the stored + // part of the hash. + if (small_table_) + return (GetCellSmallTableId(cell) << kSmallTableHashShift) | lower_part; + + return (GetCellId(cell) << kHashShift) | lower_part; +} + +// All the bits stored in the cell should match the provided hash. +bool IndexTable::IsHashMatch(const IndexCell& cell, uint32 hash) { + hash = small_table_ ? hash >> kSmallTableHashShift : hash >> kHashShift; + return GetHashValue(cell) == hash; +} + +bool IndexTable::MisplacedHash(const IndexCell& cell, uint32 hash) { + if (!extra_bits_) + return false; + + uint32 mask = (1 << extra_bits_) - 1; + hash = small_table_ ? hash >> kSmallTableHashShift : hash >> kHashShift; + return (GetHashValue(cell) & mask) != (hash & mask); +} + +} // namespace disk_cache diff --git a/net/disk_cache/v3/index_table.h b/net/disk_cache/v3/index_table.h new file mode 100644 index 0000000..5fb398a --- /dev/null +++ b/net/disk_cache/v3/index_table.h @@ -0,0 +1,279 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef NET_DISK_CACHE_V3_INDEX_TABLE_H_ +#define NET_DISK_CACHE_V3_INDEX_TABLE_H_ + +// The IndexTable class is in charge of handling all the details about the main +// index table of the cache. It provides methods to locate entries in the cache, +// create new entries and modify existing entries. It hides the fact that the +// table is backed up across multiple physical files, and that the files can +// grow and be remapped while the cache is in use. However, note that this class +// doesn't do any direct management of the backing files, and it operates only +// with the tables in memory. +// +// When the current index needs to grow, the backend is notified so that files +// are extended and remapped as needed. After that, the IndexTable should be +// re-initialized with the new structures. Note that the IndexTable instance is +// still functional while the backend performs file IO. + +#include <vector> + +#include "base/basictypes.h" +#include "base/memory/ref_counted.h" +#include "base/memory/scoped_ptr.h" +#include "base/time/time.h" +#include "net/base/net_export.h" +#include "net/disk_cache/addr.h" +#include "net/disk_cache/bitmap.h" +#include "net/disk_cache/v3/disk_format_v3.h" + +namespace net { +class IOBuffer; +} + +namespace disk_cache { + +class BackendImplV3; +struct InitResult; + +// An EntryCell represents a single entity stored by the index table. Users are +// expected to handle and store EntryCells on their own to track operations that +// they are performing with a given entity, as opposed to deal with pointers to +// individual positions on the table, given that the whole table can be moved to +// another place, and that would invalidate any pointers to individual cells in +// the table. +// However, note that it is also possible for an entity to be moved from one +// position to another, so an EntryCell may be invalid by the time a long +// operation completes. In that case, the caller should consult the table again +// using FindEntryCell(). +class NET_EXPORT_PRIVATE EntryCell { + public: + ~EntryCell(); + + bool IsValid() const; + + int32 cell_num() const { return cell_num_; } + uint32 hash() const { return hash_; } + + Addr GetAddress() const; + EntryState GetState() const; + EntryGroup GetGroup() const; + int GetReuse() const; + int GetTimestamp() const; + + void SetState(EntryState state); + void SetGroup(EntryGroup group); + void SetReuse(int count); + void SetTimestamp(int timestamp); + + static EntryCell GetEntryCellForTest(int32 cell_num, + uint32 hash, + Addr address, + IndexCell* cell, + bool small_table); + void SerializaForTest(IndexCell* destination); + + private: + friend class IndexTable; + friend class CacheDumperHelper; + + EntryCell(); + EntryCell(int32 cell_num, uint32 hash, Addr address, bool small_table); + EntryCell(int32 cell_num, + uint32 hash, + const IndexCell& cell, + bool small_table); + + void Clear() { cell_.Clear(); } + void FixSum(); + + // Returns the raw value stored on the index table. + uint32 GetLocation() const; + + // Recalculates hash_ assuming that only the low order bits are valid and the + // rest come from cell_. + uint32 RecomputeHash(); + + void Serialize(IndexCell* destination) const; + + int32 cell_num_; + uint32 hash_; + IndexCell cell_; + bool small_table_; +}; + +// Keeps a collection of EntryCells in order to be processed. +struct NET_EXPORT_PRIVATE EntrySet { + EntrySet(); + ~EntrySet(); + + int evicted_count; // The numebr of evicted entries in this set. + size_t current; // The number of the cell that is being processed. + std::vector<EntryCell> cells; +}; + +// A given entity referenced by the index table is uniquely identified by the +// combination of hash and address. +struct CellInfo { uint32 hash; Addr address; }; +typedef std::vector<CellInfo> CellList; + +// An index iterator is used to get a group of cells that share the same +// timestamp. When this structure is passed to GetNextCells(), the caller sets +// the initial timestamp and direction; whet it is used with GetOldest, the +// initial values are ignored. +struct NET_EXPORT_PRIVATE IndexIterator { + IndexIterator(); + ~IndexIterator(); + + CellList cells; + int timestamp; // The current low resolution timestamp for |cells|. + bool forward; // The direction of the iteration, in time. +}; + +// Methods that the backend has to implement to support the table. Note that the +// backend is expected to own all IndexTable instances, so it is expected to +// outlive the table. +class NET_EXPORT_PRIVATE IndexTableBackend { + public: + virtual ~IndexTableBackend() {} + + // The index has to grow. + virtual void GrowIndex() = 0; + + // Save the index to the backup file. + virtual void SaveIndex(net::IOBuffer* buffer, int buffer_len) = 0; + + // Deletes or fixes an invalid cell from the backend. + virtual void DeleteCell(EntryCell cell) = 0; + virtual void FixCell(EntryCell cell) = 0; +}; + +// The data required to initialize an index. Note that not all fields have to +// be provided when growing the tables. +struct NET_EXPORT_PRIVATE IndexTableInitData { + IndexTableInitData(); + ~IndexTableInitData(); + + IndexBitmap* index_bitmap; + IndexBucket* main_table; + IndexBucket* extra_table; + scoped_ptr<IndexHeaderV3> backup_header; + scoped_ptr<uint32[]> backup_bitmap; +}; + +// See the description at the top of this file. +class NET_EXPORT_PRIVATE IndexTable { + public: + explicit IndexTable(IndexTableBackend* backend); + ~IndexTable(); + + // Initializes the object, or re-initializes it when the backing files grow. + // Note that the only supported way to initialize this objeect is using + // pointers that come from the files being directly mapped in memory. If that + // is not the case, it must be emulated in a convincing way, for example + // making sure that the tables for re-init look the same as the tables to be + // replaced. + void Init(IndexTableInitData* params); + + // Releases the resources acquired during Init(). + void Shutdown(); + + // Locates a resouce on the index. Returns a list of all resources that match + // the provided hash. + EntrySet LookupEntries(uint32 hash); + + // Creates a new cell to store a new resource. + EntryCell CreateEntryCell(uint32 hash, Addr address); + + // Locates a particular cell. This method allows a caller to perform slow + // operations with some entries while the index evolves, by returning the + // current state of a cell. If the desired cell cannot be located, the return + // object will be invalid. + EntryCell FindEntryCell(uint32 hash, Addr address); + + // Returns an IndexTable timestamp for a given absolute time. The actual + // resolution of the timestamp should be considered an implementation detail, + // but it certainly is lower than seconds. The important part is that a group + // of cells will share the same timestamp (see IndexIterator). + int CalculateTimestamp(base::Time time); + + // Returns the equivalent time for a cell timestamp. + base::Time TimeFromTimestamp(int timestamp); + + // Updates a particular cell. + void SetSate(uint32 hash, Addr address, EntryState state); + void UpdateTime(uint32 hash, Addr address, base::Time current); + + // Saves the contents of |cell| to the table. + void Save(EntryCell* cell); + + // Returns the oldest entries for each group of entries. The initial values + // for the provided iterators are ignored. Entries are assigned to iterators + // according to their EntryGroup. + void GetOldest(IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use); + + // Returns the next group of entries for the provided iterator. This method + // does not return the cells matching the initial iterator's timestamp, + // but rather cells after (or before, depending on the iterator's |forward| + // member) that timestamp. + bool GetNextCells(IndexIterator* iterator); + + // Called each time the index should save the backup information. The caller + // can assume that anything that needs to be saved is saved when this method + // is called, and that there is only one source of timming information, and + // that source is controlled by the owner of this object. + void OnBackupTimer(); + + IndexHeaderV3* header() { return header_; } + const IndexHeaderV3* header() const { return header_; } + + private: + EntryCell FindEntryCellImpl(uint32 hash, Addr address, bool allow_deleted); + void CheckState(const EntryCell& cell); + void Write(const EntryCell& cell); + int NewExtraBucket(); + void WalkTables(int limit_time, + IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use); + void UpdateFromBucket(IndexBucket* bucket, int bucket_hash, + int limit_time, + IndexIterator* no_use, + IndexIterator* low_use, + IndexIterator* high_use); + void MoveCells(IndexBucket* old_extra_table); + void MoveSingleCell(IndexCell* current_cell, int cell_num, + int main_table_index, bool growing); + void HandleMisplacedCell(IndexCell* current_cell, int cell_num, + int main_table_index); + void CheckBucketList(int bucket_id); + + uint32 GetLocation(const IndexCell& cell); + uint32 GetHashValue(const IndexCell& cell); + uint32 GetFullHash(const IndexCell& cell, uint32 lower_part); + bool IsHashMatch(const IndexCell& cell, uint32 hash); + bool MisplacedHash(const IndexCell& cell, uint32 hash); + + IndexTableBackend* backend_; + IndexHeaderV3* header_; + scoped_ptr<Bitmap> bitmap_; + scoped_ptr<Bitmap> backup_bitmap_; + scoped_ptr<uint32[]> backup_bitmap_storage_; + scoped_ptr<IndexHeaderV3> backup_header_; + IndexBucket* main_table_; + IndexBucket* extra_table_; + uint32 mask_; // Binary mask to map a hash to the hash table. + int extra_bits_; // How many bits are in mask_ above the default value. + bool modified_; + bool small_table_; + + DISALLOW_COPY_AND_ASSIGN(IndexTable); +}; + +} // namespace disk_cache + +#endif // NET_DISK_CACHE_V3_INDEX_TABLE_H_ diff --git a/net/disk_cache/v3/index_table_unittest.cc b/net/disk_cache/v3/index_table_unittest.cc new file mode 100644 index 0000000..c9fb920 --- /dev/null +++ b/net/disk_cache/v3/index_table_unittest.cc @@ -0,0 +1,706 @@ +// Copyright 2014 The Chromium Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "base/basictypes.h" +#include "base/logging.h" +#include "net/disk_cache/addr.h" +#include "net/disk_cache/v3/disk_format_v3.h" +#include "net/disk_cache/v3/index_table.h" +#include "testing/gtest/include/gtest/gtest.h" + +using disk_cache::EntryCell; +using disk_cache::IndexCell; +using disk_cache::IndexTable; +using disk_cache::IndexTableInitData; + +namespace { + +int GetChecksum(const IndexCell& source) { + // Only the cell pointer is relevant. + disk_cache::Addr addr; + IndexCell* cell = const_cast<IndexCell*>(&source); + EntryCell entry = EntryCell::GetEntryCellForTest(0, 0, addr, cell, false); + + IndexCell result; + entry.SerializaForTest(&result); + return result.last_part >> 6; +} + +class MockIndexBackend : public disk_cache::IndexTableBackend { + public: + MockIndexBackend() : grow_called_(false), buffer_len_(-1) {} + virtual ~MockIndexBackend() {} + + bool grow_called() const { return grow_called_; } + int buffer_len() const { return buffer_len_; } + + virtual void GrowIndex() OVERRIDE { grow_called_ = true; } + virtual void SaveIndex(net::IOBuffer* buffer, int buffer_len) OVERRIDE { + buffer_len_ = buffer_len; + } + virtual void DeleteCell(EntryCell cell) OVERRIDE {} + virtual void FixCell(EntryCell cell) OVERRIDE {} + + private: + bool grow_called_; + int buffer_len_; +}; + +class TestCacheTables { + public: + // |num_entries| is the capacity of the main table. The extra table is half + // the size of the main table. + explicit TestCacheTables(int num_entries); + ~TestCacheTables() {} + + void GetInitData(IndexTableInitData* result); + void CopyFrom(const TestCacheTables& other); + base::Time start_time() const { return start_time_; } + + private: + scoped_ptr<uint64[]> main_bitmap_; + scoped_ptr<disk_cache::IndexBucket[]> main_table_; + scoped_ptr<disk_cache::IndexBucket[]> extra_table_; + base::Time start_time_; + int num_bitmap_bytes_; + + DISALLOW_COPY_AND_ASSIGN(TestCacheTables); +}; + +TestCacheTables::TestCacheTables(int num_entries) { + DCHECK_GE(num_entries, 1024); + DCHECK_EQ(num_entries, num_entries / 1024 * 1024); + main_table_.reset(new disk_cache::IndexBucket[num_entries]); + extra_table_.reset(new disk_cache::IndexBucket[num_entries / 2]); + memset(main_table_.get(), 0, num_entries * sizeof(*main_table_.get())); + memset(extra_table_.get(), 0, num_entries / 2 * sizeof(*extra_table_.get())); + + // We allow IndexBitmap smaller than a page because the code should not really + // depend on that. + num_bitmap_bytes_ = (num_entries + num_entries / 2) / 8; + size_t required_size = sizeof(disk_cache::IndexHeaderV3) + num_bitmap_bytes_; + main_bitmap_.reset(new uint64[required_size / sizeof(uint64)]); + memset(main_bitmap_.get(), 0, required_size); + + disk_cache::IndexHeaderV3* header = + reinterpret_cast<disk_cache::IndexHeaderV3*>(main_bitmap_.get()); + + header->magic = disk_cache::kIndexMagicV3; + header->version = disk_cache::kVersion3; + header->table_len = num_entries + num_entries / 2; + header->max_bucket = num_entries / 4 - 1; + + start_time_ = base::Time::Now(); + header->create_time = start_time_.ToInternalValue(); + header->base_time = + (start_time_ - base::TimeDelta::FromDays(20)).ToInternalValue(); + + if (num_entries < 64 * 1024) + header->flags = disk_cache::SMALL_CACHE; +} + +void TestCacheTables::GetInitData(IndexTableInitData* result) { + result->index_bitmap = + reinterpret_cast<disk_cache::IndexBitmap*>(main_bitmap_.get()); + + result->main_table = main_table_.get(); + result->extra_table = extra_table_.get(); + + result->backup_header.reset(new disk_cache::IndexHeaderV3); + memcpy(result->backup_header.get(), result->index_bitmap, + sizeof(result->index_bitmap->header)); + + result->backup_bitmap.reset(new uint32[num_bitmap_bytes_ / sizeof(uint32)]); + memcpy(result->backup_bitmap.get(), result->index_bitmap->bitmap, + num_bitmap_bytes_); +} + +void TestCacheTables::CopyFrom(const TestCacheTables& other) { + disk_cache::IndexBitmap* this_bitmap = + reinterpret_cast<disk_cache::IndexBitmap*>(main_bitmap_.get()); + disk_cache::IndexBitmap* other_bitmap = + reinterpret_cast<disk_cache::IndexBitmap*>(other.main_bitmap_.get()); + + DCHECK_GE(this_bitmap->header.table_len, other_bitmap->header.table_len); + DCHECK_GE(num_bitmap_bytes_, other.num_bitmap_bytes_); + + memcpy(this_bitmap->bitmap, other_bitmap->bitmap, other.num_bitmap_bytes_); + + int main_table_buckets = (other_bitmap->header.table_len * 2 / 3) / 4; + int extra_table_buckets = (other_bitmap->header.table_len * 1 / 3) / 4; + memcpy(main_table_.get(), other.main_table_.get(), + main_table_buckets * sizeof(disk_cache::IndexBucket)); + memcpy(extra_table_.get(), other.extra_table_.get(), + extra_table_buckets * sizeof(disk_cache::IndexBucket)); + + this_bitmap->header.num_entries = other_bitmap->header.num_entries; + this_bitmap->header.used_cells = other_bitmap->header.used_cells; + this_bitmap->header.max_bucket = other_bitmap->header.max_bucket; + this_bitmap->header.create_time = other_bitmap->header.create_time; + this_bitmap->header.base_time = other_bitmap->header.base_time; + this_bitmap->header.flags = other_bitmap->header.flags; + start_time_ = other.start_time_; +} + +} // namespace + +TEST(DiskCacheIndexTable, EntryCell) { + uint32 hash = 0x55aa6699; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, 0x4531); + bool small_table = true; + int cell_num = 88; + int reuse = 6; + int timestamp = 123456; + disk_cache::EntryState state = disk_cache::ENTRY_MODIFIED; + disk_cache::EntryGroup group = disk_cache::ENTRY_HIGH_USE; + + for (int i = 0; i < 4; i++) { + SCOPED_TRACE(i); + EntryCell entry = EntryCell::GetEntryCellForTest(cell_num, hash, addr, NULL, + small_table); + EXPECT_EQ(disk_cache::ENTRY_NO_USE, entry.GetGroup()); + EXPECT_EQ(disk_cache::ENTRY_NEW, entry.GetState()); + + entry.SetGroup(group); + entry.SetState(state); + entry.SetReuse(reuse); + entry.SetTimestamp(timestamp); + + EXPECT_TRUE(entry.IsValid()); + EXPECT_EQ(hash, entry.hash()); + EXPECT_EQ(cell_num, entry.cell_num()); + EXPECT_EQ(addr.value(), entry.GetAddress().value()); + + EXPECT_EQ(group, entry.GetGroup()); + EXPECT_EQ(state, entry.GetState()); + EXPECT_EQ(reuse, entry.GetReuse()); + EXPECT_EQ(timestamp, entry.GetTimestamp()); + + // Store the data and read it again. + IndexCell cell; + entry.SerializaForTest(&cell); + + EntryCell entry2 = EntryCell::GetEntryCellForTest(cell_num, hash, addr, + &cell, small_table); + + EXPECT_EQ(addr.value(), entry2.GetAddress().value()); + + EXPECT_EQ(group, entry2.GetGroup()); + EXPECT_EQ(state, entry2.GetState()); + EXPECT_EQ(reuse, entry2.GetReuse()); + EXPECT_EQ(timestamp, entry2.GetTimestamp()); + + small_table = !small_table; + if (i == 1) { + hash = ~hash; + cell_num *= 5; + state = disk_cache::ENTRY_USED; + group = disk_cache::ENTRY_EVICTED; + addr = disk_cache::Addr(disk_cache::BLOCK_EVICTED, 1, 6, 0x18a5); + reuse = 15; // 4 bits + timestamp = 0xfffff; // 20 bits. + } + } +} + +// Goes over some significant values for a cell's sum. +TEST(DiskCacheIndexTable, EntryCellSum) { + IndexCell source; + source.Clear(); + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part++; + EXPECT_EQ(1, GetChecksum(source)); + + source.Clear(); + source.last_part = 0x80; + EXPECT_EQ(0, GetChecksum(source)); + + source.last_part = 0x55; + EXPECT_EQ(3, GetChecksum(source)); + + source.first_part = 0x555555; + EXPECT_EQ(2, GetChecksum(source)); + + source.last_part = 0; + EXPECT_EQ(1, GetChecksum(source)); + + source.first_part = GG_UINT64_C(0x8000000080000000); + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part = GG_UINT64_C(0x4000000040000000); + EXPECT_EQ(2, GetChecksum(source)); + + source.first_part = GG_UINT64_C(0x200000020000000); + EXPECT_EQ(1, GetChecksum(source)); + + source.first_part = GG_UINT64_C(0x100000010010000); + EXPECT_EQ(3, GetChecksum(source)); + + source.first_part = 0x80008000; + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part = GG_UINT64_C(0x800000008000); + EXPECT_EQ(1, GetChecksum(source)); + + source.first_part = 0x8080; + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part = 0x800080; + EXPECT_EQ(1, GetChecksum(source)); + + source.first_part = 0x88; + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part = 0x808; + EXPECT_EQ(1, GetChecksum(source)); + + source.first_part = 0xA; + EXPECT_EQ(0, GetChecksum(source)); + + source.first_part = 0x22; + EXPECT_EQ(1, GetChecksum(source)); +} + +TEST(DiskCacheIndexTable, Basics) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + + IndexTable index(NULL); + index.Init(&init_data); + + // Write some entries. + disk_cache::CellList entries; + for (int i = 0; i < 250; i++) { + SCOPED_TRACE(i); + uint32 hash = i * i * 1111 + i * 11; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i * 13 + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + } + + // Read them back. + for (size_t i = 0; i < entries.size(); i++) { + SCOPED_TRACE(i); + uint32 hash = entries[i].hash; + disk_cache::Addr addr = entries[i].address; + + disk_cache::EntrySet found_entries = index.LookupEntries(hash); + ASSERT_EQ(1u, found_entries.cells.size()); + EXPECT_TRUE(found_entries.cells[0].IsValid()); + EXPECT_EQ(hash, found_entries.cells[0].hash()); + EXPECT_EQ(addr.value(), found_entries.cells[0].GetAddress().value()); + + EntryCell entry = index.FindEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + EXPECT_EQ(hash, entry.hash()); + EXPECT_EQ(addr.value(), entry.GetAddress().value()); + + // Delete the first 100 entries. + if (i < 100) + index.SetSate(hash, addr, disk_cache::ENTRY_DELETED); + } + + // See what we have now. + for (size_t i = 0; i < entries.size(); i++) { + SCOPED_TRACE(i); + uint32 hash = entries[i].hash; + disk_cache::Addr addr = entries[i].address; + + disk_cache::EntrySet found_entries = index.LookupEntries(hash); + if (i < 100) { + EXPECT_EQ(0u, found_entries.cells.size()); + } else { + ASSERT_EQ(1u, found_entries.cells.size()); + EXPECT_TRUE(found_entries.cells[0].IsValid()); + EXPECT_EQ(hash, found_entries.cells[0].hash()); + EXPECT_EQ(addr.value(), found_entries.cells[0].GetAddress().value()); + } + } +} + +// Tests handling of multiple entries with the same hash. +TEST(DiskCacheIndexTable, SameHash) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + + IndexTable index(NULL); + index.Init(&init_data); + + disk_cache::CellList entries; + uint32 hash = 0x55aa55bb; + for (int i = 0; i < 6; i++) { + SCOPED_TRACE(i); + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i * 13 + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + } + + disk_cache::EntrySet found_entries = index.LookupEntries(hash); + EXPECT_EQ(0, found_entries.evicted_count); + ASSERT_EQ(6u, found_entries.cells.size()); + + for (size_t i = 0; i < found_entries.cells.size(); i++) { + SCOPED_TRACE(i); + EXPECT_EQ(entries[i].address, found_entries.cells[i].GetAddress()); + } + + // Now verify handling of entries on different states. + index.SetSate(hash, entries[0].address, disk_cache::ENTRY_DELETED); + index.SetSate(hash, entries[1].address, disk_cache::ENTRY_DELETED); + index.SetSate(hash, entries[2].address, disk_cache::ENTRY_USED); + index.SetSate(hash, entries[3].address, disk_cache::ENTRY_USED); + index.SetSate(hash, entries[4].address, disk_cache::ENTRY_USED); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(0, found_entries.evicted_count); + ASSERT_EQ(4u, found_entries.cells.size()); + + index.SetSate(hash, entries[3].address, disk_cache::ENTRY_OPEN); + index.SetSate(hash, entries[4].address, disk_cache::ENTRY_OPEN); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(0, found_entries.evicted_count); + ASSERT_EQ(4u, found_entries.cells.size()); + + index.SetSate(hash, entries[4].address, disk_cache::ENTRY_MODIFIED); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(0, found_entries.evicted_count); + ASSERT_EQ(4u, found_entries.cells.size()); + + index.SetSate(hash, entries[1].address, disk_cache::ENTRY_FREE); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(0, found_entries.evicted_count); + ASSERT_EQ(4u, found_entries.cells.size()); + + // FindEntryCell should not see deleted entries. + EntryCell entry = index.FindEntryCell(hash, entries[0].address); + EXPECT_FALSE(entry.IsValid()); + + // A free entry is gone. + entry = index.FindEntryCell(hash, entries[1].address); + EXPECT_FALSE(entry.IsValid()); + + // Locate a used entry, and evict it. This is not really a correct operation + // in that an existing cell doesn't transition to evicted; instead a new cell + // for the evicted entry (on a different block file) should be created. Still, + // at least evicted_count would be valid. + entry = index.FindEntryCell(hash, entries[2].address); + EXPECT_TRUE(entry.IsValid()); + entry.SetGroup(disk_cache::ENTRY_EVICTED); + index.Save(&entry); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(1, found_entries.evicted_count); + ASSERT_EQ(4u, found_entries.cells.size()); + + // Now use the proper way to get an evicted entry. + disk_cache::Addr addr2(disk_cache::BLOCK_EVICTED, 1, 6, 6); // Any address. + entry = index.CreateEntryCell(hash, addr2); + EXPECT_TRUE(entry.IsValid()); + EXPECT_EQ(disk_cache::ENTRY_EVICTED, entry.GetGroup()); + + found_entries = index.LookupEntries(hash); + EXPECT_EQ(2, found_entries.evicted_count); + ASSERT_EQ(5u, found_entries.cells.size()); +} + +TEST(DiskCacheIndexTable, Timestamps) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + + IndexTable index(NULL); + index.Init(&init_data); + + // The granularity should be 1 minute. + int timestamp1 = index.CalculateTimestamp(cache.start_time()); + int timestamp2 = index.CalculateTimestamp(cache.start_time() + + base::TimeDelta::FromSeconds(59)); + EXPECT_EQ(timestamp1, timestamp2); + + int timestamp3 = index.CalculateTimestamp(cache.start_time() + + base::TimeDelta::FromSeconds(61)); + EXPECT_EQ(timestamp1 + 1, timestamp3); + + int timestamp4 = index.CalculateTimestamp(cache.start_time() + + base::TimeDelta::FromSeconds(119)); + EXPECT_EQ(timestamp1 + 1, timestamp4); + + int timestamp5 = index.CalculateTimestamp(cache.start_time() + + base::TimeDelta::FromSeconds(121)); + EXPECT_EQ(timestamp1 + 2, timestamp5); + + int timestamp6 = index.CalculateTimestamp(cache.start_time() - + base::TimeDelta::FromSeconds(30)); + EXPECT_EQ(timestamp1 - 1, timestamp6); + + // The base should be 20 days in the past. + int timestamp7 = index.CalculateTimestamp(cache.start_time() - + base::TimeDelta::FromDays(20)); + int timestamp8 = index.CalculateTimestamp(cache.start_time() - + base::TimeDelta::FromDays(35)); + EXPECT_EQ(timestamp7, timestamp8); + EXPECT_EQ(0, timestamp8); + + int timestamp9 = index.CalculateTimestamp(cache.start_time() - + base::TimeDelta::FromDays(19)); + EXPECT_NE(0, timestamp9); +} + +// Tests GetOldest and GetNextCells. +TEST(DiskCacheIndexTable, Iterations) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + + IndexTable index(NULL); + index.Init(&init_data); + + base::Time time = cache.start_time(); + + // Write some entries. + disk_cache::CellList entries; + for (int i = 0; i < 44; i++) { + SCOPED_TRACE(i); + uint32 hash = i; // The entries will be ordered on the table. + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i * 13 + 1); + if (i < 10 || i == 40) + addr = disk_cache::Addr(disk_cache::BLOCK_EVICTED, 1, 6, i * 13 + 1); + + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + + if (i < 10 || i == 40) { + // Do nothing. These are ENTRY_EVICTED by default. + } else if (i < 20 || i == 41) { + entry.SetGroup(disk_cache::ENTRY_HIGH_USE); + index.Save(&entry); + } else if (i < 30 || i == 42) { + entry.SetGroup(disk_cache::ENTRY_LOW_USE); + index.Save(&entry); + } + + // Entries [30,39] and 43 are marked as ENTRY_NO_USE (the default). + + if (!(i % 10)) + time += base::TimeDelta::FromMinutes(1); + + index.UpdateTime(hash, addr, time); + } + + // Get the oldest entries of each group. + disk_cache::IndexIterator no_use, low_use, high_use; + index.GetOldest(&no_use, &low_use, &high_use); + ASSERT_EQ(10u, no_use.cells.size()); + ASSERT_EQ(10u, low_use.cells.size()); + ASSERT_EQ(10u, high_use.cells.size()); + + EXPECT_EQ(entries[10].hash, high_use.cells[0].hash); + EXPECT_EQ(entries[19].hash, high_use.cells[9].hash); + EXPECT_EQ(entries[20].hash, low_use.cells[0].hash); + EXPECT_EQ(entries[29].hash, low_use.cells[9].hash); + EXPECT_EQ(entries[30].hash, no_use.cells[0].hash); + EXPECT_EQ(entries[39].hash, no_use.cells[9].hash); + + // Now start an iteration from the head (most recent entry). + disk_cache::IndexIterator iterator; + iterator.timestamp = index.CalculateTimestamp(time) + 1; + iterator.forward = false; // Back in time. + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(3u, iterator.cells.size()); + EXPECT_EQ(entries[41].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[42].hash, iterator.cells[1].hash); + EXPECT_EQ(entries[43].hash, iterator.cells[2].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[30].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[39].hash, iterator.cells[9].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[20].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[29].hash, iterator.cells[9].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[10].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[19].hash, iterator.cells[9].hash); + + ASSERT_FALSE(index.GetNextCells(&iterator)); + + // Now start an iteration from the tail (oldest entry). + iterator.timestamp = 0; + iterator.forward = true; + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[10].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[19].hash, iterator.cells[9].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[20].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[29].hash, iterator.cells[9].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(10u, iterator.cells.size()); + EXPECT_EQ(entries[30].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[39].hash, iterator.cells[9].hash); + + ASSERT_TRUE(index.GetNextCells(&iterator)); + ASSERT_EQ(3u, iterator.cells.size()); + EXPECT_EQ(entries[41].hash, iterator.cells[0].hash); + EXPECT_EQ(entries[42].hash, iterator.cells[1].hash); + EXPECT_EQ(entries[43].hash, iterator.cells[2].hash); +} + +// Tests doubling of the table. +TEST(DiskCacheIndexTable, Doubling) { + IndexTable index(NULL); + int size = 1024; + scoped_ptr<TestCacheTables> cache(new TestCacheTables(size)); + int entry_id = 0; + disk_cache::CellList entries; + + // Go from 1024 to 256k cells. + for (int resizes = 0; resizes <= 8; resizes++) { + scoped_ptr<TestCacheTables> old_cache(cache.Pass()); + cache.reset(new TestCacheTables(size)); + cache.get()->CopyFrom(*old_cache.get()); + + IndexTableInitData init_data; + cache.get()->GetInitData(&init_data); + index.Init(&init_data); + + // Write some entries. + for (int i = 0; i < 250; i++, entry_id++) { + SCOPED_TRACE(entry_id); + uint32 hash = entry_id * i * 321 + entry_id * 13; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, entry_id * 17 + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + } + size *= 2; + } + + // Access all the entries. + for (size_t i = 0; i < entries.size(); i++) { + SCOPED_TRACE(i); + disk_cache::EntrySet found_entries = index.LookupEntries(entries[i].hash); + ASSERT_EQ(1u, found_entries.cells.size()); + EXPECT_TRUE(found_entries.cells[0].IsValid()); + } +} + +// Tests bucket chaining when growing the index. +TEST(DiskCacheIndexTable, BucketChains) { + IndexTable index(NULL); + int size = 1024; + scoped_ptr<TestCacheTables> cache(new TestCacheTables(size)); + disk_cache::CellList entries; + + IndexTableInitData init_data; + cache.get()->GetInitData(&init_data); + index.Init(&init_data); + + // Write some entries. + for (int i = 0; i < 8; i++) { + SCOPED_TRACE(i); + uint32 hash = i * 256; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i * 7 + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + } + + // Double the size. + scoped_ptr<TestCacheTables> old_cache(cache.Pass()); + cache.reset(new TestCacheTables(size * 2)); + cache.get()->CopyFrom(*old_cache.get()); + + cache.get()->GetInitData(&init_data); + index.Init(&init_data); + + // Write more entries, starting with the upper half of the table. + for (int i = 9; i < 11; i++) { + SCOPED_TRACE(i); + uint32 hash = i * 256; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i * 7 + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + disk_cache::CellInfo info = { hash, addr }; + entries.push_back(info); + } + + // Access all the entries. + for (size_t i = 0; i < entries.size(); i++) { + SCOPED_TRACE(i); + disk_cache::EntrySet found_entries = index.LookupEntries(entries[i].hash); + ASSERT_EQ(1u, found_entries.cells.size()); + EXPECT_TRUE(found_entries.cells[0].IsValid()); + } +} + +// Tests that GrowIndex is called. +TEST(DiskCacheIndexTable, GrowIndex) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + MockIndexBackend backend; + + IndexTable index(&backend); + index.Init(&init_data); + + // Write some entries. + for (int i = 0; i < 512; i++) { + SCOPED_TRACE(i); + uint32 hash = 0; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, i + 1); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + } + + EXPECT_TRUE(backend.grow_called()); +} + +TEST(DiskCacheIndexTable, SaveIndex) { + TestCacheTables cache(1024); + IndexTableInitData init_data; + cache.GetInitData(&init_data); + MockIndexBackend backend; + + IndexTable index(&backend); + index.Init(&init_data); + + uint32 hash = 0; + disk_cache::Addr addr(disk_cache::BLOCK_ENTRIES, 1, 5, 6); + EntryCell entry = index.CreateEntryCell(hash, addr); + EXPECT_TRUE(entry.IsValid()); + + index.OnBackupTimer(); + int expected = (1024 + 512) / 8 + sizeof(disk_cache::IndexHeaderV3); + EXPECT_EQ(expected, backend.buffer_len()); +} diff --git a/net/net.gyp b/net/net.gyp index d4b3d04..7e645e4 100644 --- a/net/net.gyp +++ b/net/net.gyp @@ -460,6 +460,8 @@ 'disk_cache/v3/block_bitmaps.cc', 'disk_cache/v3/block_bitmaps.h', 'disk_cache/v3/disk_format_v3.h', + 'disk_cache/v3/index_table.cc', + 'disk_cache/v3/index_table.h', 'dns/address_sorter.h', 'dns/address_sorter_posix.cc', 'dns/address_sorter_posix.h', @@ -1673,6 +1675,7 @@ 'disk_cache/flash/segment_unittest.cc', 'disk_cache/flash/storage_unittest.cc', 'disk_cache/v3/block_bitmaps_unittest.cc', + 'disk_cache/v3/index_table_unittest.cc', 'dns/address_sorter_posix_unittest.cc', 'dns/address_sorter_unittest.cc', 'dns/dns_config_service_posix_unittest.cc', |