| // Copyright (c) 2012 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. |
| |
| // Performs basic inspection of the disk cache files with minimal disruption |
| // to the actual files (they still may change if an error is detected on the |
| // files). |
| |
| #include "net/tools/dump_cache/dump_files.h" |
| |
| #include <stdio.h> |
| |
| #include <memory> |
| #include <set> |
| #include <string> |
| |
| #include "base/command_line.h" |
| #include "base/files/file.h" |
| #include "base/files/file_enumerator.h" |
| #include "base/files/file_util.h" |
| #include "base/format_macros.h" |
| #include "base/macros.h" |
| #include "base/message_loop/message_loop.h" |
| #include "base/strings/string_number_conversions.h" |
| #include "base/strings/stringprintf.h" |
| #include "net/disk_cache/blockfile/block_files.h" |
| #include "net/disk_cache/blockfile/disk_format.h" |
| #include "net/disk_cache/blockfile/mapped_file.h" |
| #include "net/disk_cache/blockfile/stats.h" |
| #include "net/disk_cache/blockfile/storage_block-inl.h" |
| #include "net/disk_cache/blockfile/storage_block.h" |
| #include "net/url_request/view_cache_helper.h" |
| #include "starboard/memory.h" |
| #include "starboard/types.h" |
| |
| namespace { |
| |
| const base::FilePath::CharType kIndexName[] = FILE_PATH_LITERAL("index"); |
| |
| // Reads the |header_size| bytes from the beginning of file |name|. |
| bool ReadHeader(const base::FilePath& name, char* header, int header_size) { |
| base::File file(name, base::File::FLAG_OPEN | base::File::FLAG_READ); |
| if (!file.IsValid()) { |
| printf("Unable to open file %s\n", name.MaybeAsASCII().c_str()); |
| return false; |
| } |
| |
| int read = file.Read(0, header, header_size); |
| if (read != header_size) { |
| printf("Unable to read file %s\n", name.MaybeAsASCII().c_str()); |
| return false; |
| } |
| return true; |
| } |
| |
| int GetMajorVersionFromFile(const base::FilePath& name) { |
| disk_cache::IndexHeader header; |
| if (!ReadHeader(name, reinterpret_cast<char*>(&header), sizeof(header))) |
| return 0; |
| |
| return header.version >> 16; |
| } |
| |
| // Dumps the contents of the Stats record. |
| void DumpStats(const base::FilePath& path, disk_cache::CacheAddr addr) { |
| // We need a message loop, although we really don't run any task. |
| base::MessageLoopForIO loop; |
| |
| disk_cache::BlockFiles block_files(path); |
| if (!block_files.Init(false)) { |
| printf("Unable to init block files\n"); |
| return; |
| } |
| |
| disk_cache::Addr address(addr); |
| disk_cache::MappedFile* file = block_files.GetFile(address); |
| if (!file) |
| return; |
| |
| size_t length = (2 + disk_cache::Stats::kDataSizesLength) * sizeof(int32_t) + |
| disk_cache::Stats::MAX_COUNTER * sizeof(int64_t); |
| |
| size_t offset = address.start_block() * address.BlockSize() + |
| disk_cache::kBlockHeaderSize; |
| |
| std::unique_ptr<int32_t[]> buffer(new int32_t[length]); |
| if (!file->Read(buffer.get(), length, offset)) |
| return; |
| |
| printf("Stats:\nSignatrure: 0x%x\n", buffer[0]); |
| printf("Total size: %d\n", buffer[1]); |
| for (int i = 0; i < disk_cache::Stats::kDataSizesLength; i++) |
| printf("Size(%d): %d\n", i, buffer[i + 2]); |
| |
| int64_t* counters = reinterpret_cast<int64_t*>( |
| buffer.get() + 2 + disk_cache::Stats::kDataSizesLength); |
| for (int i = 0; i < disk_cache::Stats::MAX_COUNTER; i++) |
| printf("Count(%d): %" PRId64 "\n", i, *counters++); |
| printf("-------------------------\n\n"); |
| } |
| |
| // Dumps the contents of the Index-file header. |
| void DumpIndexHeader(const base::FilePath& name, |
| disk_cache::CacheAddr* stats_addr) { |
| disk_cache::IndexHeader header; |
| if (!ReadHeader(name, reinterpret_cast<char*>(&header), sizeof(header))) |
| return; |
| |
| printf("Index file:\n"); |
| printf("magic: %x\n", header.magic); |
| printf("version: %d.%d\n", header.version >> 16, header.version & 0xffff); |
| printf("entries: %d\n", header.num_entries); |
| printf("total bytes: %d\n", header.num_bytes); |
| printf("last file number: %d\n", header.last_file); |
| printf("current id: %d\n", header.this_id); |
| printf("table length: %d\n", header.table_len); |
| printf("last crash: %d\n", header.crash); |
| printf("experiment: %d\n", header.experiment); |
| printf("stats: %x\n", header.stats); |
| for (int i = 0; i < 5; i++) { |
| printf("head %d: 0x%x\n", i, header.lru.heads[i]); |
| printf("tail %d: 0x%x\n", i, header.lru.tails[i]); |
| printf("size %d: 0x%x\n", i, header.lru.sizes[i]); |
| } |
| printf("transaction: 0x%x\n", header.lru.transaction); |
| printf("operation: %d\n", header.lru.operation); |
| printf("operation list: %d\n", header.lru.operation_list); |
| printf("-------------------------\n\n"); |
| |
| if (stats_addr) |
| *stats_addr = header.stats; |
| } |
| |
| // Dumps the contents of a block-file header. |
| void DumpBlockHeader(const base::FilePath& name) { |
| disk_cache::BlockFileHeader header; |
| if (!ReadHeader(name, reinterpret_cast<char*>(&header), sizeof(header))) |
| return; |
| |
| printf("Block file: %s\n", name.BaseName().MaybeAsASCII().c_str()); |
| printf("magic: %x\n", header.magic); |
| printf("version: %d.%d\n", header.version >> 16, header.version & 0xffff); |
| printf("file id: %d\n", header.this_file); |
| printf("next file id: %d\n", header.next_file); |
| printf("entry size: %d\n", header.entry_size); |
| printf("current entries: %d\n", header.num_entries); |
| printf("max entries: %d\n", header.max_entries); |
| printf("updating: %d\n", header.updating); |
| printf("empty sz 1: %d\n", header.empty[0]); |
| printf("empty sz 2: %d\n", header.empty[1]); |
| printf("empty sz 3: %d\n", header.empty[2]); |
| printf("empty sz 4: %d\n", header.empty[3]); |
| printf("user 0: 0x%x\n", header.user[0]); |
| printf("user 1: 0x%x\n", header.user[1]); |
| printf("user 2: 0x%x\n", header.user[2]); |
| printf("user 3: 0x%x\n", header.user[3]); |
| printf("-------------------------\n\n"); |
| } |
| |
| // Simple class that interacts with the set of cache files. |
| class CacheDumper { |
| public: |
| explicit CacheDumper(const base::FilePath& path) |
| : path_(path), |
| block_files_(path), |
| index_(NULL), |
| current_hash_(0), |
| next_addr_(0) { |
| } |
| |
| bool Init(); |
| |
| // Reads an entry from disk. Return false when all entries have been already |
| // returned. |
| bool GetEntry(disk_cache::EntryStore* entry, disk_cache::CacheAddr* addr); |
| |
| // Loads a specific block from the block files. |
| bool LoadEntry(disk_cache::CacheAddr addr, disk_cache::EntryStore* entry); |
| bool LoadRankings(disk_cache::CacheAddr addr, |
| disk_cache::RankingsNode* rankings); |
| |
| // Appends the data store at |addr| to |out|. |
| bool HexDump(disk_cache::CacheAddr addr, std::string* out); |
| |
| private: |
| base::FilePath path_; |
| disk_cache::BlockFiles block_files_; |
| scoped_refptr<disk_cache::MappedFile> index_file_; |
| disk_cache::Index* index_; |
| int current_hash_; |
| disk_cache::CacheAddr next_addr_; |
| std::set<disk_cache::CacheAddr> dumped_entries_; |
| DISALLOW_COPY_AND_ASSIGN(CacheDumper); |
| }; |
| |
| bool CacheDumper::Init() { |
| if (!block_files_.Init(false)) { |
| printf("Unable to init block files\n"); |
| return false; |
| } |
| |
| base::FilePath index_name(path_.Append(kIndexName)); |
| index_file_ = new disk_cache::MappedFile; |
| index_ = reinterpret_cast<disk_cache::Index*>( |
| index_file_->Init(index_name, 0)); |
| if (!index_) { |
| printf("Unable to map index\n"); |
| return false; |
| } |
| |
| return true; |
| } |
| |
| bool CacheDumper::GetEntry(disk_cache::EntryStore* entry, |
| disk_cache::CacheAddr* addr) { |
| if (dumped_entries_.find(next_addr_) != dumped_entries_.end()) { |
| printf("Loop detected\n"); |
| next_addr_ = 0; |
| current_hash_++; |
| } |
| |
| if (next_addr_) { |
| *addr = next_addr_; |
| if (LoadEntry(next_addr_, entry)) |
| return true; |
| |
| printf("Unable to load entry at address 0x%x\n", next_addr_); |
| next_addr_ = 0; |
| current_hash_++; |
| } |
| |
| for (int i = current_hash_; i < index_->header.table_len; i++) { |
| // Yes, we'll crash if the table is shorter than expected, but only after |
| // dumping every entry that we can find. |
| if (index_->table[i]) { |
| current_hash_ = i; |
| *addr = index_->table[i]; |
| if (LoadEntry(index_->table[i], entry)) |
| return true; |
| |
| printf("Unable to load entry at address 0x%x\n", index_->table[i]); |
| } |
| } |
| return false; |
| } |
| |
| bool CacheDumper::LoadEntry(disk_cache::CacheAddr addr, |
| disk_cache::EntryStore* entry) { |
| disk_cache::Addr address(addr); |
| disk_cache::MappedFile* file = block_files_.GetFile(address); |
| if (!file) |
| return false; |
| |
| disk_cache::StorageBlock<disk_cache::EntryStore> entry_block(file, address); |
| if (!entry_block.Load()) |
| return false; |
| |
| SbMemoryCopy(entry, entry_block.Data(), sizeof(*entry)); |
| if (!entry_block.VerifyHash()) |
| printf("Self hash failed at 0x%x\n", addr); |
| |
| // Prepare for the next entry to load. |
| next_addr_ = entry->next; |
| if (next_addr_) { |
| dumped_entries_.insert(addr); |
| } else { |
| current_hash_++; |
| dumped_entries_.clear(); |
| } |
| return true; |
| } |
| |
| bool CacheDumper::LoadRankings(disk_cache::CacheAddr addr, |
| disk_cache::RankingsNode* rankings) { |
| disk_cache::Addr address(addr); |
| if (address.file_type() != disk_cache::RANKINGS) |
| return false; |
| |
| disk_cache::MappedFile* file = block_files_.GetFile(address); |
| if (!file) |
| return false; |
| |
| disk_cache::StorageBlock<disk_cache::RankingsNode> rank_block(file, address); |
| if (!rank_block.Load()) |
| return false; |
| |
| if (!rank_block.VerifyHash()) |
| printf("Self hash failed at 0x%x\n", addr); |
| |
| SbMemoryCopy(rankings, rank_block.Data(), sizeof(*rankings)); |
| return true; |
| } |
| |
| bool CacheDumper::HexDump(disk_cache::CacheAddr addr, std::string* out) { |
| disk_cache::Addr address(addr); |
| disk_cache::MappedFile* file = block_files_.GetFile(address); |
| if (!file) |
| return false; |
| |
| size_t size = address.num_blocks() * address.BlockSize(); |
| std::unique_ptr<char> buffer(new char[size]); |
| |
| size_t offset = address.start_block() * address.BlockSize() + |
| disk_cache::kBlockHeaderSize; |
| if (!file->Read(buffer.get(), size, offset)) |
| return false; |
| |
| base::StringAppendF(out, "0x%x:\n", addr); |
| net::ViewCacheHelper::HexDump(buffer.get(), size, out); |
| return true; |
| } |
| |
| std::string ToLocalTime(int64_t time_us) { |
| base::Time time = base::Time::FromInternalValue(time_us); |
| base::Time::Exploded e; |
| time.LocalExplode(&e); |
| return base::StringPrintf("%d/%d/%d %d:%d:%d.%d", e.year, e.month, |
| e.day_of_month, e.hour, e.minute, e.second, |
| e.millisecond); |
| } |
| |
| void DumpEntry(disk_cache::CacheAddr addr, |
| const disk_cache::EntryStore& entry, |
| bool verbose) { |
| std::string key; |
| static bool full_key = |
| base::CommandLine::ForCurrentProcess()->HasSwitch("full-key"); |
| if (!entry.long_key) { |
| key = std::string(entry.key, std::min(static_cast<size_t>(entry.key_len), |
| sizeof(entry.key))); |
| if (entry.key_len > 90 && !full_key) |
| key.resize(90); |
| } |
| |
| printf("Entry at 0x%x\n", addr); |
| printf("rankings: 0x%x\n", entry.rankings_node); |
| printf("key length: %d\n", entry.key_len); |
| printf("key: \"%s\"\n", key.c_str()); |
| |
| if (verbose) { |
| printf("key addr: 0x%x\n", entry.long_key); |
| printf("hash: 0x%x\n", entry.hash); |
| printf("next entry: 0x%x\n", entry.next); |
| printf("reuse count: %d\n", entry.reuse_count); |
| printf("refetch count: %d\n", entry.refetch_count); |
| printf("state: %d\n", entry.state); |
| printf("creation: %s\n", ToLocalTime(entry.creation_time).c_str()); |
| for (int i = 0; i < 4; i++) { |
| printf("data size %d: %d\n", i, entry.data_size[i]); |
| printf("data addr %d: 0x%x\n", i, entry.data_addr[i]); |
| } |
| printf("----------\n\n"); |
| } |
| } |
| |
| void DumpRankings(disk_cache::CacheAddr addr, |
| const disk_cache::RankingsNode& rankings, |
| bool verbose) { |
| printf("Rankings at 0x%x\n", addr); |
| printf("next: 0x%x\n", rankings.next); |
| printf("prev: 0x%x\n", rankings.prev); |
| printf("entry: 0x%x\n", rankings.contents); |
| |
| if (verbose) { |
| printf("dirty: %d\n", rankings.dirty); |
| if (rankings.last_used != rankings.last_modified) |
| printf("used: %s\n", ToLocalTime(rankings.last_used).c_str()); |
| printf("modified: %s\n", ToLocalTime(rankings.last_modified).c_str()); |
| printf("hash: 0x%x\n", rankings.self_hash); |
| printf("----------\n\n"); |
| } else { |
| printf("\n"); |
| } |
| } |
| |
| void PrintCSVHeader() { |
| printf( |
| "entry,rankings,next,prev,rank-contents,chain,reuse,key," |
| "d0,d1,d2,d3\n"); |
| } |
| |
| void DumpCSV(disk_cache::CacheAddr addr, |
| const disk_cache::EntryStore& entry, |
| const disk_cache::RankingsNode& rankings) { |
| printf("0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x,0x%x\n", addr, |
| entry.rankings_node, rankings.next, rankings.prev, rankings.contents, |
| entry.next, entry.reuse_count, entry.long_key, entry.data_addr[0], |
| entry.data_addr[1], entry.data_addr[2], entry.data_addr[3]); |
| |
| if (addr != rankings.contents) |
| printf("Broken entry\n"); |
| } |
| |
| bool CanDump(disk_cache::CacheAddr addr) { |
| disk_cache::Addr address(addr); |
| return address.is_initialized() && address.is_block_file(); |
| } |
| |
| } // namespace. |
| |
| // ----------------------------------------------------------------------- |
| |
| int GetMajorVersion(const base::FilePath& input_path) { |
| base::FilePath index_name(input_path.Append(kIndexName)); |
| |
| int version = GetMajorVersionFromFile(index_name); |
| if (!version) |
| return 0; |
| |
| base::FilePath data_name(input_path.Append(FILE_PATH_LITERAL("data_0"))); |
| if (version != GetMajorVersionFromFile(data_name)) |
| return 0; |
| |
| data_name = input_path.Append(FILE_PATH_LITERAL("data_1")); |
| if (version != GetMajorVersionFromFile(data_name)) |
| return 0; |
| |
| data_name = input_path.Append(FILE_PATH_LITERAL("data_2")); |
| if (version != GetMajorVersionFromFile(data_name)) |
| return 0; |
| |
| data_name = input_path.Append(FILE_PATH_LITERAL("data_3")); |
| if (version != GetMajorVersionFromFile(data_name)) |
| return 0; |
| |
| return version; |
| } |
| |
| // Dumps the headers of all files. |
| int DumpHeaders(const base::FilePath& input_path) { |
| base::FilePath index_name(input_path.Append(kIndexName)); |
| disk_cache::CacheAddr stats_addr = 0; |
| DumpIndexHeader(index_name, &stats_addr); |
| |
| base::FileEnumerator iter(input_path, false, |
| base::FileEnumerator::FILES, |
| FILE_PATH_LITERAL("data_*")); |
| for (base::FilePath file = iter.Next(); !file.empty(); file = iter.Next()) |
| DumpBlockHeader(file); |
| |
| DumpStats(input_path, stats_addr); |
| return 0; |
| } |
| |
| // Dumps all entries from the cache. |
| int DumpContents(const base::FilePath& input_path) { |
| bool print_csv = base::CommandLine::ForCurrentProcess()->HasSwitch("csv"); |
| if (!print_csv) |
| DumpIndexHeader(input_path.Append(kIndexName), nullptr); |
| |
| // We need a message loop, although we really don't run any task. |
| base::MessageLoopForIO loop; |
| CacheDumper dumper(input_path); |
| if (!dumper.Init()) |
| return -1; |
| |
| if (print_csv) |
| PrintCSVHeader(); |
| |
| disk_cache::EntryStore entry; |
| disk_cache::CacheAddr addr; |
| bool verbose = base::CommandLine::ForCurrentProcess()->HasSwitch("v"); |
| while (dumper.GetEntry(&entry, &addr)) { |
| if (!print_csv) |
| DumpEntry(addr, entry, verbose); |
| disk_cache::RankingsNode rankings; |
| if (!dumper.LoadRankings(entry.rankings_node, &rankings)) |
| continue; |
| |
| if (print_csv) |
| DumpCSV(addr, entry, rankings); |
| else |
| DumpRankings(entry.rankings_node, rankings, verbose); |
| } |
| |
| printf("Done.\n"); |
| |
| return 0; |
| } |
| |
| int DumpLists(const base::FilePath& input_path) { |
| base::FilePath index_name(input_path.Append(kIndexName)); |
| disk_cache::IndexHeader header; |
| if (!ReadHeader(index_name, reinterpret_cast<char*>(&header), sizeof(header))) |
| return -1; |
| |
| // We need a message loop, although we really don't run any task. |
| base::MessageLoopForIO loop; |
| CacheDumper dumper(input_path); |
| if (!dumper.Init()) |
| return -1; |
| |
| printf("list, addr, next, prev, entry\n"); |
| |
| const int kMaxLength = 1 * 1000 * 1000; |
| for (int i = 0; i < 5; i++) { |
| int32_t size = header.lru.sizes[i]; |
| if (size < 0 || size > kMaxLength) { |
| printf("Wrong size %d\n", size); |
| size = kMaxLength; |
| } |
| |
| disk_cache::CacheAddr addr = header.lru.tails[i]; |
| int count = 0; |
| for (; size && addr; size--) { |
| count++; |
| disk_cache::RankingsNode rankings; |
| if (!dumper.LoadRankings(addr, &rankings)) { |
| printf("Failed to load node at 0x%x\n", addr); |
| break; |
| } |
| printf("%d, 0x%x, 0x%x, 0x%x, 0x%x\n", i, addr, rankings.next, |
| rankings.prev, rankings.contents); |
| |
| if (rankings.prev == addr) |
| break; |
| |
| addr = rankings.prev; |
| } |
| printf("%d nodes found, %d reported\n", count, header.lru.sizes[i]); |
| } |
| |
| printf("Done.\n"); |
| return 0; |
| } |
| |
| int DumpEntryAt(const base::FilePath& input_path, const std::string& at) { |
| disk_cache::CacheAddr addr; |
| if (!base::HexStringToUInt(at, &addr)) |
| return -1; |
| |
| if (!CanDump(addr)) |
| return -1; |
| |
| base::FilePath index_name(input_path.Append(kIndexName)); |
| disk_cache::IndexHeader header; |
| if (!ReadHeader(index_name, reinterpret_cast<char*>(&header), sizeof(header))) |
| return -1; |
| |
| // We need a message loop, although we really don't run any task. |
| base::MessageLoopForIO loop; |
| CacheDumper dumper(input_path); |
| if (!dumper.Init()) |
| return -1; |
| |
| disk_cache::CacheAddr entry_addr = 0; |
| disk_cache::CacheAddr rankings_addr = 0; |
| disk_cache::Addr address(addr); |
| |
| disk_cache::RankingsNode rankings; |
| if (address.file_type() == disk_cache::RANKINGS) { |
| if (dumper.LoadRankings(addr, &rankings)) { |
| rankings_addr = addr; |
| addr = rankings.contents; |
| address = disk_cache::Addr(addr); |
| } |
| } |
| |
| disk_cache::EntryStore entry = {}; |
| if (address.file_type() == disk_cache::BLOCK_256 && |
| dumper.LoadEntry(addr, &entry)) { |
| entry_addr = addr; |
| DumpEntry(addr, entry, true); |
| if (!rankings_addr && dumper.LoadRankings(entry.rankings_node, &rankings)) |
| rankings_addr = entry.rankings_node; |
| } |
| |
| bool verbose = base::CommandLine::ForCurrentProcess()->HasSwitch("v"); |
| |
| std::string hex_dump; |
| if (!rankings_addr || verbose) |
| dumper.HexDump(addr, &hex_dump); |
| |
| if (rankings_addr) |
| DumpRankings(rankings_addr, rankings, true); |
| |
| if (entry_addr && verbose) { |
| if (entry.long_key && CanDump(entry.long_key)) |
| dumper.HexDump(entry.long_key, &hex_dump); |
| |
| for (int i = 0; i < 4; i++) { |
| if (entry.data_addr[i] && CanDump(entry.data_addr[i])) |
| dumper.HexDump(entry.data_addr[i], &hex_dump); |
| } |
| } |
| |
| printf("%s\n", hex_dump.c_str()); |
| printf("Done.\n"); |
| return 0; |
| } |
| |
| int DumpAllocation(const base::FilePath& file) { |
| disk_cache::BlockFileHeader header; |
| if (!ReadHeader(file, reinterpret_cast<char*>(&header), sizeof(header))) |
| return -1; |
| |
| std::string out; |
| net::ViewCacheHelper::HexDump(reinterpret_cast<char*>(&header.allocation_map), |
| sizeof(header.allocation_map), &out); |
| printf("%s\n", out.c_str()); |
| return 0; |
| } |