/* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ #ifndef CacheIndex__h__ #define CacheIndex__h__ #include "CacheLog.h" #include "CacheFileIOManager.h" #include "nsIRunnable.h" #include "CacheHashUtils.h" #include "nsICacheStorageService.h" #include "nsICacheEntry.h" #include "nsILoadContextInfo.h" #include "nsTHashtable.h" #include "nsThreadUtils.h" #include "nsWeakReference.h" #include "mozilla/SHA1.h" #include "mozilla/StaticMutex.h" #include "mozilla/StaticPtr.h" #include "mozilla/EndianUtils.h" #include "mozilla/TimeStamp.h" class nsIFile; class nsIDirectoryEnumerator; class nsITimer; #ifdef DEBUG #define DEBUG_STATS 1 #endif namespace mozilla { namespace net { class CacheFileMetadata; class FileOpenHelper; class CacheIndexIterator; typedef struct { // Version of the index. The index must be ignored and deleted when the file // on disk was written with a newer version. uint32_t mVersion; // Timestamp of time when the last successful write of the index started. // During update process we use this timestamp for a quick validation of entry // files. If last modified time of the file is lower than this timestamp, we // skip parsing of such file since the information in index should be up to // date. uint32_t mTimeStamp; // We set this flag as soon as possible after parsing index during startup // and clean it after we write journal to disk during shutdown. We ignore the // journal and start update process whenever this flag is set during index // parsing. uint32_t mIsDirty; } CacheIndexHeader; static_assert( sizeof(CacheIndexHeader::mVersion) + sizeof(CacheIndexHeader::mTimeStamp) + sizeof(CacheIndexHeader::mIsDirty) == sizeof(CacheIndexHeader), "Unexpected sizeof(CacheIndexHeader)!"); struct CacheIndexRecord { SHA1Sum::Hash mHash; uint32_t mFrecency; OriginAttrsHash mOriginAttrsHash; uint32_t mExpirationTime; /* * 1000 0000 0000 0000 0000 0000 0000 0000 : initialized * 0100 0000 0000 0000 0000 0000 0000 0000 : anonymous * 0010 0000 0000 0000 0000 0000 0000 0000 : removed * 0001 0000 0000 0000 0000 0000 0000 0000 : dirty * 0000 1000 0000 0000 0000 0000 0000 0000 : fresh * 0000 0100 0000 0000 0000 0000 0000 0000 : pinned * 0000 0011 0000 0000 0000 0000 0000 0000 : reserved * 0000 0000 1111 1111 1111 1111 1111 1111 : file size (in kB) */ uint32_t mFlags; CacheIndexRecord() : mFrecency(0) , mOriginAttrsHash(0) , mExpirationTime(nsICacheEntry::NO_EXPIRATION_TIME) , mFlags(0) {} }; static_assert( sizeof(CacheIndexRecord::mHash) + sizeof(CacheIndexRecord::mFrecency) + sizeof(CacheIndexRecord::mOriginAttrsHash) + sizeof(CacheIndexRecord::mExpirationTime) + sizeof(CacheIndexRecord::mFlags) == sizeof(CacheIndexRecord), "Unexpected sizeof(CacheIndexRecord)!"); class CacheIndexEntry : public PLDHashEntryHdr { public: typedef const SHA1Sum::Hash& KeyType; typedef const SHA1Sum::Hash* KeyTypePointer; explicit CacheIndexEntry(KeyTypePointer aKey) { MOZ_COUNT_CTOR(CacheIndexEntry); mRec = new CacheIndexRecord(); LOG(("CacheIndexEntry::CacheIndexEntry() - Created record [rec=%p]", mRec.get())); memcpy(&mRec->mHash, aKey, sizeof(SHA1Sum::Hash)); } CacheIndexEntry(const CacheIndexEntry& aOther) { NS_NOTREACHED("CacheIndexEntry copy constructor is forbidden!"); } ~CacheIndexEntry() { MOZ_COUNT_DTOR(CacheIndexEntry); LOG(("CacheIndexEntry::~CacheIndexEntry() - Deleting record [rec=%p]", mRec.get())); } // KeyEquals(): does this entry match this key? bool KeyEquals(KeyTypePointer aKey) const { return memcmp(&mRec->mHash, aKey, sizeof(SHA1Sum::Hash)) == 0; } // KeyToPointer(): Convert KeyType to KeyTypePointer static KeyTypePointer KeyToPointer(KeyType aKey) { return &aKey; } // HashKey(): calculate the hash number static PLDHashNumber HashKey(KeyTypePointer aKey) { return (reinterpret_cast<const uint32_t *>(aKey))[0]; } // ALLOW_MEMMOVE can we move this class with memmove(), or do we have // to use the copy constructor? enum { ALLOW_MEMMOVE = true }; bool operator==(const CacheIndexEntry& aOther) const { return KeyEquals(&aOther.mRec->mHash); } CacheIndexEntry& operator=(const CacheIndexEntry& aOther) { MOZ_ASSERT(memcmp(&mRec->mHash, &aOther.mRec->mHash, sizeof(SHA1Sum::Hash)) == 0); mRec->mFrecency = aOther.mRec->mFrecency; mRec->mExpirationTime = aOther.mRec->mExpirationTime; mRec->mOriginAttrsHash = aOther.mRec->mOriginAttrsHash; mRec->mFlags = aOther.mRec->mFlags; return *this; } void InitNew() { mRec->mFrecency = 0; mRec->mExpirationTime = nsICacheEntry::NO_EXPIRATION_TIME; mRec->mOriginAttrsHash = 0; mRec->mFlags = 0; } void Init(OriginAttrsHash aOriginAttrsHash, bool aAnonymous, bool aPinned) { MOZ_ASSERT(mRec->mFrecency == 0); MOZ_ASSERT(mRec->mExpirationTime == nsICacheEntry::NO_EXPIRATION_TIME); MOZ_ASSERT(mRec->mOriginAttrsHash == 0); // When we init the entry it must be fresh and may be dirty MOZ_ASSERT((mRec->mFlags & ~kDirtyMask) == kFreshMask); mRec->mOriginAttrsHash = aOriginAttrsHash; mRec->mFlags |= kInitializedMask; if (aAnonymous) { mRec->mFlags |= kAnonymousMask; } if (aPinned) { mRec->mFlags |= kPinnedMask; } } const SHA1Sum::Hash * Hash() const { return &mRec->mHash; } bool IsInitialized() const { return !!(mRec->mFlags & kInitializedMask); } mozilla::net::OriginAttrsHash OriginAttrsHash() const { return mRec->mOriginAttrsHash; } bool Anonymous() const { return !!(mRec->mFlags & kAnonymousMask); } bool IsRemoved() const { return !!(mRec->mFlags & kRemovedMask); } void MarkRemoved() { mRec->mFlags |= kRemovedMask; } bool IsDirty() const { return !!(mRec->mFlags & kDirtyMask); } void MarkDirty() { mRec->mFlags |= kDirtyMask; } void ClearDirty() { mRec->mFlags &= ~kDirtyMask; } bool IsFresh() const { return !!(mRec->mFlags & kFreshMask); } void MarkFresh() { mRec->mFlags |= kFreshMask; } bool IsPinned() const { return !!(mRec->mFlags & kPinnedMask); } void SetFrecency(uint32_t aFrecency) { mRec->mFrecency = aFrecency; } uint32_t GetFrecency() const { return mRec->mFrecency; } void SetExpirationTime(uint32_t aExpirationTime) { mRec->mExpirationTime = aExpirationTime; } uint32_t GetExpirationTime() const { return mRec->mExpirationTime; } // Sets filesize in kilobytes. void SetFileSize(uint32_t aFileSize) { if (aFileSize > kFileSizeMask) { LOG(("CacheIndexEntry::SetFileSize() - FileSize is too large, " "truncating to %u", kFileSizeMask)); aFileSize = kFileSizeMask; } mRec->mFlags &= ~kFileSizeMask; mRec->mFlags |= aFileSize; } // Returns filesize in kilobytes. uint32_t GetFileSize() const { return GetFileSize(mRec); } static uint32_t GetFileSize(CacheIndexRecord *aRec) { return aRec->mFlags & kFileSizeMask; } static uint32_t IsPinned(CacheIndexRecord *aRec) { return aRec->mFlags & kPinnedMask; } bool IsFileEmpty() const { return GetFileSize() == 0; } void WriteToBuf(void *aBuf) { uint8_t* ptr = static_cast<uint8_t*>(aBuf); memcpy(ptr, mRec->mHash, sizeof(SHA1Sum::Hash)); ptr += sizeof(SHA1Sum::Hash); NetworkEndian::writeUint32(ptr, mRec->mFrecency); ptr += sizeof(uint32_t); NetworkEndian::writeUint64(ptr, mRec->mOriginAttrsHash); ptr += sizeof(uint64_t); NetworkEndian::writeUint32(ptr, mRec->mExpirationTime); ptr += sizeof(uint32_t); // Dirty and fresh flags should never go to disk, since they make sense only // during current session. NetworkEndian::writeUint32(ptr, mRec->mFlags & ~(kDirtyMask | kFreshMask)); } void ReadFromBuf(void *aBuf) { const uint8_t* ptr = static_cast<const uint8_t*>(aBuf); MOZ_ASSERT(memcmp(&mRec->mHash, ptr, sizeof(SHA1Sum::Hash)) == 0); ptr += sizeof(SHA1Sum::Hash); mRec->mFrecency = NetworkEndian::readUint32(ptr); ptr += sizeof(uint32_t); mRec->mOriginAttrsHash = NetworkEndian::readUint64(ptr); ptr += sizeof(uint64_t); mRec->mExpirationTime = NetworkEndian::readUint32(ptr); ptr += sizeof(uint32_t); mRec->mFlags = NetworkEndian::readUint32(ptr); } void Log() const { LOG(("CacheIndexEntry::Log() [this=%p, hash=%08x%08x%08x%08x%08x, fresh=%u," " initialized=%u, removed=%u, dirty=%u, anonymous=%u, " "originAttrsHash=%llx, frecency=%u, expirationTime=%u, size=%u]", this, LOGSHA1(mRec->mHash), IsFresh(), IsInitialized(), IsRemoved(), IsDirty(), Anonymous(), OriginAttrsHash(), GetFrecency(), GetExpirationTime(), GetFileSize())); } static bool RecordMatchesLoadContextInfo(CacheIndexRecord *aRec, nsILoadContextInfo *aInfo) { if (!aInfo->IsPrivate() && GetOriginAttrsHash(*aInfo->OriginAttributesPtr()) == aRec->mOriginAttrsHash && aInfo->IsAnonymous() == !!(aRec->mFlags & kAnonymousMask)) { return true; } return false; } // Memory reporting size_t SizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const { return mallocSizeOf(mRec.get()); } size_t SizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const { return mallocSizeOf(this) + SizeOfExcludingThis(mallocSizeOf); } private: friend class CacheIndexEntryUpdate; friend class CacheIndex; friend class CacheIndexEntryAutoManage; static const uint32_t kInitializedMask = 0x80000000; static const uint32_t kAnonymousMask = 0x40000000; // This flag is set when the entry was removed. We need to keep this // information in memory until we write the index file. static const uint32_t kRemovedMask = 0x20000000; // This flag is set when the information in memory is not in sync with the // information in index file on disk. static const uint32_t kDirtyMask = 0x10000000; // This flag is set when the information about the entry is fresh, i.e. // we've created or opened this entry during this session, or we've seen // this entry during update or build process. static const uint32_t kFreshMask = 0x08000000; // Indicates a pinned entry. static const uint32_t kPinnedMask = 0x04000000; static const uint32_t kReservedMask = 0x03000000; // FileSize in kilobytes static const uint32_t kFileSizeMask = 0x00FFFFFF; nsAutoPtr<CacheIndexRecord> mRec; }; class CacheIndexEntryUpdate : public CacheIndexEntry { public: explicit CacheIndexEntryUpdate(CacheIndexEntry::KeyTypePointer aKey) : CacheIndexEntry(aKey) , mUpdateFlags(0) { MOZ_COUNT_CTOR(CacheIndexEntryUpdate); LOG(("CacheIndexEntryUpdate::CacheIndexEntryUpdate()")); } ~CacheIndexEntryUpdate() { MOZ_COUNT_DTOR(CacheIndexEntryUpdate); LOG(("CacheIndexEntryUpdate::~CacheIndexEntryUpdate()")); } CacheIndexEntryUpdate& operator=(const CacheIndexEntry& aOther) { MOZ_ASSERT(memcmp(&mRec->mHash, &aOther.mRec->mHash, sizeof(SHA1Sum::Hash)) == 0); mUpdateFlags = 0; *(static_cast<CacheIndexEntry *>(this)) = aOther; return *this; } void InitNew() { mUpdateFlags = kFrecencyUpdatedMask | kExpirationUpdatedMask | kFileSizeUpdatedMask; CacheIndexEntry::InitNew(); } void SetFrecency(uint32_t aFrecency) { mUpdateFlags |= kFrecencyUpdatedMask; CacheIndexEntry::SetFrecency(aFrecency); } void SetExpirationTime(uint32_t aExpirationTime) { mUpdateFlags |= kExpirationUpdatedMask; CacheIndexEntry::SetExpirationTime(aExpirationTime); } void SetFileSize(uint32_t aFileSize) { mUpdateFlags |= kFileSizeUpdatedMask; CacheIndexEntry::SetFileSize(aFileSize); } void ApplyUpdate(CacheIndexEntry *aDst) { MOZ_ASSERT(memcmp(&mRec->mHash, &aDst->mRec->mHash, sizeof(SHA1Sum::Hash)) == 0); if (mUpdateFlags & kFrecencyUpdatedMask) { aDst->mRec->mFrecency = mRec->mFrecency; } if (mUpdateFlags & kExpirationUpdatedMask) { aDst->mRec->mExpirationTime = mRec->mExpirationTime; } aDst->mRec->mOriginAttrsHash = mRec->mOriginAttrsHash; if (mUpdateFlags & kFileSizeUpdatedMask) { aDst->mRec->mFlags = mRec->mFlags; } else { // Copy all flags except file size. aDst->mRec->mFlags &= kFileSizeMask; aDst->mRec->mFlags |= (mRec->mFlags & ~kFileSizeMask); } } private: static const uint32_t kFrecencyUpdatedMask = 0x00000001; static const uint32_t kExpirationUpdatedMask = 0x00000002; static const uint32_t kFileSizeUpdatedMask = 0x00000004; uint32_t mUpdateFlags; }; class CacheIndexStats { public: CacheIndexStats() : mCount(0) , mNotInitialized(0) , mRemoved(0) , mDirty(0) , mFresh(0) , mEmpty(0) , mSize(0) #ifdef DEBUG , mStateLogged(false) , mDisableLogging(false) #endif { } bool operator==(const CacheIndexStats& aOther) const { return #ifdef DEBUG aOther.mStateLogged == mStateLogged && #endif aOther.mCount == mCount && aOther.mNotInitialized == mNotInitialized && aOther.mRemoved == mRemoved && aOther.mDirty == mDirty && aOther.mFresh == mFresh && aOther.mEmpty == mEmpty && aOther.mSize == mSize; } #ifdef DEBUG void DisableLogging() { mDisableLogging = true; } #endif void Log() { LOG(("CacheIndexStats::Log() [count=%u, notInitialized=%u, removed=%u, " "dirty=%u, fresh=%u, empty=%u, size=%u]", mCount, mNotInitialized, mRemoved, mDirty, mFresh, mEmpty, mSize)); } void Clear() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::Clear() - state logged!"); mCount = 0; mNotInitialized = 0; mRemoved = 0; mDirty = 0; mFresh = 0; mEmpty = 0; mSize = 0; } #ifdef DEBUG bool StateLogged() { return mStateLogged; } #endif uint32_t Count() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::Count() - state logged!"); return mCount; } uint32_t Dirty() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::Dirty() - state logged!"); return mDirty; } uint32_t Fresh() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::Fresh() - state logged!"); return mFresh; } uint32_t ActiveEntriesCount() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::ActiveEntriesCount() - state " "logged!"); return mCount - mRemoved - mNotInitialized - mEmpty; } uint32_t Size() { MOZ_ASSERT(!mStateLogged, "CacheIndexStats::Size() - state logged!"); return mSize; } void BeforeChange(const CacheIndexEntry *aEntry) { #ifdef DEBUG_STATS if (!mDisableLogging) { LOG(("CacheIndexStats::BeforeChange()")); Log(); } #endif MOZ_ASSERT(!mStateLogged, "CacheIndexStats::BeforeChange() - state " "logged!"); #ifdef DEBUG mStateLogged = true; #endif if (aEntry) { MOZ_ASSERT(mCount); mCount--; if (aEntry->IsDirty()) { MOZ_ASSERT(mDirty); mDirty--; } if (aEntry->IsFresh()) { MOZ_ASSERT(mFresh); mFresh--; } if (aEntry->IsRemoved()) { MOZ_ASSERT(mRemoved); mRemoved--; } else { if (!aEntry->IsInitialized()) { MOZ_ASSERT(mNotInitialized); mNotInitialized--; } else { if (aEntry->IsFileEmpty()) { MOZ_ASSERT(mEmpty); mEmpty--; } else { MOZ_ASSERT(mSize >= aEntry->GetFileSize()); mSize -= aEntry->GetFileSize(); } } } } } void AfterChange(const CacheIndexEntry *aEntry) { MOZ_ASSERT(mStateLogged, "CacheIndexStats::AfterChange() - state not " "logged!"); #ifdef DEBUG mStateLogged = false; #endif if (aEntry) { ++mCount; if (aEntry->IsDirty()) { mDirty++; } if (aEntry->IsFresh()) { mFresh++; } if (aEntry->IsRemoved()) { mRemoved++; } else { if (!aEntry->IsInitialized()) { mNotInitialized++; } else { if (aEntry->IsFileEmpty()) { mEmpty++; } else { mSize += aEntry->GetFileSize(); } } } } #ifdef DEBUG_STATS if (!mDisableLogging) { LOG(("CacheIndexStats::AfterChange()")); Log(); } #endif } private: uint32_t mCount; uint32_t mNotInitialized; uint32_t mRemoved; uint32_t mDirty; uint32_t mFresh; uint32_t mEmpty; uint32_t mSize; #ifdef DEBUG // We completely remove the data about an entry from the stats in // BeforeChange() and set this flag to true. The entry is then modified, // deleted or created and the data is again put into the stats and this flag // set to false. Statistics must not be read during this time since the // information is not correct. bool mStateLogged; // Disables logging in this instance of CacheIndexStats bool mDisableLogging; #endif }; class CacheIndex : public CacheFileIOListener , public nsIRunnable { public: NS_DECL_THREADSAFE_ISUPPORTS NS_DECL_NSIRUNNABLE CacheIndex(); static nsresult Init(nsIFile *aCacheDirectory); static nsresult PreShutdown(); static nsresult Shutdown(); // Following methods can be called only on IO thread. // Add entry to the index. The entry shouldn't be present in index. This // method is called whenever a new handle for a new entry file is created. The // newly created entry is not initialized and it must be either initialized // with InitEntry() or removed with RemoveEntry(). static nsresult AddEntry(const SHA1Sum::Hash *aHash); // Inform index about an existing entry that should be present in index. This // method is called whenever a new handle for an existing entry file is // created. Like in case of AddEntry(), either InitEntry() or RemoveEntry() // must be called on the entry, since the entry is not initizlized if the // index is outdated. static nsresult EnsureEntryExists(const SHA1Sum::Hash *aHash); // Initialize the entry. It MUST be present in index. Call to AddEntry() or // EnsureEntryExists() must precede the call to this method. static nsresult InitEntry(const SHA1Sum::Hash *aHash, OriginAttrsHash aOriginAttrsHash, bool aAnonymous, bool aPinned); // Remove entry from index. The entry should be present in index. static nsresult RemoveEntry(const SHA1Sum::Hash *aHash); // Update some information in entry. The entry MUST be present in index and // MUST be initialized. Call to AddEntry() or EnsureEntryExists() and to // InitEntry() must precede the call to this method. // Pass nullptr if the value didn't change. static nsresult UpdateEntry(const SHA1Sum::Hash *aHash, const uint32_t *aFrecency, const uint32_t *aExpirationTime, const uint32_t *aSize); // Remove all entries from the index. Called when clearing the whole cache. static nsresult RemoveAll(); enum EntryStatus { EXISTS = 0, DOES_NOT_EXIST = 1, DO_NOT_KNOW = 2 }; // Returns status of the entry in index for the given key. It can be called // on any thread. // If _pinned is non-null, it's filled with pinning status of the entry. static nsresult HasEntry(const nsACString &aKey, EntryStatus *_retval, bool *_pinned = nullptr); static nsresult HasEntry(const SHA1Sum::Hash &hash, EntryStatus *_retval, bool *_pinned = nullptr); // Returns a hash of the least important entry that should be evicted if the // cache size is over limit and also returns a total number of all entries in // the index minus the number of forced valid entries and unpinned entries // that we encounter when searching (see below) static nsresult GetEntryForEviction(bool aIgnoreEmptyEntries, SHA1Sum::Hash *aHash, uint32_t *aCnt); // Checks if a cache entry is currently forced valid. Used to prevent an entry // (that has been forced valid) from being evicted when the cache size reaches // its limit. static bool IsForcedValidEntry(const SHA1Sum::Hash *aHash); // Returns cache size in kB. static nsresult GetCacheSize(uint32_t *_retval); // Synchronously returns the disk occupation and number of entries per-context. // Callable on any thread. static nsresult GetCacheStats(nsILoadContextInfo *aInfo, uint32_t *aSize, uint32_t *aCount); // Asynchronously gets the disk cache size, used for display in the UI. static nsresult AsyncGetDiskConsumption(nsICacheStorageConsumptionObserver* aObserver); // Returns an iterator that returns entries matching a given context that were // present in the index at the time this method was called. If aAddNew is true // then the iterator will also return entries created after this call. // NOTE: When some entry is removed from index it is removed also from the // iterator regardless what aAddNew was passed. static nsresult GetIterator(nsILoadContextInfo *aInfo, bool aAddNew, CacheIndexIterator **_retval); // Returns true if we _think_ that the index is up to date. I.e. the state is // READY or WRITING and mIndexNeedsUpdate as well as mShuttingDown is false. static nsresult IsUpToDate(bool *_retval); // Called from CacheStorageService::Clear() and CacheFileContextEvictor::EvictEntries(), // sets a flag that blocks notification to AsyncGetDiskConsumption. static void OnAsyncEviction(bool aEvicting); // Memory reporting static size_t SizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf); static size_t SizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf); private: friend class CacheIndexEntryAutoManage; friend class FileOpenHelper; friend class CacheIndexIterator; virtual ~CacheIndex(); NS_IMETHOD OnFileOpened(CacheFileHandle *aHandle, nsresult aResult) override; nsresult OnFileOpenedInternal(FileOpenHelper *aOpener, CacheFileHandle *aHandle, nsresult aResult); NS_IMETHOD OnDataWritten(CacheFileHandle *aHandle, const char *aBuf, nsresult aResult) override; NS_IMETHOD OnDataRead(CacheFileHandle *aHandle, char *aBuf, nsresult aResult) override; NS_IMETHOD OnFileDoomed(CacheFileHandle *aHandle, nsresult aResult) override; NS_IMETHOD OnEOFSet(CacheFileHandle *aHandle, nsresult aResult) override; NS_IMETHOD OnFileRenamed(CacheFileHandle *aHandle, nsresult aResult) override; nsresult InitInternal(nsIFile *aCacheDirectory); void PreShutdownInternal(); // This method returns false when index is not initialized or is shut down. bool IsIndexUsable(); // This method checks whether the entry has the same values of // originAttributes and isAnonymous. We don't expect to find a collision // since these values are part of the key that we hash and we use a strong // hash function. static bool IsCollision(CacheIndexEntry *aEntry, OriginAttrsHash aOriginAttrsHash, bool aAnonymous); // Checks whether any of the information about the entry has changed. static bool HasEntryChanged(CacheIndexEntry *aEntry, const uint32_t *aFrecency, const uint32_t *aExpirationTime, const uint32_t *aSize); // Merge all pending operations from mPendingUpdates into mIndex. void ProcessPendingOperations(); // Following methods perform writing of the index file. // // The index is written periodically, but not earlier than once in // kMinDumpInterval and there must be at least kMinUnwrittenChanges // differences between index on disk and in memory. Index is always first // written to a temporary file and the old index file is replaced when the // writing process succeeds. // // Starts writing of index when both limits (minimal delay between writes and // minimum number of changes in index) were exceeded. bool WriteIndexToDiskIfNeeded(); // Starts writing of index file. void WriteIndexToDisk(); // Serializes part of mIndex hashtable to the write buffer a writes the buffer // to the file. void WriteRecords(); // Finalizes writing process. void FinishWrite(bool aSucceeded); // Following methods perform writing of the journal during shutdown. All these // methods must be called only during shutdown since they write/delete files // directly on the main thread instead of using CacheFileIOManager that does // it asynchronously on IO thread. Journal contains only entries that are // dirty, i.e. changes that are not present in the index file on the disk. // When the log is written successfully, the dirty flag in index file is // cleared. nsresult GetFile(const nsACString &aName, nsIFile **_retval); nsresult RemoveFile(const nsACString &aName); void RemoveAllIndexFiles(); void RemoveJournalAndTempFile(); // Writes journal to the disk and clears dirty flag in index header. nsresult WriteLogToDisk(); // Following methods perform reading of the index from the disk. // // Index is read at startup just after initializing the CacheIndex. There are // 3 files used when manipulating with index: index file, journal file and // a temporary file. All files contain the hash of the data, so we can check // whether the content is valid and complete. Index file contains also a dirty // flag in the index header which is unset on a clean shutdown. During opening // and reading of the files we determine the status of the whole index from // the states of the separate files. Following table shows all possible // combinations: // // index, journal, tmpfile // M * * - index is missing -> BUILD // I * * - index is invalid -> BUILD // D * * - index is dirty -> UPDATE // C M * - index is dirty -> UPDATE // C I * - unexpected state -> UPDATE // C V E - unexpected state -> UPDATE // C V M - index is up to date -> READY // // where the letters mean: // * - any state // E - file exists // M - file is missing // I - data is invalid (parsing failed or hash didn't match) // D - dirty (data in index file is correct, but dirty flag is set) // C - clean (index file is clean) // V - valid (data in journal file is correct) // // Note: We accept the data from journal only when the index is up to date as // a whole (i.e. C,V,M state). // // We rename the journal file to the temporary file as soon as possible after // initial test to ensure that we start update process on the next startup if // FF crashes during parsing of the index. // // Initiates reading index from disk. void ReadIndexFromDisk(); // Starts reading data from index file. void StartReadingIndex(); // Parses data read from index file. void ParseRecords(); // Starts reading data from journal file. void StartReadingJournal(); // Parses data read from journal file. void ParseJournal(); // Merges entries from journal into mIndex. void MergeJournal(); // In debug build this method checks that we have no fresh entry in mIndex // after we finish reading index and before we process pending operations. void EnsureNoFreshEntry(); // In debug build this method is called after processing pending operations // to make sure mIndexStats contains correct information. void EnsureCorrectStats(); // Finalizes reading process. void FinishRead(bool aSucceeded); // Following methods perform updating and building of the index. // Timer callback that starts update or build process. static void DelayedUpdate(nsITimer *aTimer, void *aClosure); void DelayedUpdateLocked(); // Posts timer event that start update or build process. nsresult ScheduleUpdateTimer(uint32_t aDelay); nsresult SetupDirectoryEnumerator(); void InitEntryFromDiskData(CacheIndexEntry *aEntry, CacheFileMetadata *aMetaData, int64_t aFileSize); // Returns true when either a timer is scheduled or event is posted. bool IsUpdatePending(); // Iterates through all files in entries directory that we didn't create/open // during this session, parses them and adds the entries to the index. void BuildIndex(); bool StartUpdatingIndexIfNeeded(bool aSwitchingToReadyState = false); // Starts update or build process or fires a timer when it is too early after // startup. void StartUpdatingIndex(bool aRebuild); // Iterates through all files in entries directory that we didn't create/open // during this session and theirs last modified time is newer than timestamp // in the index header. Parses the files and adds the entries to the index. void UpdateIndex(); // Finalizes update or build process. void FinishUpdate(bool aSucceeded); void RemoveNonFreshEntries(); enum EState { // Initial state in which the index is not usable // Possible transitions: // -> READING INITIAL = 0, // Index is being read from the disk. // Possible transitions: // -> INITIAL - We failed to dispatch a read event. // -> BUILDING - No or corrupted index file was found. // -> UPDATING - No or corrupted journal file was found. // - Dirty flag was set in index header. // -> READY - Index was read successfully or was interrupted by // pre-shutdown. // -> SHUTDOWN - This could happen only in case of pre-shutdown failure. READING = 1, // Index is being written to the disk. // Possible transitions: // -> READY - Writing of index finished or was interrupted by // pre-shutdown.. // -> UPDATING - Writing of index finished, but index was found outdated // during writing. // -> SHUTDOWN - This could happen only in case of pre-shutdown failure. WRITING = 2, // Index is being build. // Possible transitions: // -> READY - Building of index finished or was interrupted by // pre-shutdown. // -> SHUTDOWN - This could happen only in case of pre-shutdown failure. BUILDING = 3, // Index is being updated. // Possible transitions: // -> READY - Updating of index finished or was interrupted by // pre-shutdown. // -> SHUTDOWN - This could happen only in case of pre-shutdown failure. UPDATING = 4, // Index is ready. // Possible transitions: // -> UPDATING - Index was found outdated. // -> SHUTDOWN - Index is shutting down. READY = 5, // Index is shutting down. SHUTDOWN = 6 }; static char const * StateString(EState aState); void ChangeState(EState aNewState); void NotifyAsyncGetDiskConsumptionCallbacks(); // Allocates and releases buffer used for reading and writing index. void AllocBuffer(); void ReleaseBuffer(); // Methods used by CacheIndexEntryAutoManage to keep the iterators up to date. void AddRecordToIterators(CacheIndexRecord *aRecord); void RemoveRecordFromIterators(CacheIndexRecord *aRecord); void ReplaceRecordInIterators(CacheIndexRecord *aOldRecord, CacheIndexRecord *aNewRecord); // Memory reporting (private part) size_t SizeOfExcludingThisInternal(mozilla::MallocSizeOf mallocSizeOf) const; static mozilla::StaticRefPtr<CacheIndex> gInstance; static StaticMutex sLock; nsCOMPtr<nsIFile> mCacheDirectory; EState mState; // Timestamp of time when the index was initialized. We use it to delay // initial update or build of index. TimeStamp mStartTime; // Set to true in PreShutdown(), it is checked on variaous places to prevent // starting any process (write, update, etc.) during shutdown. bool mShuttingDown; // When set to true, update process should start as soon as possible. This // flag is set whenever we find some inconsistency which would be fixed by // update process. The flag is checked always when switching to READY state. // To make sure we start the update process as soon as possible, methods that // set this flag should also call StartUpdatingIndexIfNeeded() to cover the // case when we are currently in READY state. bool mIndexNeedsUpdate; // Set at the beginning of RemoveAll() which clears the whole index. When // removing all entries we must stop any pending reading, writing, updating or // building operation. This flag is checked at various places and it prevents // we won't start another operation (e.g. canceling reading of the index would // normally start update or build process) bool mRemovingAll; // Whether the index file on disk exists and is valid. bool mIndexOnDiskIsValid; // When something goes wrong during updating or building process, we don't // mark index clean (and also don't write journal) to ensure that update or // build will be initiated on the next start. bool mDontMarkIndexClean; // Timestamp value from index file. It is used during update process to skip // entries that were last modified before this timestamp. uint32_t mIndexTimeStamp; // Timestamp of last time the index was dumped to disk. // NOTE: The index might not be necessarily dumped at this time. The value // is used to schedule next dump of the index. TimeStamp mLastDumpTime; // Timer of delayed update/build. nsCOMPtr<nsITimer> mUpdateTimer; // True when build or update event is posted bool mUpdateEventPending; // Helper members used when reading/writing index from/to disk. // Contains number of entries that should be skipped: // - in hashtable when writing index because they were already written // - in index file when reading index because they were already read uint32_t mSkipEntries; // Number of entries that should be written to disk. This is number of entries // in hashtable that are initialized and are not marked as removed when writing // begins. uint32_t mProcessEntries; char *mRWBuf; uint32_t mRWBufSize; uint32_t mRWBufPos; RefPtr<CacheHash> mRWHash; // True if read or write operation is pending. It is used to ensure that // mRWBuf is not freed until OnDataRead or OnDataWritten is called. bool mRWPending; // Reading of journal succeeded if true. bool mJournalReadSuccessfully; // Handle used for writing and reading index file. RefPtr<CacheFileHandle> mIndexHandle; // Handle used for reading journal file. RefPtr<CacheFileHandle> mJournalHandle; // Used to check the existence of the file during reading process. RefPtr<CacheFileHandle> mTmpHandle; RefPtr<FileOpenHelper> mIndexFileOpener; RefPtr<FileOpenHelper> mJournalFileOpener; RefPtr<FileOpenHelper> mTmpFileOpener; // Directory enumerator used when building and updating index. nsCOMPtr<nsIDirectoryEnumerator> mDirEnumerator; // Main index hashtable. nsTHashtable<CacheIndexEntry> mIndex; // We cannot add, remove or change any entry in mIndex in states READING and // WRITING. We track all changes in mPendingUpdates during these states. nsTHashtable<CacheIndexEntryUpdate> mPendingUpdates; // Contains information statistics for mIndex + mPendingUpdates. CacheIndexStats mIndexStats; // When reading journal, we must first parse the whole file and apply the // changes iff the journal was read successfully. mTmpJournal is used to store // entries from the journal file. We throw away all these entries if parsing // of the journal fails or the hash does not match. nsTHashtable<CacheIndexEntry> mTmpJournal; // FrecencyArray maintains order of entry records for eviction. Ideally, the // records would be ordered by frecency all the time, but since this would be // quite expensive, we allow certain amount of entries to be out of order. // When the frecency is updated the new value is always bigger than the old // one. Instead of keeping updated entries at the same position, we move them // at the end of the array. This protects recently updated entries from // eviction. The array is sorted once we hit the limit of maximum unsorted // entries. class FrecencyArray { class Iterator { public: explicit Iterator(nsTArray<CacheIndexRecord *> *aRecs) : mRecs(aRecs) , mIdx(0) { while (!Done() && !(*mRecs)[mIdx]) { mIdx++; } } bool Done() const { return mIdx == mRecs->Length(); } CacheIndexRecord* Get() const { MOZ_ASSERT(!Done()); return (*mRecs)[mIdx]; } void Next() { MOZ_ASSERT(!Done()); ++mIdx; while (!Done() && !(*mRecs)[mIdx]) { mIdx++; } } private: nsTArray<CacheIndexRecord *> *mRecs; uint32_t mIdx; }; public: Iterator Iter() { return Iterator(&mRecs); } FrecencyArray() : mUnsortedElements(0) , mRemovedElements(0) {} // Methods used by CacheIndexEntryAutoManage to keep the array up to date. void AppendRecord(CacheIndexRecord *aRecord); void RemoveRecord(CacheIndexRecord *aRecord); void ReplaceRecord(CacheIndexRecord *aOldRecord, CacheIndexRecord *aNewRecord); void SortIfNeeded(); size_t Length() const { return mRecs.Length() - mRemovedElements; } void Clear() { mRecs.Clear(); } private: friend class CacheIndex; nsTArray<CacheIndexRecord *> mRecs; uint32_t mUnsortedElements; // Instead of removing elements from the array immediately, we null them out // and the iterator skips them when accessing the array. The null pointers // are placed at the end during sorting and we strip them out all at once. // This saves moving a lot of memory in nsTArray::RemoveElementsAt. uint32_t mRemovedElements; }; FrecencyArray mFrecencyArray; nsTArray<CacheIndexIterator *> mIterators; // This flag is true iff we are between CacheStorageService:Clear() and processing // all contexts to be evicted. It will make UI to show "calculating" instead of // any intermediate cache size. bool mAsyncGetDiskConsumptionBlocked; class DiskConsumptionObserver : public Runnable { public: static DiskConsumptionObserver* Init(nsICacheStorageConsumptionObserver* aObserver) { nsWeakPtr observer = do_GetWeakReference(aObserver); if (!observer) return nullptr; return new DiskConsumptionObserver(observer); } void OnDiskConsumption(int64_t aSize) { mSize = aSize; NS_DispatchToMainThread(this); } private: explicit DiskConsumptionObserver(nsWeakPtr const &aWeakObserver) : mObserver(aWeakObserver) { } virtual ~DiskConsumptionObserver() { if (mObserver && !NS_IsMainThread()) { NS_ReleaseOnMainThread(mObserver.forget()); } } NS_IMETHOD Run() override { MOZ_ASSERT(NS_IsMainThread()); nsCOMPtr<nsICacheStorageConsumptionObserver> observer = do_QueryReferent(mObserver); mObserver = nullptr; if (observer) { observer->OnNetworkCacheDiskConsumption(mSize); } return NS_OK; } nsWeakPtr mObserver; int64_t mSize; }; // List of async observers that want to get disk consumption information nsTArray<RefPtr<DiskConsumptionObserver> > mDiskConsumptionObservers; }; } // namespace net } // namespace mozilla #endif