gecko-dev/netwerk/cache2/CacheFile.cpp

2040 строки
53 KiB
C++

/* 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/. */
#include "CacheLog.h"
#include "CacheFile.h"
#include "CacheFileChunk.h"
#include "CacheFileInputStream.h"
#include "CacheFileOutputStream.h"
#include "nsThreadUtils.h"
#include "mozilla/DebugOnly.h"
#include <algorithm>
#include "nsComponentManagerUtils.h"
#include "nsProxyRelease.h"
#include "mozilla/Telemetry.h"
// When CACHE_CHUNKS is defined we always cache unused chunks in mCacheChunks.
// When it is not defined, we always release the chunks ASAP, i.e. we cache
// unused chunks only when:
// - CacheFile is memory-only
// - CacheFile is still waiting for the handle
// - the chunk is preloaded
//#define CACHE_CHUNKS
namespace mozilla {
namespace net {
class NotifyCacheFileListenerEvent : public nsRunnable {
public:
NotifyCacheFileListenerEvent(CacheFileListener *aCallback,
nsresult aResult,
bool aIsNew)
: mCallback(aCallback)
, mRV(aResult)
, mIsNew(aIsNew)
{
LOG(("NotifyCacheFileListenerEvent::NotifyCacheFileListenerEvent() "
"[this=%p]", this));
MOZ_COUNT_CTOR(NotifyCacheFileListenerEvent);
}
protected:
~NotifyCacheFileListenerEvent()
{
LOG(("NotifyCacheFileListenerEvent::~NotifyCacheFileListenerEvent() "
"[this=%p]", this));
MOZ_COUNT_DTOR(NotifyCacheFileListenerEvent);
}
public:
NS_IMETHOD Run()
{
LOG(("NotifyCacheFileListenerEvent::Run() [this=%p]", this));
mCallback->OnFileReady(mRV, mIsNew);
return NS_OK;
}
protected:
nsCOMPtr<CacheFileListener> mCallback;
nsresult mRV;
bool mIsNew;
};
class NotifyChunkListenerEvent : public nsRunnable {
public:
NotifyChunkListenerEvent(CacheFileChunkListener *aCallback,
nsresult aResult,
uint32_t aChunkIdx,
CacheFileChunk *aChunk)
: mCallback(aCallback)
, mRV(aResult)
, mChunkIdx(aChunkIdx)
, mChunk(aChunk)
{
LOG(("NotifyChunkListenerEvent::NotifyChunkListenerEvent() [this=%p]",
this));
MOZ_COUNT_CTOR(NotifyChunkListenerEvent);
}
protected:
~NotifyChunkListenerEvent()
{
LOG(("NotifyChunkListenerEvent::~NotifyChunkListenerEvent() [this=%p]",
this));
MOZ_COUNT_DTOR(NotifyChunkListenerEvent);
}
public:
NS_IMETHOD Run()
{
LOG(("NotifyChunkListenerEvent::Run() [this=%p]", this));
mCallback->OnChunkAvailable(mRV, mChunkIdx, mChunk);
return NS_OK;
}
protected:
nsCOMPtr<CacheFileChunkListener> mCallback;
nsresult mRV;
uint32_t mChunkIdx;
nsRefPtr<CacheFileChunk> mChunk;
};
class DoomFileHelper : public CacheFileIOListener
{
public:
NS_DECL_THREADSAFE_ISUPPORTS
explicit DoomFileHelper(CacheFileListener *aListener)
: mListener(aListener)
{
MOZ_COUNT_CTOR(DoomFileHelper);
}
NS_IMETHOD OnFileOpened(CacheFileHandle *aHandle, nsresult aResult) MOZ_OVERRIDE
{
MOZ_CRASH("DoomFileHelper::OnFileOpened should not be called!");
return NS_ERROR_UNEXPECTED;
}
NS_IMETHOD OnDataWritten(CacheFileHandle *aHandle, const char *aBuf,
nsresult aResult) MOZ_OVERRIDE
{
MOZ_CRASH("DoomFileHelper::OnDataWritten should not be called!");
return NS_ERROR_UNEXPECTED;
}
NS_IMETHOD OnDataRead(CacheFileHandle *aHandle, char *aBuf, nsresult aResult) MOZ_OVERRIDE
{
MOZ_CRASH("DoomFileHelper::OnDataRead should not be called!");
return NS_ERROR_UNEXPECTED;
}
NS_IMETHOD OnFileDoomed(CacheFileHandle *aHandle, nsresult aResult) MOZ_OVERRIDE
{
if (mListener)
mListener->OnFileDoomed(aResult);
return NS_OK;
}
NS_IMETHOD OnEOFSet(CacheFileHandle *aHandle, nsresult aResult) MOZ_OVERRIDE
{
MOZ_CRASH("DoomFileHelper::OnEOFSet should not be called!");
return NS_ERROR_UNEXPECTED;
}
NS_IMETHOD OnFileRenamed(CacheFileHandle *aHandle, nsresult aResult) MOZ_OVERRIDE
{
MOZ_CRASH("DoomFileHelper::OnFileRenamed should not be called!");
return NS_ERROR_UNEXPECTED;
}
private:
virtual ~DoomFileHelper()
{
MOZ_COUNT_DTOR(DoomFileHelper);
}
nsCOMPtr<CacheFileListener> mListener;
};
NS_IMPL_ISUPPORTS(DoomFileHelper, CacheFileIOListener)
NS_IMPL_ADDREF(CacheFile)
NS_IMPL_RELEASE(CacheFile)
NS_INTERFACE_MAP_BEGIN(CacheFile)
NS_INTERFACE_MAP_ENTRY(mozilla::net::CacheFileChunkListener)
NS_INTERFACE_MAP_ENTRY(mozilla::net::CacheFileIOListener)
NS_INTERFACE_MAP_ENTRY(mozilla::net::CacheFileMetadataListener)
NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports,
mozilla::net::CacheFileChunkListener)
NS_INTERFACE_MAP_END_THREADSAFE
CacheFile::CacheFile()
: mLock("CacheFile.mLock")
, mOpeningFile(false)
, mReady(false)
, mMemoryOnly(false)
, mOpenAsMemoryOnly(false)
, mPriority(false)
, mDataAccessed(false)
, mDataIsDirty(false)
, mWritingMetadata(false)
, mPreloadWithoutInputStreams(true)
, mPreloadChunkCount(0)
, mStatus(NS_OK)
, mDataSize(-1)
, mOutput(nullptr)
{
LOG(("CacheFile::CacheFile() [this=%p]", this));
}
CacheFile::~CacheFile()
{
LOG(("CacheFile::~CacheFile() [this=%p]", this));
MutexAutoLock lock(mLock);
if (!mMemoryOnly && mReady) {
// mReady flag indicates we have metadata plus in a valid state.
WriteMetadataIfNeededLocked(true);
}
}
nsresult
CacheFile::Init(const nsACString &aKey,
bool aCreateNew,
bool aMemoryOnly,
bool aPriority,
CacheFileListener *aCallback)
{
MOZ_ASSERT(!mListener);
MOZ_ASSERT(!mHandle);
nsresult rv;
mKey = aKey;
mOpenAsMemoryOnly = mMemoryOnly = aMemoryOnly;
mPriority = aPriority;
// Some consumers (at least nsHTTPCompressConv) assume that Read() can read
// such amount of data that was announced by Available().
// CacheFileInputStream::Available() uses also preloaded chunks to compute
// number of available bytes in the input stream, so we have to make sure the
// preloadChunkCount won't change during CacheFile's lifetime since otherwise
// we could potentially release some cached chunks that was used to calculate
// available bytes but would not be available later during call to
// CacheFileInputStream::Read().
mPreloadChunkCount = CacheObserver::PreloadChunkCount();
LOG(("CacheFile::Init() [this=%p, key=%s, createNew=%d, memoryOnly=%d, "
"priority=%d, listener=%p]", this, mKey.get(), aCreateNew, aMemoryOnly,
aPriority, aCallback));
if (mMemoryOnly) {
MOZ_ASSERT(!aCallback);
mMetadata = new CacheFileMetadata(mOpenAsMemoryOnly, mKey);
mReady = true;
mDataSize = mMetadata->Offset();
return NS_OK;
}
else {
uint32_t flags;
if (aCreateNew) {
MOZ_ASSERT(!aCallback);
flags = CacheFileIOManager::CREATE_NEW;
// make sure we can use this entry immediately
mMetadata = new CacheFileMetadata(mOpenAsMemoryOnly, mKey);
mReady = true;
mDataSize = mMetadata->Offset();
} else {
flags = CacheFileIOManager::CREATE;
}
if (mPriority) {
flags |= CacheFileIOManager::PRIORITY;
}
mOpeningFile = true;
mListener = aCallback;
rv = CacheFileIOManager::OpenFile(mKey, flags, this);
if (NS_FAILED(rv)) {
mListener = nullptr;
mOpeningFile = false;
if (aCreateNew) {
NS_WARNING("Forcing memory-only entry since OpenFile failed");
LOG(("CacheFile::Init() - CacheFileIOManager::OpenFile() failed "
"synchronously. We can continue in memory-only mode since "
"aCreateNew == true. [this=%p]", this));
mMemoryOnly = true;
}
else if (rv == NS_ERROR_NOT_INITIALIZED) {
NS_WARNING("Forcing memory-only entry since CacheIOManager isn't "
"initialized.");
LOG(("CacheFile::Init() - CacheFileIOManager isn't initialized, "
"initializing entry as memory-only. [this=%p]", this));
mMemoryOnly = true;
mMetadata = new CacheFileMetadata(mOpenAsMemoryOnly, mKey);
mReady = true;
mDataSize = mMetadata->Offset();
nsRefPtr<NotifyCacheFileListenerEvent> ev;
ev = new NotifyCacheFileListenerEvent(aCallback, NS_OK, true);
rv = NS_DispatchToCurrentThread(ev);
NS_ENSURE_SUCCESS(rv, rv);
}
else {
NS_ENSURE_SUCCESS(rv, rv);
}
}
}
return NS_OK;
}
nsresult
CacheFile::OnChunkRead(nsresult aResult, CacheFileChunk *aChunk)
{
CacheFileAutoLock lock(this);
nsresult rv;
uint32_t index = aChunk->Index();
LOG(("CacheFile::OnChunkRead() [this=%p, rv=0x%08x, chunk=%p, idx=%u]",
this, aResult, aChunk, index));
if (NS_FAILED(aResult)) {
SetError(aResult);
}
if (HaveChunkListeners(index)) {
rv = NotifyChunkListeners(index, aResult, aChunk);
NS_ENSURE_SUCCESS(rv, rv);
}
return NS_OK;
}
nsresult
CacheFile::OnChunkWritten(nsresult aResult, CacheFileChunk *aChunk)
{
// In case the chunk was reused, made dirty and released between calls to
// CacheFileChunk::Write() and CacheFile::OnChunkWritten(), we must write
// the chunk to the disk again. When the chunk is unused and is dirty simply
// addref and release (outside the lock) the chunk which ensures that
// CacheFile::DeactivateChunk() will be called again.
nsRefPtr<CacheFileChunk> deactivateChunkAgain;
CacheFileAutoLock lock(this);
nsresult rv;
LOG(("CacheFile::OnChunkWritten() [this=%p, rv=0x%08x, chunk=%p, idx=%u]",
this, aResult, aChunk, aChunk->Index()));
MOZ_ASSERT(!mMemoryOnly);
MOZ_ASSERT(!mOpeningFile);
MOZ_ASSERT(mHandle);
if (NS_FAILED(aResult)) {
SetError(aResult);
}
if (NS_SUCCEEDED(aResult) && !aChunk->IsDirty()) {
// update hash value in metadata
mMetadata->SetHash(aChunk->Index(), aChunk->Hash());
}
// notify listeners if there is any
if (HaveChunkListeners(aChunk->Index())) {
// don't release the chunk since there are some listeners queued
rv = NotifyChunkListeners(aChunk->Index(), aResult, aChunk);
if (NS_SUCCEEDED(rv)) {
MOZ_ASSERT(aChunk->mRefCnt != 2);
return NS_OK;
}
}
if (aChunk->mRefCnt != 2) {
LOG(("CacheFile::OnChunkWritten() - Chunk is still used [this=%p, chunk=%p,"
" refcnt=%d]", this, aChunk, aChunk->mRefCnt.get()));
return NS_OK;
}
if (aChunk->IsDirty()) {
LOG(("CacheFile::OnChunkWritten() - Unused chunk is dirty. We must go "
"through deactivation again. [this=%p, chunk=%p]", this, aChunk));
deactivateChunkAgain = aChunk;
return NS_OK;
}
bool keepChunk = false;
if (NS_SUCCEEDED(aResult)) {
keepChunk = ShouldCacheChunk(aChunk->Index());
LOG(("CacheFile::OnChunkWritten() - %s unused chunk [this=%p, chunk=%p]",
keepChunk ? "Caching" : "Releasing", this, aChunk));
} else {
LOG(("CacheFile::OnChunkWritten() - Releasing failed chunk [this=%p, "
"chunk=%p]", this, aChunk));
}
RemoveChunkInternal(aChunk, keepChunk);
WriteMetadataIfNeededLocked();
return NS_OK;
}
nsresult
CacheFile::OnChunkAvailable(nsresult aResult, uint32_t aChunkIdx,
CacheFileChunk *aChunk)
{
MOZ_CRASH("CacheFile::OnChunkAvailable should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OnChunkUpdated(CacheFileChunk *aChunk)
{
MOZ_CRASH("CacheFile::OnChunkUpdated should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OnFileOpened(CacheFileHandle *aHandle, nsresult aResult)
{
nsresult rv;
// Using an 'auto' class to perform doom or fail the listener
// outside the CacheFile's lock.
class AutoFailDoomListener
{
public:
explicit AutoFailDoomListener(CacheFileHandle *aHandle)
: mHandle(aHandle)
, mAlreadyDoomed(false)
{}
~AutoFailDoomListener()
{
if (!mListener)
return;
if (mHandle) {
if (mAlreadyDoomed) {
mListener->OnFileDoomed(mHandle, NS_OK);
} else {
CacheFileIOManager::DoomFile(mHandle, mListener);
}
} else {
mListener->OnFileDoomed(nullptr, NS_ERROR_NOT_AVAILABLE);
}
}
CacheFileHandle* mHandle;
nsCOMPtr<CacheFileIOListener> mListener;
bool mAlreadyDoomed;
} autoDoom(aHandle);
nsCOMPtr<CacheFileListener> listener;
bool isNew = false;
nsresult retval = NS_OK;
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mOpeningFile);
MOZ_ASSERT((NS_SUCCEEDED(aResult) && aHandle) ||
(NS_FAILED(aResult) && !aHandle));
MOZ_ASSERT((mListener && !mMetadata) || // !createNew
(!mListener && mMetadata)); // createNew
MOZ_ASSERT(!mMemoryOnly || mMetadata); // memory-only was set on new entry
LOG(("CacheFile::OnFileOpened() [this=%p, rv=0x%08x, handle=%p]",
this, aResult, aHandle));
mOpeningFile = false;
autoDoom.mListener.swap(mDoomAfterOpenListener);
if (mMemoryOnly) {
// We can be here only in case the entry was initilized as createNew and
// SetMemoryOnly() was called.
// Just don't store the handle into mHandle and exit
autoDoom.mAlreadyDoomed = true;
return NS_OK;
}
else if (NS_FAILED(aResult)) {
if (mMetadata) {
// This entry was initialized as createNew, just switch to memory-only
// mode.
NS_WARNING("Forcing memory-only entry since OpenFile failed");
LOG(("CacheFile::OnFileOpened() - CacheFileIOManager::OpenFile() "
"failed asynchronously. We can continue in memory-only mode since "
"aCreateNew == true. [this=%p]", this));
mMemoryOnly = true;
return NS_OK;
}
else if (aResult == NS_ERROR_FILE_INVALID_PATH) {
// CacheFileIOManager doesn't have mCacheDirectory, switch to
// memory-only mode.
NS_WARNING("Forcing memory-only entry since CacheFileIOManager doesn't "
"have mCacheDirectory.");
LOG(("CacheFile::OnFileOpened() - CacheFileIOManager doesn't have "
"mCacheDirectory, initializing entry as memory-only. [this=%p]",
this));
mMemoryOnly = true;
mMetadata = new CacheFileMetadata(mOpenAsMemoryOnly, mKey);
mReady = true;
mDataSize = mMetadata->Offset();
isNew = true;
retval = NS_OK;
}
else {
// CacheFileIOManager::OpenFile() failed for another reason.
isNew = false;
retval = aResult;
}
mListener.swap(listener);
}
else {
mHandle = aHandle;
if (NS_FAILED(mStatus)) {
CacheFileIOManager::DoomFile(mHandle, nullptr);
}
if (mMetadata) {
InitIndexEntry();
// The entry was initialized as createNew, don't try to read metadata.
mMetadata->SetHandle(mHandle);
// Write all cached chunks, otherwise they may stay unwritten.
mCachedChunks.Enumerate(&CacheFile::WriteAllCachedChunks, this);
return NS_OK;
}
}
}
if (listener) {
listener->OnFileReady(retval, isNew);
return NS_OK;
}
MOZ_ASSERT(NS_SUCCEEDED(aResult));
MOZ_ASSERT(!mMetadata);
MOZ_ASSERT(mListener);
mMetadata = new CacheFileMetadata(mHandle, mKey);
rv = mMetadata->ReadMetadata(this);
if (NS_FAILED(rv)) {
mListener.swap(listener);
listener->OnFileReady(rv, false);
}
return NS_OK;
}
nsresult
CacheFile::OnDataWritten(CacheFileHandle *aHandle, const char *aBuf,
nsresult aResult)
{
MOZ_CRASH("CacheFile::OnDataWritten should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OnDataRead(CacheFileHandle *aHandle, char *aBuf, nsresult aResult)
{
MOZ_CRASH("CacheFile::OnDataRead should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OnMetadataRead(nsresult aResult)
{
MOZ_ASSERT(mListener);
LOG(("CacheFile::OnMetadataRead() [this=%p, rv=0x%08x]", this, aResult));
bool isNew = false;
if (NS_SUCCEEDED(aResult)) {
mReady = true;
mDataSize = mMetadata->Offset();
if (mDataSize == 0 && mMetadata->ElementsSize() == 0) {
isNew = true;
mMetadata->MarkDirty();
} else {
CacheFileAutoLock lock(this);
PreloadChunks(0);
}
InitIndexEntry();
}
nsCOMPtr<CacheFileListener> listener;
mListener.swap(listener);
listener->OnFileReady(aResult, isNew);
return NS_OK;
}
nsresult
CacheFile::OnMetadataWritten(nsresult aResult)
{
CacheFileAutoLock lock(this);
LOG(("CacheFile::OnMetadataWritten() [this=%p, rv=0x%08x]", this, aResult));
MOZ_ASSERT(mWritingMetadata);
mWritingMetadata = false;
MOZ_ASSERT(!mMemoryOnly);
MOZ_ASSERT(!mOpeningFile);
if (NS_FAILED(aResult)) {
// TODO close streams with an error ???
}
if (mOutput || mInputs.Length() || mChunks.Count())
return NS_OK;
if (IsDirty())
WriteMetadataIfNeededLocked();
if (!mWritingMetadata) {
LOG(("CacheFile::OnMetadataWritten() - Releasing file handle [this=%p]",
this));
CacheFileIOManager::ReleaseNSPRHandle(mHandle);
}
return NS_OK;
}
nsresult
CacheFile::OnFileDoomed(CacheFileHandle *aHandle, nsresult aResult)
{
nsCOMPtr<CacheFileListener> listener;
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mListener);
LOG(("CacheFile::OnFileDoomed() [this=%p, rv=0x%08x, handle=%p]",
this, aResult, aHandle));
mListener.swap(listener);
}
listener->OnFileDoomed(aResult);
return NS_OK;
}
nsresult
CacheFile::OnEOFSet(CacheFileHandle *aHandle, nsresult aResult)
{
MOZ_CRASH("CacheFile::OnEOFSet should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OnFileRenamed(CacheFileHandle *aHandle, nsresult aResult)
{
MOZ_CRASH("CacheFile::OnFileRenamed should not be called!");
return NS_ERROR_UNEXPECTED;
}
nsresult
CacheFile::OpenInputStream(nsIInputStream **_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mHandle || mMemoryOnly || mOpeningFile);
if (!mReady) {
LOG(("CacheFile::OpenInputStream() - CacheFile is not ready [this=%p]",
this));
return NS_ERROR_NOT_AVAILABLE;
}
if (NS_FAILED(mStatus)) {
LOG(("CacheFile::OpenInputStream() - CacheFile is in a failure state "
"[this=%p, status=0x%08x]", this, mStatus));
// Don't allow opening the input stream when this CacheFile is in
// a failed state. This is the only way to protect consumers correctly
// from reading a broken entry. When the file is in the failed state,
// it's also doomed, so reopening the entry won't make any difference -
// data will still be inaccessible anymore. Note that for just doomed
// files, we must allow reading the data.
return mStatus;
}
// Once we open input stream we no longer allow preloading of chunks without
// input stream, i.e. we will no longer keep first few chunks preloaded when
// the last input stream is closed.
mPreloadWithoutInputStreams = false;
CacheFileInputStream *input = new CacheFileInputStream(this);
LOG(("CacheFile::OpenInputStream() - Creating new input stream %p [this=%p]",
input, this));
mInputs.AppendElement(input);
NS_ADDREF(input);
mDataAccessed = true;
NS_ADDREF(*_retval = input);
return NS_OK;
}
nsresult
CacheFile::OpenOutputStream(CacheOutputCloseListener *aCloseListener, nsIOutputStream **_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mHandle || mMemoryOnly || mOpeningFile);
if (!mReady) {
LOG(("CacheFile::OpenOutputStream() - CacheFile is not ready [this=%p]",
this));
return NS_ERROR_NOT_AVAILABLE;
}
if (mOutput) {
LOG(("CacheFile::OpenOutputStream() - We already have output stream %p "
"[this=%p]", mOutput, this));
return NS_ERROR_NOT_AVAILABLE;
}
// Once we open output stream we no longer allow preloading of chunks without
// input stream. There is no reason to believe that some input stream will be
// opened soon. Otherwise we would cache unused chunks of all newly created
// entries until the CacheFile is destroyed.
mPreloadWithoutInputStreams = false;
mOutput = new CacheFileOutputStream(this, aCloseListener);
LOG(("CacheFile::OpenOutputStream() - Creating new output stream %p "
"[this=%p]", mOutput, this));
mDataAccessed = true;
NS_ADDREF(*_retval = mOutput);
return NS_OK;
}
nsresult
CacheFile::SetMemoryOnly()
{
LOG(("CacheFile::SetMemoryOnly() mMemoryOnly=%d [this=%p]",
mMemoryOnly, this));
if (mMemoryOnly)
return NS_OK;
MOZ_ASSERT(mReady);
if (!mReady) {
LOG(("CacheFile::SetMemoryOnly() - CacheFile is not ready [this=%p]",
this));
return NS_ERROR_NOT_AVAILABLE;
}
if (mDataAccessed) {
LOG(("CacheFile::SetMemoryOnly() - Data was already accessed [this=%p]", this));
return NS_ERROR_NOT_AVAILABLE;
}
// TODO what to do when this isn't a new entry and has an existing metadata???
mMemoryOnly = true;
return NS_OK;
}
nsresult
CacheFile::Doom(CacheFileListener *aCallback)
{
LOG(("CacheFile::Doom() [this=%p, listener=%p]", this, aCallback));
CacheFileAutoLock lock(this);
return DoomLocked(aCallback);
}
nsresult
CacheFile::DoomLocked(CacheFileListener *aCallback)
{
MOZ_ASSERT(mHandle || mMemoryOnly || mOpeningFile);
LOG(("CacheFile::DoomLocked() [this=%p, listener=%p]", this, aCallback));
nsresult rv = NS_OK;
if (mMemoryOnly) {
return NS_ERROR_FILE_NOT_FOUND;
}
if (mHandle && mHandle->IsDoomed()) {
return NS_ERROR_FILE_NOT_FOUND;
}
nsCOMPtr<CacheFileIOListener> listener;
if (aCallback || !mHandle) {
listener = new DoomFileHelper(aCallback);
}
if (mHandle) {
rv = CacheFileIOManager::DoomFile(mHandle, listener);
} else if (mOpeningFile) {
mDoomAfterOpenListener = listener;
}
return rv;
}
nsresult
CacheFile::ThrowMemoryCachedData()
{
CacheFileAutoLock lock(this);
LOG(("CacheFile::ThrowMemoryCachedData() [this=%p]", this));
if (mMemoryOnly) {
// This method should not be called when the CacheFile was initialized as
// memory-only, but it can be called when CacheFile end up as memory-only
// due to e.g. IO failure since CacheEntry doesn't know it.
LOG(("CacheFile::ThrowMemoryCachedData() - Ignoring request because the "
"entry is memory-only. [this=%p]", this));
return NS_ERROR_NOT_AVAILABLE;
}
if (mOpeningFile) {
// mayhemer, note: we shouldn't get here, since CacheEntry prevents loading
// entries from being purged.
LOG(("CacheFile::ThrowMemoryCachedData() - Ignoring request because the "
"entry is still opening the file [this=%p]", this));
return NS_ERROR_ABORT;
}
// We cannot release all cached chunks since we need to keep preloaded chunks
// in memory. See initialization of mPreloadChunkCount for explanation.
mCachedChunks.Enumerate(&CacheFile::CleanUpCachedChunks, this);
return NS_OK;
}
nsresult
CacheFile::GetElement(const char *aKey, char **_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
const char *value;
value = mMetadata->GetElement(aKey);
if (!value)
return NS_ERROR_NOT_AVAILABLE;
*_retval = NS_strdup(value);
return NS_OK;
}
nsresult
CacheFile::SetElement(const char *aKey, const char *aValue)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
PostWriteTimer();
return mMetadata->SetElement(aKey, aValue);
}
nsresult
CacheFile::VisitMetaData(nsICacheEntryMetaDataVisitor *aVisitor)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
MOZ_ASSERT(mReady);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->Visit(aVisitor);
}
nsresult
CacheFile::ElementsSize(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
if (!mMetadata)
return NS_ERROR_NOT_AVAILABLE;
*_retval = mMetadata->ElementsSize();
return NS_OK;
}
nsresult
CacheFile::SetExpirationTime(uint32_t aExpirationTime)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
PostWriteTimer();
if (mHandle && !mHandle->IsDoomed())
CacheFileIOManager::UpdateIndexEntry(mHandle, nullptr, &aExpirationTime);
return mMetadata->SetExpirationTime(aExpirationTime);
}
nsresult
CacheFile::GetExpirationTime(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->GetExpirationTime(_retval);
}
nsresult
CacheFile::SetFrecency(uint32_t aFrecency)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
PostWriteTimer();
if (mHandle && !mHandle->IsDoomed())
CacheFileIOManager::UpdateIndexEntry(mHandle, &aFrecency, nullptr);
return mMetadata->SetFrecency(aFrecency);
}
nsresult
CacheFile::GetFrecency(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->GetFrecency(_retval);
}
nsresult
CacheFile::GetLastModified(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->GetLastModified(_retval);
}
nsresult
CacheFile::GetLastFetched(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->GetLastFetched(_retval);
}
nsresult
CacheFile::GetFetchCount(uint32_t *_retval)
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
return mMetadata->GetFetchCount(_retval);
}
nsresult
CacheFile::OnFetched()
{
CacheFileAutoLock lock(this);
MOZ_ASSERT(mMetadata);
NS_ENSURE_TRUE(mMetadata, NS_ERROR_UNEXPECTED);
PostWriteTimer();
return mMetadata->OnFetched();
}
void
CacheFile::Lock()
{
mLock.Lock();
}
void
CacheFile::Unlock()
{
nsTArray<nsISupports*> objs;
objs.SwapElements(mObjsToRelease);
mLock.Unlock();
for (uint32_t i = 0; i < objs.Length(); i++)
objs[i]->Release();
}
void
CacheFile::AssertOwnsLock() const
{
mLock.AssertCurrentThreadOwns();
}
void
CacheFile::ReleaseOutsideLock(nsISupports *aObject)
{
AssertOwnsLock();
mObjsToRelease.AppendElement(aObject);
}
nsresult
CacheFile::GetChunk(uint32_t aIndex, ECallerType aCaller,
CacheFileChunkListener *aCallback, CacheFileChunk **_retval)
{
CacheFileAutoLock lock(this);
return GetChunkLocked(aIndex, aCaller, aCallback, _retval);
}
nsresult
CacheFile::GetChunkLocked(uint32_t aIndex, ECallerType aCaller,
CacheFileChunkListener *aCallback,
CacheFileChunk **_retval)
{
AssertOwnsLock();
LOG(("CacheFile::GetChunkLocked() [this=%p, idx=%u, caller=%d, listener=%p]",
this, aIndex, aCaller, aCallback));
MOZ_ASSERT(mReady);
MOZ_ASSERT(mHandle || mMemoryOnly || mOpeningFile);
MOZ_ASSERT((aCaller == READER && aCallback) ||
(aCaller == WRITER && !aCallback) ||
(aCaller == PRELOADER && !aCallback));
// Preload chunks from disk when this is disk backed entry and the listener
// is reader.
bool preload = !mMemoryOnly && (aCaller == READER);
nsresult rv;
nsRefPtr<CacheFileChunk> chunk;
if (mChunks.Get(aIndex, getter_AddRefs(chunk))) {
LOG(("CacheFile::GetChunkLocked() - Found chunk %p in mChunks [this=%p]",
chunk.get(), this));
// Preloader calls this method to preload only non-loaded chunks.
MOZ_ASSERT(aCaller != PRELOADER, "Unexpected!");
// We might get failed chunk between releasing the lock in
// CacheFileChunk::OnDataWritten/Read and CacheFile::OnChunkWritten/Read
rv = chunk->GetStatus();
if (NS_FAILED(rv)) {
SetError(rv);
LOG(("CacheFile::GetChunkLocked() - Found failed chunk in mChunks "
"[this=%p]", this));
return rv;
}
if (chunk->IsReady() || aCaller == WRITER) {
chunk.swap(*_retval);
} else {
rv = QueueChunkListener(aIndex, aCallback);
NS_ENSURE_SUCCESS(rv, rv);
}
if (preload) {
PreloadChunks(aIndex + 1);
}
return NS_OK;
}
if (mCachedChunks.Get(aIndex, getter_AddRefs(chunk))) {
LOG(("CacheFile::GetChunkLocked() - Reusing cached chunk %p [this=%p]",
chunk.get(), this));
// Preloader calls this method to preload only non-loaded chunks.
MOZ_ASSERT(aCaller != PRELOADER, "Unexpected!");
mChunks.Put(aIndex, chunk);
mCachedChunks.Remove(aIndex);
chunk->mFile = this;
chunk->mActiveChunk = true;
MOZ_ASSERT(chunk->IsReady());
chunk.swap(*_retval);
if (preload) {
PreloadChunks(aIndex + 1);
}
return NS_OK;
}
int64_t off = aIndex * kChunkSize;
if (off < mDataSize) {
// We cannot be here if this is memory only entry since the chunk must exist
MOZ_ASSERT(!mMemoryOnly);
if (mMemoryOnly) {
// If this ever really happen it is better to fail rather than crashing on
// a null handle.
LOG(("CacheFile::GetChunkLocked() - Unexpected state! Offset < mDataSize "
"for memory-only entry. [this=%p, off=%lld, mDataSize=%lld]",
this, off, mDataSize));
return NS_ERROR_UNEXPECTED;
}
chunk = new CacheFileChunk(this, aIndex, aCaller == WRITER);
mChunks.Put(aIndex, chunk);
chunk->mActiveChunk = true;
LOG(("CacheFile::GetChunkLocked() - Reading newly created chunk %p from "
"the disk [this=%p]", chunk.get(), this));
// Read the chunk from the disk
rv = chunk->Read(mHandle, std::min(static_cast<uint32_t>(mDataSize - off),
static_cast<uint32_t>(kChunkSize)),
mMetadata->GetHash(aIndex), this);
if (NS_WARN_IF(NS_FAILED(rv))) {
RemoveChunkInternal(chunk, false);
return rv;
}
if (aCaller == WRITER) {
chunk.swap(*_retval);
} else if (aCaller != PRELOADER) {
rv = QueueChunkListener(aIndex, aCallback);
NS_ENSURE_SUCCESS(rv, rv);
}
if (preload) {
PreloadChunks(aIndex + 1);
}
return NS_OK;
} else if (off == mDataSize) {
if (aCaller == WRITER) {
// this listener is going to write to the chunk
chunk = new CacheFileChunk(this, aIndex, true);
mChunks.Put(aIndex, chunk);
chunk->mActiveChunk = true;
LOG(("CacheFile::GetChunkLocked() - Created new empty chunk %p [this=%p]",
chunk.get(), this));
chunk->InitNew();
mMetadata->SetHash(aIndex, chunk->Hash());
if (HaveChunkListeners(aIndex)) {
rv = NotifyChunkListeners(aIndex, NS_OK, chunk);
NS_ENSURE_SUCCESS(rv, rv);
}
chunk.swap(*_retval);
return NS_OK;
}
} else {
if (aCaller == WRITER) {
// this chunk was requested by writer, but we need to fill the gap first
// Fill with zero the last chunk if it is incomplete
if (mDataSize % kChunkSize) {
rv = PadChunkWithZeroes(mDataSize / kChunkSize);
NS_ENSURE_SUCCESS(rv, rv);
MOZ_ASSERT(!(mDataSize % kChunkSize));
}
uint32_t startChunk = mDataSize / kChunkSize;
if (mMemoryOnly) {
// We need to create all missing CacheFileChunks if this is memory-only
// entry
for (uint32_t i = startChunk ; i < aIndex ; i++) {
rv = PadChunkWithZeroes(i);
NS_ENSURE_SUCCESS(rv, rv);
}
} else {
// We don't need to create CacheFileChunk for other empty chunks unless
// there is some input stream waiting for this chunk.
if (startChunk != aIndex) {
// Make sure the file contains zeroes at the end of the file
rv = CacheFileIOManager::TruncateSeekSetEOF(mHandle,
startChunk * kChunkSize,
aIndex * kChunkSize,
nullptr);
NS_ENSURE_SUCCESS(rv, rv);
}
for (uint32_t i = startChunk ; i < aIndex ; i++) {
if (HaveChunkListeners(i)) {
rv = PadChunkWithZeroes(i);
NS_ENSURE_SUCCESS(rv, rv);
} else {
mMetadata->SetHash(i, kEmptyChunkHash);
mDataSize = (i + 1) * kChunkSize;
}
}
}
MOZ_ASSERT(mDataSize == off);
rv = GetChunkLocked(aIndex, WRITER, nullptr, getter_AddRefs(chunk));
NS_ENSURE_SUCCESS(rv, rv);
chunk.swap(*_retval);
return NS_OK;
}
}
// We can be here only if the caller is reader since writer always create a
// new chunk above and preloader calls this method to preload only chunks that
// are not loaded but that do exist.
MOZ_ASSERT(aCaller == READER, "Unexpected!");
if (mOutput) {
// the chunk doesn't exist but mOutput may create it
rv = QueueChunkListener(aIndex, aCallback);
NS_ENSURE_SUCCESS(rv, rv);
} else {
return NS_ERROR_NOT_AVAILABLE;
}
return NS_OK;
}
void
CacheFile::PreloadChunks(uint32_t aIndex)
{
AssertOwnsLock();
uint32_t limit = aIndex + mPreloadChunkCount;
for (uint32_t i = aIndex; i < limit; ++i) {
int64_t off = i * kChunkSize;
if (off >= mDataSize) {
// This chunk is beyond EOF.
return;
}
if (mChunks.GetWeak(i) || mCachedChunks.GetWeak(i)) {
// This chunk is already in memory or is being read right now.
continue;
}
LOG(("CacheFile::PreloadChunks() - Preloading chunk [this=%p, idx=%u]",
this, i));
nsRefPtr<CacheFileChunk> chunk;
GetChunkLocked(i, PRELOADER, nullptr, getter_AddRefs(chunk));
// We've checked that we don't have this chunk, so no chunk must be
// returned.
MOZ_ASSERT(!chunk);
}
}
bool
CacheFile::ShouldCacheChunk(uint32_t aIndex)
{
AssertOwnsLock();
#ifdef CACHE_CHUNKS
// We cache all chunks.
return true;
#else
if (mPreloadChunkCount != 0 && mInputs.Length() == 0 &&
mPreloadWithoutInputStreams && aIndex < mPreloadChunkCount) {
// We don't have any input stream yet, but it is likely that some will be
// opened soon. Keep first mPreloadChunkCount chunks in memory. The
// condition is here instead of in MustKeepCachedChunk() since these
// chunks should be preloaded and can be kept in memory as an optimization,
// but they can be released at any time until they are considered as
// preloaded chunks for any input stream.
return true;
}
// Cache only chunks that we really need to keep.
return MustKeepCachedChunk(aIndex);
#endif
}
bool
CacheFile::MustKeepCachedChunk(uint32_t aIndex)
{
AssertOwnsLock();
// We must keep the chunk when this is memory only entry or we don't have
// a handle yet.
if (mMemoryOnly || mOpeningFile) {
return true;
}
if (mPreloadChunkCount == 0) {
// Preloading of chunks is disabled
return false;
}
// Check whether this chunk should be considered as preloaded chunk for any
// existing input stream.
// maxPos is the position of the last byte in the given chunk
int64_t maxPos = static_cast<int64_t>(aIndex + 1) * kChunkSize - 1;
// minPos is the position of the first byte in a chunk that precedes the given
// chunk by mPreloadChunkCount chunks
int64_t minPos;
if (mPreloadChunkCount >= aIndex) {
minPos = 0;
} else {
minPos = static_cast<int64_t>(aIndex - mPreloadChunkCount) * kChunkSize;
}
for (uint32_t i = 0; i < mInputs.Length(); ++i) {
int64_t inputPos = mInputs[i]->GetPosition();
if (inputPos >= minPos && inputPos <= maxPos) {
return true;
}
}
return false;
}
nsresult
CacheFile::DeactivateChunk(CacheFileChunk *aChunk)
{
nsresult rv;
// Avoid lock reentrancy by increasing the RefCnt
nsRefPtr<CacheFileChunk> chunk = aChunk;
{
CacheFileAutoLock lock(this);
LOG(("CacheFile::DeactivateChunk() [this=%p, chunk=%p, idx=%u]",
this, aChunk, aChunk->Index()));
MOZ_ASSERT(mReady);
MOZ_ASSERT((mHandle && !mMemoryOnly && !mOpeningFile) ||
(!mHandle && mMemoryOnly && !mOpeningFile) ||
(!mHandle && !mMemoryOnly && mOpeningFile));
if (aChunk->mRefCnt != 2) {
LOG(("CacheFile::DeactivateChunk() - Chunk is still used [this=%p, "
"chunk=%p, refcnt=%d]", this, aChunk, aChunk->mRefCnt.get()));
// somebody got the reference before the lock was acquired
return NS_OK;
}
#ifdef DEBUG
{
// We can be here iff the chunk is in the hash table
nsRefPtr<CacheFileChunk> chunkCheck;
mChunks.Get(chunk->Index(), getter_AddRefs(chunkCheck));
MOZ_ASSERT(chunkCheck == chunk);
// We also shouldn't have any queued listener for this chunk
ChunkListeners *listeners;
mChunkListeners.Get(chunk->Index(), &listeners);
MOZ_ASSERT(!listeners);
}
#endif
if (NS_FAILED(chunk->GetStatus())) {
SetError(chunk->GetStatus());
}
if (NS_FAILED(mStatus)) {
// Don't write any chunk to disk since this entry will be doomed
LOG(("CacheFile::DeactivateChunk() - Releasing chunk because of status "
"[this=%p, chunk=%p, mStatus=0x%08x]", this, chunk.get(), mStatus));
RemoveChunkInternal(chunk, false);
return mStatus;
}
if (chunk->IsDirty() && !mMemoryOnly && !mOpeningFile) {
LOG(("CacheFile::DeactivateChunk() - Writing dirty chunk to the disk "
"[this=%p]", this));
mDataIsDirty = true;
rv = chunk->Write(mHandle, this);
if (NS_FAILED(rv)) {
LOG(("CacheFile::DeactivateChunk() - CacheFileChunk::Write() failed "
"synchronously. Removing it. [this=%p, chunk=%p, rv=0x%08x]",
this, chunk.get(), rv));
RemoveChunkInternal(chunk, false);
SetError(rv);
return rv;
}
// Chunk will be removed in OnChunkWritten if it is still unused
// chunk needs to be released under the lock to be able to rely on
// CacheFileChunk::mRefCnt in CacheFile::OnChunkWritten()
chunk = nullptr;
return NS_OK;
}
bool keepChunk = ShouldCacheChunk(aChunk->Index());
LOG(("CacheFile::DeactivateChunk() - %s unused chunk [this=%p, chunk=%p]",
keepChunk ? "Caching" : "Releasing", this, chunk.get()));
RemoveChunkInternal(chunk, keepChunk);
if (!mMemoryOnly)
WriteMetadataIfNeededLocked();
}
return NS_OK;
}
void
CacheFile::RemoveChunkInternal(CacheFileChunk *aChunk, bool aCacheChunk)
{
AssertOwnsLock();
aChunk->mActiveChunk = false;
ReleaseOutsideLock(static_cast<CacheFileChunkListener *>(
aChunk->mFile.forget().take()));
if (aCacheChunk) {
mCachedChunks.Put(aChunk->Index(), aChunk);
}
mChunks.Remove(aChunk->Index());
}
int64_t
CacheFile::BytesFromChunk(uint32_t aIndex)
{
AssertOwnsLock();
if (!mDataSize)
return 0;
// Index of the last existing chunk.
uint32_t lastChunk = (mDataSize - 1) / kChunkSize;
if (aIndex > lastChunk)
return 0;
// We can use only preloaded chunks for the given stream to calculate
// available bytes if this is an entry stored on disk, since only those
// chunks are guaranteed not to be released.
uint32_t maxPreloadedChunk;
if (mMemoryOnly) {
maxPreloadedChunk = lastChunk;
} else {
maxPreloadedChunk = std::min(aIndex + mPreloadChunkCount, lastChunk);
}
uint32_t i;
for (i = aIndex; i <= maxPreloadedChunk; ++i) {
CacheFileChunk * chunk;
chunk = mChunks.GetWeak(i);
if (chunk) {
MOZ_ASSERT(i == lastChunk || chunk->mDataSize == kChunkSize);
if (chunk->IsReady()) {
continue;
}
// don't search this chunk in cached
break;
}
chunk = mCachedChunks.GetWeak(i);
if (chunk) {
MOZ_ASSERT(i == lastChunk || chunk->mDataSize == kChunkSize);
continue;
}
break;
}
// theoretic bytes in advance
int64_t advance = int64_t(i - aIndex) * kChunkSize;
// real bytes till the end of the file
int64_t tail = mDataSize - (aIndex * kChunkSize);
return std::min(advance, tail);
}
static uint32_t
StatusToTelemetryEnum(nsresult aStatus)
{
if (NS_SUCCEEDED(aStatus)) {
return 0;
}
switch (aStatus) {
case NS_BASE_STREAM_CLOSED:
return 0; // Log this as a success
case NS_ERROR_OUT_OF_MEMORY:
return 2;
case NS_ERROR_FILE_DISK_FULL:
return 3;
case NS_ERROR_FILE_CORRUPTED:
return 4;
case NS_ERROR_FILE_NOT_FOUND:
return 5;
case NS_BINDING_ABORTED:
return 6;
default:
return 1; // other error
}
NS_NOTREACHED("We should never get here");
}
nsresult
CacheFile::RemoveInput(CacheFileInputStream *aInput, nsresult aStatus)
{
CacheFileAutoLock lock(this);
LOG(("CacheFile::RemoveInput() [this=%p, input=%p, status=0x%08x]", this,
aInput, aStatus));
DebugOnly<bool> found;
found = mInputs.RemoveElement(aInput);
MOZ_ASSERT(found);
ReleaseOutsideLock(static_cast<nsIInputStream*>(aInput));
if (!mMemoryOnly)
WriteMetadataIfNeededLocked();
// If the input didn't read all data, there might be left some preloaded
// chunks that won't be used anymore.
mCachedChunks.Enumerate(&CacheFile::CleanUpCachedChunks, this);
Telemetry::Accumulate(Telemetry::NETWORK_CACHE_V2_INPUT_STREAM_STATUS,
StatusToTelemetryEnum(aStatus));
return NS_OK;
}
nsresult
CacheFile::RemoveOutput(CacheFileOutputStream *aOutput, nsresult aStatus)
{
AssertOwnsLock();
LOG(("CacheFile::RemoveOutput() [this=%p, output=%p, status=0x%08x]", this,
aOutput, aStatus));
if (mOutput != aOutput) {
LOG(("CacheFile::RemoveOutput() - This output was already removed, ignoring"
" call [this=%p]", this));
return NS_OK;
}
mOutput = nullptr;
// Cancel all queued chunk and update listeners that cannot be satisfied
NotifyListenersAboutOutputRemoval();
if (!mMemoryOnly)
WriteMetadataIfNeededLocked();
// Make sure the CacheFile status is set to a failure when the output stream
// is closed with a fatal error. This way we propagate correctly and w/o any
// windows the failure state of this entry to end consumers.
if (NS_SUCCEEDED(mStatus) && NS_FAILED(aStatus) && aStatus != NS_BASE_STREAM_CLOSED) {
mStatus = aStatus;
}
// Notify close listener as the last action
aOutput->NotifyCloseListener();
Telemetry::Accumulate(Telemetry::NETWORK_CACHE_V2_OUTPUT_STREAM_STATUS,
StatusToTelemetryEnum(aStatus));
return NS_OK;
}
nsresult
CacheFile::NotifyChunkListener(CacheFileChunkListener *aCallback,
nsIEventTarget *aTarget,
nsresult aResult,
uint32_t aChunkIdx,
CacheFileChunk *aChunk)
{
LOG(("CacheFile::NotifyChunkListener() [this=%p, listener=%p, target=%p, "
"rv=0x%08x, idx=%u, chunk=%p]", this, aCallback, aTarget, aResult,
aChunkIdx, aChunk));
nsresult rv;
nsRefPtr<NotifyChunkListenerEvent> ev;
ev = new NotifyChunkListenerEvent(aCallback, aResult, aChunkIdx, aChunk);
if (aTarget)
rv = aTarget->Dispatch(ev, NS_DISPATCH_NORMAL);
else
rv = NS_DispatchToCurrentThread(ev);
NS_ENSURE_SUCCESS(rv, rv);
return NS_OK;
}
nsresult
CacheFile::QueueChunkListener(uint32_t aIndex,
CacheFileChunkListener *aCallback)
{
LOG(("CacheFile::QueueChunkListener() [this=%p, idx=%u, listener=%p]",
this, aIndex, aCallback));
AssertOwnsLock();
MOZ_ASSERT(aCallback);
ChunkListenerItem *item = new ChunkListenerItem();
item->mTarget = CacheFileIOManager::IOTarget();
if (!item->mTarget) {
LOG(("CacheFile::QueueChunkListener() - Cannot get Cache I/O thread! Using "
"main thread for callback."));
item->mTarget = do_GetMainThread();
}
item->mCallback = aCallback;
ChunkListeners *listeners;
if (!mChunkListeners.Get(aIndex, &listeners)) {
listeners = new ChunkListeners();
mChunkListeners.Put(aIndex, listeners);
}
listeners->mItems.AppendElement(item);
return NS_OK;
}
nsresult
CacheFile::NotifyChunkListeners(uint32_t aIndex, nsresult aResult,
CacheFileChunk *aChunk)
{
LOG(("CacheFile::NotifyChunkListeners() [this=%p, idx=%u, rv=0x%08x, "
"chunk=%p]", this, aIndex, aResult, aChunk));
AssertOwnsLock();
nsresult rv, rv2;
ChunkListeners *listeners;
mChunkListeners.Get(aIndex, &listeners);
MOZ_ASSERT(listeners);
rv = NS_OK;
for (uint32_t i = 0 ; i < listeners->mItems.Length() ; i++) {
ChunkListenerItem *item = listeners->mItems[i];
rv2 = NotifyChunkListener(item->mCallback, item->mTarget, aResult, aIndex,
aChunk);
if (NS_FAILED(rv2) && NS_SUCCEEDED(rv))
rv = rv2;
delete item;
}
mChunkListeners.Remove(aIndex);
return rv;
}
bool
CacheFile::HaveChunkListeners(uint32_t aIndex)
{
ChunkListeners *listeners;
mChunkListeners.Get(aIndex, &listeners);
return !!listeners;
}
void
CacheFile::NotifyListenersAboutOutputRemoval()
{
LOG(("CacheFile::NotifyListenersAboutOutputRemoval() [this=%p]", this));
AssertOwnsLock();
// First fail all chunk listeners that wait for non-existent chunk
mChunkListeners.Enumerate(&CacheFile::FailListenersIfNonExistentChunk,
this);
// Fail all update listeners
mChunks.Enumerate(&CacheFile::FailUpdateListeners, this);
}
bool
CacheFile::DataSize(int64_t* aSize)
{
CacheFileAutoLock lock(this);
if (mOutput)
return false;
*aSize = mDataSize;
return true;
}
bool
CacheFile::IsDoomed()
{
if (!mHandle)
return false;
return mHandle->IsDoomed();
}
bool
CacheFile::IsWriteInProgress()
{
// Returns true when there is a potentially unfinished write operation.
// Not using lock for performance reasons. mMetadata is never released
// during life time of CacheFile.
bool result = false;
if (!mMemoryOnly) {
result = mDataIsDirty ||
(mMetadata && mMetadata->IsDirty()) ||
mWritingMetadata;
}
result = result ||
mOpeningFile ||
mOutput ||
mChunks.Count();
return result;
}
bool
CacheFile::IsDirty()
{
return mDataIsDirty || mMetadata->IsDirty();
}
void
CacheFile::WriteMetadataIfNeeded()
{
LOG(("CacheFile::WriteMetadataIfNeeded() [this=%p]", this));
CacheFileAutoLock lock(this);
if (!mMemoryOnly)
WriteMetadataIfNeededLocked();
}
void
CacheFile::WriteMetadataIfNeededLocked(bool aFireAndForget)
{
// When aFireAndForget is set to true, we are called from dtor.
// |this| must not be referenced after this method returns!
LOG(("CacheFile::WriteMetadataIfNeededLocked() [this=%p]", this));
nsresult rv;
AssertOwnsLock();
MOZ_ASSERT(!mMemoryOnly);
if (!mMetadata) {
MOZ_CRASH("Must have metadata here");
return;
}
if (NS_FAILED(mStatus))
return;
if (!IsDirty() || mOutput || mInputs.Length() || mChunks.Count() ||
mWritingMetadata || mOpeningFile)
return;
if (!aFireAndForget) {
// if aFireAndForget is set, we are called from dtor. Write
// scheduler hard-refers CacheFile otherwise, so we cannot be here.
CacheFileIOManager::UnscheduleMetadataWrite(this);
}
LOG(("CacheFile::WriteMetadataIfNeededLocked() - Writing metadata [this=%p]",
this));
rv = mMetadata->WriteMetadata(mDataSize, aFireAndForget ? nullptr : this);
if (NS_SUCCEEDED(rv)) {
mWritingMetadata = true;
mDataIsDirty = false;
} else {
LOG(("CacheFile::WriteMetadataIfNeededLocked() - Writing synchronously "
"failed [this=%p]", this));
// TODO: close streams with error
SetError(rv);
}
}
void
CacheFile::PostWriteTimer()
{
if (mMemoryOnly)
return;
LOG(("CacheFile::PostWriteTimer() [this=%p]", this));
CacheFileIOManager::ScheduleMetadataWrite(this);
}
PLDHashOperator
CacheFile::WriteAllCachedChunks(const uint32_t& aIdx,
nsRefPtr<CacheFileChunk>& aChunk,
void* aClosure)
{
CacheFile *file = static_cast<CacheFile*>(aClosure);
LOG(("CacheFile::WriteAllCachedChunks() [this=%p, idx=%u, chunk=%p]",
file, aIdx, aChunk.get()));
file->mChunks.Put(aIdx, aChunk);
aChunk->mFile = file;
aChunk->mActiveChunk = true;
MOZ_ASSERT(aChunk->IsReady());
NS_ADDREF(aChunk);
file->ReleaseOutsideLock(aChunk);
return PL_DHASH_REMOVE;
}
PLDHashOperator
CacheFile::FailListenersIfNonExistentChunk(
const uint32_t& aIdx,
nsAutoPtr<ChunkListeners>& aListeners,
void* aClosure)
{
CacheFile *file = static_cast<CacheFile*>(aClosure);
LOG(("CacheFile::FailListenersIfNonExistentChunk() [this=%p, idx=%u]",
file, aIdx));
nsRefPtr<CacheFileChunk> chunk;
file->mChunks.Get(aIdx, getter_AddRefs(chunk));
if (chunk) {
MOZ_ASSERT(!chunk->IsReady());
return PL_DHASH_NEXT;
}
for (uint32_t i = 0 ; i < aListeners->mItems.Length() ; i++) {
ChunkListenerItem *item = aListeners->mItems[i];
file->NotifyChunkListener(item->mCallback, item->mTarget,
NS_ERROR_NOT_AVAILABLE, aIdx, nullptr);
delete item;
}
return PL_DHASH_REMOVE;
}
PLDHashOperator
CacheFile::FailUpdateListeners(
const uint32_t& aIdx,
nsRefPtr<CacheFileChunk>& aChunk,
void* aClosure)
{
#ifdef PR_LOGGING
CacheFile *file = static_cast<CacheFile*>(aClosure);
#endif
LOG(("CacheFile::FailUpdateListeners() [this=%p, idx=%u]",
file, aIdx));
if (aChunk->IsReady()) {
aChunk->NotifyUpdateListeners();
}
return PL_DHASH_NEXT;
}
PLDHashOperator
CacheFile::CleanUpCachedChunks(const uint32_t& aIdx,
nsRefPtr<CacheFileChunk>& aChunk,
void* aClosure)
{
CacheFile *file = static_cast<CacheFile*>(aClosure);
LOG(("CacheFile::CleanUpCachedChunks() [this=%p, idx=%u, chunk=%p]", file,
aIdx, aChunk.get()));
if (file->MustKeepCachedChunk(aIdx)) {
LOG(("CacheFile::CleanUpCachedChunks() - Keeping chunk"));
return PL_DHASH_NEXT;
}
LOG(("CacheFile::CleanUpCachedChunks() - Removing chunk"));
return PL_DHASH_REMOVE;
}
nsresult
CacheFile::PadChunkWithZeroes(uint32_t aChunkIdx)
{
AssertOwnsLock();
// This method is used to pad last incomplete chunk with zeroes or create
// a new chunk full of zeroes
MOZ_ASSERT(mDataSize / kChunkSize == aChunkIdx);
nsresult rv;
nsRefPtr<CacheFileChunk> chunk;
rv = GetChunkLocked(aChunkIdx, WRITER, nullptr, getter_AddRefs(chunk));
NS_ENSURE_SUCCESS(rv, rv);
LOG(("CacheFile::PadChunkWithZeroes() - Zeroing hole in chunk %d, range %d-%d"
" [this=%p]", aChunkIdx, chunk->DataSize(), kChunkSize - 1, this));
rv = chunk->EnsureBufSize(kChunkSize);
if (NS_FAILED(rv)) {
ReleaseOutsideLock(chunk.forget().take());
SetError(rv);
return rv;
}
memset(chunk->BufForWriting() + chunk->DataSize(), 0, kChunkSize - chunk->DataSize());
chunk->UpdateDataSize(chunk->DataSize(), kChunkSize - chunk->DataSize(),
false);
ReleaseOutsideLock(chunk.forget().take());
return NS_OK;
}
void
CacheFile::SetError(nsresult aStatus)
{
AssertOwnsLock();
if (NS_SUCCEEDED(mStatus)) {
mStatus = aStatus;
if (mHandle) {
CacheFileIOManager::DoomFile(mHandle, nullptr);
}
}
}
nsresult
CacheFile::InitIndexEntry()
{
MOZ_ASSERT(mHandle);
if (mHandle->IsDoomed())
return NS_OK;
nsresult rv;
rv = CacheFileIOManager::InitIndexEntry(mHandle,
mMetadata->AppId(),
mMetadata->IsAnonymous(),
mMetadata->IsInBrowser());
NS_ENSURE_SUCCESS(rv, rv);
uint32_t expTime;
mMetadata->GetExpirationTime(&expTime);
uint32_t frecency;
mMetadata->GetFrecency(&frecency);
rv = CacheFileIOManager::UpdateIndexEntry(mHandle, &frecency, &expTime);
NS_ENSURE_SUCCESS(rv, rv);
return NS_OK;
}
// Memory reporting
namespace { // anon
size_t
CollectChunkSize(uint32_t const & aIdx,
nsRefPtr<mozilla::net::CacheFileChunk> const & aChunk,
mozilla::MallocSizeOf mallocSizeOf, void* aClosure)
{
return aChunk->SizeOfIncludingThis(mallocSizeOf);
}
} // anon
size_t
CacheFile::SizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const
{
CacheFileAutoLock lock(const_cast<CacheFile*>(this));
size_t n = 0;
n += mKey.SizeOfExcludingThisIfUnshared(mallocSizeOf);
n += mChunks.SizeOfExcludingThis(CollectChunkSize, mallocSizeOf);
n += mCachedChunks.SizeOfExcludingThis(CollectChunkSize, mallocSizeOf);
if (mMetadata) {
n += mMetadata->SizeOfIncludingThis(mallocSizeOf);
}
// Input streams are not elsewhere reported.
n += mInputs.SizeOfExcludingThis(mallocSizeOf);
for (uint32_t i = 0; i < mInputs.Length(); ++i) {
n += mInputs[i]->SizeOfIncludingThis(mallocSizeOf);
}
// Output streams are not elsewhere reported.
if (mOutput) {
n += mOutput->SizeOfIncludingThis(mallocSizeOf);
}
// The listeners are usually classes reported just above.
n += mChunkListeners.SizeOfExcludingThis(nullptr, mallocSizeOf);
n += mObjsToRelease.SizeOfExcludingThis(mallocSizeOf);
// mHandle reported directly from CacheFileIOManager.
return n;
}
size_t
CacheFile::SizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const
{
return mallocSizeOf(this) + SizeOfExcludingThis(mallocSizeOf);
}
} // net
} // mozilla