/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* vim:set ts=2 sw=2 sts=2 et cindent: */ /* 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 "AudioDestinationNode.h" #include "AlignmentUtils.h" #include "AudibilityMonitor.h" #include "AudioChannelService.h" #include "AudioContext.h" #include "AudioNodeEngine.h" #include "AudioNodeTrack.h" #include "CubebUtils.h" #include "MediaTrackGraph.h" #include "mozilla/StaticPrefs_dom.h" #include "mozilla/dom/AudioDestinationNodeBinding.h" #include "mozilla/dom/BaseAudioContextBinding.h" #include "mozilla/dom/OfflineAudioCompletionEvent.h" #include "mozilla/dom/Promise.h" #include "mozilla/dom/ScriptSettings.h" #include "mozilla/dom/WakeLock.h" #include "mozilla/dom/power/PowerManagerService.h" #include "mozilla/Telemetry.h" #include "mozilla/TelemetryHistogramEnums.h" #include "nsContentUtils.h" #include "nsIInterfaceRequestorUtils.h" #include "nsIScriptObjectPrincipal.h" #include "nsServiceManagerUtils.h" #include "Tracing.h" extern mozilla::LazyLogModule gAudioChannelLog; #define AUDIO_CHANNEL_LOG(msg, ...) \ MOZ_LOG(gAudioChannelLog, LogLevel::Debug, (msg, ##__VA_ARGS__)) namespace mozilla::dom { namespace { class OnCompleteTask final : public Runnable { public: OnCompleteTask(AudioContext* aAudioContext, AudioBuffer* aRenderedBuffer) : Runnable("dom::OfflineDestinationNodeEngine::OnCompleteTask"), mAudioContext(aAudioContext), mRenderedBuffer(aRenderedBuffer) {} NS_IMETHOD Run() override { OfflineAudioCompletionEventInit param; param.mRenderedBuffer = mRenderedBuffer; RefPtr event = OfflineAudioCompletionEvent::Constructor(mAudioContext, u"complete"_ns, param); mAudioContext->DispatchTrustedEvent(event); return NS_OK; } private: RefPtr mAudioContext; RefPtr mRenderedBuffer; }; } // anonymous namespace class OfflineDestinationNodeEngine final : public AudioNodeEngine { public: explicit OfflineDestinationNodeEngine(AudioDestinationNode* aNode) : AudioNodeEngine(aNode), mWriteIndex(0), mNumberOfChannels(aNode->ChannelCount()), mLength(aNode->Length()), mSampleRate(aNode->Context()->SampleRate()), mBufferAllocated(false) {} void ProcessBlock(AudioNodeTrack* aTrack, GraphTime aFrom, const AudioBlock& aInput, AudioBlock* aOutput, bool* aFinished) override { TRACE("OfflineDestinationNodeEngine::ProcessBlock"); // Do this just for the sake of political correctness; this output // will not go anywhere. *aOutput = aInput; // The output buffer is allocated lazily, on the rendering thread, when // non-null input is received. if (!mBufferAllocated && !aInput.IsNull()) { // These allocations might fail if content provides a huge number of // channels or size, but it's OK since we'll deal with the failure // gracefully. mBuffer = ThreadSharedFloatArrayBufferList::Create(mNumberOfChannels, mLength, fallible); if (mBuffer && mWriteIndex) { // Zero leading for any null chunks that were skipped. for (uint32_t i = 0; i < mNumberOfChannels; ++i) { float* channelData = mBuffer->GetDataForWrite(i); PodZero(channelData, mWriteIndex); } } mBufferAllocated = true; } // Skip copying if there is no buffer. uint32_t outputChannelCount = mBuffer ? mNumberOfChannels : 0; // Record our input buffer MOZ_ASSERT(mWriteIndex < mLength, "How did this happen?"); const uint32_t duration = std::min(WEBAUDIO_BLOCK_SIZE, mLength - mWriteIndex); const uint32_t inputChannelCount = aInput.ChannelCount(); for (uint32_t i = 0; i < outputChannelCount; ++i) { float* outputData = mBuffer->GetDataForWrite(i) + mWriteIndex; if (aInput.IsNull() || i >= inputChannelCount) { PodZero(outputData, duration); } else { const float* inputBuffer = static_cast(aInput.mChannelData[i]); if (duration == WEBAUDIO_BLOCK_SIZE && IS_ALIGNED16(inputBuffer)) { // Use the optimized version of the copy with scale operation AudioBlockCopyChannelWithScale(inputBuffer, aInput.mVolume, outputData); } else { if (aInput.mVolume == 1.0f) { PodCopy(outputData, inputBuffer, duration); } else { for (uint32_t j = 0; j < duration; ++j) { outputData[j] = aInput.mVolume * inputBuffer[j]; } } } } } mWriteIndex += duration; if (mWriteIndex >= mLength) { NS_ASSERTION(mWriteIndex == mLength, "Overshot length"); // Go to finished state. When the graph's current time eventually reaches // the end of the track, then the main thread will be notified and we'll // shut down the AudioContext. *aFinished = true; } } bool IsActive() const override { // Keep processing to track track time, which is used for all timelines // associated with the same AudioContext. return true; } already_AddRefed CreateAudioBuffer(AudioContext* aContext) { MOZ_ASSERT(NS_IsMainThread()); // Create the input buffer ErrorResult rv; RefPtr renderedBuffer = AudioBuffer::Create(aContext->GetOwner(), mNumberOfChannels, mLength, mSampleRate, mBuffer.forget(), rv); if (rv.Failed()) { rv.SuppressException(); return nullptr; } return renderedBuffer.forget(); } size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const override { size_t amount = AudioNodeEngine::SizeOfExcludingThis(aMallocSizeOf); if (mBuffer) { amount += mBuffer->SizeOfIncludingThis(aMallocSizeOf); } return amount; } size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const override { return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); } private: // The input to the destination node is recorded in mBuffer. // When this buffer fills up with mLength frames, the buffered input is sent // to the main thread in order to dispatch OfflineAudioCompletionEvent. RefPtr mBuffer; // An index representing the next offset in mBuffer to be written to. uint32_t mWriteIndex; uint32_t mNumberOfChannels; // How many frames the OfflineAudioContext intends to produce. uint32_t mLength; float mSampleRate; bool mBufferAllocated; }; class DestinationNodeEngine final : public AudioNodeEngine { public: explicit DestinationNodeEngine(AudioDestinationNode* aNode) : AudioNodeEngine(aNode), mSampleRate(CubebUtils::PreferredSampleRate()), mVolume(1.0f), mAudibilityMonitor( mSampleRate, StaticPrefs::dom_media_silence_duration_for_audibility()), mSuspended(false), mIsAudible(false) { MOZ_ASSERT(aNode); } void ProcessBlock(AudioNodeTrack* aTrack, GraphTime aFrom, const AudioBlock& aInput, AudioBlock* aOutput, bool* aFinished) override { TRACE("DestinationNodeEngine::ProcessBlock"); *aOutput = aInput; aOutput->mVolume *= mVolume; if (mSuspended) { return; } mAudibilityMonitor.Process(aInput); bool isAudible = mAudibilityMonitor.RecentlyAudible() && aOutput->mVolume > 0.0; if (isAudible != mIsAudible) { mIsAudible = isAudible; RefPtr track = aTrack; auto r = [track, isAudible]() -> void { MOZ_ASSERT(NS_IsMainThread()); RefPtr node = track->Engine()->NodeMainThread(); if (node) { RefPtr destinationNode = static_cast(node.get()); destinationNode->NotifyDataAudibleStateChanged(isAudible); } }; aTrack->Graph()->DispatchToMainThreadStableState(NS_NewRunnableFunction( "dom::WebAudioAudibleStateChangedRunnable", r)); } } bool IsActive() const override { // Keep processing to track track time, which is used for all timelines // associated with the same AudioContext. If there are no other engines // for the AudioContext, then this could return false to suspend the // track, but the track is blocked anyway through // AudioDestinationNode::SetIsOnlyNodeForContext(). return true; } void SetDoubleParameter(uint32_t aIndex, double aParam) override { if (aIndex == VOLUME) { mVolume = static_cast(aParam); } } void SetInt32Parameter(uint32_t aIndex, int32_t aParam) override { if (aIndex == SUSPENDED) { mSuspended = !!aParam; if (mSuspended) { mIsAudible = false; } } } enum Parameters { VOLUME, SUSPENDED, }; size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const override { return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); } private: int mSampleRate; float mVolume; AudibilityMonitor mAudibilityMonitor; bool mSuspended; bool mIsAudible; }; NS_IMPL_CYCLE_COLLECTION_INHERITED(AudioDestinationNode, AudioNode, mAudioChannelAgent, mOfflineRenderingPromise) NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(AudioDestinationNode) NS_INTERFACE_MAP_ENTRY(nsIAudioChannelAgentCallback) NS_INTERFACE_MAP_END_INHERITING(AudioNode) NS_IMPL_ADDREF_INHERITED(AudioDestinationNode, AudioNode) NS_IMPL_RELEASE_INHERITED(AudioDestinationNode, AudioNode) const AudioNodeTrack::Flags kTrackFlags = AudioNodeTrack::NEED_MAIN_THREAD_CURRENT_TIME | AudioNodeTrack::NEED_MAIN_THREAD_ENDED | AudioNodeTrack::EXTERNAL_OUTPUT; AudioDestinationNode::AudioDestinationNode(AudioContext* aContext, bool aIsOffline, uint32_t aNumberOfChannels, uint32_t aLength) : AudioNode(aContext, aNumberOfChannels, ChannelCountMode::Explicit, ChannelInterpretation::Speakers), mFramesToProduce(aLength), mIsOffline(aIsOffline), mCreatedTime(TimeStamp::Now()) { if (aIsOffline) { // The track is created on demand to avoid creating a graph thread that // may not be used. return; } // GetParentObject can return nullptr here. This will end up creating another // MediaTrackGraph MediaTrackGraph* graph = MediaTrackGraph::GetInstance( MediaTrackGraph::AUDIO_THREAD_DRIVER, aContext->GetParentObject(), aContext->SampleRate(), MediaTrackGraph::DEFAULT_OUTPUT_DEVICE); AudioNodeEngine* engine = new DestinationNodeEngine(this); mTrack = AudioNodeTrack::Create(aContext, engine, kTrackFlags, graph); mTrack->AddMainThreadListener(this); // null key is fine: only one output per mTrack mTrack->AddAudioOutput(nullptr); } void AudioDestinationNode::Init() { // The reason we don't do that in ctor is because we have to keep AudioContext // holding a strong reference to the destination node first. If we don't do // that, initializing the agent would cause an unexpected destroy of the // destination node when destroying the local weak reference inside // `InitWithWeakCallback()`. if (!mIsOffline) { CreateAndStartAudioChannelAgent(); } } void AudioDestinationNode::Close() { DestroyAudioChannelAgentIfExists(); ReleaseAudioWakeLockIfExists(); } void AudioDestinationNode::CreateAndStartAudioChannelAgent() { MOZ_ASSERT(!mIsOffline); MOZ_ASSERT(!mAudioChannelAgent); AudioChannelAgent* agent = new AudioChannelAgent(); nsresult rv = agent->InitWithWeakCallback(GetOwner(), this); if (NS_WARN_IF(NS_FAILED(rv))) { AUDIO_CHANNEL_LOG("Failed to init audio channel agent"); return; } AudibleState state = IsAudible() ? AudibleState::eAudible : AudibleState::eNotAudible; rv = agent->NotifyStartedPlaying(state); if (NS_WARN_IF(NS_FAILED(rv))) { AUDIO_CHANNEL_LOG("Failed to start audio channel agent"); return; } mAudioChannelAgent = agent; mAudioChannelAgent->PullInitialUpdate(); } AudioDestinationNode::~AudioDestinationNode() { MOZ_ASSERT(!mAudioChannelAgent); MOZ_ASSERT(!mWakeLock); MOZ_ASSERT(!mCaptureTrackPort); } size_t AudioDestinationNode::SizeOfExcludingThis( MallocSizeOf aMallocSizeOf) const { size_t amount = AudioNode::SizeOfExcludingThis(aMallocSizeOf); // Might be useful in the future: // - mAudioChannelAgent return amount; } size_t AudioDestinationNode::SizeOfIncludingThis( MallocSizeOf aMallocSizeOf) const { return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); } AudioNodeTrack* AudioDestinationNode::Track() { if (mTrack) { return mTrack; } AudioContext* context = Context(); if (!context) { // This node has been unlinked. return nullptr; } MOZ_ASSERT(mIsOffline, "Realtime tracks are created in constructor"); // GetParentObject can return nullptr here when the document has been // unlinked. MediaTrackGraph* graph = MediaTrackGraph::CreateNonRealtimeInstance( context->SampleRate(), context->GetParentObject()); AudioNodeEngine* engine = new OfflineDestinationNodeEngine(this); mTrack = AudioNodeTrack::Create(context, engine, kTrackFlags, graph); mTrack->AddMainThreadListener(this); return mTrack; } void AudioDestinationNode::DestroyAudioChannelAgentIfExists() { if (mAudioChannelAgent) { mAudioChannelAgent->NotifyStoppedPlaying(); mAudioChannelAgent = nullptr; if (IsCapturingAudio()) { StopAudioCapturingTrack(); } } } void AudioDestinationNode::DestroyMediaTrack() { Close(); if (!mTrack) { return; } Context()->ShutdownWorklet(); mTrack->RemoveMainThreadListener(this); AudioNode::DestroyMediaTrack(); } void AudioDestinationNode::NotifyMainThreadTrackEnded() { MOZ_ASSERT(NS_IsMainThread()); MOZ_ASSERT(mTrack->IsEnded()); if (mIsOffline && GetAbstractMainThread()) { GetAbstractMainThread()->Dispatch(NewRunnableMethod( "dom::AudioDestinationNode::FireOfflineCompletionEvent", this, &AudioDestinationNode::FireOfflineCompletionEvent)); } } void AudioDestinationNode::FireOfflineCompletionEvent() { AudioContext* context = Context(); context->OfflineClose(); OfflineDestinationNodeEngine* engine = static_cast(Track()->Engine()); RefPtr renderedBuffer = engine->CreateAudioBuffer(context); if (!renderedBuffer) { return; } ResolvePromise(renderedBuffer); context->Dispatch(do_AddRef(new OnCompleteTask(context, renderedBuffer))); context->OnStateChanged(nullptr, AudioContextState::Closed); mOfflineRenderingRef.Drop(this); } void AudioDestinationNode::ResolvePromise(AudioBuffer* aRenderedBuffer) { MOZ_ASSERT(NS_IsMainThread()); MOZ_ASSERT(mIsOffline); mOfflineRenderingPromise->MaybeResolve(aRenderedBuffer); } uint32_t AudioDestinationNode::MaxChannelCount() const { return Context()->MaxChannelCount(); } void AudioDestinationNode::SetChannelCount(uint32_t aChannelCount, ErrorResult& aRv) { if (aChannelCount > MaxChannelCount()) { aRv.ThrowIndexSizeError( nsPrintfCString("%u is larger than maxChannelCount", aChannelCount)); return; } if (aChannelCount == ChannelCount()) { return; } AudioNode::SetChannelCount(aChannelCount, aRv); } void AudioDestinationNode::Mute() { MOZ_ASSERT(Context() && !Context()->IsOffline()); SendDoubleParameterToTrack(DestinationNodeEngine::VOLUME, 0.0f); } void AudioDestinationNode::Unmute() { MOZ_ASSERT(Context() && !Context()->IsOffline()); SendDoubleParameterToTrack(DestinationNodeEngine::VOLUME, 1.0f); } void AudioDestinationNode::Suspend() { SendInt32ParameterToTrack(DestinationNodeEngine::SUSPENDED, 1); } void AudioDestinationNode::Resume() { SendInt32ParameterToTrack(DestinationNodeEngine::SUSPENDED, 0); } void AudioDestinationNode::NotifyAudioContextStateChanged() { UpdateFinalAudibleStateIfNeeded(AudibleChangedReasons::ePauseStateChanged); } void AudioDestinationNode::OfflineShutdown() { MOZ_ASSERT(Context() && Context()->IsOffline(), "Should only be called on a valid OfflineAudioContext"); mOfflineRenderingRef.Drop(this); } JSObject* AudioDestinationNode::WrapObject(JSContext* aCx, JS::Handle aGivenProto) { return AudioDestinationNode_Binding::Wrap(aCx, this, aGivenProto); } void AudioDestinationNode::StartRendering(Promise* aPromise) { mOfflineRenderingPromise = aPromise; mOfflineRenderingRef.Take(this); Track()->Graph()->StartNonRealtimeProcessing(mFramesToProduce); } NS_IMETHODIMP AudioDestinationNode::WindowVolumeChanged(float aVolume, bool aMuted) { MOZ_ASSERT(mAudioChannelAgent); if (!mTrack) { return NS_OK; } AUDIO_CHANNEL_LOG( "AudioDestinationNode %p WindowVolumeChanged, " "aVolume = %f, aMuted = %s\n", this, aVolume, aMuted ? "true" : "false"); mAudioChannelVolume = aMuted ? 0.0f : aVolume; mTrack->SetAudioOutputVolume(nullptr, mAudioChannelVolume); UpdateFinalAudibleStateIfNeeded(AudibleChangedReasons::eVolumeChanged); return NS_OK; } NS_IMETHODIMP AudioDestinationNode::WindowSuspendChanged(nsSuspendedTypes aSuspend) { MOZ_ASSERT(mAudioChannelAgent); if (!mTrack) { return NS_OK; } const bool shouldDisable = aSuspend == nsISuspendedTypes::SUSPENDED_BLOCK; if (mAudioChannelDisabled == shouldDisable) { return NS_OK; } mAudioChannelDisabled = shouldDisable; AUDIO_CHANNEL_LOG( "AudioDestinationNode %p WindowSuspendChanged, shouldDisable = %d\n", this, mAudioChannelDisabled); DisabledTrackMode disabledMode = mAudioChannelDisabled ? DisabledTrackMode::SILENCE_BLACK : DisabledTrackMode::ENABLED; mTrack->SetDisabledTrackMode(disabledMode); UpdateFinalAudibleStateIfNeeded(AudibleChangedReasons::ePauseStateChanged); return NS_OK; } NS_IMETHODIMP AudioDestinationNode::WindowAudioCaptureChanged(bool aCapture) { MOZ_ASSERT(mAudioChannelAgent); if (!mTrack) { return NS_OK; } nsCOMPtr ownerWindow = GetOwner(); if (!ownerWindow) { return NS_OK; } if (aCapture == IsCapturingAudio()) { return NS_OK; } if (aCapture) { StartAudioCapturingTrack(); } else { StopAudioCapturingTrack(); } return NS_OK; } bool AudioDestinationNode::IsCapturingAudio() const { return mCaptureTrackPort != nullptr; } void AudioDestinationNode::StartAudioCapturingTrack() { MOZ_ASSERT(!IsCapturingAudio()); nsCOMPtr window = Context()->GetParentObject(); uint64_t id = window->WindowID(); mCaptureTrackPort = mTrack->Graph()->ConnectToCaptureTrack(id, mTrack); } void AudioDestinationNode::StopAudioCapturingTrack() { MOZ_ASSERT(IsCapturingAudio()); mCaptureTrackPort->Destroy(); mCaptureTrackPort = nullptr; } void AudioDestinationNode::CreateAudioWakeLockIfNeeded() { if (!mWakeLock && IsAudible()) { RefPtr pmService = power::PowerManagerService::GetInstance(); NS_ENSURE_TRUE_VOID(pmService); ErrorResult rv; mWakeLock = pmService->NewWakeLock(u"audio-playing"_ns, GetOwner(), rv); } } void AudioDestinationNode::ReleaseAudioWakeLockIfExists() { if (mWakeLock) { IgnoredErrorResult rv; mWakeLock->Unlock(rv); mWakeLock = nullptr; } } void AudioDestinationNode::NotifyDataAudibleStateChanged(bool aAudible) { MOZ_ASSERT(!mIsOffline); AUDIO_CHANNEL_LOG( "AudioDestinationNode %p NotifyDataAudibleStateChanged, audible=%d", this, aAudible); if (mDurationBeforeFirstTimeAudible.IsZero()) { MOZ_ASSERT(aAudible); mDurationBeforeFirstTimeAudible = TimeStamp::Now() - mCreatedTime; Telemetry::Accumulate(Telemetry::WEB_AUDIO_BECOMES_AUDIBLE_TIME, mDurationBeforeFirstTimeAudible.ToSeconds()); } mIsDataAudible = aAudible; UpdateFinalAudibleStateIfNeeded(AudibleChangedReasons::eDataAudibleChanged); } void AudioDestinationNode::UpdateFinalAudibleStateIfNeeded( AudibleChangedReasons aReason) { // The audio context has been closed and we've destroyed the agent. if (!mAudioChannelAgent) { return; } const bool newAudibleState = IsAudible(); if (mFinalAudibleState == newAudibleState) { return; } AUDIO_CHANNEL_LOG("AudioDestinationNode %p Final audible state=%d", this, newAudibleState); mFinalAudibleState = newAudibleState; AudibleState state = mFinalAudibleState ? AudibleState::eAudible : AudibleState::eNotAudible; mAudioChannelAgent->NotifyStartedAudible(state, aReason); if (mFinalAudibleState) { CreateAudioWakeLockIfNeeded(); } else { ReleaseAudioWakeLockIfExists(); } } bool AudioDestinationNode::IsAudible() const { // The desitionation node will be regarded as audible if all following // conditions are true. // (1) data audible state : both audio input and output are audible // (2) window audible state : the tab isn't muted by tab sound indicator // (3) audio context state : audio context should be running return Context()->State() == AudioContextState::Running && mIsDataAudible && mAudioChannelVolume != 0.0; } } // namespace mozilla::dom