gecko-dev/dom/workers/WorkerPrivate.h
Andrew Sutherland a212459d62 Bug 1544232 - Limit lifetime extension of SWs by SWs to the sender's lifetime. r=edenchuang
This patch introduces an explicit concept of lifetimes with mechanisms
in place so that actions taken by Clients (windows or non-ServiceWorker
orkers) will extend the lifetime of a ServiceWorker, but a ServiceWorker
cannot extend the life of another ServiceWorker.

The areas of concern are:
- ServiceWorker.postMessage: With ServiceWorkers exposed on workers and
  the ability to access other registrations via ServiceWorkerContainer
  being exposed, ServiceWorkers can message other ServiceWorkers.  It's
  essential that they never be allowed to give a ServiceWorker a
  lifetime longer than their own.
- ServiceWorkerRegistration.update(): Requesting an update of a
  registration should not allow any installed/updated ServiceWorker to
  have a lifetime longer than the ServiceWorker creating the request.
- ServiceWorkerContainer.register(): Requesting the installation of a
  new ServiceWorker should likewise constrain the lifetime of the newly
  installed ServiceWorker.

Note that in cases where we would potentially spawn a ServiceWorker,
whether it be in response to postMessage or as part of an install or
update, a key criteria is whether the lifetime extends far enough into
the future for us to believe the ServiceWorker can accomplish anything.
Currently we have a constant of 5 seconds against a normal full
lifetime of 30 seconds (before 30 second grace period).  So once a SW
has < 5 seconds of lifetime left, it won't be able to spawn a SW.  Note
that in the case of install/update, we do not prevent the creation of
the job at this time, instead the job will fail during the check script
evaluation step as failure to spawn the ServiceWorker is equivalent to
a script load failure.

A somewhat ugly part of this implementation is that because Bug 1853706
is not yet implemented, our actors that are fundamentally associated
with a global don't have an inherent understanding of their relationship
to that global.  So we approximate that by:
- For postMessage, we always have a ServiceWorkerDescriptor if we are
  being messaged by a ServiceWorker, allowing us direct lookup.
- ServiceWorkerRegistration.update(): In a previous patch in the stack
  we had ServiceWorkerRegistrationProxy latch the ClientInfo of its
  owning global when it was created.  Note that in the case of a
  ServiceWorker's own registration, this will be created at startup
  before the worker hits the execution ready state.
  - Note that because we have at most one live
    ServiceWorkerRegistration per global at a time, and the
    registration is fundamentally associated with the
    ServiceWorkerGlobalScope, that registration and its proxy will
    remain alive for the duration of the global.
- ServiceWorkerContainer.register(): We already were sending the client
  info along with the register call (as well as all other calls on the
  container).

Looking up the ServiceWorker from its client is not something that was
really intended.  This is further complicated by ServiceWorkerManager
being authoritative for ServiceWorkers on the parent process main thread
whereas the ClientManagerService is authoritative on PBackground and
actor-centric, making sketchy multi-threaded maps not really an option.

Looking up the ServiceWorker from a ServiceWorkerDescriptor is intended,
but the primary intent in those cases is so that the recipient of such a
descriptor can easily create a ServiceWorker instance that is
live-updating (by way of its owning ServiceWorkerRegistration; we don't
have IPC actors directly for ServiceWorkers, just the registration).
Adding the descriptor to clients until Bug 1853706 is implemented would
be an exceedingly ugly workaround because it would greatly complicate
the existing plumbing code, and a lot of the code is confusing enough
as-is.

This patch initially adopted an approach of encoding the scope of a
ServiceWorker as its client URL, but it turns out web extension
ServiceWorker support (reasonably) assumed the URL would be the script
URL so the original behavior was restored and when performing our
lookup we just check all registrations associated with the given
origin.  This is okay because register and update calls are inherently
expensive, rare operations and the overhead of the additional checks is
marginal.  Additionally, we can remove this logic once Bug 1853706 is
implemented.

As part of that initial scope tunneling was that, as noted above, we
do sample the ClientInfo for a ServiceWorker's own registration before
the worker is execution-ready.  And prior to this patch, we only would
populate the URL during execution-ready because for most globals, we
can't possibly know the URL when the ClientSource is created.  However,
for ServiceWorkers we can.  Because we also want to know what the id of
the ServiceWorker client would be, we also change the creation of the
ServiceWorker ClientSource so that it uses a ClientInfo created by the
authoritative ServiceWorkerPrivate in its Initialize method.

A minor retained hack is that because the worker scriptloader propagates
its CSP structure onto its ClientInfo (but not its ClientSource, which
feels weird, but makes sense) and that does get sent via register(), we
do also need to normalize the ClientInfo in the parent when we do
equality checking to have it ignore the CSP.

Differential Revision: https://phabricator.services.mozilla.com/D180915
2024-10-24 03:02:42 +00:00

1759 lines
58 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* 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 mozilla_dom_workers_workerprivate_h__
#define mozilla_dom_workers_workerprivate_h__
#include <bitset>
#include "MainThreadUtils.h"
#include "ScriptLoader.h"
#include "js/ContextOptions.h"
#include "mozilla/Atomics.h"
#include "mozilla/Attributes.h"
#include "mozilla/AutoRestore.h"
#include "mozilla/BasePrincipal.h"
#include "mozilla/CondVar.h"
#include "mozilla/DOMEventTargetHelper.h"
#include "mozilla/Maybe.h"
#include "mozilla/MozPromise.h"
#include "mozilla/OriginTrials.h"
#include "mozilla/RelativeTimeline.h"
#include "mozilla/Result.h"
#include "mozilla/StorageAccess.h"
#include "mozilla/ThreadBound.h"
#include "mozilla/ThreadSafeWeakPtr.h"
#include "mozilla/UniquePtr.h"
#include "mozilla/UseCounter.h"
#include "mozilla/dom/ClientSource.h"
#include "mozilla/dom/FlippedOnce.h"
#include "mozilla/dom/Timeout.h"
#include "mozilla/dom/quota/CheckedUnsafePtr.h"
#include "mozilla/dom/Worker.h"
#include "mozilla/dom/WorkerBinding.h"
#include "mozilla/dom/WorkerCommon.h"
#include "mozilla/dom/WorkerLoadInfo.h"
#include "mozilla/dom/WorkerStatus.h"
#include "mozilla/dom/workerinternals/JSSettings.h"
#include "mozilla/dom/workerinternals/Queue.h"
#include "mozilla/dom/JSExecutionManager.h"
#include "mozilla/net/NeckoChannelParams.h"
#include "mozilla/StaticPrefs_extensions.h"
#include "nsContentUtils.h"
#include "nsIChannel.h"
#include "nsIContentPolicy.h"
#include "nsIEventTarget.h"
#include "nsILoadInfo.h"
#include "nsRFPService.h"
#include "nsTObserverArray.h"
#include "stdint.h"
class nsIContentSecurityPolicy;
class nsIThreadInternal;
namespace JS {
struct RuntimeStats;
}
namespace mozilla {
class ThrottledEventQueue;
namespace dom {
class RemoteWorkerChild;
// If you change this, the corresponding list in nsIWorkerDebugger.idl needs
// to be updated too. And histograms enum for worker use counters uses the same
// order of worker kind. Please also update dom/base/usecounters.py.
enum WorkerKind : uint8_t {
WorkerKindDedicated,
WorkerKindShared,
WorkerKindService
};
class ClientInfo;
class ClientSource;
class Function;
class JSExecutionManager;
class MessagePort;
class UniqueMessagePortId;
class PerformanceStorage;
class StrongWorkerRef;
class TimeoutHandler;
class WorkerControlRunnable;
class WorkerCSPEventListener;
class WorkerDebugger;
class WorkerDebuggerGlobalScope;
class WorkerErrorReport;
class WorkerEventTarget;
class WorkerGlobalScope;
class WorkerParentRef;
class WorkerRef;
class WorkerRunnable;
class WorkerDebuggeeRunnable;
class WorkerThread;
class WorkerThreadRunnable;
// SharedMutex is a small wrapper around an (internal) reference-counted Mutex
// object. It exists to avoid changing a lot of code to use Mutex* instead of
// Mutex&.
class MOZ_CAPABILITY("mutex") SharedMutex {
using Mutex = mozilla::Mutex;
class MOZ_CAPABILITY("mutex") RefCountedMutex final : public Mutex {
public:
explicit RefCountedMutex(const char* aName) : Mutex(aName) {}
NS_INLINE_DECL_THREADSAFE_REFCOUNTING(RefCountedMutex)
private:
~RefCountedMutex() = default;
};
const RefPtr<RefCountedMutex> mMutex;
public:
explicit SharedMutex(const char* aName)
: mMutex(new RefCountedMutex(aName)) {}
SharedMutex(const SharedMutex& aOther) = default;
operator Mutex&() MOZ_RETURN_CAPABILITY(this) { return *mMutex; }
operator const Mutex&() const MOZ_RETURN_CAPABILITY(this) { return *mMutex; }
// We need these to make thread-safety analysis work
void Lock() MOZ_CAPABILITY_ACQUIRE() { mMutex->Lock(); }
void Unlock() MOZ_CAPABILITY_RELEASE() { mMutex->Unlock(); }
// We can assert we own 'this', but we can't assert we hold mMutex
void AssertCurrentThreadOwns() const
MOZ_ASSERT_CAPABILITY(this) MOZ_NO_THREAD_SAFETY_ANALYSIS {
mMutex->AssertCurrentThreadOwns();
}
};
nsString ComputeWorkerPrivateId();
class WorkerPrivate final
: public RelativeTimeline,
public SupportsCheckedUnsafePtr<CheckIf<DiagnosticAssertEnabled>> {
public:
// Callback invoked on the parent thread when the worker's cancellation is
// about to be requested. This covers both calls to
// WorkerPrivate::Cancel() by the owner as well as self-initiated cancellation
// due to top-level script evaluation failing or close() being invoked on the
// global scope for Dedicated and Shared workers, but not Service Workers as
// they do not expose a close() method.
//
// ### Parent-Initiated Cancellation
//
// When WorkerPrivate::Cancel is invoked on the parent thread (by the binding
// exposed Worker::Terminate), this callback is invoked synchronously inside
// that call.
//
// ### Worker Self-Cancellation
//
// When a worker initiates self-cancellation, the worker's notification to the
// parent thread is a non-blocking, async mechanism triggered by
// `WorkerPrivate::DispatchCancelingRunnable`.
//
// Self-cancellation races a normally scheduled runnable against a timer that
// is scheduled against the parent. The 2 paths initiated by
// DispatchCancelingRunnable are:
//
// 1. A CancelingRunnable is dispatched at the worker's normal event target to
// wait for the event loop to be clear of runnables. When the
// CancelingRunnable runs it will dispatch a CancelingOnParentRunnable to
// its parent which is a normal, non-control WorkerDebuggeeRunnable to
// ensure that any postMessages to the parent or similar events get a
// chance to be processed prior to cancellation. The timer scheduled in
// the next bullet will not be canceled unless
//
// 2. A CancelingWithTimeoutOnParentRunnable control runnable is dispatched
// to the parent to schedule a timer which will (also) fire on the parent
// thread. This handles the case where the worker does not yield
// control-flow, and so the normal runnable scheduled above does not get to
// run in a timely fashion. Because this is a control runnable, if the
// parent is a worker then the runnable will be processed with urgency.
// However, if the worker is top-level, then the control-like throttled
// WorkerPrivate::mMainThreadEventTarget will end up getting used which is
// nsIRunnablePriority::PRIORITY_MEDIUMHIGH and distinct from the
// mMainThreadDebuggeeEventTarget which most runnables (like postMessage)
// use.
//
// The timer will explicitly use the control event target if the parent is
// a worker and the implicit event target (via `NS_NewTimer()`) otherwise.
// The callback is CancelingTimerCallback which just calls
// WorkerPrivate::Cancel.
using CancellationCallback = std::function<void(bool aEverRan)>;
// Callback invoked on the parent just prior to dropping the worker thread's
// strong reference that keeps the WorkerPrivate alive while the worker thread
// is running. This does not provide a guarantee that the underlying thread
// has fully shutdown, just that the worker logic has fully shutdown.
//
// ### Details
//
// The last thing the worker thread's WorkerThreadPrimaryRunnable does before
// initiating the shutdown of the underlying thread is call ScheduleDeletion.
// ScheduleDeletion dispatches a runnable to the parent to notify it that the
// worker has completed its work and will never touch the WorkerPrivate again
// and that the strong self-reference can be dropped.
//
// For parents that are themselves workers, this will be done by
// WorkerFinishedRunnable which is a WorkerControlRunnable, ensuring that this
// is processed in a timely fashion. For main-thread parents,
// TopLevelWorkerFinishedRunnable will be used and sent via
// mMainThreadEventTargetForMessaging which is a weird ThrottledEventQueue
// which does not provide any ordering guarantees relative to
// mMainThreadDebuggeeEventTarget, so if you want those, you need to enhance
// things.
using TerminationCallback = std::function<void(void)>;
struct LocationInfo {
nsCString mHref;
nsCString mProtocol;
nsCString mHost;
nsCString mHostname;
nsCString mPort;
nsCString mPathname;
nsCString mSearch;
nsCString mHash;
nsString mOrigin;
};
NS_INLINE_DECL_REFCOUNTING(WorkerPrivate)
static already_AddRefed<WorkerPrivate> Constructor(
JSContext* aCx, const nsAString& aScriptURL, bool aIsChromeWorker,
WorkerKind aWorkerKind, RequestCredentials aRequestCredentials,
const WorkerType aWorkerType, const nsAString& aWorkerName,
const nsACString& aServiceWorkerScope, WorkerLoadInfo* aLoadInfo,
ErrorResult& aRv, nsString aId = u""_ns,
CancellationCallback&& aCancellationCallback = {},
TerminationCallback&& aTerminationCallback = {});
enum LoadGroupBehavior { InheritLoadGroup, OverrideLoadGroup };
static nsresult GetLoadInfo(
JSContext* aCx, nsPIDOMWindowInner* aWindow, WorkerPrivate* aParent,
const nsAString& aScriptURL, const enum WorkerType& aWorkerType,
const RequestCredentials& aCredentials, bool aIsChromeWorker,
LoadGroupBehavior aLoadGroupBehavior, WorkerKind aWorkerKind,
WorkerLoadInfo* aLoadInfo);
void Traverse(nsCycleCollectionTraversalCallback& aCb);
void ClearSelfAndParentEventTargetRef() {
AssertIsOnParentThread();
MOZ_ASSERT(mSelfRef);
if (mTerminationCallback) {
mTerminationCallback();
mTerminationCallback = nullptr;
}
mParentEventTargetRef = nullptr;
mSelfRef = nullptr;
}
// May be called on any thread...
bool Start();
// Called on the parent thread.
bool Notify(WorkerStatus aStatus);
bool Cancel() { return Notify(Canceling); }
bool Close() MOZ_REQUIRES(mMutex);
// The passed principal must be the Worker principal in case of a
// ServiceWorker and the loading principal for any other type.
static void OverrideLoadInfoLoadGroup(WorkerLoadInfo& aLoadInfo,
nsIPrincipal* aPrincipal);
bool IsDebuggerRegistered() MOZ_NO_THREAD_SAFETY_ANALYSIS {
AssertIsOnMainThread();
// No need to lock here since this is only ever modified by the same thread.
return mDebuggerRegistered; // would give a thread-safety warning
}
bool ExtensionAPIAllowed() {
return (
StaticPrefs::extensions_backgroundServiceWorker_enabled_AtStartup() &&
mExtensionAPIAllowed);
}
void SetIsDebuggerRegistered(bool aDebuggerRegistered) {
AssertIsOnMainThread();
MutexAutoLock lock(mMutex);
MOZ_ASSERT(mDebuggerRegistered != aDebuggerRegistered);
mDebuggerRegistered = aDebuggerRegistered;
mCondVar.Notify();
}
// Mark worker private as running in the background tab
// for further throttling
void SetIsRunningInBackground();
void SetIsRunningInForeground();
bool ChangeBackgroundStateInternal(bool aIsBackground);
// returns true, if worker is running in the background tab
bool IsRunningInBackground() const { return mIsInBackground; }
void WaitForIsDebuggerRegistered(bool aDebuggerRegistered) {
AssertIsOnParentThread();
// Yield so that the main thread won't be blocked.
AutoYieldJSThreadExecution yield;
MOZ_ASSERT(!NS_IsMainThread());
MutexAutoLock lock(mMutex);
while (mDebuggerRegistered != aDebuggerRegistered) {
mCondVar.Wait();
}
}
nsresult SetIsDebuggerReady(bool aReady);
WorkerDebugger* Debugger() const {
AssertIsOnMainThread();
MOZ_ASSERT(mDebugger);
return mDebugger;
}
const OriginTrials& Trials() const { return mLoadInfo.mTrials; }
void SetDebugger(WorkerDebugger* aDebugger) {
AssertIsOnMainThread();
MOZ_ASSERT(mDebugger != aDebugger);
mDebugger = aDebugger;
}
JS::UniqueChars AdoptDefaultLocale() {
MOZ_ASSERT(mDefaultLocale,
"the default locale must have been successfully set for anyone "
"to be trying to adopt it");
return std::move(mDefaultLocale);
}
/**
* Invoked by WorkerThreadPrimaryRunnable::Run if it already called
* SetWorkerPrivateInWorkerThread but has to bail out on initialization before
* calling DoRunLoop because PBackground failed to initialize or something
* like that. Note that there's currently no point earlier than this that
* failure can be reported.
*
* When this happens, the worker will need to be deleted, plus the call to
* SetWorkerPrivateInWorkerThread will have scheduled all the
* mPreStartRunnables which need to be cleaned up after, as well as any
* scheduled control runnables. We're somewhat punting on debugger runnables
* for now, which may leak, but the intent is to moot this whole scenario via
* shutdown blockers, so we don't want the extra complexity right now.
*/
void RunLoopNeverRan();
MOZ_CAN_RUN_SCRIPT
void DoRunLoop(JSContext* aCx);
void UnrootGlobalScopes();
MOZ_CAN_RUN_SCRIPT bool InterruptCallback(JSContext* aCx);
bool IsOnCurrentThread();
void CloseInternal();
bool FreezeInternal();
bool ThawInternal();
void PropagateStorageAccessPermissionGrantedInternal();
void TraverseTimeouts(nsCycleCollectionTraversalCallback& aCallback);
void UnlinkTimeouts();
bool AddChildWorker(WorkerPrivate& aChildWorker);
void RemoveChildWorker(WorkerPrivate& aChildWorker);
void PostMessageToParent(JSContext* aCx, JS::Handle<JS::Value> aMessage,
const Sequence<JSObject*>& aTransferable,
ErrorResult& aRv);
void PostMessageToParentMessagePort(JSContext* aCx,
JS::Handle<JS::Value> aMessage,
const Sequence<JSObject*>& aTransferable,
ErrorResult& aRv);
MOZ_CAN_RUN_SCRIPT void EnterDebuggerEventLoop();
void LeaveDebuggerEventLoop();
void PostMessageToDebugger(const nsAString& aMessage);
void SetDebuggerImmediate(Function& aHandler, ErrorResult& aRv);
void ReportErrorToDebugger(const nsACString& aFilename, uint32_t aLineno,
const nsAString& aMessage);
bool NotifyInternal(WorkerStatus aStatus);
void ReportError(JSContext* aCx, JS::ConstUTF8CharsZ aToStringResult,
JSErrorReport* aReport);
static void ReportErrorToConsole(
uint32_t aErrorFlags, const nsCString& aCategory,
nsContentUtils::PropertiesFile aFile, const nsCString& aMessageName,
const nsTArray<nsString>& aParams = nsTArray<nsString>(),
const mozilla::SourceLocation& aLocation =
mozilla::JSCallingLocation::Get());
int32_t SetTimeout(JSContext* aCx, TimeoutHandler* aHandler, int32_t aTimeout,
bool aIsInterval, Timeout::Reason aReason,
ErrorResult& aRv);
void ClearTimeout(int32_t aId, Timeout::Reason aReason);
MOZ_CAN_RUN_SCRIPT bool RunExpiredTimeouts(JSContext* aCx);
bool RescheduleTimeoutTimer(JSContext* aCx);
void UpdateContextOptionsInternal(JSContext* aCx,
const JS::ContextOptions& aContextOptions);
void UpdateLanguagesInternal(const nsTArray<nsString>& aLanguages);
void UpdateJSWorkerMemoryParameterInternal(JSContext* aCx, JSGCParamKey key,
Maybe<uint32_t> aValue);
enum WorkerRanOrNot { WorkerNeverRan = 0, WorkerRan };
void ScheduleDeletion(WorkerRanOrNot aRanOrNot);
bool CollectRuntimeStats(JS::RuntimeStats* aRtStats, bool aAnonymize);
#ifdef JS_GC_ZEAL
void UpdateGCZealInternal(JSContext* aCx, uint8_t aGCZeal,
uint32_t aFrequency);
#endif
void SetLowMemoryStateInternal(JSContext* aCx, bool aState);
void GarbageCollectInternal(JSContext* aCx, bool aShrinking,
bool aCollectChildren);
void CycleCollectInternal(bool aCollectChildren);
void OfflineStatusChangeEventInternal(bool aIsOffline);
void MemoryPressureInternal();
typedef MozPromise<uint64_t, nsresult, true> JSMemoryUsagePromise;
RefPtr<JSMemoryUsagePromise> GetJSMemoryUsage();
void SetFetchHandlerWasAdded() {
MOZ_ASSERT(IsServiceWorker());
AssertIsOnWorkerThread();
mFetchHandlerWasAdded = true;
}
bool FetchHandlerWasAdded() const {
MOZ_ASSERT(IsServiceWorker());
AssertIsOnWorkerThread();
return mFetchHandlerWasAdded;
}
JSContext* GetJSContext() const MOZ_NO_THREAD_SAFETY_ANALYSIS {
// mJSContext is only modified on the worker thread, so workerthread code
// can safely read it without a lock
AssertIsOnWorkerThread();
return mJSContext;
}
WorkerGlobalScope* GlobalScope() const {
auto data = mWorkerThreadAccessible.Access();
return data->mScope;
}
WorkerDebuggerGlobalScope* DebuggerGlobalScope() const {
auto data = mWorkerThreadAccessible.Access();
return data->mDebuggerScope;
}
// Get the global associated with the current nested event loop. Will return
// null if we're not in a nested event loop or that nested event loop does not
// have an associated global.
nsIGlobalObject* GetCurrentEventLoopGlobal() const {
auto data = mWorkerThreadAccessible.Access();
return data->mCurrentEventLoopGlobal;
}
nsICSPEventListener* CSPEventListener() const;
void SetThread(WorkerThread* aThread);
void SetWorkerPrivateInWorkerThread(WorkerThread* aThread);
void ResetWorkerPrivateInWorkerThread();
bool IsOnWorkerThread() const;
void AssertIsOnWorkerThread() const
#ifdef DEBUG
;
#else
{
}
#endif
// This may block!
void BeginCTypesCall();
// This may block!
void EndCTypesCall();
void BeginCTypesCallback();
void EndCTypesCallback();
bool ConnectMessagePort(JSContext* aCx, UniqueMessagePortId& aIdentifier);
WorkerGlobalScope* GetOrCreateGlobalScope(JSContext* aCx);
WorkerDebuggerGlobalScope* CreateDebuggerGlobalScope(JSContext* aCx);
bool RegisterBindings(JSContext* aCx, JS::Handle<JSObject*> aGlobal);
bool RegisterDebuggerBindings(JSContext* aCx, JS::Handle<JSObject*> aGlobal);
bool OnLine() const {
auto data = mWorkerThreadAccessible.Access();
return data->mOnLine;
}
void StopSyncLoop(nsIEventTarget* aSyncLoopTarget, nsresult aResult);
bool MaybeStopSyncLoop(nsIEventTarget* aSyncLoopTarget, nsresult aResult);
void ShutdownModuleLoader();
void ClearPreStartRunnables();
MOZ_CAN_RUN_SCRIPT void ProcessSingleDebuggerRunnable();
void ClearDebuggerEventQueue();
void OnProcessNextEvent();
void AfterProcessNextEvent();
void AssertValidSyncLoop(nsIEventTarget* aSyncLoopTarget)
#ifdef DEBUG
;
#else
{
}
#endif
void AssertIsNotPotentiallyLastGCCCRunning() {
#ifdef MOZ_DIAGNOSTIC_ASSERT_ENABLED
auto data = mWorkerThreadAccessible.Access();
MOZ_DIAGNOSTIC_ASSERT(!data->mIsPotentiallyLastGCCCRunning);
#endif
}
void SetWorkerScriptExecutedSuccessfully() {
AssertIsOnWorkerThread();
// Should only be called once!
MOZ_ASSERT(!mWorkerScriptExecutedSuccessfully);
mWorkerScriptExecutedSuccessfully = true;
}
// Only valid after CompileScriptRunnable has finished running!
bool WorkerScriptExecutedSuccessfully() const {
AssertIsOnWorkerThread();
return mWorkerScriptExecutedSuccessfully;
}
// Get the event target to use when dispatching to the main thread
// from this Worker thread. This may be the main thread itself or
// a ThrottledEventQueue to the main thread.
nsISerialEventTarget* MainThreadEventTargetForMessaging();
nsresult DispatchToMainThreadForMessaging(
nsIRunnable* aRunnable, uint32_t aFlags = NS_DISPATCH_NORMAL);
nsresult DispatchToMainThreadForMessaging(
already_AddRefed<nsIRunnable> aRunnable,
uint32_t aFlags = NS_DISPATCH_NORMAL);
nsISerialEventTarget* MainThreadEventTarget();
nsresult DispatchToMainThread(nsIRunnable* aRunnable,
uint32_t aFlags = NS_DISPATCH_NORMAL);
nsresult DispatchToMainThread(already_AddRefed<nsIRunnable> aRunnable,
uint32_t aFlags = NS_DISPATCH_NORMAL);
nsresult DispatchDebuggeeToMainThread(
already_AddRefed<WorkerRunnable> aRunnable,
uint32_t aFlags = NS_DISPATCH_NORMAL);
// Get an event target that will dispatch runnables as control runnables on
// the worker thread. Implement nsICancelableRunnable if you wish to take
// action on cancelation.
nsISerialEventTarget* ControlEventTarget();
// Get an event target that will attempt to dispatch a normal WorkerRunnable,
// but if that fails will then fall back to a control runnable.
nsISerialEventTarget* HybridEventTarget();
void DumpCrashInformation(nsACString& aString);
ClientType GetClientType() const;
bool EnsureCSPEventListener();
void EnsurePerformanceStorage();
bool GetExecutionGranted() const;
void SetExecutionGranted(bool aGranted);
void ScheduleTimeSliceExpiration(uint32_t aDelay);
void CancelTimeSliceExpiration();
JSExecutionManager* GetExecutionManager() const;
void SetExecutionManager(JSExecutionManager* aManager);
void ExecutionReady();
PerformanceStorage* GetPerformanceStorage();
bool IsAcceptingEvents() MOZ_EXCLUDES(mMutex) {
AssertIsOnParentThread();
MutexAutoLock lock(mMutex);
return mParentStatus < Canceling;
}
// This method helps know if the Worker is already at Dead status.
// Note that it is racy. The status may change after the function returns.
bool IsDead() MOZ_EXCLUDES(mMutex) {
MutexAutoLock lock(mMutex);
return mStatus == Dead;
}
WorkerStatus ParentStatusProtected() {
AssertIsOnParentThread();
MutexAutoLock lock(mMutex);
return mParentStatus;
}
WorkerStatus ParentStatus() const MOZ_REQUIRES(mMutex) {
mMutex.AssertCurrentThreadOwns();
return mParentStatus;
}
Worker* ParentEventTargetRef() const {
MOZ_DIAGNOSTIC_ASSERT(mParentEventTargetRef);
return mParentEventTargetRef;
}
void SetParentEventTargetRef(Worker* aParentEventTargetRef) {
MOZ_DIAGNOSTIC_ASSERT(aParentEventTargetRef);
MOZ_DIAGNOSTIC_ASSERT(!mParentEventTargetRef);
mParentEventTargetRef = aParentEventTargetRef;
}
// Check whether this worker is a secure context. For use from the parent
// thread only; the canonical "is secure context" boolean is stored on the
// compartment of the worker global. The only reason we don't
// AssertIsOnParentThread() here is so we can assert that this value matches
// the one on the compartment, which has to be done from the worker thread.
bool IsSecureContext() const { return mIsSecureContext; }
// Check whether we're running in automation.
bool IsInAutomation() const { return mIsInAutomation; }
bool IsPrivilegedAddonGlobal() const { return mIsPrivilegedAddonGlobal; }
TimeStamp CreationTimeStamp() const { return mCreationTimeStamp; }
DOMHighResTimeStamp CreationTime() const { return mCreationTimeHighRes; }
DOMHighResTimeStamp TimeStampToDOMHighRes(const TimeStamp& aTimeStamp) const {
MOZ_ASSERT(!aTimeStamp.IsNull());
TimeDuration duration = aTimeStamp - mCreationTimeStamp;
return duration.ToMilliseconds();
}
LocationInfo& GetLocationInfo() { return mLocationInfo; }
void CopyJSSettings(workerinternals::JSSettings& aSettings) {
mozilla::MutexAutoLock lock(mMutex);
aSettings = mJSSettings;
}
void CopyJSRealmOptions(JS::RealmOptions& aOptions) {
mozilla::MutexAutoLock lock(mMutex);
aOptions = IsChromeWorker() ? mJSSettings.chromeRealmOptions
: mJSSettings.contentRealmOptions;
}
// The ability to be a chrome worker is orthogonal to the type of
// worker [Dedicated|Shared|Service].
bool IsChromeWorker() const { return mIsChromeWorker; }
// TODO: Invariants require that the parent worker out-live any child
// worker, so WorkerPrivate* should be safe in the moment of calling.
// We would like to have stronger type-system annotated/enforced handling.
WorkerPrivate* GetParent() const { return mParent; }
// Returns the top level worker. It can be the current worker if it's the top
// level one.
WorkerPrivate* GetTopLevelWorker() const {
WorkerPrivate const* wp = this;
while (wp->GetParent()) {
wp = wp->GetParent();
}
return const_cast<WorkerPrivate*>(wp);
}
bool IsFrozen() const {
AssertIsOnParentThread();
return mParentFrozen;
}
bool IsParentWindowPaused() const {
AssertIsOnParentThread();
return mParentWindowPaused;
}
// When we debug a worker, we want to disconnect the window and the worker
// communication. This happens calling this method.
// Note: this method doesn't suspend the worker! Use Freeze/Thaw instead.
void ParentWindowPaused();
void ParentWindowResumed();
const nsString& ScriptURL() const { return mScriptURL; }
const nsString& WorkerName() const { return mWorkerName; }
RequestCredentials WorkerCredentials() const { return mCredentialsMode; }
enum WorkerType WorkerType() const { return mWorkerType; }
WorkerKind Kind() const { return mWorkerKind; }
bool IsDedicatedWorker() const { return mWorkerKind == WorkerKindDedicated; }
bool IsSharedWorker() const { return mWorkerKind == WorkerKindShared; }
bool IsServiceWorker() const { return mWorkerKind == WorkerKindService; }
nsContentPolicyType ContentPolicyType() const {
return ContentPolicyType(mWorkerKind);
}
static nsContentPolicyType ContentPolicyType(WorkerKind aWorkerKind) {
switch (aWorkerKind) {
case WorkerKindDedicated:
return nsIContentPolicy::TYPE_INTERNAL_WORKER;
case WorkerKindShared:
return nsIContentPolicy::TYPE_INTERNAL_SHARED_WORKER;
case WorkerKindService:
return nsIContentPolicy::TYPE_INTERNAL_SERVICE_WORKER;
default:
MOZ_ASSERT_UNREACHABLE("Invalid worker type");
return nsIContentPolicy::TYPE_INVALID;
}
}
nsIScriptContext* GetScriptContext() const {
AssertIsOnMainThread();
return mLoadInfo.mScriptContext;
}
const nsCString& Domain() const { return mLoadInfo.mDomain; }
bool IsFromWindow() const { return mLoadInfo.mFromWindow; }
nsLoadFlags GetLoadFlags() const { return mLoadInfo.mLoadFlags; }
uint64_t WindowID() const { return mLoadInfo.mWindowID; }
uint64_t AssociatedBrowsingContextID() const {
return mLoadInfo.mAssociatedBrowsingContextID;
}
uint64_t ServiceWorkerID() const { return GetServiceWorkerDescriptor().Id(); }
const nsCString& ServiceWorkerScope() const {
return GetServiceWorkerDescriptor().Scope();
}
// This value should never change after the script load completes. Before
// then, it may only be called on the main thread.
nsIURI* GetBaseURI() const { return mLoadInfo.mBaseURI; }
void SetBaseURI(nsIURI* aBaseURI);
nsIURI* GetResolvedScriptURI() const { return mLoadInfo.mResolvedScriptURI; }
const nsString& ServiceWorkerCacheName() const {
MOZ_DIAGNOSTIC_ASSERT(IsServiceWorker());
AssertIsOnMainThread();
return mLoadInfo.mServiceWorkerCacheName;
}
const ServiceWorkerDescriptor& GetServiceWorkerDescriptor() const {
MOZ_DIAGNOSTIC_ASSERT(IsServiceWorker());
MOZ_DIAGNOSTIC_ASSERT(mLoadInfo.mServiceWorkerDescriptor.isSome());
return mLoadInfo.mServiceWorkerDescriptor.ref();
}
const ServiceWorkerRegistrationDescriptor&
GetServiceWorkerRegistrationDescriptor() const {
MOZ_DIAGNOSTIC_ASSERT(IsServiceWorker());
MOZ_DIAGNOSTIC_ASSERT(
mLoadInfo.mServiceWorkerRegistrationDescriptor.isSome());
return mLoadInfo.mServiceWorkerRegistrationDescriptor.ref();
}
const ClientInfo& GetSourceInfo() const {
MOZ_DIAGNOSTIC_ASSERT(IsServiceWorker());
MOZ_DIAGNOSTIC_ASSERT(mLoadInfo.mSourceInfo.isSome());
return mLoadInfo.mSourceInfo.ref();
}
void UpdateServiceWorkerState(ServiceWorkerState aState) {
MOZ_DIAGNOSTIC_ASSERT(IsServiceWorker());
MOZ_DIAGNOSTIC_ASSERT(mLoadInfo.mServiceWorkerDescriptor.isSome());
return mLoadInfo.mServiceWorkerDescriptor.ref().SetState(aState);
}
const Maybe<ServiceWorkerDescriptor>& GetParentController() const {
return mLoadInfo.mParentController;
}
const ChannelInfo& GetChannelInfo() const { return mLoadInfo.mChannelInfo; }
void SetChannelInfo(const ChannelInfo& aChannelInfo) {
AssertIsOnMainThread();
MOZ_ASSERT(!mLoadInfo.mChannelInfo.IsInitialized());
MOZ_ASSERT(aChannelInfo.IsInitialized());
mLoadInfo.mChannelInfo = aChannelInfo;
}
void InitChannelInfo(nsIChannel* aChannel) {
mLoadInfo.mChannelInfo.InitFromChannel(aChannel);
}
void InitChannelInfo(const ChannelInfo& aChannelInfo) {
mLoadInfo.mChannelInfo = aChannelInfo;
}
nsIPrincipal* GetPrincipal() const { return mLoadInfo.mPrincipal; }
nsIPrincipal* GetLoadingPrincipal() const {
return mLoadInfo.mLoadingPrincipal;
}
nsIPrincipal* GetPartitionedPrincipal() const {
return mLoadInfo.mPartitionedPrincipal;
}
nsIPrincipal* GetEffectiveStoragePrincipal() const;
nsILoadGroup* GetLoadGroup() const {
AssertIsOnMainThread();
return mLoadInfo.mLoadGroup;
}
bool UsesSystemPrincipal() const {
return GetPrincipal()->IsSystemPrincipal();
}
bool UsesAddonOrExpandedAddonPrincipal() const {
return GetPrincipal()->GetIsAddonOrExpandedAddonPrincipal();
}
const mozilla::ipc::PrincipalInfo& GetPrincipalInfo() const {
return *mLoadInfo.mPrincipalInfo;
}
const mozilla::ipc::PrincipalInfo& GetPartitionedPrincipalInfo() const {
return *mLoadInfo.mPartitionedPrincipalInfo;
}
const mozilla::ipc::PrincipalInfo& GetEffectiveStoragePrincipalInfo() const;
already_AddRefed<nsIChannel> ForgetWorkerChannel() {
AssertIsOnMainThread();
return mLoadInfo.mChannel.forget();
}
nsPIDOMWindowInner* GetWindow() const {
AssertIsOnMainThread();
return mLoadInfo.mWindow;
}
nsPIDOMWindowInner* GetAncestorWindow() const;
void EvictFromBFCache();
nsIContentSecurityPolicy* GetCsp() const {
AssertIsOnMainThread();
return mLoadInfo.mCSP;
}
void SetCsp(nsIContentSecurityPolicy* aCSP);
nsresult SetCSPFromHeaderValues(const nsACString& aCSPHeaderValue,
const nsACString& aCSPReportOnlyHeaderValue);
void StoreCSPOnClient();
const mozilla::ipc::CSPInfo& GetCSPInfo() const {
return *mLoadInfo.mCSPInfo;
}
void UpdateReferrerInfoFromHeader(
const nsACString& aReferrerPolicyHeaderValue);
nsIReferrerInfo* GetReferrerInfo() const { return mLoadInfo.mReferrerInfo; }
ReferrerPolicy GetReferrerPolicy() const {
return mLoadInfo.mReferrerInfo->ReferrerPolicy();
}
void SetReferrerInfo(nsIReferrerInfo* aReferrerInfo) {
mLoadInfo.mReferrerInfo = aReferrerInfo;
}
bool IsEvalAllowed() const { return mLoadInfo.mEvalAllowed; }
void SetEvalAllowed(bool aAllowed) { mLoadInfo.mEvalAllowed = aAllowed; }
bool GetReportEvalCSPViolations() const {
return mLoadInfo.mReportEvalCSPViolations;
}
void SetReportEvalCSPViolations(bool aReport) {
mLoadInfo.mReportEvalCSPViolations = aReport;
}
bool IsWasmEvalAllowed() const { return mLoadInfo.mWasmEvalAllowed; }
void SetWasmEvalAllowed(bool aAllowed) {
mLoadInfo.mWasmEvalAllowed = aAllowed;
}
bool GetReportWasmEvalCSPViolations() const {
return mLoadInfo.mReportWasmEvalCSPViolations;
}
void SetReportWasmEvalCSPViolations(bool aReport) {
mLoadInfo.mReportWasmEvalCSPViolations = aReport;
}
bool XHRParamsAllowed() const { return mLoadInfo.mXHRParamsAllowed; }
void SetXHRParamsAllowed(bool aAllowed) {
mLoadInfo.mXHRParamsAllowed = aAllowed;
}
mozilla::StorageAccess StorageAccess() const {
AssertIsOnWorkerThread();
if (mLoadInfo.mUsingStorageAccess) {
return mozilla::StorageAccess::eAllow;
}
return mLoadInfo.mStorageAccess;
}
bool UseRegularPrincipal() const {
AssertIsOnWorkerThread();
return mLoadInfo.mUseRegularPrincipal;
}
bool UsingStorageAccess() const {
AssertIsOnWorkerThread();
return mLoadInfo.mUsingStorageAccess;
}
nsICookieJarSettings* CookieJarSettings() const {
// Any thread.
MOZ_ASSERT(mLoadInfo.mCookieJarSettings);
return mLoadInfo.mCookieJarSettings;
}
const net::CookieJarSettingsArgs& CookieJarSettingsArgs() const {
MOZ_ASSERT(mLoadInfo.mCookieJarSettings);
return mLoadInfo.mCookieJarSettingsArgs;
}
const OriginAttributes& GetOriginAttributes() const {
return mLoadInfo.mOriginAttributes;
}
// Determine if the SW testing per-window flag is set by devtools
bool ServiceWorkersTestingInWindow() const {
return mLoadInfo.mServiceWorkersTestingInWindow;
}
// Determine if the worker was created under a third-party context.
bool IsThirdPartyContext() const { return mLoadInfo.mIsThirdPartyContext; }
bool IsWatchedByDevTools() const { return mLoadInfo.mWatchedByDevTools; }
bool ShouldResistFingerprinting(RFPTarget aTarget) const;
const Maybe<RFPTarget>& GetOverriddenFingerprintingSettings() const {
return mLoadInfo.mOverriddenFingerprintingSettings;
}
RemoteWorkerChild* GetRemoteWorkerController();
void SetRemoteWorkerController(RemoteWorkerChild* aController);
RefPtr<GenericPromise> SetServiceWorkerSkipWaitingFlag();
// We can assume that an nsPIDOMWindow will be available for Freeze, Thaw
// as these are only used for globals going in and out of the bfcache.
bool Freeze(const nsPIDOMWindowInner* aWindow);
bool Thaw(const nsPIDOMWindowInner* aWindow);
void PropagateStorageAccessPermissionGranted();
void NotifyStorageKeyUsed();
void EnableDebugger();
void DisableDebugger();
already_AddRefed<WorkerRunnable> MaybeWrapAsWorkerRunnable(
already_AddRefed<nsIRunnable> aRunnable);
bool ProxyReleaseMainThreadObjects();
void SetLowMemoryState(bool aState);
void GarbageCollect(bool aShrinking);
void CycleCollect();
nsresult SetPrincipalsAndCSPOnMainThread(nsIPrincipal* aPrincipal,
nsIPrincipal* aPartitionedPrincipal,
nsILoadGroup* aLoadGroup,
nsIContentSecurityPolicy* aCsp);
nsresult SetPrincipalsAndCSPFromChannel(nsIChannel* aChannel);
bool FinalChannelPrincipalIsValid(nsIChannel* aChannel);
#ifdef MOZ_DIAGNOSTIC_ASSERT_ENABLED
bool PrincipalURIMatchesScriptURL();
#endif
void UpdateOverridenLoadGroup(nsILoadGroup* aBaseLoadGroup);
void WorkerScriptLoaded();
Document* GetDocument() const;
void MemoryPressure();
void UpdateContextOptions(const JS::ContextOptions& aContextOptions);
void UpdateLanguages(const nsTArray<nsString>& aLanguages);
void UpdateJSWorkerMemoryParameter(JSGCParamKey key, Maybe<uint32_t> value);
#ifdef JS_GC_ZEAL
void UpdateGCZeal(uint8_t aGCZeal, uint32_t aFrequency);
#endif
void OfflineStatusChangeEvent(bool aIsOffline);
nsresult Dispatch(already_AddRefed<WorkerRunnable> aRunnable,
nsIEventTarget* aSyncLoopTarget = nullptr);
nsresult DispatchControlRunnable(
already_AddRefed<WorkerRunnable> aWorkerRunnable);
nsresult DispatchDebuggerRunnable(
already_AddRefed<WorkerRunnable> aDebuggerRunnable);
nsresult DispatchToParent(already_AddRefed<WorkerRunnable> aRunnable);
bool IsOnParentThread() const;
void DebuggerInterruptRequest();
#ifdef DEBUG
void AssertIsOnParentThread() const;
void AssertInnerWindowIsCorrect() const;
#else
void AssertIsOnParentThread() const {}
void AssertInnerWindowIsCorrect() const {}
#endif
#ifdef MOZ_DIAGNOSTIC_ASSERT_ENABLED
bool PrincipalIsValid() const;
#endif
void StartCancelingTimer();
const nsAString& Id();
const nsID& AgentClusterId() const { return mAgentClusterId; }
bool IsSharedMemoryAllowed() const;
// https://whatpr.org/html/4734/structured-data.html#cross-origin-isolated
bool CrossOriginIsolated() const;
void SetUseCounter(UseCounterWorker aUseCounter) {
MOZ_ASSERT(!mReportedUseCounters);
MOZ_ASSERT(aUseCounter > UseCounterWorker::Unknown);
AssertIsOnWorkerThread();
mUseCounters[static_cast<size_t>(aUseCounter)] = true;
}
/**
* COEP Methods
*
* If browser.tabs.remote.useCrossOriginEmbedderPolicy=false, these methods
* will, depending on the return type, return a value that will avoid
* assertion failures or a value that won't block loads.
*/
nsILoadInfo::CrossOriginEmbedderPolicy GetEmbedderPolicy() const;
// Fails if a policy has already been set or if `aPolicy` violates the owner's
// policy, if an owner exists.
mozilla::Result<Ok, nsresult> SetEmbedderPolicy(
nsILoadInfo::CrossOriginEmbedderPolicy aPolicy);
// `aRequest` is the request loading the worker and must be QI-able to
// `nsIChannel*`. It's used to verify that the worker can indeed inherit its
// owner's COEP (when an owner exists).
//
// TODO: remove `aRequest`; currently, it's required because instances may not
// always know its final, resolved script URL or have access internally to
// `aRequest`.
void InheritOwnerEmbedderPolicyOrNull(nsIRequest* aRequest);
// Requires a policy to already have been set.
bool MatchEmbedderPolicy(
nsILoadInfo::CrossOriginEmbedderPolicy aPolicy) const;
nsILoadInfo::CrossOriginEmbedderPolicy GetOwnerEmbedderPolicy() const;
void SetCCCollectedAnything(bool collectedAnything);
bool isLastCCCollectedAnything();
uint32_t GetCurrentTimerNestingLevel() const {
auto data = mWorkerThreadAccessible.Access();
return data->mCurrentTimerNestingLevel;
}
void IncreaseTopLevelWorkerFinishedRunnableCount() {
++mTopLevelWorkerFinishedRunnableCount;
}
void DecreaseTopLevelWorkerFinishedRunnableCount() {
--mTopLevelWorkerFinishedRunnableCount;
}
void IncreaseWorkerFinishedRunnableCount() { ++mWorkerFinishedRunnableCount; }
void DecreaseWorkerFinishedRunnableCount() { --mWorkerFinishedRunnableCount; }
void RunShutdownTasks();
bool CancelBeforeWorkerScopeConstructed() const {
auto data = mWorkerThreadAccessible.Access();
return data->mCancelBeforeWorkerScopeConstructed;
}
enum class CCFlag : uint8_t {
EligibleForWorkerRef,
IneligibleForWorkerRef,
EligibleForChildWorker,
IneligibleForChildWorker,
EligibleForTimeout,
IneligibleForTimeout,
CheckBackgroundActors,
};
// When create/release a StrongWorkerRef, child worker, and timeout, this
// method is used to setup if mParentEventTargetRef can get into
// cycle-collection.
// When this method is called, it will also checks if any background actor
// should block the mParentEventTargetRef cycle-collection when there is no
// StrongWorkerRef/ChildWorker/Timeout.
// Worker thread only.
void UpdateCCFlag(const CCFlag);
// This is used in WorkerPrivate::Traverse() to checking if
// mParentEventTargetRef should get into cycle-collection.
// Parent thread only method.
bool IsEligibleForCC();
// A method which adjusts the count of background actors which should not
// block WorkerPrivate::mParentEventTargetRef cycle-collection.
// Worker thread only.
void AdjustNonblockingCCBackgroundActorCount(int32_t aCount);
RefPtr<WorkerParentRef> GetWorkerParentRef() const;
bool MayContinueRunning() {
AssertIsOnWorkerThread();
WorkerStatus status;
{
MutexAutoLock lock(mMutex);
status = mStatus;
}
if (status < Canceling) {
return true;
}
return false;
}
private:
WorkerPrivate(
WorkerPrivate* aParent, const nsAString& aScriptURL, bool aIsChromeWorker,
WorkerKind aWorkerKind, RequestCredentials aRequestCredentials,
enum WorkerType aWorkerType, const nsAString& aWorkerName,
const nsACString& aServiceWorkerScope, WorkerLoadInfo& aLoadInfo,
nsString&& aId, const nsID& aAgentClusterId,
const nsILoadInfo::CrossOriginOpenerPolicy aAgentClusterOpenerPolicy,
CancellationCallback&& aCancellationCallback,
TerminationCallback&& aTerminationCallback);
~WorkerPrivate();
struct AgentClusterIdAndCoop {
nsID mId;
nsILoadInfo::CrossOriginOpenerPolicy mCoop;
};
static AgentClusterIdAndCoop ComputeAgentClusterIdAndCoop(
WorkerPrivate* aParent, WorkerKind aWorkerKind, WorkerLoadInfo* aLoadInfo,
bool aIsChromeWorker);
void CancelAllTimeouts();
enum class ProcessAllControlRunnablesResult {
// We did not process anything.
Nothing,
// We did process something, states may have changed, but we can keep
// executing script.
MayContinue,
// We did process something, and should not continue executing script.
Abort
};
ProcessAllControlRunnablesResult ProcessAllControlRunnables() {
MutexAutoLock lock(mMutex);
return ProcessAllControlRunnablesLocked();
}
ProcessAllControlRunnablesResult ProcessAllControlRunnablesLocked()
MOZ_REQUIRES(mMutex);
void EnableMemoryReporter();
void DisableMemoryReporter();
void WaitForWorkerEvents() MOZ_REQUIRES(mMutex);
// If the worker shutdown status is equal or greater then aFailStatus, this
// operation will fail and nullptr will be returned. See WorkerStatus.h for
// more information about the correct value to use.
already_AddRefed<nsISerialEventTarget> CreateNewSyncLoop(
WorkerStatus aFailStatus);
nsresult RunCurrentSyncLoop();
nsresult DestroySyncLoop(uint32_t aLoopIndex);
void InitializeGCTimers();
enum GCTimerMode { PeriodicTimer = 0, IdleTimer, NoTimer };
void SetGCTimerMode(GCTimerMode aMode);
public:
void CancelGCTimers() { SetGCTimerMode(NoTimer); }
private:
void ShutdownGCTimers();
friend class WorkerRef;
bool AddWorkerRef(WorkerRef* aWorkerRefer, WorkerStatus aFailStatus);
void RemoveWorkerRef(WorkerRef* aWorkerRef);
void NotifyWorkerRefs(WorkerStatus aStatus);
bool HasActiveWorkerRefs() {
auto data = mWorkerThreadAccessible.Access();
return !(data->mChildWorkers.IsEmpty() && data->mTimeouts.IsEmpty() &&
data->mWorkerRefs.IsEmpty());
}
friend class WorkerEventTarget;
nsresult RegisterShutdownTask(nsITargetShutdownTask* aTask);
nsresult UnregisterShutdownTask(nsITargetShutdownTask* aTask);
// Internal logic to dispatch a runnable. This is separate from Dispatch()
// to allow runnables to be atomically dispatched in bulk.
nsresult DispatchLockHeld(already_AddRefed<WorkerRunnable> aRunnable,
nsIEventTarget* aSyncLoopTarget,
const MutexAutoLock& aProofOfLock)
MOZ_REQUIRES(mMutex);
// This method dispatches a simple runnable that starts the shutdown procedure
// after a self.close(). This method is called after a ClearMainEventQueue()
// to be sure that the canceling runnable is the only one in the queue. We
// need this async operation to be sure that all the current JS code is
// executed.
void DispatchCancelingRunnable();
bool GetUseCounter(UseCounterWorker aUseCounter) {
MOZ_ASSERT(aUseCounter > UseCounterWorker::Unknown);
AssertIsOnWorkerThread();
return mUseCounters[static_cast<size_t>(aUseCounter)];
}
void ReportUseCounters();
UniquePtr<ClientSource> CreateClientSource();
// This method is called when corresponding script loader processes the COEP
// header for the worker.
// This method should be called only once in the main thread.
// After this method is called the COEP value owner(window/parent worker) is
// cached in mOwnerEmbedderPolicy such that it can be accessed in other
// threads, i.e. WorkerThread.
void EnsureOwnerEmbedderPolicy();
class EventTarget;
friend class EventTarget;
friend class AutoSyncLoopHolder;
struct TimeoutInfo;
class MemoryReporter;
friend class MemoryReporter;
friend class mozilla::dom::WorkerThread;
SharedMutex mMutex;
mozilla::CondVar mCondVar MOZ_GUARDED_BY(mMutex);
// We cannot make this CheckedUnsafePtr<WorkerPrivate> as this would violate
// our static assert
MOZ_NON_OWNING_REF WorkerPrivate* const mParent;
const nsString mScriptURL;
// This is the worker name for shared workers and dedicated workers.
const nsString mWorkerName;
const RequestCredentials mCredentialsMode;
enum WorkerType mWorkerType;
const WorkerKind mWorkerKind;
// The worker is owned by its thread, which is represented here. This is set
// in Constructor() and emptied by WorkerFinishedRunnable, and conditionally
// traversed by the cycle collector if no other things preventing shutdown.
//
// There are 4 ways a worker can be terminated:
// 1. GC/CC - When the worker is in idle state (busycount == 0), it allows to
// traverse the 'hidden' mParentEventTargetRef pointer. This is the exposed
// Worker webidl object. Doing this, CC will be able to detect a cycle and
// Unlink is called. In Unlink, Worker calls Cancel().
// 2. Worker::Cancel() is called - the shutdown procedure starts immediately.
// 3. WorkerScope::Close() is called - Similar to point 2.
// 4. xpcom-shutdown notification - We call Kill().
RefPtr<Worker> mParentEventTargetRef;
RefPtr<WorkerPrivate> mSelfRef;
CancellationCallback mCancellationCallback;
// The termination callback is passed into the constructor on the parent
// thread and invoked by `ClearSelfAndParentEventTargetRef` just before it
// drops its self-ref.
TerminationCallback mTerminationCallback;
// The lifetime of these objects within LoadInfo is managed explicitly;
// they do not need to be cycle collected.
WorkerLoadInfo mLoadInfo;
LocationInfo mLocationInfo;
// Protected by mMutex.
workerinternals::JSSettings mJSSettings MOZ_GUARDED_BY(mMutex);
WorkerDebugger* mDebugger;
workerinternals::Queue<WorkerRunnable*, 4> mControlQueue;
workerinternals::Queue<WorkerRunnable*, 4> mDebuggerQueue
MOZ_GUARDED_BY(mMutex);
// Touched on multiple threads, protected with mMutex. Only modified on the
// worker thread
JSContext* mJSContext MOZ_GUARDED_BY(mMutex);
// mThread is only modified on the Worker thread, before calling DoRunLoop
RefPtr<WorkerThread> mThread MOZ_GUARDED_BY(mMutex);
// mPRThread is only modified on another thread in ScheduleWorker(), and is
// constant for the duration of DoRunLoop. Static mutex analysis doesn't help
// here
PRThread* mPRThread;
// Accessed from main thread
RefPtr<ThrottledEventQueue> mMainThreadEventTargetForMessaging;
RefPtr<ThrottledEventQueue> mMainThreadEventTarget;
// Accessed from worker thread and destructing thread
RefPtr<WorkerEventTarget> mWorkerControlEventTarget;
RefPtr<WorkerEventTarget> mWorkerHybridEventTarget;
// A pauseable queue for WorkerDebuggeeRunnables directed at the main thread.
// See WorkerDebuggeeRunnable for details.
RefPtr<ThrottledEventQueue> mMainThreadDebuggeeEventTarget;
struct SyncLoopInfo {
explicit SyncLoopInfo(EventTarget* aEventTarget);
RefPtr<EventTarget> mEventTarget;
nsresult mResult;
bool mCompleted;
#ifdef DEBUG
bool mHasRun;
#endif
};
// This is only modified on the worker thread, but in DEBUG builds
// AssertValidSyncLoop function iterates it on other threads. Therefore
// modifications are done with mMutex held *only* in DEBUG builds.
nsTArray<UniquePtr<SyncLoopInfo>> mSyncLoopStack;
nsCOMPtr<nsITimer> mCancelingTimer;
// fired on the main thread if the worker script fails to load
nsCOMPtr<nsIRunnable> mLoadFailedRunnable;
RefPtr<PerformanceStorage> mPerformanceStorage;
RefPtr<WorkerCSPEventListener> mCSPEventListener;
// Protected by mMutex.
nsTArray<RefPtr<WorkerThreadRunnable>> mPreStartRunnables
MOZ_GUARDED_BY(mMutex);
// Only touched on the parent thread. Used for both SharedWorker and
// ServiceWorker RemoteWorkers.
RefPtr<RemoteWorkerChild> mRemoteWorkerController;
JS::UniqueChars mDefaultLocale; // nulled during worker JSContext init
TimeStamp mKillTime;
WorkerStatus mParentStatus MOZ_GUARDED_BY(mMutex);
WorkerStatus mStatus MOZ_GUARDED_BY(mMutex);
TimeStamp mCreationTimeStamp;
DOMHighResTimeStamp mCreationTimeHighRes;
// Flags for use counters used directly by this worker.
static_assert(sizeof(UseCounterWorker) <= sizeof(size_t),
"UseCounterWorker is too big");
static_assert(UseCounterWorker::Count >= static_cast<UseCounterWorker>(0),
"Should be non-negative value and safe to cast to unsigned");
std::bitset<static_cast<size_t>(UseCounterWorker::Count)> mUseCounters;
bool mReportedUseCounters;
// This is created while creating the WorkerPrivate, so it's safe to be
// touched on any thread.
const nsID mAgentClusterId;
// Things touched on worker thread only.
struct WorkerThreadAccessible {
explicit WorkerThreadAccessible(WorkerPrivate* aParent);
RefPtr<WorkerGlobalScope> mScope;
RefPtr<WorkerDebuggerGlobalScope> mDebuggerScope;
// We cannot make this CheckedUnsafePtr<WorkerPrivate> as this would violate
// our static assert
nsTArray<WorkerPrivate*> mChildWorkers;
nsTObserverArray<WorkerRef*> mWorkerRefs;
nsTArray<UniquePtr<TimeoutInfo>> mTimeouts;
nsCOMPtr<nsITimer> mTimer;
nsCOMPtr<nsITimerCallback> mTimerRunnable;
nsCOMPtr<nsITimer> mPeriodicGCTimer;
nsCOMPtr<nsITimer> mIdleGCTimer;
RefPtr<MemoryReporter> mMemoryReporter;
// While running a nested event loop, whether a sync loop or a debugger
// event loop we want to keep track of which global is running it, if any,
// so runnables that run off that event loop can get at that information. In
// practice this only matters for various worker debugger runnables running
// against sandboxes, because all other runnables know which globals they
// belong to already. We could also address this by threading the relevant
// global through the chains of runnables involved, but we'd need to thread
// it through some runnables that run on the main thread, and that would
// require some care to make sure things get released on the correct thread,
// which we'd rather avoid. This member is only accessed on the worker
// thread.
nsCOMPtr<nsIGlobalObject> mCurrentEventLoopGlobal;
// Timer that triggers an interrupt on expiration of the current time slice
nsCOMPtr<nsITimer> mTSTimer;
// Execution manager used to regulate execution for this worker.
RefPtr<JSExecutionManager> mExecutionManager;
// Used to relinguish clearance for CTypes Callbacks.
nsTArray<AutoYieldJSThreadExecution> mYieldJSThreadExecution;
uint32_t mNumWorkerRefsPreventingShutdownStart;
uint32_t mDebuggerEventLoopLevel;
// This is the count of background actors that binding with IPCWorkerRefs.
// This count would be used in WorkerPrivate::UpdateCCFlag for checking if
// CC should be blocked by background actors.
uint32_t mNonblockingCCBackgroundActorCount;
uint32_t mErrorHandlerRecursionCount;
int32_t mNextTimeoutId;
// Tracks the current setTimeout/setInterval nesting level.
// When there isn't a TimeoutHandler on the stack, this will be 0.
// Whenever setTimeout/setInterval are called, a new TimeoutInfo will be
// created with a nesting level one more than the current nesting level,
// saturating at the kClampTimeoutNestingLevel.
//
// When RunExpiredTimeouts is run, it sets this value to the
// TimeoutInfo::mNestingLevel for the duration of
// the WorkerScriptTimeoutHandler::Call which will explicitly trigger a
// microtask checkpoint so that any immediately-resolved promises will
// still see the nesting level.
uint32_t mCurrentTimerNestingLevel;
bool mFrozen;
// This flag is set by the debugger interrupt control runnable to indicate
// that we want to process debugger runnables as part of control runnable
// processing, which is something we don't normally want to do. The flag is
// cleared after processing the debugger runnables.
bool mDebuggerInterruptRequested;
bool mTimerRunning;
bool mRunningExpiredTimeouts;
bool mPeriodicGCTimerRunning;
bool mIdleGCTimerRunning;
bool mOnLine;
bool mJSThreadExecutionGranted;
bool mCCCollectedAnything;
FlippedOnce<false> mDeletionScheduled;
FlippedOnce<false> mCancelBeforeWorkerScopeConstructed;
FlippedOnce<false> mPerformedShutdownAfterLastContentTaskExecuted;
#ifdef MOZ_DIAGNOSTIC_ASSERT_ENABLED
bool mIsPotentiallyLastGCCCRunning = false;
#endif
};
ThreadBound<WorkerThreadAccessible> mWorkerThreadAccessible;
class MOZ_RAII AutoPushEventLoopGlobal {
public:
AutoPushEventLoopGlobal(WorkerPrivate* aWorkerPrivate, JSContext* aCx);
~AutoPushEventLoopGlobal();
private:
nsCOMPtr<nsIGlobalObject> mOldEventLoopGlobal;
#ifdef DEBUG
// This is used to checking if we are on the right stack while push the
// mOldEventLoopGlobal back.
nsCOMPtr<nsIGlobalObject> mNewEventLoopGlobal;
#endif
};
friend class AutoPushEventLoopGlobal;
uint32_t mPostSyncLoopOperations;
// List of operations to do at the end of the last sync event loop.
enum {
eDispatchCancelingRunnable = 0x02,
};
bool mParentWindowPaused;
bool mWorkerScriptExecutedSuccessfully;
bool mFetchHandlerWasAdded;
bool mMainThreadObjectsForgotten;
bool mIsChromeWorker;
bool mParentFrozen;
// In order to ensure that the debugger can interrupt a busy worker,
// including when atomics are used, we reuse the worker's existing
// JS Interrupt facility used by control runnables.
// Rather that triggering an interrupt immediately when a debugger runnable
// is dispatched, we instead start a timer that will fire if we haven't
// already processed the runnable, targeting the timer's callback at the
// control event target. This allows existing runnables that were going to
// finish in a timely fashion to finish.
nsCOMPtr<nsITimer> mDebuggerInterruptTimer MOZ_GUARDED_BY(mMutex);
// mIsSecureContext is set once in our constructor; after that it can be read
// from various threads.
//
// It's a bit unfortunate that we have to have an out-of-band boolean for
// this, but we need access to this state from the parent thread, and we can't
// use our global object's secure state there.
const bool mIsSecureContext;
bool mDebuggerRegistered MOZ_GUARDED_BY(mMutex);
mozilla::Atomic<bool> mIsInBackground;
// During registration, this worker may be marked as not being ready to
// execute debuggee runnables or content.
//
// Protected by mMutex.
bool mDebuggerReady;
nsTArray<RefPtr<WorkerRunnable>> mDelayedDebuggeeRunnables;
// Whether this worker should have access to the WebExtension API bindings
// (currently only the Extension Background ServiceWorker declared in the
// extension manifest is allowed to access any WebExtension API bindings).
// This default to false, and it is eventually set to true by
// RemoteWorkerChild::ExecWorkerOnMainThread if the needed conditions
// are met.
bool mExtensionAPIAllowed;
// mIsInAutomation is true when we're running in test automation.
// We expose some extra testing functions in that case.
bool mIsInAutomation;
nsString mId;
// This is used to check if it's allowed to share the memory across the agent
// cluster.
const nsILoadInfo::CrossOriginOpenerPolicy mAgentClusterOpenerPolicy;
// Member variable of this class rather than the worker global scope because
// it's received on the main thread, but the global scope is thread-bound
// to the worker thread, so storing the value in the global scope would
// involve sacrificing the thread-bound-ness or using a WorkerRunnable, and
// there isn't a strong reason to store it on the global scope other than
// better consistency with the COEP spec.
Maybe<nsILoadInfo::CrossOriginEmbedderPolicy> mEmbedderPolicy;
Maybe<nsILoadInfo::CrossOriginEmbedderPolicy> mOwnerEmbedderPolicy;
/* Privileged add-on flag extracted from the AddonPolicy on the nsIPrincipal
* on the main thread when constructing a top-level worker. The flag is
* propagated to nested workers. The flag is only allowed to take effect in
* extension processes and is forbidden in content scripts in content
* processes. The flag may be read on either the parent/owner thread as well
* as on the worker thread itself. When bug 1443925 is fixed allowing
* nsIPrincipal to be used OMT, it may be possible to remove this flag. */
bool mIsPrivilegedAddonGlobal;
Atomic<uint32_t> mTopLevelWorkerFinishedRunnableCount;
Atomic<uint32_t> mWorkerFinishedRunnableCount;
nsTArray<nsCOMPtr<nsITargetShutdownTask>> mShutdownTasks
MOZ_GUARDED_BY(mMutex);
bool mShutdownTasksRun MOZ_GUARDED_BY(mMutex) = false;
bool mCCFlagSaysEligible MOZ_GUARDED_BY(mMutex){true};
// The flag indicates if the worke is idle for events in the main event loop.
bool mWorkerLoopIsIdle MOZ_GUARDED_BY(mMutex){false};
// This flag is used to ensure we only call NotifyStorageKeyUsed once per
// global.
bool hasNotifiedStorageKeyUsed{false};
RefPtr<WorkerParentRef> mParentRef;
};
class AutoSyncLoopHolder {
RefPtr<StrongWorkerRef> mWorkerRef;
nsCOMPtr<nsISerialEventTarget> mTarget;
uint32_t mIndex;
public:
// See CreateNewSyncLoop() for more information about the correct value to use
// for aFailStatus.
AutoSyncLoopHolder(WorkerPrivate* aWorkerPrivate, WorkerStatus aFailStatus,
const char* const aName = "AutoSyncLoopHolder");
~AutoSyncLoopHolder();
nsresult Run();
nsISerialEventTarget* GetSerialEventTarget() const;
};
/**
* WorkerParentRef is a RefPtr<WorkerPrivate> wrapper for cross-thread access.
* WorkerPrivate needs to be accessed in multiple threads; for example,
* in WorkerParentThreadRunnable, the associated WorkerPrivate must be accessed
* in the worker thread when creating/dispatching and in the parent thread when
* executing. Unfortunately, RefPtr can not be used on this WorkerPrivate since
* it is not a thread-safe ref-counted object.
*
* Instead of using a raw pointer and a complicated mechanism to ensure the
* WorkerPrivate's accessibility. WorkerParentRef is used to resolve the
* problem. WorkerParentRef has a RefPtr<WorkerPrivate> mWorkerPrivate
* initialized on the parent thread when WorkerPrivate::Constructor().
* WorkerParentRef is a thread-safe ref-counted object that can be copied at
* any thread by WorkerPrivate::GetWorkerParentRef() and propagated to other
* threads. In the target thread, call WorkerParentRef::Private() to get the
* reference for WorkerPrivate or get a nullptr if the Worker has shut down.
*
* Since currently usage cases, WorkerParentRef::Private() will assert to be on
* the parent thread.
*/
class WorkerParentRef final {
public:
NS_INLINE_DECL_THREADSAFE_REFCOUNTING(WorkerParentRef);
explicit WorkerParentRef(RefPtr<WorkerPrivate>& aWorkerPrivate);
const RefPtr<WorkerPrivate>& Private() const;
void DropWorkerPrivate();
private:
~WorkerParentRef();
RefPtr<WorkerPrivate> mWorkerPrivate;
};
} // namespace dom
} // namespace mozilla
#endif /* mozilla_dom_workers_workerprivate_h__ */