StructuredClone.cpp

mozilla-central/js/src/vm/StructuredClone.cpp (file symbol)

Enable keyboard shortcuts

Source code

File a bug in Core :: JavaScript Engine

Revision control

Copy as Markdown

Other Tools

/* -*- 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/. */

/*

 * This file implements the structured data algorithms of

 * https://html.spec.whatwg.org/multipage/structured-data.html

 * The spec is in two parts:

 * -   StructuredSerialize examines a JS value and produces a graph of Records.

 * -   StructuredDeserialize walks the Records and produces a new JS value.

 * The differences between our implementation and the spec are minor:

 * -   We call the two phases "write" and "read".

 * -   Our algorithms use an explicit work stack, rather than recursion.

 * -   Serialized data is a flat array of bytes, not a (possibly cyclic) graph

 *     of "Records".

 * -   As a consequence, we handle non-treelike object graphs differently.

 *     We serialize objects that appear in multiple places in the input as

 *     backreferences, using sequential integer indexes.

 *     See `JSStructuredCloneReader::allObjs`, our take on the "memory" map

 *     in the spec's StructuredDeserialize.

*/

#include "js/StructuredClone.h"

#include "mozilla/Casting.h"

#include "mozilla/CheckedInt.h"

#include "mozilla/EndianUtils.h"

#include "mozilla/FloatingPoint.h"

#include "mozilla/Maybe.h"

#include "mozilla/ScopeExit.h"

#include <algorithm>

#include <memory>

#include <utility>

#include "jsdate.h"

#include "builtin/DataViewObject.h"

#include "builtin/MapObject.h"

#include "gc/GC.h"           // AutoSelectGCHeap

#include "js/Array.h"        // JS::GetArrayLength, JS::IsArrayObject

#include "js/ArrayBuffer.h"  // JS::{ArrayBufferHasData,DetachArrayBuffer,IsArrayBufferObject,New{,Mapped}ArrayBufferWithContents,ReleaseMappedArrayBufferContents}

#include "js/ColumnNumber.h"  // JS::ColumnNumberOneOrigin, JS::TaggedColumnNumberOneOrigin

#include "js/Date.h"

#include "js/experimental/TypedData.h"  // JS_NewDataView, JS_New{{Ui,I}nt{8,16,32},Float{32,64},Uint8Clamped,Big{Ui,I}nt64}ArrayWithBuffer

#include "js/friend/ErrorMessages.h"    // js::GetErrorMessage, JSMSG_*

#include "js/GCAPI.h"

#include "js/GCHashTable.h"

#include "js/Object.h"              // JS::GetBuiltinClass

#include "js/PropertyAndElement.h"  // JS_GetElement

#include "js/RegExpFlags.h"         // JS::RegExpFlag, JS::RegExpFlags

#include "js/ScalarType.h"          // js::Scalar::Type

#include "js/SharedArrayBuffer.h"   // JS::IsSharedArrayBufferObject

#include "js/Wrapper.h"

#include "util/DifferentialTesting.h"

#include "vm/BigIntType.h"

#include "vm/ErrorObject.h"

#include "vm/JSContext.h"

#include "vm/PlainObject.h"  // js::PlainObject

#include "vm/RegExpObject.h"

#include "vm/SavedFrame.h"

#include "vm/SharedArrayObject.h"

#include "vm/TypedArrayObject.h"

#include "wasm/WasmJS.h"

#include "vm/ArrayObject-inl.h"

#include "vm/Compartment-inl.h"

#include "vm/ErrorObject-inl.h"

#include "vm/InlineCharBuffer-inl.h"

#include "vm/JSContext-inl.h"

#include "vm/JSObject-inl.h"

#include "vm/NativeObject-inl.h"

#include "vm/ObjectOperations-inl.h"

#include "vm/Realm-inl.h"

using namespace js;

using JS::CanonicalizeNaN;

using JS::GetBuiltinClass;

using JS::RegExpFlag;

using JS::RegExpFlags;

using JS::RootedValueVector;

using mozilla::AssertedCast;

using mozilla::BitwiseCast;

using mozilla::Maybe;

using mozilla::NativeEndian;

using mozilla::NumbersAreIdentical;

// When you make updates here, make sure you consider whether you need to bump

// the value of JS_STRUCTURED_CLONE_VERSION in js/public/StructuredClone.h.  You

// will likely need to increment the version if anything at all changes in the

// serialization format.

//

// Note that SCTAG_END_OF_KEYS is written into the serialized form and should

// have a stable ID, it need not be at the end of the list and should not be

// used for sizing data structures.

enum StructuredDataType : uint32_t {

  // Structured data types provided by the engine

  SCTAG_FLOAT_MAX = 0xFFF00000,

  SCTAG_HEADER = 0xFFF10000,

  SCTAG_NULL = 0xFFFF0000,

  SCTAG_UNDEFINED,

  SCTAG_BOOLEAN,

  SCTAG_INT32,

  SCTAG_STRING,

  SCTAG_DATE_OBJECT,

  SCTAG_REGEXP_OBJECT,

  SCTAG_ARRAY_OBJECT,

  SCTAG_OBJECT_OBJECT,

  SCTAG_ARRAY_BUFFER_OBJECT_V2,  // Old version, for backwards compatibility.

  SCTAG_BOOLEAN_OBJECT,

  SCTAG_STRING_OBJECT,

  SCTAG_NUMBER_OBJECT,

  SCTAG_BACK_REFERENCE_OBJECT,

  SCTAG_DO_NOT_USE_1,           // Required for backwards compatibility

  SCTAG_DO_NOT_USE_2,           // Required for backwards compatibility

  SCTAG_TYPED_ARRAY_OBJECT_V2,  // Old version, for backwards compatibility.

  SCTAG_MAP_OBJECT,

  SCTAG_SET_OBJECT,

  SCTAG_END_OF_KEYS,

  SCTAG_DO_NOT_USE_3,         // Required for backwards compatibility

  SCTAG_DATA_VIEW_OBJECT_V2,  // Old version, for backwards compatibility.

  SCTAG_SAVED_FRAME_OBJECT,

  // No new tags before principals.

  SCTAG_JSPRINCIPALS,

  SCTAG_NULL_JSPRINCIPALS,

  SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_SYSTEM,

  SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_NOT_SYSTEM,

  SCTAG_SHARED_ARRAY_BUFFER_OBJECT,

  SCTAG_SHARED_WASM_MEMORY_OBJECT,

  SCTAG_BIGINT,

  SCTAG_BIGINT_OBJECT,

  SCTAG_ARRAY_BUFFER_OBJECT,

  SCTAG_TYPED_ARRAY_OBJECT,

  SCTAG_DATA_VIEW_OBJECT,

  SCTAG_ERROR_OBJECT,

  SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT,

  SCTAG_GROWABLE_SHARED_ARRAY_BUFFER_OBJECT,

  SCTAG_TYPED_ARRAY_V1_MIN = 0xFFFF0100,

  SCTAG_TYPED_ARRAY_V1_INT8 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Int8,

  SCTAG_TYPED_ARRAY_V1_UINT8 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Uint8,

  SCTAG_TYPED_ARRAY_V1_INT16 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Int16,

  SCTAG_TYPED_ARRAY_V1_UINT16 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Uint16,

  SCTAG_TYPED_ARRAY_V1_INT32 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Int32,

  SCTAG_TYPED_ARRAY_V1_UINT32 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Uint32,

  SCTAG_TYPED_ARRAY_V1_FLOAT32 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Float32,

  SCTAG_TYPED_ARRAY_V1_FLOAT64 = SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Float64,

  SCTAG_TYPED_ARRAY_V1_UINT8_CLAMPED =

      SCTAG_TYPED_ARRAY_V1_MIN + Scalar::Uint8Clamped,

  // BigInt64 and BigUint64 are not supported in the v1 format.

  SCTAG_TYPED_ARRAY_V1_MAX = SCTAG_TYPED_ARRAY_V1_UINT8_CLAMPED,

  // Define a separate range of numbers for Transferable-only tags, since

  // they are not used for persistent clone buffers and therefore do not

  // require bumping JS_STRUCTURED_CLONE_VERSION.

  SCTAG_TRANSFER_MAP_HEADER = 0xFFFF0200,

  SCTAG_TRANSFER_MAP_PENDING_ENTRY,

  SCTAG_TRANSFER_MAP_ARRAY_BUFFER,

  SCTAG_TRANSFER_MAP_STORED_ARRAY_BUFFER,

  SCTAG_TRANSFER_MAP_END_OF_BUILTIN_TYPES,

  SCTAG_END_OF_BUILTIN_TYPES

};

/*

 * Format of transfer map:

 *   <SCTAG_TRANSFER_MAP_HEADER, TransferableMapHeader(UNREAD|TRANSFERRED)>

 *   numTransferables (64 bits)

 *   array of:

 *     <SCTAG_TRANSFER_MAP_*, TransferableOwnership>

 *     pointer (64 bits)

 *     extraData (64 bits), eg byte length for ArrayBuffers

*/

// Data associated with an SCTAG_TRANSFER_MAP_HEADER that tells whether the

// contents have been read out yet or not.

enum TransferableMapHeader { SCTAG_TM_UNREAD = 0, SCTAG_TM_TRANSFERRED };

static inline uint64_t PairToUInt64(uint32_t tag, uint32_t data) {

  return uint64_t(data) | (uint64_t(tag) << 32);

namespace js {

template <typename T, typename AllocPolicy>

struct BufferIterator {

  using BufferList = mozilla::BufferList<AllocPolicy>;

  explicit BufferIterator(const BufferList& buffer)

      : mBuffer(buffer), mIter(buffer.Iter()) {

    static_assert(8 % sizeof(T) == 0);

  explicit BufferIterator(const JSStructuredCloneData& data)

      : mBuffer(data.bufList_), mIter(data.Start()) {}

  BufferIterator& operator=(const BufferIterator& other) {

    MOZ_ASSERT(&mBuffer == &other.mBuffer);

    mIter = other.mIter;

    return *this;

  [[nodiscard]] bool advance(size_t size = sizeof(T)) {

    return mIter.AdvanceAcrossSegments(mBuffer, size);

  BufferIterator operator++(int) {

    BufferIterator ret = *this;

    if (!advance(sizeof(T))) {

      MOZ_ASSERT(false, "Failed to read StructuredCloneData. Data incomplete");

    return ret;

  BufferIterator& operator+=(size_t size) {

    if (!advance(size)) {

      MOZ_ASSERT(false, "Failed to read StructuredCloneData. Data incomplete");

    return *this;

  size_t operator-(const BufferIterator& other) const {

    MOZ_ASSERT(&mBuffer == &other.mBuffer);

    return mBuffer.RangeLength(other.mIter, mIter);

  bool operator==(const BufferIterator& other) const {

    return mBuffer.Start() == other.mBuffer.Start() && mIter == other.mIter;

  bool operator!=(const BufferIterator& other) const {

    return !(*this == other);

  bool done() const { return mIter.Done(); }

  [[nodiscard]] bool readBytes(char* outData, size_t size) {

    return mBuffer.ReadBytes(mIter, outData, size);

  void write(const T& data) {

    MOZ_ASSERT(mIter.HasRoomFor(sizeof(T)));

    *reinterpret_cast<T*>(mIter.Data()) = data;

  T peek() const {

    MOZ_ASSERT(mIter.HasRoomFor(sizeof(T)));

    return *reinterpret_cast<T*>(mIter.Data());

  bool canPeek() const { return mIter.HasRoomFor(sizeof(T)); }

  const BufferList& mBuffer;

  typename BufferList::IterImpl mIter;

};

SharedArrayRawBufferRefs& SharedArrayRawBufferRefs::operator=(

    SharedArrayRawBufferRefs&& other) {

  takeOwnership(std::move(other));

  return *this;

SharedArrayRawBufferRefs::~SharedArrayRawBufferRefs() { releaseAll(); }

bool SharedArrayRawBufferRefs::acquire(JSContext* cx,

                                       SharedArrayRawBuffer* rawbuf) {

  if (!refs_.append(rawbuf)) {

    ReportOutOfMemory(cx);

    return false;

  if (!rawbuf->addReference()) {

    refs_.popBack();

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_SAB_REFCNT_OFLO);

    return false;

  return true;

bool SharedArrayRawBufferRefs::acquireAll(

    JSContext* cx, const SharedArrayRawBufferRefs& that) {

  if (!refs_.reserve(refs_.length() + that.refs_.length())) {

    ReportOutOfMemory(cx);

    return false;

  for (auto ref : that.refs_) {

    if (!ref->addReference()) {

      JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                                JSMSG_SC_SAB_REFCNT_OFLO);

      return false;

    MOZ_ALWAYS_TRUE(refs_.append(ref));

  return true;

void SharedArrayRawBufferRefs::takeOwnership(SharedArrayRawBufferRefs&& other) {

  MOZ_ASSERT(refs_.empty());

  refs_ = std::move(other.refs_);

void SharedArrayRawBufferRefs::releaseAll() {

  for (auto ref : refs_) {

    ref->dropReference();

  refs_.clear();

// SCOutput provides an interface to write raw data -- eg uint64_ts, doubles,

// arrays of bytes -- into a structured clone data output stream. It also knows

// how to free any transferable data within that stream.

//

// Note that it contains a full JSStructuredCloneData object, which holds the

// callbacks necessary to read/write/transfer/free the data. For the purpose of

// this class, only the freeTransfer callback is relevant; the rest of the

// callbacks are used by the higher-level JSStructuredCloneWriter interface.

struct SCOutput {

 public:

  using Iter = BufferIterator<uint64_t, SystemAllocPolicy>;

  SCOutput(JSContext* cx, JS::StructuredCloneScope scope);

  JSContext* context() const { return cx; }

  JS::StructuredCloneScope scope() const { return buf.scope(); }

  void sameProcessScopeRequired() { buf.sameProcessScopeRequired(); }

  [[nodiscard]] bool write(uint64_t u);

  [[nodiscard]] bool writePair(uint32_t tag, uint32_t data);

  [[nodiscard]] bool writeDouble(double d);

  [[nodiscard]] bool writeBytes(const void* p, size_t nbytes);

  [[nodiscard]] bool writeChars(const Latin1Char* p, size_t nchars);

  [[nodiscard]] bool writeChars(const char16_t* p, size_t nchars);

  template <class T>

  [[nodiscard]] bool writeArray(const T* p, size_t nelems);

  void setCallbacks(const JSStructuredCloneCallbacks* callbacks, void* closure,

                    OwnTransferablePolicy policy) {

    buf.setCallbacks(callbacks, closure, policy);

  void extractBuffer(JSStructuredCloneData* data) { *data = std::move(buf); }

  uint64_t tell() const { return buf.Size(); }

  uint64_t count() const { return buf.Size() / sizeof(uint64_t); }

  Iter iter() { return Iter(buf); }

  size_t offset(Iter dest) { return dest - iter(); }

  JSContext* cx;

  JSStructuredCloneData buf;

};

class SCInput {

 public:

  using BufferIterator = js::BufferIterator<uint64_t, SystemAllocPolicy>;

  SCInput(JSContext* cx, const JSStructuredCloneData& data);

  JSContext* context() const { return cx; }

  static void getPtr(uint64_t data, void** ptr);

  static void getPair(uint64_t data, uint32_t* tagp, uint32_t* datap);

  [[nodiscard]] bool read(uint64_t* p);

  [[nodiscard]] bool readPair(uint32_t* tagp, uint32_t* datap);

  [[nodiscard]] bool readDouble(double* p);

  [[nodiscard]] bool readBytes(void* p, size_t nbytes);

  [[nodiscard]] bool readChars(Latin1Char* p, size_t nchars);

  [[nodiscard]] bool readChars(char16_t* p, size_t nchars);

  [[nodiscard]] bool readPtr(void**);

  [[nodiscard]] bool get(uint64_t* p);

  [[nodiscard]] bool getPair(uint32_t* tagp, uint32_t* datap);

  const BufferIterator& tell() const { return point; }

  void seekTo(const BufferIterator& pos) { point = pos; }

  [[nodiscard]] bool seekBy(size_t pos) {

    if (!point.advance(pos)) {

      reportTruncated();

      return false;

    return true;

  template <class T>

  [[nodiscard]] bool readArray(T* p, size_t nelems);

  bool reportTruncated() {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA, "truncated");

    return false;

 private:

  void staticAssertions() {

    static_assert(sizeof(char16_t) == 2);

    static_assert(sizeof(uint32_t) == 4);

  JSContext* cx;

  BufferIterator point;

};

}  // namespace js

struct JSStructuredCloneReader {

 public:

  explicit JSStructuredCloneReader(SCInput& in, JS::StructuredCloneScope scope,

                                   const JS::CloneDataPolicy& cloneDataPolicy,

                                   const JSStructuredCloneCallbacks* cb,

                                   void* cbClosure);

  SCInput& input() { return in; }

  bool read(MutableHandleValue vp, size_t nbytes);

 private:

  JSContext* context() { return in.context(); }

  bool readHeader();

  bool readTransferMap();

  [[nodiscard]] bool readUint32(uint32_t* num);

  enum ShouldAtomizeStrings : bool {

    DontAtomizeStrings = false,

    AtomizeStrings = true

};

  template <typename CharT>

  JSString* readStringImpl(uint32_t nchars, ShouldAtomizeStrings atomize);

  JSString* readString(uint32_t data, ShouldAtomizeStrings atomize);

  BigInt* readBigInt(uint32_t data);

  [[nodiscard]] bool readTypedArray(uint32_t arrayType, uint64_t nelems,

                                    MutableHandleValue vp, bool v1Read = false);

  [[nodiscard]] bool readDataView(uint64_t byteLength, MutableHandleValue vp);

  [[nodiscard]] bool readArrayBuffer(StructuredDataType type, uint32_t data,

                                     MutableHandleValue vp);

  [[nodiscard]] bool readV1ArrayBuffer(uint32_t arrayType, uint32_t nelems,

                                       MutableHandleValue vp);

  [[nodiscard]] bool readSharedArrayBuffer(StructuredDataType type,

                                           MutableHandleValue vp);

  [[nodiscard]] bool readSharedWasmMemory(uint32_t nbytes,

                                          MutableHandleValue vp);

  // A serialized SavedFrame contains primitive values in a header followed by

  // an optional parent frame that is read recursively.

  [[nodiscard]] JSObject* readSavedFrameHeader(uint32_t principalsTag);

  [[nodiscard]] bool readSavedFrameFields(Handle<SavedFrame*> frameObj,

                                          HandleValue parent, bool* state);

  // A serialized Error contains primitive values in a header followed by

  // 'cause', 'errors', and 'stack' fields that are read recursively.

  [[nodiscard]] JSObject* readErrorHeader(uint32_t type);

  [[nodiscard]] bool readErrorFields(Handle<ErrorObject*> errorObj,

                                     HandleValue cause, bool* state);

  [[nodiscard]] bool readMapField(Handle<MapObject*> mapObj, HandleValue key);

  [[nodiscard]] bool readObjectField(HandleObject obj, HandleValue key);

  [[nodiscard]] bool startRead(

      MutableHandleValue vp,

      ShouldAtomizeStrings atomizeStrings = DontAtomizeStrings);

  SCInput& in;

  // The widest scope that the caller will accept, where

  // SameProcess is the widest (it can store anything it wants)

  // and DifferentProcess is the narrowest (it cannot contain pointers and must

  // be valid cross-process.)

  JS::StructuredCloneScope allowedScope;

  const JS::CloneDataPolicy cloneDataPolicy;

  // Stack of objects with properties remaining to be read.

  RootedValueVector objs;

  // Maintain a stack of state values for the `objs` stack. Since this is only

  // needed for a very small subset of objects (those with a known set of

  // object children), the state information is stored as a stack of

  // <object, state> pairs where the object determines which element of the

  // `objs` stack that it corresponds to. So when reading from the `objs` stack,

  // the state will be retrieved only if the top object on `objState` matches

  // the top object of `objs`.

//

  // Currently, the only state needed is a boolean indicating whether the fields

  // have been read yet.

  Rooted<GCVector<std::pair<HeapPtr<JSObject*>, bool>, 8>> objState;

  // Array of all objects read during this deserialization, for resolving

  // backreferences.

//

  // For backreferences to work correctly, objects must be added to this

  // array in exactly the order expected by the version of the Writer that

  // created the serialized data, even across years and format versions. This

  // is usually no problem, since both algorithms do a single linear pass

  // over the serialized data. There is one hitch; see readTypedArray.

//

  // The values in this vector are objects, except it can temporarily have

  // one `undefined` placeholder value (the readTypedArray hack).

  RootedValueVector allObjs;

  size_t numItemsRead;

  // The user defined callbacks that will be used for cloning.

  const JSStructuredCloneCallbacks* callbacks;

  // Any value passed to JS_ReadStructuredClone.

  void* closure;

  // The heap to use for allocating common GC things. This starts out as the

  // nursery (the default) but may switch to the tenured heap if nursery

  // collection occurs, as nursery allocation is pointless after the

  // deserialized root object is tenured.

//

  // This is only used for the most common kind, e.g. plain objects, strings

  // and a couple of others.

  AutoSelectGCHeap gcHeap;

  friend bool JS_ReadString(JSStructuredCloneReader* r,

                            JS::MutableHandleString str);

  friend bool JS_ReadTypedArray(JSStructuredCloneReader* r,

                                MutableHandleValue vp);

  // Provide a way to detect whether any of the clone data is never used. When

  // "tail" data (currently, this is only stored data for Transferred

  // ArrayBuffers in the DifferentProcess scope) is read, record the first and

  // last positions. At the end of deserialization, make sure there's nothing

  // between the end of the main data and the beginning of the tail, nor after

  // the end of the tail.

  mozilla::Maybe<SCInput::BufferIterator> tailStartPos;

  mozilla::Maybe<SCInput::BufferIterator> tailEndPos;

};

struct JSStructuredCloneWriter {

 public:

  explicit JSStructuredCloneWriter(JSContext* cx,

                                   JS::StructuredCloneScope scope,

                                   const JS::CloneDataPolicy& cloneDataPolicy,

                                   const JSStructuredCloneCallbacks* cb,

                                   void* cbClosure, const Value& tVal)

      : out(cx, scope),

        callbacks(cb),

        closure(cbClosure),

        objs(cx),

        counts(cx),

        objectEntries(cx),

        otherEntries(cx),

        memory(cx),

        transferable(cx, tVal),

        transferableObjects(cx, TransferableObjectsList(cx)),

        cloneDataPolicy(cloneDataPolicy) {

    out.setCallbacks(cb, cbClosure,

                     OwnTransferablePolicy::OwnsTransferablesIfAny);

  bool init() {

    return parseTransferable() && writeHeader() && writeTransferMap();

  bool write(HandleValue v);

  SCOutput& output() { return out; }

  void extractBuffer(JSStructuredCloneData* newData) {

    out.extractBuffer(newData);

 private:

  JSStructuredCloneWriter() = delete;

  JSStructuredCloneWriter(const JSStructuredCloneWriter&) = delete;

  JSContext* context() { return out.context(); }

  bool writeHeader();

  bool writeTransferMap();

  bool writeString(uint32_t tag, JSString* str);

  bool writeBigInt(uint32_t tag, BigInt* bi);

  bool writeArrayBuffer(HandleObject obj);

  bool writeTypedArray(HandleObject obj);

  bool writeDataView(HandleObject obj);

  bool writeSharedArrayBuffer(HandleObject obj);

  bool writeSharedWasmMemory(HandleObject obj);

  bool startObject(HandleObject obj, bool* backref);

  bool writePrimitive(HandleValue v);

  bool startWrite(HandleValue v);

  bool traverseObject(HandleObject obj, ESClass cls);

  bool traverseMap(HandleObject obj);

  bool traverseSet(HandleObject obj);

  bool traverseSavedFrame(HandleObject obj);

  bool traverseError(HandleObject obj);

  template <typename... Args>

  bool reportDataCloneError(uint32_t errorId, Args&&... aArgs);

  bool parseTransferable();

  bool transferOwnership();

  inline void checkStack();

  SCOutput out;

  // The user defined callbacks that will be used to signal cloning, in some

  // cases.

  const JSStructuredCloneCallbacks* callbacks;

  // Any value passed to the callbacks.

  void* closure;

  // Vector of objects with properties remaining to be written.

//

  // NB: These can span multiple compartments, so the compartment must be

  // entered before any manipulation is performed.

  RootedValueVector objs;

  // counts[i] is the number of entries of objs[i] remaining to be written.

  // counts.length() == objs.length() and sum(counts) == entries.length().

  Vector<size_t> counts;

  // For JSObject: Property IDs as value

  RootedIdVector objectEntries;

  // For Map: Key followed by value

  // For Set: Key

  // For SavedFrame: parent SavedFrame

  // For Error: cause, errors, stack

  RootedValueVector otherEntries;

  // The "memory" list described in the HTML5 internal structured cloning

  // algorithm.  memory is a superset of objs; items are never removed from

  // Memory until a serialization operation is finished

  using CloneMemory = GCHashMap<JSObject*, uint32_t,

                                StableCellHasher<JSObject*>, SystemAllocPolicy>;

  Rooted<CloneMemory> memory;

  // Set of transferable objects

  RootedValue transferable;

  using TransferableObjectsList = GCVector<JSObject*>;

  Rooted<TransferableObjectsList> transferableObjects;

  const JS::CloneDataPolicy cloneDataPolicy;

  friend bool JS_WriteString(JSStructuredCloneWriter* w, HandleString str);

  friend bool JS_WriteTypedArray(JSStructuredCloneWriter* w, HandleValue v);

  friend bool JS_ObjectNotWritten(JSStructuredCloneWriter* w, HandleObject obj);

};

JS_PUBLIC_API uint64_t js::GetSCOffset(JSStructuredCloneWriter* writer) {

  MOZ_ASSERT(writer);

  return writer->output().count() * sizeof(uint64_t);

static_assert(SCTAG_END_OF_BUILTIN_TYPES <= JS_SCTAG_USER_MIN);

static_assert(JS_SCTAG_USER_MIN <= JS_SCTAG_USER_MAX);

static_assert(Scalar::Int8 == 0);

template <typename... Args>

static void ReportDataCloneError(JSContext* cx,

                                 const JSStructuredCloneCallbacks* callbacks,

                                 uint32_t errorId, void* closure,

                                 Args&&... aArgs) {

  unsigned errorNumber;

  switch (errorId) {

    case JS_SCERR_DUP_TRANSFERABLE:

      errorNumber = JSMSG_SC_DUP_TRANSFERABLE;

      break;

    case JS_SCERR_TRANSFERABLE:

      errorNumber = JSMSG_SC_NOT_TRANSFERABLE;

      break;

    case JS_SCERR_UNSUPPORTED_TYPE:

      errorNumber = JSMSG_SC_UNSUPPORTED_TYPE;

      break;

    case JS_SCERR_SHMEM_TRANSFERABLE:

      errorNumber = JSMSG_SC_SHMEM_TRANSFERABLE;

      break;

    case JS_SCERR_TYPED_ARRAY_DETACHED:

      errorNumber = JSMSG_TYPED_ARRAY_DETACHED;

      break;

    case JS_SCERR_WASM_NO_TRANSFER:

      errorNumber = JSMSG_WASM_NO_TRANSFER;

      break;

    case JS_SCERR_NOT_CLONABLE:

      errorNumber = JSMSG_SC_NOT_CLONABLE;

      break;

    case JS_SCERR_NOT_CLONABLE_WITH_COOP_COEP:

      errorNumber = JSMSG_SC_NOT_CLONABLE_WITH_COOP_COEP;

      break;

    default:

      MOZ_CRASH("Unkown errorId");

      break;

  if (callbacks && callbacks->reportError) {

    MOZ_RELEASE_ASSERT(!cx->isExceptionPending());

    JSErrorReport report;

    report.errorNumber = errorNumber;

    // Get js error message if it's possible and propagate it through callback.

    if (JS_ExpandErrorArgumentsASCII(cx, GetErrorMessage, errorNumber, &report,

                                     std::forward<Args>(aArgs)...) &&

        report.message()) {

      callbacks->reportError(cx, errorId, closure, report.message().c_str());

    } else {

      ReportOutOfMemory(cx);

      callbacks->reportError(cx, errorId, closure, "");

    return;

  JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr, errorNumber,

                            std::forward<Args>(aArgs)...);

bool WriteStructuredClone(JSContext* cx, HandleValue v,

                          JSStructuredCloneData* bufp,

                          JS::StructuredCloneScope scope,

                          const JS::CloneDataPolicy& cloneDataPolicy,

                          const JSStructuredCloneCallbacks* cb, void* cbClosure,

                          const Value& transferable) {

  JSStructuredCloneWriter w(cx, scope, cloneDataPolicy, cb, cbClosure,

                            transferable);

  if (!w.init()) {

    return false;

  if (!w.write(v)) {

    return false;

  w.extractBuffer(bufp);

  return true;

bool ReadStructuredClone(JSContext* cx, const JSStructuredCloneData& data,

                         JS::StructuredCloneScope scope, MutableHandleValue vp,

                         const JS::CloneDataPolicy& cloneDataPolicy,

                         const JSStructuredCloneCallbacks* cb,

                         void* cbClosure) {

  if (data.Size() % 8) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA, "misaligned");

    return false;

  SCInput in(cx, data);

  JSStructuredCloneReader r(in, scope, cloneDataPolicy, cb, cbClosure);

  return r.read(vp, data.Size());

static bool StructuredCloneHasTransferObjects(

    const JSStructuredCloneData& data) {

  if (data.Size() < sizeof(uint64_t)) {

    return false;

  uint64_t u;

  BufferIterator<uint64_t, SystemAllocPolicy> iter(data);

  MOZ_ALWAYS_TRUE(iter.readBytes(reinterpret_cast<char*>(&u), sizeof(u)));

  uint32_t tag = uint32_t(u >> 32);

  return (tag == SCTAG_TRANSFER_MAP_HEADER);

namespace js {

SCInput::SCInput(JSContext* cx, const JSStructuredCloneData& data)

    : cx(cx), point(data) {

  static_assert(JSStructuredCloneData::BufferList::kSegmentAlignment % 8 == 0,

                "structured clone buffer reads should be aligned");

  MOZ_ASSERT(data.Size() % 8 == 0);

bool SCInput::read(uint64_t* p) {

  if (!point.canPeek()) {

    *p = 0;  // initialize to shut GCC up

    return reportTruncated();

  *p = NativeEndian::swapFromLittleEndian(point.peek());

  MOZ_ALWAYS_TRUE(point.advance());

  return true;

bool SCInput::readPair(uint32_t* tagp, uint32_t* datap) {

  uint64_t u;

  bool ok = read(&u);

  if (ok) {

    *tagp = uint32_t(u >> 32);

    *datap = uint32_t(u);

  return ok;

bool SCInput::get(uint64_t* p) {

  if (!point.canPeek()) {

    return reportTruncated();

  *p = NativeEndian::swapFromLittleEndian(point.peek());

  return true;

bool SCInput::getPair(uint32_t* tagp, uint32_t* datap) {

  uint64_t u = 0;

  if (!get(&u)) {

    return false;

  *tagp = uint32_t(u >> 32);

  *datap = uint32_t(u);

  return true;

void SCInput::getPair(uint64_t data, uint32_t* tagp, uint32_t* datap) {

  uint64_t u = NativeEndian::swapFromLittleEndian(data);

  *tagp = uint32_t(u >> 32);

  *datap = uint32_t(u);

bool SCInput::readDouble(double* p) {

  uint64_t u;

  if (!read(&u)) {

    return false;

  *p = CanonicalizeNaN(mozilla::BitwiseCast<double>(u));

  return true;

template <typename T>

static void swapFromLittleEndianInPlace(T* ptr, size_t nelems) {

  if (nelems > 0) {

    NativeEndian::swapFromLittleEndianInPlace(ptr, nelems);

template <>

void swapFromLittleEndianInPlace(uint8_t* ptr, size_t nelems) {}

// Data is packed into an integral number of uint64_t words. Compute the

// padding required to finish off the final word.

static size_t ComputePadding(size_t nelems, size_t elemSize) {

  // We want total length mod 8, where total length is nelems * sizeof(T),

  // but that might overflow. So reduce nelems to nelems mod 8, since we are

  // going to be doing a mod 8 later anyway.

  size_t leftoverLength = (nelems % sizeof(uint64_t)) * elemSize;

  return (-leftoverLength) & (sizeof(uint64_t) - 1);

template <class T>

bool SCInput::readArray(T* p, size_t nelems) {

  if (!nelems) {

    return true;

  static_assert(sizeof(uint64_t) % sizeof(T) == 0);

  // Fail if nelems is so huge that computing the full size will overflow.

  mozilla::CheckedInt<size_t> size =

      mozilla::CheckedInt<size_t>(nelems) * sizeof(T);

  if (!size.isValid()) {

    return reportTruncated();

  if (!point.readBytes(reinterpret_cast<char*>(p), size.value())) {

    // To avoid any way in which uninitialized data could escape, zero the array

    // if filling it failed.

    std::uninitialized_fill_n(p, nelems, 0);

    return reportTruncated();

  swapFromLittleEndianInPlace(p, nelems);

  point += ComputePadding(nelems, sizeof(T));

  return true;

bool SCInput::readBytes(void* p, size_t nbytes) {

  return readArray((uint8_t*)p, nbytes);

bool SCInput::readChars(Latin1Char* p, size_t nchars) {

  static_assert(sizeof(Latin1Char) == sizeof(uint8_t),

                "Latin1Char must fit in 1 byte");

  return readBytes(p, nchars);

bool SCInput::readChars(char16_t* p, size_t nchars) {

  MOZ_ASSERT(sizeof(char16_t) == sizeof(uint16_t));

  return readArray((uint16_t*)p, nchars);

void SCInput::getPtr(uint64_t data, void** ptr) {

  *ptr = reinterpret_cast<void*>(NativeEndian::swapFromLittleEndian(data));

bool SCInput::readPtr(void** p) {

  uint64_t u;

  if (!read(&u)) {

    return false;

  *p = reinterpret_cast<void*>(u);

  return true;

SCOutput::SCOutput(JSContext* cx, JS::StructuredCloneScope scope)

    : cx(cx), buf(scope) {}

bool SCOutput::write(uint64_t u) {

  uint64_t v = NativeEndian::swapToLittleEndian(u);

  if (!buf.AppendBytes(reinterpret_cast<char*>(&v), sizeof(u))) {

    ReportOutOfMemory(context());

    return false;

  return true;

bool SCOutput::writePair(uint32_t tag, uint32_t data) {

  // As it happens, the tag word appears after the data word in the output.

  // This is because exponents occupy the last 2 bytes of doubles on the

  // little-endian platforms we care most about.

//

  // For example, TrueValue() is written using writePair(SCTAG_BOOLEAN, 1).

  // PairToUInt64 produces the number 0xFFFF000200000001.

  // That is written out as the bytes 01 00 00 00 02 00 FF FF.

  return write(PairToUInt64(tag, data));

static inline double ReinterpretPairAsDouble(uint32_t tag, uint32_t data) {

  return BitwiseCast<double>(PairToUInt64(tag, data));

bool SCOutput::writeDouble(double d) {

  return write(BitwiseCast<uint64_t>(CanonicalizeNaN(d)));

template <class T>

bool SCOutput::writeArray(const T* p, size_t nelems) {

  static_assert(8 % sizeof(T) == 0);

  static_assert(sizeof(uint64_t) % sizeof(T) == 0);

  if (nelems == 0) {

    return true;

  for (size_t i = 0; i < nelems; i++) {

    T value = NativeEndian::swapToLittleEndian(p[i]);

    if (!buf.AppendBytes(reinterpret_cast<char*>(&value), sizeof(value))) {

      return false;

  // Zero-pad to 8 bytes boundary.

  size_t padbytes = ComputePadding(nelems, sizeof(T));

  char zeroes[sizeof(uint64_t)] = {0};

  if (!buf.AppendBytes(zeroes, padbytes)) {

    return false;

  return true;

template <>

bool SCOutput::writeArray<uint8_t>(const uint8_t* p, size_t nelems) {

  if (nelems == 0) {

    return true;

  if (!buf.AppendBytes(reinterpret_cast<const char*>(p), nelems)) {

    return false;

  // zero-pad to 8 bytes boundary

  size_t padbytes = ComputePadding(nelems, 1);

  char zeroes[sizeof(uint64_t)] = {0};

  if (!buf.AppendBytes(zeroes, padbytes)) {

    return false;

  return true;

bool SCOutput::writeBytes(const void* p, size_t nbytes) {

  return writeArray((const uint8_t*)p, nbytes);

bool SCOutput::writeChars(const char16_t* p, size_t nchars) {

  static_assert(sizeof(char16_t) == sizeof(uint16_t),

                "required so that treating char16_t[] memory as uint16_t[] "

                "memory is permissible");

  return writeArray((const uint16_t*)p, nchars);

bool SCOutput::writeChars(const Latin1Char* p, size_t nchars) {

  static_assert(sizeof(Latin1Char) == sizeof(uint8_t),

                "Latin1Char must fit in 1 byte");

  return writeBytes(p, nchars);

}  // namespace js

JSStructuredCloneData::~JSStructuredCloneData() { discardTransferables(); }

// If the buffer contains Transferables, free them. Note that custom

// Transferables will use the JSStructuredCloneCallbacks::freeTransfer() to

// delete their transferables.

void JSStructuredCloneData::discardTransferables() {

  if (!Size()) {

    return;

  if (ownTransferables_ != OwnTransferablePolicy::OwnsTransferablesIfAny) {

    return;

  // DifferentProcess clones cannot contain pointers, so nothing needs to be

  // released.

  if (scope() == JS::StructuredCloneScope::DifferentProcess) {

    return;

  FreeTransferStructuredCloneOp freeTransfer = nullptr;

  if (callbacks_) {

    freeTransfer = callbacks_->freeTransfer;

  auto point = BufferIterator<uint64_t, SystemAllocPolicy>(*this);

  if (point.done()) {

    return;  // Empty buffer

  uint32_t tag, data;

  MOZ_RELEASE_ASSERT(point.canPeek());

  SCInput::getPair(point.peek(), &tag, &data);

  MOZ_ALWAYS_TRUE(point.advance());

  if (tag == SCTAG_HEADER) {

    if (point.done()) {

      return;

    MOZ_RELEASE_ASSERT(point.canPeek());

    SCInput::getPair(point.peek(), &tag, &data);

    MOZ_ALWAYS_TRUE(point.advance());

  if (tag != SCTAG_TRANSFER_MAP_HEADER) {

    return;

  if (TransferableMapHeader(data) == SCTAG_TM_TRANSFERRED) {

    return;

  // freeTransfer should not GC

  JS::AutoSuppressGCAnalysis nogc;

  if (point.done()) {

    return;

  MOZ_RELEASE_ASSERT(point.canPeek());

  uint64_t numTransferables = NativeEndian::swapFromLittleEndian(point.peek());

  MOZ_ALWAYS_TRUE(point.advance());

  while (numTransferables--) {

    if (!point.canPeek()) {

      return;

    uint32_t ownership;

    SCInput::getPair(point.peek(), &tag, &ownership);

    MOZ_ALWAYS_TRUE(point.advance());

    MOZ_ASSERT(tag >= SCTAG_TRANSFER_MAP_PENDING_ENTRY);

    if (!point.canPeek()) {

      return;

    void* content;

    SCInput::getPtr(point.peek(), &content);

    MOZ_ALWAYS_TRUE(point.advance());

    if (!point.canPeek()) {

      return;

    uint64_t extraData = NativeEndian::swapFromLittleEndian(point.peek());

    MOZ_ALWAYS_TRUE(point.advance());

    if (ownership < JS::SCTAG_TMO_FIRST_OWNED) {

      continue;

    if (ownership == JS::SCTAG_TMO_ALLOC_DATA) {

      js_free(content);

    } else if (ownership == JS::SCTAG_TMO_MAPPED_DATA) {

      JS::ReleaseMappedArrayBufferContents(content, extraData);

    } else if (freeTransfer) {

      freeTransfer(tag, JS::TransferableOwnership(ownership), content,

                   extraData, closure_);

    } else {

      MOZ_ASSERT(false, "unknown ownership");

static_assert(JSString::MAX_LENGTH < UINT32_MAX);

bool JSStructuredCloneWriter::parseTransferable() {

  // NOTE: The transferables set is tested for non-emptiness at various

  //       junctures in structured cloning, so this set must be initialized

  //       by this method in all non-error cases.

  MOZ_ASSERT(transferableObjects.empty(),

             "parseTransferable called with stale data");

  if (transferable.isNull() || transferable.isUndefined()) {

    return true;

  if (!transferable.isObject()) {

    return reportDataCloneError(JS_SCERR_TRANSFERABLE);

  JSContext* cx = context();

  RootedObject array(cx, &transferable.toObject());

  bool isArray;

  if (!JS::IsArrayObject(cx, array, &isArray)) {

    return false;

  if (!isArray) {

    return reportDataCloneError(JS_SCERR_TRANSFERABLE);

  uint32_t length;

  if (!JS::GetArrayLength(cx, array, &length)) {

    return false;

  // Initialize the set for the provided array's length.

  if (!transferableObjects.reserve(length)) {

    return false;

  if (length == 0) {

    return true;

  RootedValue v(context());

  RootedObject tObj(context());

  for (uint32_t i = 0; i < length; ++i) {

    if (!CheckForInterrupt(cx)) {

      return false;

    if (!JS_GetElement(cx, array, i, &v)) {

      return false;

    if (!v.isObject()) {

      return reportDataCloneError(JS_SCERR_TRANSFERABLE);

    tObj = &v.toObject();

    RootedObject unwrappedObj(cx, CheckedUnwrapStatic(tObj));

    if (!unwrappedObj) {

      ReportAccessDenied(cx);

      return false;

    // Shared memory cannot be transferred because it is not possible (nor

    // desirable) to detach the memory in agents that already hold a

    // reference to it.

    if (unwrappedObj->is<SharedArrayBufferObject>()) {

      return reportDataCloneError(JS_SCERR_SHMEM_TRANSFERABLE);

    else if (unwrappedObj->is<WasmMemoryObject>()) {

      if (unwrappedObj->as<WasmMemoryObject>().isShared()) {

        return reportDataCloneError(JS_SCERR_SHMEM_TRANSFERABLE);

    // External array buffers may be able to be transferred in the future,

    // but that is not currently implemented.

    else if (unwrappedObj->is<ArrayBufferObject>()) {

      if (unwrappedObj->as<ArrayBufferObject>().isExternal()) {

        return reportDataCloneError(JS_SCERR_TRANSFERABLE);

    else {

      if (!out.buf.callbacks_ || !out.buf.callbacks_->canTransfer) {

        return reportDataCloneError(JS_SCERR_TRANSFERABLE);

      JSAutoRealm ar(cx, unwrappedObj);

      bool sameProcessScopeRequired = false;

      if (!out.buf.callbacks_->canTransfer(

              cx, unwrappedObj, &sameProcessScopeRequired, out.buf.closure_)) {

        return reportDataCloneError(JS_SCERR_TRANSFERABLE);

      if (sameProcessScopeRequired) {

        output().sameProcessScopeRequired();

    // No duplicates allowed

    if (std::find(transferableObjects.begin(), transferableObjects.end(),

                  tObj) != transferableObjects.end()) {

      return reportDataCloneError(JS_SCERR_DUP_TRANSFERABLE);

    if (!transferableObjects.append(tObj)) {

      return false;

  return true;

template <typename... Args>

bool JSStructuredCloneWriter::reportDataCloneError(uint32_t errorId,

                                                   Args&&... aArgs) {

  ReportDataCloneError(context(), out.buf.callbacks_, errorId, out.buf.closure_,

                       std::forward<Args>(aArgs)...);

  return false;

bool JSStructuredCloneWriter::writeString(uint32_t tag, JSString* str) {

  JSLinearString* linear = str->ensureLinear(context());

  if (!linear) {

    return false;

#if FUZZING_JS_FUZZILLI

  if (js::SupportDifferentialTesting()) {

    // TODO we could always output a twoByteChar string

    return true;

#endif

  static_assert(JSString::MAX_LENGTH <= INT32_MAX,

                "String length must fit in 31 bits");

  uint32_t length = linear->length();

  uint32_t lengthAndEncoding =

      length | (uint32_t(linear->hasLatin1Chars()) << 31);

  if (!out.writePair(tag, lengthAndEncoding)) {

    return false;

  JS::AutoCheckCannotGC nogc;

  return linear->hasLatin1Chars()

             ? out.writeChars(linear->latin1Chars(nogc), length)

             : out.writeChars(linear->twoByteChars(nogc), length);

bool JSStructuredCloneWriter::writeBigInt(uint32_t tag, BigInt* bi) {

  bool signBit = bi->isNegative();

  size_t length = bi->digitLength();

  // The length must fit in 31 bits to leave room for a sign bit.

  if (length > size_t(INT32_MAX)) {

    return false;

  uint32_t lengthAndSign = length | (static_cast<uint32_t>(signBit) << 31);

  if (!out.writePair(tag, lengthAndSign)) {

    return false;

  return out.writeArray(bi->digits().data(), length);

inline void JSStructuredCloneWriter::checkStack() {

#ifdef DEBUG

  // To avoid making serialization O(n^2), limit stack-checking at 10.

  const size_t MAX = 10;

  size_t limit = std::min(counts.length(), MAX);

  MOZ_ASSERT(objs.length() == counts.length());

  size_t total = 0;

  for (size_t i = 0; i < limit; i++) {

    MOZ_ASSERT(total + counts[i] >= total);

    total += counts[i];

  if (counts.length() <= MAX) {

    MOZ_ASSERT(total == objectEntries.length() + otherEntries.length());

  } else {

    MOZ_ASSERT(total <= objectEntries.length() + otherEntries.length());

  size_t j = objs.length();

  for (size_t i = 0; i < limit; i++) {

    --j;

    MOZ_ASSERT(memory.has(&objs[j].toObject()));

#endif

/*

 * Write out a typed array. Note that post-v1 structured clone buffers do not

 * perform endianness conversion on stored data, so multibyte typed arrays

 * cannot be deserialized into a different endianness machine. Endianness

 * conversion would prevent sharing ArrayBuffers: if you have Int8Array and

 * Int16Array views of the same ArrayBuffer, should the data bytes be

 * byte-swapped when writing or not? The Int8Array requires them to not be

 * swapped; the Int16Array requires that they are.

*/

bool JSStructuredCloneWriter::writeTypedArray(HandleObject obj) {

  Rooted<TypedArrayObject*> tarr(context(),

                                 obj->maybeUnwrapAs<TypedArrayObject>());

  JSAutoRealm ar(context(), tarr);

#ifdef FUZZING_JS_FUZZILLI

  if (js::SupportDifferentialTesting() && !tarr->hasBuffer()) {

    // fake oom because differential testing will fail

    fprintf(stderr, "[unhandlable oom]");

    _exit(-1);

    return false;

#endif

  if (!TypedArrayObject::ensureHasBuffer(context(), tarr)) {

    return false;

  if (!out.writePair(SCTAG_TYPED_ARRAY_OBJECT, uint32_t(tarr->type()))) {

    return false;

  mozilla::Maybe<size_t> nelems = tarr->length();

  if (!nelems) {

    return reportDataCloneError(JS_SCERR_TYPED_ARRAY_DETACHED);

  // Auto-length TypedArrays are tagged by storing `-1` for the length. We still

  // need to query the length to check for detached or out-of-bounds lengths.

  bool isAutoLength = tarr->is<ResizableTypedArrayObject>() &&

                      tarr->as<ResizableTypedArrayObject>().isAutoLength();

  uint64_t length = isAutoLength ? uint64_t(-1) : uint64_t(*nelems);

  if (!out.write(length)) {

    return false;

  // Write out the ArrayBuffer tag and contents

  RootedValue val(context(), tarr->bufferValue());

  if (!startWrite(val)) {

    return false;

  uint64_t byteOffset = tarr->byteOffset().valueOr(0);

  return out.write(byteOffset);

bool JSStructuredCloneWriter::writeDataView(HandleObject obj) {

  Rooted<DataViewObject*> view(context(), obj->maybeUnwrapAs<DataViewObject>());

  JSAutoRealm ar(context(), view);

  if (!out.writePair(SCTAG_DATA_VIEW_OBJECT, 0)) {

    return false;

  mozilla::Maybe<size_t> byteLength = view->byteLength();

  if (!byteLength) {

    return reportDataCloneError(JS_SCERR_TYPED_ARRAY_DETACHED);

  // Auto-length DataViews are tagged by storing `-1` for the length. We still

  // need to query the length to check for detached or out-of-bounds lengths.

  bool isAutoLength = view->is<ResizableDataViewObject>() &&

                      view->as<ResizableDataViewObject>().isAutoLength();

  uint64_t length = isAutoLength ? uint64_t(-1) : uint64_t(*byteLength);

  if (!out.write(length)) {

    return false;

  // Write out the ArrayBuffer tag and contents

  RootedValue val(context(), view->bufferValue());

  if (!startWrite(val)) {

    return false;

  uint64_t byteOffset = view->byteOffset().valueOr(0);

  return out.write(byteOffset);

bool JSStructuredCloneWriter::writeArrayBuffer(HandleObject obj) {

  Rooted<ArrayBufferObject*> buffer(context(),

                                    obj->maybeUnwrapAs<ArrayBufferObject>());

  JSAutoRealm ar(context(), buffer);

  StructuredDataType type = !buffer->isResizable()

                                ? SCTAG_ARRAY_BUFFER_OBJECT

                                : SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT;

  if (!out.writePair(type, 0)) {

    return false;

  uint64_t byteLength = buffer->byteLength();

  if (!out.write(byteLength)) {

    return false;

  if (buffer->isResizable()) {

    uint64_t maxByteLength =

        buffer->as<ResizableArrayBufferObject>().maxByteLength();

    if (!out.write(maxByteLength)) {

      return false;

  return out.writeBytes(buffer->dataPointer(), byteLength);

bool JSStructuredCloneWriter::writeSharedArrayBuffer(HandleObject obj) {

  MOZ_ASSERT(obj->canUnwrapAs<SharedArrayBufferObject>());

  if (!cloneDataPolicy.areSharedMemoryObjectsAllowed()) {

    auto error = context()->realm()->creationOptions().getCoopAndCoepEnabled()

                     ? JS_SCERR_NOT_CLONABLE_WITH_COOP_COEP

                     : JS_SCERR_NOT_CLONABLE;

    reportDataCloneError(error, "SharedArrayBuffer");

    return false;

  output().sameProcessScopeRequired();

  // We must not transmit SAB pointers (including for WebAssembly.Memory)

  // cross-process.  The cloneDataPolicy should have guarded against this;

  // since it did not then throw, with a very explicit message.

  if (output().scope() > JS::StructuredCloneScope::SameProcess) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_SHMEM_POLICY);

    return false;

  Rooted<SharedArrayBufferObject*> sharedArrayBuffer(

      context(), obj->maybeUnwrapAs<SharedArrayBufferObject>());

  SharedArrayRawBuffer* rawbuf = sharedArrayBuffer->rawBufferObject();

  if (!out.buf.refsHeld_.acquire(context(), rawbuf)) {

    return false;

  // We must serialize the length so that the buffer object arrives in the

  // receiver with the same length, and not with the length read from the

  // rawbuf - that length can be different, and it can change at any time.

  StructuredDataType type = !sharedArrayBuffer->isGrowable()

                                ? SCTAG_SHARED_ARRAY_BUFFER_OBJECT

                                : SCTAG_GROWABLE_SHARED_ARRAY_BUFFER_OBJECT;

  intptr_t p = reinterpret_cast<intptr_t>(rawbuf);

  uint64_t byteLength = sharedArrayBuffer->byteLengthOrMaxByteLength();

  if (!(out.writePair(type, /* unused data word */ 0) &&

        out.writeBytes(&byteLength, sizeof(byteLength)) &&

        out.writeBytes(&p, sizeof(p)))) {

    return false;

  if (callbacks && callbacks->sabCloned &&

      !callbacks->sabCloned(context(), /*receiving=*/false, closure)) {

    return false;

  return true;

bool JSStructuredCloneWriter::writeSharedWasmMemory(HandleObject obj) {

  MOZ_ASSERT(obj->canUnwrapAs<WasmMemoryObject>());

  // Check the policy here so that we can report a sane error.

  if (!cloneDataPolicy.areSharedMemoryObjectsAllowed()) {

    auto error = context()->realm()->creationOptions().getCoopAndCoepEnabled()

                     ? JS_SCERR_NOT_CLONABLE_WITH_COOP_COEP

                     : JS_SCERR_NOT_CLONABLE;

    reportDataCloneError(error, "WebAssembly.Memory");

    return false;

  // If this changes, might need to change what we write.

  MOZ_ASSERT(WasmMemoryObject::RESERVED_SLOTS == 3);

  Rooted<WasmMemoryObject*> memoryObj(context(),

                                      &obj->unwrapAs<WasmMemoryObject>());

  Rooted<SharedArrayBufferObject*> sab(

      context(), &memoryObj->buffer().as<SharedArrayBufferObject>());

  return out.writePair(SCTAG_SHARED_WASM_MEMORY_OBJECT, 0) &&

         out.writePair(SCTAG_BOOLEAN, memoryObj->isHuge()) &&

         writeSharedArrayBuffer(sab);

bool JSStructuredCloneWriter::startObject(HandleObject obj, bool* backref) {

  // Handle cycles in the object graph.

  CloneMemory::AddPtr p = memory.lookupForAdd(obj);

  if ((*backref = p.found())) {

    return out.writePair(SCTAG_BACK_REFERENCE_OBJECT, p->value());

  if (!memory.add(p, obj, memory.count())) {

    ReportOutOfMemory(context());

    return false;

  if (memory.count() == UINT32_MAX) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_NEED_DIET, "object graph to serialize");

    return false;

  return true;

static bool TryAppendNativeProperties(JSContext* cx, HandleObject obj,

                                      MutableHandleIdVector entries,

                                      size_t* properties, bool* optimized) {

  *optimized = false;

  if (!obj->is<NativeObject>()) {

    return true;

  Handle<NativeObject*> nobj = obj.as<NativeObject>();

  if (nobj->isIndexed() || nobj->is<TypedArrayObject>() ||

      nobj->getClass()->getNewEnumerate() || nobj->getClass()->getEnumerate()) {

    return true;

  *optimized = true;

  size_t count = 0;

  // We iterate from the last to the first property, so the property names

  // are already in reverse order.

  for (ShapePropertyIter<NoGC> iter(nobj->shape()); !iter.done(); iter++) {

    jsid id = iter->key();

    // Ignore symbols and non-enumerable properties.

    if (!iter->enumerable() || id.isSymbol()) {

      continue;

    MOZ_ASSERT(id.isString());

    if (!entries.append(id)) {

      return false;

    count++;

  // Add dense element ids in reverse order.

  for (uint32_t i = nobj->getDenseInitializedLength(); i > 0; --i) {

    if (nobj->getDenseElement(i - 1).isMagic(JS_ELEMENTS_HOLE)) {

      continue;

    if (!entries.append(PropertyKey::Int(i - 1))) {

      return false;

    count++;

  *properties = count;

  return true;

// Objects are written as a "preorder" traversal of the object graph: object

// "headers" (the class tag and any data needed for initial construction) are

// visited first, then the children are recursed through (where children are

// properties, Set or Map entries, etc.). So for example

//

//     obj1 = { key1: { key1.1: val1.1, key1.2: val1.2 }, key2: {} }

//

// would be stored as:

//

//     <Object tag for obj1>

//       <key1 data>

//       <Object tag for key1's value>

//         <key1.1 data>

//         <val1.1 data>

//         <key1.2 data>

//         <val1.2 data>

//       <end-of-children marker for key1's value>

//       <key2 data>

//       <Object tag for key2's value>

//       <end-of-children marker for key2's value>

//     <end-of-children marker for obj1>

//

// This nests nicely (ie, an entire recursive value starts with its tag and

// ends with its end-of-children marker) and so it can be presented indented.

// But see traverseMap below for how this looks different for Maps.

bool JSStructuredCloneWriter::traverseObject(HandleObject obj, ESClass cls) {

  size_t count;

  bool optimized = false;

  if (!js::SupportDifferentialTesting()) {

    if (!TryAppendNativeProperties(context(), obj, &objectEntries, &count,

                                   &optimized)) {

      return false;

  if (!optimized) {

    // Get enumerable property ids and put them in reverse order so that they

    // will come off the stack in forward order.

    RootedIdVector properties(context());

    if (!GetPropertyKeys(context(), obj, JSITER_OWNONLY, &properties)) {

      return false;

    for (size_t i = properties.length(); i > 0; --i) {

      jsid id = properties[i - 1];

      MOZ_ASSERT(id.isString() || id.isInt());

      if (!objectEntries.append(id)) {

        return false;

    count = properties.length();

  // Push obj and count to the stack.

  if (!objs.append(ObjectValue(*obj)) || !counts.append(count)) {

    return false;

  checkStack();

#if DEBUG

  ESClass cls2;

  if (!GetBuiltinClass(context(), obj, &cls2)) {

    return false;

  MOZ_ASSERT(cls2 == cls);

#endif

  // Write the header for obj.

  if (cls == ESClass::Array) {

    uint32_t length = 0;

    if (!JS::GetArrayLength(context(), obj, &length)) {

      return false;

    return out.writePair(SCTAG_ARRAY_OBJECT,

                         NativeEndian::swapToLittleEndian(length));

  return out.writePair(SCTAG_OBJECT_OBJECT, 0);

// Use the same basic setup as for traverseObject, but now keys can themselves

// be complex objects. Keys and values are visited first via startWrite(), then

// the key's children (if any) are handled, then the value's children.

//

//     m = new Map();

//     m.set(key1 = ..., value1 = ...)

//

// where key1 and value2 are both objects would be stored as

//

//     <Map tag>

//     <key1 class tag>

//     <value1 class tag>

//     ...key1 fields...

//     <end-of-children marker for key1>

//     ...value1 fields...

//     <end-of-children marker for value1>

//     <end-of-children marker for Map>

//

// Notice how the end-of-children marker for key1 is sandwiched between the

// value1 beginning and end.

bool JSStructuredCloneWriter::traverseMap(HandleObject obj) {

  Rooted<GCVector<Value>> newEntries(context(), GCVector<Value>(context()));

    // If there is no wrapper, the compartment munging is a no-op.

    RootedObject unwrapped(context(), obj->maybeUnwrapAs<MapObject>());

    MOZ_ASSERT(unwrapped);

    JSAutoRealm ar(context(), unwrapped);

    if (!MapObject::getKeysAndValuesInterleaved(unwrapped, &newEntries)) {

      return false;

  if (!context()->compartment()->wrap(context(), &newEntries)) {

    return false;

  for (size_t i = newEntries.length(); i > 0; --i) {

    if (!otherEntries.append(newEntries[i - 1])) {

      return false;

  // Push obj and count to the stack.

  if (!objs.append(ObjectValue(*obj)) || !counts.append(newEntries.length())) {

    return false;

  checkStack();

  // Write the header for obj.

  return out.writePair(SCTAG_MAP_OBJECT, 0);

// Similar to traverseMap, only there is a single value instead of a key and

// value, and thus no interleaving is possible: a value will be fully emitted

// before the next value is begun.

bool JSStructuredCloneWriter::traverseSet(HandleObject obj) {

  Rooted<GCVector<Value>> keys(context(), GCVector<Value>(context()));

    // If there is no wrapper, the compartment munging is a no-op.

    RootedObject unwrapped(context(), obj->maybeUnwrapAs<SetObject>());

    MOZ_ASSERT(unwrapped);

    JSAutoRealm ar(context(), unwrapped);

    if (!SetObject::keys(context(), unwrapped, &keys)) {

      return false;

  if (!context()->compartment()->wrap(context(), &keys)) {

    return false;

  for (size_t i = keys.length(); i > 0; --i) {

    if (!otherEntries.append(keys[i - 1])) {

      return false;

  // Push obj and count to the stack.

  if (!objs.append(ObjectValue(*obj)) || !counts.append(keys.length())) {

    return false;

  checkStack();

  // Write the header for obj.

  return out.writePair(SCTAG_SET_OBJECT, 0);

bool JSStructuredCloneWriter::traverseSavedFrame(HandleObject obj) {

  Rooted<SavedFrame*> savedFrame(context(), obj->maybeUnwrapAs<SavedFrame>());

  MOZ_ASSERT(savedFrame);

  RootedObject parent(context(), savedFrame->getParent());

  if (!context()->compartment()->wrap(context(), &parent)) {

    return false;

  if (!objs.append(ObjectValue(*obj)) ||

      !otherEntries.append(parent ? ObjectValue(*parent) : NullValue()) ||

      !counts.append(1)) {

    return false;

  checkStack();

  // Write the SavedFrame tag and the SavedFrame's principals.

  if (savedFrame->getPrincipals() ==

      &ReconstructedSavedFramePrincipals::IsSystem) {

    if (!out.writePair(SCTAG_SAVED_FRAME_OBJECT,

                       SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_SYSTEM)) {

      return false;

};

  } else if (savedFrame->getPrincipals() ==

             &ReconstructedSavedFramePrincipals::IsNotSystem) {

    if (!out.writePair(

            SCTAG_SAVED_FRAME_OBJECT,

            SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_NOT_SYSTEM)) {

      return false;

  } else {

    if (auto principals = savedFrame->getPrincipals()) {

      if (!out.writePair(SCTAG_SAVED_FRAME_OBJECT, SCTAG_JSPRINCIPALS) ||

          !principals->write(context(), this)) {

        return false;

    } else {

      if (!out.writePair(SCTAG_SAVED_FRAME_OBJECT, SCTAG_NULL_JSPRINCIPALS)) {

        return false;

  // Write the SavedFrame's reserved slots, except for the parent, which is

  // queued on objs for further traversal.

  RootedValue val(context());

  val = BooleanValue(savedFrame->getMutedErrors());

  if (!writePrimitive(val)) {

    return false;

  context()->markAtom(savedFrame->getSource());

  val = StringValue(savedFrame->getSource());

  if (!writePrimitive(val)) {

    return false;

  val = NumberValue(savedFrame->getLine());

  if (!writePrimitive(val)) {

    return false;

  val = NumberValue(*savedFrame->getColumn().addressOfValueForTranscode());

  if (!writePrimitive(val)) {

    return false;

  auto name = savedFrame->getFunctionDisplayName();

  if (name) {

    context()->markAtom(name);

  val = name ? StringValue(name) : NullValue();

  if (!writePrimitive(val)) {

    return false;

  auto cause = savedFrame->getAsyncCause();

  if (cause) {

    context()->markAtom(cause);

  val = cause ? StringValue(cause) : NullValue();

  if (!writePrimitive(val)) {

    return false;

  return true;

// https://html.spec.whatwg.org/multipage/structured-data.html#structuredserializeinternal

// 2.7.3 StructuredSerializeInternal ( value, forStorage [ , memory ] )

//

// Step 17. Otherwise, if value has an [[ErrorData]] internal slot and

//          value is not a platform object, then:

//

// Note: This contains custom extensions for handling non-standard properties.

bool JSStructuredCloneWriter::traverseError(HandleObject obj) {

  JSContext* cx = context();

  // 1. Let name be ? Get(value, "name").

  RootedValue name(cx);

  if (!GetProperty(cx, obj, obj, cx->names().name, &name)) {

    return false;

  // 2. If name is not one of "Error", "EvalError", "RangeError",

  // "ReferenceError", "SyntaxError", "TypeError", or "URIError",

  // (not yet specified: or "AggregateError")

  // then set name to "Error".

  JSExnType type = JSEXN_ERR;

  if (name.isString()) {

    JSLinearString* linear = name.toString()->ensureLinear(cx);

    if (!linear) {

      return false;

    if (EqualStrings(linear, cx->names().Error)) {

      type = JSEXN_ERR;

    } else if (EqualStrings(linear, cx->names().EvalError)) {

      type = JSEXN_EVALERR;

    } else if (EqualStrings(linear, cx->names().RangeError)) {

      type = JSEXN_RANGEERR;

    } else if (EqualStrings(linear, cx->names().ReferenceError)) {

      type = JSEXN_REFERENCEERR;

    } else if (EqualStrings(linear, cx->names().SyntaxError)) {

      type = JSEXN_SYNTAXERR;

    } else if (EqualStrings(linear, cx->names().TypeError)) {

      type = JSEXN_TYPEERR;

    } else if (EqualStrings(linear, cx->names().URIError)) {

      type = JSEXN_URIERR;

    } else if (EqualStrings(linear, cx->names().AggregateError)) {

      type = JSEXN_AGGREGATEERR;

  // 3. Let valueMessageDesc be ? value.[[GetOwnProperty]]("message").

  RootedId messageId(cx, NameToId(cx->names().message));

  Rooted<Maybe<PropertyDescriptor>> messageDesc(cx);

  if (!GetOwnPropertyDescriptor(cx, obj, messageId, &messageDesc)) {

    return false;

  // 4. Let message be undefined if IsDataDescriptor(valueMessageDesc) is false,

  //    and ? ToString(valueMessageDesc.[[Value]]) otherwise.

  RootedString message(cx);

  if (messageDesc.isSome() && messageDesc->isDataDescriptor()) {

    RootedValue messageVal(cx, messageDesc->value());

    message = ToString<CanGC>(cx, messageVal);

    if (!message) {

      return false;

  // 5. Set serialized to { [[Type]]: "Error", [[Name]]: name, [[Message]]:

  // message }.

  if (!objs.append(ObjectValue(*obj))) {

    return false;

  Rooted<ErrorObject*> unwrapped(cx, obj->maybeUnwrapAs<ErrorObject>());

  MOZ_ASSERT(unwrapped);

  // Non-standard: Serialize |stack|.

  // The Error stack property is saved as SavedFrames.

  RootedValue stack(cx, NullValue());

  RootedObject stackObj(cx, unwrapped->stack());

  if (stackObj) {

    MOZ_ASSERT(stackObj->canUnwrapAs<SavedFrame>());

    stack.setObject(*stackObj);

    if (!cx->compartment()->wrap(cx, &stack)) {

      return false;

  if (!otherEntries.append(stack)) {

    return false;

  // Serialize |errors|

  if (type == JSEXN_AGGREGATEERR) {

    RootedValue errors(cx);

    if (!GetProperty(cx, obj, obj, cx->names().errors, &errors)) {

      return false;

    if (!otherEntries.append(errors)) {

      return false;

  } else {

    if (!otherEntries.append(NullValue())) {

      return false;

  // Non-standard: Serialize |cause|. Because this property

  // might be missing we also write "hasCause" later.

  RootedId causeId(cx, NameToId(cx->names().cause));

  Rooted<Maybe<PropertyDescriptor>> causeDesc(cx);

  if (!GetOwnPropertyDescriptor(cx, obj, causeId, &causeDesc)) {

    return false;

  Rooted<Maybe<Value>> cause(cx);

  if (causeDesc.isSome() && causeDesc->isDataDescriptor()) {

    cause = mozilla::Some(causeDesc->value());

  if (!cx->compartment()->wrap(cx, &cause)) {

    return false;

  if (!otherEntries.append(cause.get().valueOr(NullValue()))) {

    return false;

  // |cause| + |errors| + |stack|, pushed in reverse order

  if (!counts.append(3)) {

    return false;

  checkStack();

  if (!out.writePair(SCTAG_ERROR_OBJECT, type)) {

    return false;

  RootedValue val(cx, message ? StringValue(message) : NullValue());

  if (!writePrimitive(val)) {

    return false;

  // hasCause

  val = BooleanValue(cause.isSome());

  if (!writePrimitive(val)) {

    return false;

  // Non-standard: Also serialize fileName, lineNumber and columnNumber.

    JSAutoRealm ar(cx, unwrapped);

    val = StringValue(unwrapped->fileName(cx));

  if (!cx->compartment()->wrap(cx, &val) || !writePrimitive(val)) {

    return false;

  val = Int32Value(unwrapped->lineNumber());

  if (!writePrimitive(val)) {

    return false;

  val = Int32Value(*unwrapped->columnNumber().addressOfValueForTranscode());

  return writePrimitive(val);

bool JSStructuredCloneWriter::writePrimitive(HandleValue v) {

  MOZ_ASSERT(v.isPrimitive());

  context()->check(v);

  if (v.isString()) {

    return writeString(SCTAG_STRING, v.toString());

  } else if (v.isInt32()) {

    if (js::SupportDifferentialTesting()) {

      return out.writeDouble(v.toInt32());

    return out.writePair(SCTAG_INT32, v.toInt32());

  } else if (v.isDouble()) {

    return out.writeDouble(v.toDouble());

  } else if (v.isBoolean()) {

    return out.writePair(SCTAG_BOOLEAN, v.toBoolean());

  } else if (v.isNull()) {

    return out.writePair(SCTAG_NULL, 0);

  } else if (v.isUndefined()) {

    return out.writePair(SCTAG_UNDEFINED, 0);

  } else if (v.isBigInt()) {

    return writeBigInt(SCTAG_BIGINT, v.toBigInt());

  return reportDataCloneError(JS_SCERR_UNSUPPORTED_TYPE);

bool JSStructuredCloneWriter::startWrite(HandleValue v) {

  context()->check(v);

  if (v.isPrimitive()) {

    return writePrimitive(v);

  if (!v.isObject()) {

    return reportDataCloneError(JS_SCERR_UNSUPPORTED_TYPE);

  RootedObject obj(context(), &v.toObject());

  bool backref;

  if (!startObject(obj, &backref)) {

    return false;

  if (backref) {

    return true;

  ESClass cls;

  if (!GetBuiltinClass(context(), obj, &cls)) {

    return false;

  switch (cls) {

    case ESClass::Object:

    case ESClass::Array:

      return traverseObject(obj, cls);

    case ESClass::Number: {

      RootedValue unboxed(context());

      if (!Unbox(context(), obj, &unboxed)) {

        return false;

      return out.writePair(SCTAG_NUMBER_OBJECT, 0) &&

             out.writeDouble(unboxed.toNumber());

    case ESClass::String: {

      RootedValue unboxed(context());

      if (!Unbox(context(), obj, &unboxed)) {

        return false;

      return writeString(SCTAG_STRING_OBJECT, unboxed.toString());

    case ESClass::Boolean: {

      RootedValue unboxed(context());

      if (!Unbox(context(), obj, &unboxed)) {

        return false;

      return out.writePair(SCTAG_BOOLEAN_OBJECT, unboxed.toBoolean());

    case ESClass::RegExp: {

      RegExpShared* re = RegExpToShared(context(), obj);

      if (!re) {

        return false;

      return out.writePair(SCTAG_REGEXP_OBJECT, re->getFlags().value()) &&

             writeString(SCTAG_STRING, re->getSource());

    case ESClass::ArrayBuffer: {

      if (JS::IsArrayBufferObject(obj) && JS::ArrayBufferHasData(obj)) {

        return writeArrayBuffer(obj);

      break;

    case ESClass::SharedArrayBuffer:

      if (JS::IsSharedArrayBufferObject(obj)) {

        return writeSharedArrayBuffer(obj);

      break;

    case ESClass::Date: {

      RootedValue unboxed(context());

      if (!Unbox(context(), obj, &unboxed)) {

        return false;

      return out.writePair(SCTAG_DATE_OBJECT, 0) &&

             out.writeDouble(unboxed.toNumber());

    case ESClass::Set:

      return traverseSet(obj);

    case ESClass::Map:

      return traverseMap(obj);

    case ESClass::Error:

      return traverseError(obj);

    case ESClass::BigInt: {

      RootedValue unboxed(context());

      if (!Unbox(context(), obj, &unboxed)) {

        return false;

      return writeBigInt(SCTAG_BIGINT_OBJECT, unboxed.toBigInt());

    case ESClass::Promise:

    case ESClass::MapIterator:

    case ESClass::SetIterator:

    case ESClass::Arguments:

    case ESClass::Function:

      break;

#ifdef ENABLE_RECORD_TUPLE

    case ESClass::Record:

    case ESClass::Tuple:

      MOZ_CRASH("Record and Tuple are not supported");

#endif

    case ESClass::Other: {

      if (obj->canUnwrapAs<TypedArrayObject>()) {

        return writeTypedArray(obj);

      if (obj->canUnwrapAs<DataViewObject>()) {

        return writeDataView(obj);

      if (wasm::IsSharedWasmMemoryObject(obj)) {

        return writeSharedWasmMemory(obj);

      if (obj->canUnwrapAs<SavedFrame>()) {

        return traverseSavedFrame(obj);

      break;

  if (out.buf.callbacks_ && out.buf.callbacks_->write) {

    bool sameProcessScopeRequired = false;

    if (!out.buf.callbacks_->write(context(), this, obj,

                                   &sameProcessScopeRequired,

                                   out.buf.closure_)) {

      return false;

    if (sameProcessScopeRequired) {

      output().sameProcessScopeRequired();

    return true;

  return reportDataCloneError(JS_SCERR_UNSUPPORTED_TYPE);

bool JSStructuredCloneWriter::writeHeader() {

  return out.writePair(SCTAG_HEADER, (uint32_t)output().scope());

bool JSStructuredCloneWriter::writeTransferMap() {

  if (transferableObjects.empty()) {

    return true;

  if (!out.writePair(SCTAG_TRANSFER_MAP_HEADER, (uint32_t)SCTAG_TM_UNREAD)) {

    return false;

  if (!out.write(transferableObjects.length())) {

    return false;

  RootedObject obj(context());

  for (auto* o : transferableObjects) {

    obj = o;

    if (!memory.put(obj, memory.count())) {

      ReportOutOfMemory(context());

      return false;

    // Emit a placeholder pointer.  We defer stealing the data until later

    // (and, if necessary, detaching this object if it's an ArrayBuffer).

    if (!out.writePair(SCTAG_TRANSFER_MAP_PENDING_ENTRY,

                       JS::SCTAG_TMO_UNFILLED)) {

      return false;

    if (!out.write(0)) {  // Pointer to ArrayBuffer contents.

      return false;

    if (!out.write(0)) {  // extraData

      return false;

  return true;

bool JSStructuredCloneWriter::transferOwnership() {

  if (transferableObjects.empty()) {

    return true;

  // Walk along the transferables and the transfer map at the same time,

  // grabbing out pointers from the transferables and stuffing them into the

  // transfer map.

  auto point = out.iter();

  MOZ_RELEASE_ASSERT(point.canPeek());

  MOZ_ASSERT(uint32_t(NativeEndian::swapFromLittleEndian(point.peek()) >> 32) ==

             SCTAG_HEADER);

  point++;

  MOZ_RELEASE_ASSERT(point.canPeek());

  MOZ_ASSERT(uint32_t(NativeEndian::swapFromLittleEndian(point.peek()) >> 32) ==

             SCTAG_TRANSFER_MAP_HEADER);

  point++;

  MOZ_RELEASE_ASSERT(point.canPeek());

  MOZ_ASSERT(NativeEndian::swapFromLittleEndian(point.peek()) ==

             transferableObjects.length());

  point++;

  JSContext* cx = context();

  RootedObject obj(cx);

  JS::StructuredCloneScope scope = output().scope();

  for (auto* o : transferableObjects) {

    obj = o;

    uint32_t tag;

    JS::TransferableOwnership ownership;

    void* content;

    uint64_t extraData;

#if DEBUG

    SCInput::getPair(point.peek(), &tag, (uint32_t*)&ownership);

    MOZ_ASSERT(tag == SCTAG_TRANSFER_MAP_PENDING_ENTRY);

    MOZ_ASSERT(ownership == JS::SCTAG_TMO_UNFILLED);

#endif

    ESClass cls;

    if (!GetBuiltinClass(cx, obj, &cls)) {

      return false;

    if (cls == ESClass::ArrayBuffer) {

      tag = SCTAG_TRANSFER_MAP_ARRAY_BUFFER;

      // The current setup of the array buffer inheritance hierarchy doesn't

      // lend itself well to generic manipulation via proxies.

      Rooted<ArrayBufferObject*> arrayBuffer(

          cx, obj->maybeUnwrapAs<ArrayBufferObject>());

      JSAutoRealm ar(cx, arrayBuffer);

      if (arrayBuffer->isDetached()) {

        reportDataCloneError(JS_SCERR_TYPED_ARRAY_DETACHED);

        return false;

      if (arrayBuffer->isPreparedForAsmJS()) {

        reportDataCloneError(JS_SCERR_WASM_NO_TRANSFER);

        return false;

      if (scope == JS::StructuredCloneScope::DifferentProcess ||

          scope == JS::StructuredCloneScope::DifferentProcessForIndexedDB ||

          arrayBuffer->isResizable()) {

        // Write Transferred ArrayBuffers in DifferentProcess scope at

        // the end of the clone buffer, and store the offset within the

        // buffer to where the ArrayBuffer was written. Note that this

        // will invalidate the current position iterator.

//

        // Resizable ArrayBuffers need to store two extra data, the byte length

        // and the maximum byte length, but the current transferables format

        // supports only a single additional datum. Therefore resizable buffers

        // currently go through this slower code path.

        size_t pointOffset = out.offset(point);

        tag = SCTAG_TRANSFER_MAP_STORED_ARRAY_BUFFER;

        ownership = JS::SCTAG_TMO_UNOWNED;

        content = nullptr;

        extraData = out.tell() -

                    pointOffset;  // Offset from tag to current end of buffer

        if (!writeArrayBuffer(arrayBuffer)) {

          ReportOutOfMemory(cx);

          return false;

        // Must refresh the point iterator after its collection has

        // been modified.

        point = out.iter();

        point += pointOffset;

        if (!JS::DetachArrayBuffer(cx, arrayBuffer)) {

          return false;

      } else {

        size_t nbytes = arrayBuffer->byteLength();

        using BufferContents = ArrayBufferObject::BufferContents;

        BufferContents bufContents =

            ArrayBufferObject::extractStructuredCloneContents(cx, arrayBuffer);

        if (!bufContents) {

          return false;  // out of memory

        content = bufContents.data();

        if (bufContents.kind() == ArrayBufferObject::MAPPED) {

          ownership = JS::SCTAG_TMO_MAPPED_DATA;

        } else {

          MOZ_ASSERT(

              bufContents.kind() ==

                      ArrayBufferObject::MALLOCED_ARRAYBUFFER_CONTENTS_ARENA ||

                  bufContents.kind() ==

                      ArrayBufferObject::MALLOCED_UNKNOWN_ARENA,

              "failing to handle new ArrayBuffer kind?");

          ownership = JS::SCTAG_TMO_ALLOC_DATA;

        extraData = nbytes;

    } else {

      if (!out.buf.callbacks_ || !out.buf.callbacks_->writeTransfer) {

        return reportDataCloneError(JS_SCERR_TRANSFERABLE);

      if (!out.buf.callbacks_->writeTransfer(cx, obj, out.buf.closure_, &tag,

                                             &ownership, &content,

                                             &extraData)) {

        return false;

      MOZ_ASSERT(tag > SCTAG_TRANSFER_MAP_PENDING_ENTRY);

    point.write(NativeEndian::swapToLittleEndian(PairToUInt64(tag, ownership)));

    MOZ_ALWAYS_TRUE(point.advance());

    point.write(

        NativeEndian::swapToLittleEndian(reinterpret_cast<uint64_t>(content)));

    MOZ_ALWAYS_TRUE(point.advance());

    point.write(NativeEndian::swapToLittleEndian(extraData));

    MOZ_ALWAYS_TRUE(point.advance());

#if DEBUG

  // Make sure there aren't any more transfer map entries after the expected

  // number we read out.

  if (!point.done()) {

    uint32_t tag, data;

    SCInput::getPair(point.peek(), &tag, &data);

    MOZ_ASSERT(tag < SCTAG_TRANSFER_MAP_HEADER ||

               tag >= SCTAG_TRANSFER_MAP_END_OF_BUILTIN_TYPES);

#endif

  return true;

bool JSStructuredCloneWriter::write(HandleValue v) {

  if (!startWrite(v)) {

    return false;

  RootedObject obj(context());

  RootedValue key(context());

  RootedValue val(context());

  RootedId id(context());

  RootedValue cause(context());

  RootedValue errors(context());

  RootedValue stack(context());

  while (!counts.empty()) {

    obj = &objs.back().toObject();

    context()->check(obj);

    if (counts.back()) {

      counts.back()--;

      ESClass cls;

      if (!GetBuiltinClass(context(), obj, &cls)) {

        return false;

      if (cls == ESClass::Map) {

        key = otherEntries.popCopy();

        checkStack();

        counts.back()--;

        val = otherEntries.popCopy();

        checkStack();

        if (!startWrite(key) || !startWrite(val)) {

          return false;

      } else if (cls == ESClass::Set || obj->canUnwrapAs<SavedFrame>()) {

        key = otherEntries.popCopy();

        checkStack();

        if (!startWrite(key)) {

          return false;

      } else if (cls == ESClass::Error) {

        cause = otherEntries.popCopy();

        checkStack();

        counts.back()--;

        errors = otherEntries.popCopy();

        checkStack();

        counts.back()--;

        stack = otherEntries.popCopy();

        checkStack();

        if (!startWrite(cause) || !startWrite(errors) || !startWrite(stack)) {

          return false;

      } else {

        id = objectEntries.popCopy();

        key = IdToValue(id);

        checkStack();

        // If obj still has an own property named id, write it out.

        bool found;

        if (GetOwnPropertyPure(context(), obj, id, val.address(), &found)) {

          if (found) {

            if (!writePrimitive(key) || !startWrite(val)) {

              return false;

          continue;

        if (!HasOwnProperty(context(), obj, id, &found)) {

          return false;

        if (found) {

#if FUZZING_JS_FUZZILLI

          // supress calls into user code

          if (js::SupportDifferentialTesting()) {

            fprintf(stderr, "Differential testing: cannot call GetProperty\n");

            return false;

#endif

          if (!writePrimitive(key) ||

              !GetProperty(context(), obj, obj, id, &val) || !startWrite(val)) {

            return false;

    } else {

      if (!out.writePair(SCTAG_END_OF_KEYS, 0)) {

        return false;

      objs.popBack();

      counts.popBack();

  memory.clear();

  return transferOwnership();

JSStructuredCloneReader::JSStructuredCloneReader(

    SCInput& in, JS::StructuredCloneScope scope,

    const JS::CloneDataPolicy& cloneDataPolicy,

    const JSStructuredCloneCallbacks* cb, void* cbClosure)

    : in(in),

      allowedScope(scope),

      cloneDataPolicy(cloneDataPolicy),

      objs(in.context()),

      objState(in.context(), in.context()),

      allObjs(in.context()),

      numItemsRead(0),

      callbacks(cb),

      closure(cbClosure),

      gcHeap(in.context()) {

  // Avoid the need to bounds check by keeping a never-matching element at the

  // base of the `objState` stack. This append() will always succeed because

  // the objState vector has a nonzero MinInlineCapacity.

  MOZ_ALWAYS_TRUE(objState.append(std::make_pair(nullptr, true)));

template <typename CharT>

JSString* JSStructuredCloneReader::readStringImpl(

    uint32_t nchars, ShouldAtomizeStrings atomize) {

  InlineCharBuffer<CharT> chars;

  if (!chars.maybeAlloc(context(), nchars) ||

      !in.readChars(chars.get(), nchars)) {

    return nullptr;

  if (atomize) {

    return chars.toAtom(context(), nchars);

  return chars.toStringDontDeflate(context(), nchars, gcHeap);

JSString* JSStructuredCloneReader::readString(uint32_t data,

                                              ShouldAtomizeStrings atomize) {

  uint32_t nchars = data & BitMask(31);

  bool latin1 = data & (1 << 31);

  if (nchars > JSString::MAX_LENGTH) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA, "string length");

    return nullptr;

  return latin1 ? readStringImpl<Latin1Char>(nchars, atomize)

                : readStringImpl<char16_t>(nchars, atomize);

[[nodiscard]] bool JSStructuredCloneReader::readUint32(uint32_t* num) {

  Rooted<Value> lineVal(context());

  if (!startRead(&lineVal)) {

    return false;

  if (!lineVal.isInt32()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA, "integer required");

    return false;

  *num = uint32_t(lineVal.toInt32());

  return true;

BigInt* JSStructuredCloneReader::readBigInt(uint32_t data) {

  size_t length = data & BitMask(31);

  bool isNegative = data & (1 << 31);

  if (length == 0) {

    return BigInt::zero(context());

  RootedBigInt result(context(), BigInt::createUninitialized(

                                     context(), length, isNegative, gcHeap));

  if (!result) {

    return nullptr;

  if (!in.readArray(result->digits().data(), length)) {

    return nullptr;

  return JS::BigInt::destructivelyTrimHighZeroDigits(context(), result);

static uint32_t TagToV1ArrayType(uint32_t tag) {

  MOZ_ASSERT(tag >= SCTAG_TYPED_ARRAY_V1_MIN &&

             tag <= SCTAG_TYPED_ARRAY_V1_MAX);

  return tag - SCTAG_TYPED_ARRAY_V1_MIN;

bool JSStructuredCloneReader::readTypedArray(uint32_t arrayType,

                                             uint64_t nelems,

                                             MutableHandleValue vp,

                                             bool v1Read) {

  if (arrayType > (v1Read ? Scalar::Uint8Clamped : Scalar::BigUint64)) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "unhandled typed array element type");

    return false;

  // Push a placeholder onto the allObjs list to stand in for the typed array.

  uint32_t placeholderIndex = allObjs.length();

  Value dummy = UndefinedValue();

  if (!allObjs.append(dummy)) {

    return false;

  // Auto-length TypedArrays are tagged by using `-1` for their length.

  bool isAutoLength = nelems == uint64_t(-1);

  // Zero |nelems| if it was only used as a tag.

  if (isAutoLength) {

    nelems = 0;

  // Read the ArrayBuffer object and its contents (but no properties)

  RootedValue v(context());

  uint64_t byteOffset;

  if (v1Read) {

    MOZ_ASSERT(!isAutoLength, "v1Read can't produce auto-length TypedArrays");

    if (!readV1ArrayBuffer(arrayType, nelems, &v)) {

      return false;

    byteOffset = 0;

  } else {

    if (!startRead(&v)) {

      return false;

    if (!in.read(&byteOffset)) {

      return false;

  // Ensure invalid 64-bit values won't be truncated below.

  if (nelems > ArrayBufferObject::ByteLengthLimit ||

      byteOffset > ArrayBufferObject::ByteLengthLimit) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid typed array length or offset");

    return false;

  if (!v.isObject() || !v.toObject().is<ArrayBufferObjectMaybeShared>()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "typed array must be backed by an ArrayBuffer");

    return false;

  RootedObject buffer(context(), &v.toObject());

  RootedObject obj(context(), nullptr);

  // Negative values represent an absent length parameter.

  int64_t length = isAutoLength ? -1 : int64_t(nelems);

  switch (arrayType) {

#define CREATE_FROM_BUFFER(ExternalType, NativeType, Name)             \

  case Scalar::Name:                                                   \

    obj = JS::TypedArray<Scalar::Name>::fromBuffer(context(), buffer,  \

                                                   byteOffset, length) \

              .asObject();                                             \

    break;

    JS_FOR_EACH_TYPED_ARRAY(CREATE_FROM_BUFFER)

#undef CREATE_FROM_BUFFER

    default:

      MOZ_CRASH("Can't happen: arrayType range checked above");

  if (!obj) {

    return false;

  vp.setObject(*obj);

  allObjs[placeholderIndex].set(vp);

  return true;

bool JSStructuredCloneReader::readDataView(uint64_t byteLength,

                                           MutableHandleValue vp) {

  // Push a placeholder onto the allObjs list to stand in for the DataView.

  uint32_t placeholderIndex = allObjs.length();

  Value dummy = UndefinedValue();

  if (!allObjs.append(dummy)) {

    return false;

  // Read the ArrayBuffer object and its contents (but no properties).

  RootedValue v(context());

  if (!startRead(&v)) {

    return false;

  if (!v.isObject() || !v.toObject().is<ArrayBufferObjectMaybeShared>()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "DataView must be backed by an ArrayBuffer");

    return false;

  // Read byteOffset.

  uint64_t byteOffset;

  if (!in.read(&byteOffset)) {

    return false;

  // Auto-length DataViews are tagged by using `-1` for their byte length.

  bool isAutoLength = byteLength == uint64_t(-1);

  // Zero |byteLength| if it was only used as a tag.

  if (isAutoLength) {

    byteLength = 0;

  // Ensure invalid 64-bit values won't be truncated below.

  if (byteLength > ArrayBufferObject::ByteLengthLimit ||

      byteOffset > ArrayBufferObject::ByteLengthLimit) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid DataView length or offset");

    return false;

  RootedObject buffer(context(), &v.toObject());

  RootedObject obj(context());

  if (!isAutoLength) {

    obj = JS_NewDataView(context(), buffer, byteOffset, byteLength);

  } else {

    obj = js::NewDataView(context(), buffer, byteOffset);

  if (!obj) {

    return false;

  vp.setObject(*obj);

  allObjs[placeholderIndex].set(vp);

  return true;

bool JSStructuredCloneReader::readArrayBuffer(StructuredDataType type,

                                              uint32_t data,

                                              MutableHandleValue vp) {

  // V2 stores the length in |data|. The current version stores the

  // length separately to allow larger length values.

  uint64_t nbytes = 0;

  uint64_t maxbytes = 0;

  if (type == SCTAG_ARRAY_BUFFER_OBJECT) {

    if (!in.read(&nbytes)) {

      return false;

  } else if (type == SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT) {

    if (!in.read(&nbytes)) {

      return false;

    if (!in.read(&maxbytes)) {

      return false;

  } else {

    MOZ_ASSERT(type == SCTAG_ARRAY_BUFFER_OBJECT_V2);

    nbytes = data;

  // The maximum ArrayBuffer size depends on the platform, and we cast to size_t

  // below, so we have to check this here.

  if (nbytes > ArrayBufferObject::ByteLengthLimit ||

      maxbytes > ArrayBufferObject::ByteLengthLimit) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_BAD_ARRAY_LENGTH);

    return false;

  JSObject* obj;

  if (type != SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT) {

    MOZ_ASSERT(maxbytes == 0);

    obj = ArrayBufferObject::createZeroed(context(), size_t(nbytes));

  } else {

    obj = ResizableArrayBufferObject::createZeroed(context(), size_t(nbytes),

                                                   size_t(maxbytes));

  if (!obj) {

    return false;

  vp.setObject(*obj);

  ArrayBufferObject& buffer = obj->as<ArrayBufferObject>();

  MOZ_ASSERT(buffer.byteLength() == nbytes);

  return in.readArray(buffer.dataPointer(), nbytes);

bool JSStructuredCloneReader::readSharedArrayBuffer(StructuredDataType type,

                                                    MutableHandleValue vp) {

  MOZ_ASSERT(type == SCTAG_SHARED_ARRAY_BUFFER_OBJECT ||

             type == SCTAG_GROWABLE_SHARED_ARRAY_BUFFER_OBJECT);

  if (!cloneDataPolicy.areIntraClusterClonableSharedObjectsAllowed() ||

      !cloneDataPolicy.areSharedMemoryObjectsAllowed()) {

    auto error = context()->realm()->creationOptions().getCoopAndCoepEnabled()

                     ? JS_SCERR_NOT_CLONABLE_WITH_COOP_COEP

                     : JS_SCERR_NOT_CLONABLE;

    ReportDataCloneError(context(), callbacks, error, closure,

                         "SharedArrayBuffer");

    return false;

  uint64_t byteLength;

  if (!in.readBytes(&byteLength, sizeof(byteLength))) {

    return in.reportTruncated();

  // The maximum ArrayBuffer size depends on the platform, and we cast to size_t

  // below, so we have to check this here.

  if (byteLength > ArrayBufferObject::ByteLengthLimit) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_BAD_ARRAY_LENGTH);

    return false;

  intptr_t p;

  if (!in.readBytes(&p, sizeof(p))) {

    return in.reportTruncated();

  bool isGrowable = type == SCTAG_GROWABLE_SHARED_ARRAY_BUFFER_OBJECT;

  SharedArrayRawBuffer* rawbuf = reinterpret_cast<SharedArrayRawBuffer*>(p);

  MOZ_RELEASE_ASSERT(isGrowable == rawbuf->isGrowable());

  // There's no guarantee that the receiving agent has enabled shared memory

  // even if the transmitting agent has done so.  Ideally we'd check at the

  // transmission point, but that's tricky, and it will be a very rare problem

  // in any case.  Just fail at the receiving end if we can't handle it.

  if (!context()

           ->realm()

           ->creationOptions()

           .getSharedMemoryAndAtomicsEnabled()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_SAB_DISABLED);

    return false;

  // The new object will have a new reference to the rawbuf.

  if (!rawbuf->addReference()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_SAB_REFCNT_OFLO);

    return false;

  RootedObject obj(context());

  if (!isGrowable) {

    obj = SharedArrayBufferObject::New(context(), rawbuf, byteLength);

  } else {

    obj = SharedArrayBufferObject::NewGrowable(context(), rawbuf, byteLength);

  if (!obj) {

    rawbuf->dropReference();

    return false;

  // `rawbuf` is now owned by `obj`.

  if (callbacks && callbacks->sabCloned &&

      !callbacks->sabCloned(context(), /*receiving=*/true, closure)) {

    return false;

  vp.setObject(*obj);

  return true;

bool JSStructuredCloneReader::readSharedWasmMemory(uint32_t nbytes,

                                                   MutableHandleValue vp) {

  JSContext* cx = context();

  if (nbytes != 0) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid shared wasm memory tag");

    return false;

  if (!cloneDataPolicy.areIntraClusterClonableSharedObjectsAllowed() ||

      !cloneDataPolicy.areSharedMemoryObjectsAllowed()) {

    auto error = context()->realm()->creationOptions().getCoopAndCoepEnabled()

                     ? JS_SCERR_NOT_CLONABLE_WITH_COOP_COEP

                     : JS_SCERR_NOT_CLONABLE;

    ReportDataCloneError(cx, callbacks, error, closure, "WebAssembly.Memory");

    return false;

  // Read the isHuge flag

  RootedValue isHuge(cx);

  if (!startRead(&isHuge)) {

    return false;

  // Read the SharedArrayBuffer object.

  RootedValue payload(cx);

  if (!startRead(&payload)) {

    return false;

  if (!payload.isObject() ||

      !payload.toObject().is<SharedArrayBufferObject>() ||

      payload.toObject().as<SharedArrayBufferObject>().isGrowable()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "shared wasm memory must be backed by a "

                              "non-growable SharedArrayBuffer");

    return false;

  Rooted<ArrayBufferObjectMaybeShared*> sab(

      cx, &payload.toObject().as<SharedArrayBufferObject>());

  // Construct the memory.

  RootedObject proto(

      cx, GlobalObject::getOrCreatePrototype(cx, JSProto_WasmMemory));

  if (!proto) {

    return false;

  RootedObject memory(

      cx, WasmMemoryObject::create(cx, sab, isHuge.toBoolean(), proto));

  if (!memory) {

    return false;

  vp.setObject(*memory);

  return true;

/*

 * Read in the data for a structured clone version 1 ArrayBuffer, performing

 * endianness-conversion while reading.

*/

bool JSStructuredCloneReader::readV1ArrayBuffer(uint32_t arrayType,

                                                uint32_t nelems,

                                                MutableHandleValue vp) {

  if (arrayType > Scalar::Uint8Clamped) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid TypedArray type");

    return false;

  mozilla::CheckedInt<size_t> nbytes =

      mozilla::CheckedInt<size_t>(nelems) *

      TypedArrayElemSize(static_cast<Scalar::Type>(arrayType));

  if (!nbytes.isValid() || nbytes.value() > UINT32_MAX) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid typed array size");

    return false;

  JSObject* obj = ArrayBufferObject::createZeroed(context(), nbytes.value());

  if (!obj) {

    return false;

  vp.setObject(*obj);

  ArrayBufferObject& buffer = obj->as<ArrayBufferObject>();

  MOZ_ASSERT(buffer.byteLength() == nbytes);

  switch (arrayType) {

    case Scalar::Int8:

    case Scalar::Uint8:

    case Scalar::Uint8Clamped:

      return in.readArray((uint8_t*)buffer.dataPointer(), nelems);

    case Scalar::Int16:

    case Scalar::Uint16:

      return in.readArray((uint16_t*)buffer.dataPointer(), nelems);

    case Scalar::Int32:

    case Scalar::Uint32:

    case Scalar::Float32:

      return in.readArray((uint32_t*)buffer.dataPointer(), nelems);

    case Scalar::Float64:

    case Scalar::BigInt64:

    case Scalar::BigUint64:

      return in.readArray((uint64_t*)buffer.dataPointer(), nelems);

    default:

      MOZ_CRASH("Can't happen: arrayType range checked by caller");

static bool PrimitiveToObject(JSContext* cx, MutableHandleValue vp) {

  JSObject* obj = js::PrimitiveToObject(cx, vp);

  if (!obj) {

    return false;

  vp.setObject(*obj);

  return true;

bool JSStructuredCloneReader::startRead(MutableHandleValue vp,

                                        ShouldAtomizeStrings atomizeStrings) {

  uint32_t tag, data;

  bool alreadAppended = false;

  if (!in.readPair(&tag, &data)) {

    return false;

  numItemsRead++;

  switch (tag) {

    case SCTAG_NULL:

      vp.setNull();

      break;

    case SCTAG_UNDEFINED:

      vp.setUndefined();

      break;

    case SCTAG_INT32:

      vp.setInt32(data);

      break;

    case SCTAG_BOOLEAN:

    case SCTAG_BOOLEAN_OBJECT:

      vp.setBoolean(!!data);

      if (tag == SCTAG_BOOLEAN_OBJECT && !PrimitiveToObject(context(), vp)) {

        return false;

      break;

    case SCTAG_STRING:

    case SCTAG_STRING_OBJECT: {

      JSString* str = readString(data, atomizeStrings);

      if (!str) {

        return false;

      vp.setString(str);

      if (tag == SCTAG_STRING_OBJECT && !PrimitiveToObject(context(), vp)) {

        return false;

      break;

    case SCTAG_NUMBER_OBJECT: {

      double d;

      if (!in.readDouble(&d)) {

        return false;

      vp.setDouble(CanonicalizeNaN(d));

      if (!PrimitiveToObject(context(), vp)) {

        return false;

      break;

    case SCTAG_BIGINT:

    case SCTAG_BIGINT_OBJECT: {

      RootedBigInt bi(context(), readBigInt(data));

      if (!bi) {

        return false;

      vp.setBigInt(bi);

      if (tag == SCTAG_BIGINT_OBJECT && !PrimitiveToObject(context(), vp)) {

        return false;

      break;

    case SCTAG_DATE_OBJECT: {

      double d;

      if (!in.readDouble(&d)) {

        return false;

      JS::ClippedTime t = JS::TimeClip(d);

      if (!NumbersAreIdentical(d, t.toDouble())) {

        JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                  JSMSG_SC_BAD_SERIALIZED_DATA, "date");

        return false;

      JSObject* obj = NewDateObjectMsec(context(), t);

      if (!obj) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_REGEXP_OBJECT: {

      if ((data & RegExpFlag::AllFlags) != data) {

        JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                  JSMSG_SC_BAD_SERIALIZED_DATA, "regexp");

        return false;

      RegExpFlags flags(AssertedCast<uint8_t>(data));

      uint32_t tag2, stringData;

      if (!in.readPair(&tag2, &stringData)) {

        return false;

      if (tag2 != SCTAG_STRING) {

        JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                  JSMSG_SC_BAD_SERIALIZED_DATA, "regexp");

        return false;

      JSString* str = readString(stringData, AtomizeStrings);

      if (!str) {

        return false;

      Rooted<JSAtom*> atom(context(), &str->asAtom());

      NewObjectKind kind =

          gcHeap == gc::Heap::Tenured ? TenuredObject : GenericObject;

      RegExpObject* reobj = RegExpObject::create(context(), atom, flags, kind);

      if (!reobj) {

        return false;

      vp.setObject(*reobj);

      break;

    case SCTAG_ARRAY_OBJECT:

    case SCTAG_OBJECT_OBJECT: {

      NewObjectKind kind =

          gcHeap == gc::Heap::Tenured ? TenuredObject : GenericObject;

      JSObject* obj;

      if (tag == SCTAG_ARRAY_OBJECT) {

        obj = NewDenseUnallocatedArray(

            context(), NativeEndian::swapFromLittleEndian(data), kind);

      } else {

        obj = NewPlainObject(context(), kind);

      if (!obj || !objs.append(ObjectValue(*obj))) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_BACK_REFERENCE_OBJECT: {

      if (data >= allObjs.length() || !allObjs[data].isObject()) {

        JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                  JSMSG_SC_BAD_SERIALIZED_DATA,

                                  "invalid back reference in input");

        return false;

      vp.set(allObjs[data]);

      return true;

    case SCTAG_TRANSFER_MAP_HEADER:

    case SCTAG_TRANSFER_MAP_PENDING_ENTRY:

      // We should be past all the transfer map tags.

      JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                JSMSG_SC_BAD_SERIALIZED_DATA, "invalid input");

      return false;

    case SCTAG_ARRAY_BUFFER_OBJECT_V2:

    case SCTAG_ARRAY_BUFFER_OBJECT:

    case SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT:

      if (!readArrayBuffer(StructuredDataType(tag), data, vp)) {

        return false;

      break;

    case SCTAG_SHARED_ARRAY_BUFFER_OBJECT:

    case SCTAG_GROWABLE_SHARED_ARRAY_BUFFER_OBJECT:

      if (!readSharedArrayBuffer(StructuredDataType(tag), vp)) {

        return false;

      break;

    case SCTAG_SHARED_WASM_MEMORY_OBJECT:

      if (!readSharedWasmMemory(data, vp)) {

        return false;

      break;

    case SCTAG_TYPED_ARRAY_OBJECT_V2: {

      // readTypedArray adds the array to allObjs.

      // V2 stores the length (nelems) in |data| and the arrayType separately.

      uint64_t arrayType;

      if (!in.read(&arrayType)) {

        return false;

      uint64_t nelems = data;

      return readTypedArray(arrayType, nelems, vp);

    case SCTAG_TYPED_ARRAY_OBJECT: {

      // readTypedArray adds the array to allObjs.

      // The current version stores the array type in |data| and the length

      // (nelems) separately to support large TypedArrays.

      uint32_t arrayType = data;

      uint64_t nelems;

      if (!in.read(&nelems)) {

        return false;

      return readTypedArray(arrayType, nelems, vp);

    case SCTAG_DATA_VIEW_OBJECT_V2: {

      // readDataView adds the array to allObjs.

      uint64_t byteLength = data;

      return readDataView(byteLength, vp);

    case SCTAG_DATA_VIEW_OBJECT: {

      // readDataView adds the array to allObjs.

      uint64_t byteLength;

      if (!in.read(&byteLength)) {

        return false;

      return readDataView(byteLength, vp);

    case SCTAG_MAP_OBJECT: {

      JSObject* obj = MapObject::create(context());

      if (!obj || !objs.append(ObjectValue(*obj))) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_SET_OBJECT: {

      JSObject* obj = SetObject::create(context());

      if (!obj || !objs.append(ObjectValue(*obj))) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_SAVED_FRAME_OBJECT: {

      auto* obj = readSavedFrameHeader(data);

      if (!obj || !objs.append(ObjectValue(*obj)) ||

          !objState.append(std::make_pair(obj, false))) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_ERROR_OBJECT: {

      auto* obj = readErrorHeader(data);

      if (!obj || !objs.append(ObjectValue(*obj)) ||

          !objState.append(std::make_pair(obj, false))) {

        return false;

      vp.setObject(*obj);

      break;

    case SCTAG_END_OF_KEYS:

      JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                JSMSG_SC_BAD_SERIALIZED_DATA,

                                "truncated input");

      return false;

      break;

    default: {

      if (tag <= SCTAG_FLOAT_MAX) {

        double d = ReinterpretPairAsDouble(tag, data);

        vp.setNumber(CanonicalizeNaN(d));

        break;

      if (SCTAG_TYPED_ARRAY_V1_MIN <= tag && tag <= SCTAG_TYPED_ARRAY_V1_MAX) {

        // A v1-format typed array

        // readTypedArray adds the array to allObjs

        return readTypedArray(TagToV1ArrayType(tag), data, vp, true);

      if (!callbacks || !callbacks->read) {

        JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                  JSMSG_SC_BAD_SERIALIZED_DATA,

                                  "unsupported type");

        return false;

      // callbacks->read() might read other objects from the buffer.

      // In startWrite we always write the object itself before calling

      // the custom function. We should do the same here to keep

      // indexing consistent.

      uint32_t placeholderIndex = allObjs.length();

      Value dummy = UndefinedValue();

      if (!allObjs.append(dummy)) {

        return false;

      JSObject* obj =

          callbacks->read(context(), this, cloneDataPolicy, tag, data, closure);

      if (!obj) {

        return false;

      vp.setObject(*obj);

      allObjs[placeholderIndex].set(vp);

      alreadAppended = true;

  if (!alreadAppended && vp.isObject() && !allObjs.append(vp)) {

    return false;

  return true;

bool JSStructuredCloneReader::readHeader() {

  uint32_t tag, data;

  if (!in.getPair(&tag, &data)) {

    return in.reportTruncated();

  JS::StructuredCloneScope storedScope;

  if (tag == SCTAG_HEADER) {

    MOZ_ALWAYS_TRUE(in.readPair(&tag, &data));

    storedScope = JS::StructuredCloneScope(data);

  } else {

    // Old structured clone buffer. We must have read it from disk.

    storedScope = JS::StructuredCloneScope::DifferentProcessForIndexedDB;

  // Backward compatibility with old structured clone buffers. Value '0' was

  // used for SameProcessSameThread scope.

  if ((int)storedScope == 0) {

    storedScope = JS::StructuredCloneScope::SameProcess;

  if (storedScope < JS::StructuredCloneScope::SameProcess ||

      storedScope > JS::StructuredCloneScope::DifferentProcessForIndexedDB) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid structured clone scope");

    return false;

  if (allowedScope == JS::StructuredCloneScope::DifferentProcessForIndexedDB) {

    // Bug 1434308 and bug 1458320 - the scopes stored in old IndexedDB

    // clones are incorrect. Treat them as if they were DifferentProcess.

    allowedScope = JS::StructuredCloneScope::DifferentProcess;

    return true;

  if (storedScope < allowedScope) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "incompatible structured clone scope");

    return false;

  return true;

bool JSStructuredCloneReader::readTransferMap() {

  JSContext* cx = context();

  auto headerPos = in.tell();

  uint32_t tag, data;

  if (!in.getPair(&tag, &data)) {

    return in.reportTruncated();

  if (tag != SCTAG_TRANSFER_MAP_HEADER ||

      TransferableMapHeader(data) == SCTAG_TM_TRANSFERRED) {

    return true;

  uint64_t numTransferables;

  MOZ_ALWAYS_TRUE(in.readPair(&tag, &data));

  if (!in.read(&numTransferables)) {

    return false;

  for (uint64_t i = 0; i < numTransferables; i++) {

    auto pos = in.tell();

    if (!in.readPair(&tag, &data)) {

      return false;

    if (tag == SCTAG_TRANSFER_MAP_PENDING_ENTRY) {

      ReportDataCloneError(cx, callbacks, JS_SCERR_TRANSFERABLE, closure);

      return false;

    RootedObject obj(cx);

    void* content;

    if (!in.readPtr(&content)) {

      return false;

    uint64_t extraData;

    if (!in.read(&extraData)) {

      return false;

    if (tag == SCTAG_TRANSFER_MAP_ARRAY_BUFFER) {

      if (allowedScope == JS::StructuredCloneScope::DifferentProcess ||

          allowedScope ==

              JS::StructuredCloneScope::DifferentProcessForIndexedDB) {

        // Transferred ArrayBuffers in a DifferentProcess clone buffer

        // are treated as if they weren't Transferred at all. We should

        // only see SCTAG_TRANSFER_MAP_STORED_ARRAY_BUFFER.

        ReportDataCloneError(cx, callbacks, JS_SCERR_TRANSFERABLE, closure);

        return false;

      MOZ_RELEASE_ASSERT(extraData <= ArrayBufferObject::ByteLengthLimit);

      size_t nbytes = extraData;

      MOZ_ASSERT(data == JS::SCTAG_TMO_ALLOC_DATA ||

                 data == JS::SCTAG_TMO_MAPPED_DATA);

      if (data == JS::SCTAG_TMO_ALLOC_DATA) {

        // When the ArrayBuffer can't be allocated, |content| will be free'ed

        // in `JSStructuredCloneData::discardTransferables()`.

        obj = JS::NewArrayBufferWithContents(

            cx, nbytes, content,

            JS::NewArrayBufferOutOfMemory::CallerMustFreeMemory);

      } else if (data == JS::SCTAG_TMO_MAPPED_DATA) {

        obj = JS::NewMappedArrayBufferWithContents(cx, nbytes, content);

    } else if (tag == SCTAG_TRANSFER_MAP_STORED_ARRAY_BUFFER) {

      auto savedPos = in.tell();

      auto guard = mozilla::MakeScopeExit([&] { in.seekTo(savedPos); });

      in.seekTo(pos);

      if (!in.seekBy(static_cast<size_t>(extraData))) {

        return false;

      if (tailStartPos.isNothing()) {

        tailStartPos = mozilla::Some(in.tell());

      uint32_t tag, data;

      if (!in.readPair(&tag, &data)) {

        return false;

      if (tag != SCTAG_ARRAY_BUFFER_OBJECT_V2 &&

          tag != SCTAG_ARRAY_BUFFER_OBJECT &&

          tag != SCTAG_RESIZABLE_ARRAY_BUFFER_OBJECT) {

        ReportDataCloneError(cx, callbacks, JS_SCERR_TRANSFERABLE, closure);

        return false;

      RootedValue val(cx);

      if (!readArrayBuffer(StructuredDataType(tag), data, &val)) {

        return false;

      obj = &val.toObject();

      tailEndPos = mozilla::Some(in.tell());

    } else {

      if (!callbacks || !callbacks->readTransfer) {

        ReportDataCloneError(cx, callbacks, JS_SCERR_TRANSFERABLE, closure);

        return false;

      if (!callbacks->readTransfer(cx, this, cloneDataPolicy, tag, content,

                                   extraData, closure, &obj)) {

        if (!cx->isExceptionPending()) {

          ReportDataCloneError(cx, callbacks, JS_SCERR_TRANSFERABLE, closure);

        return false;

      MOZ_ASSERT(obj);

      MOZ_ASSERT(!cx->isExceptionPending());

    // On failure, the buffer will still own the data (since its ownership

    // will not get set to SCTAG_TMO_UNOWNED), so the data will be freed by

    // DiscardTransferables.

    if (!obj) {

      return false;

    // Mark the SCTAG_TRANSFER_MAP_* entry as no longer owned by the input

    // buffer.

    pos.write(PairToUInt64(tag, JS::SCTAG_TMO_UNOWNED));

    MOZ_ASSERT(!pos.done());

    if (!allObjs.append(ObjectValue(*obj))) {

      return false;

  // Mark the whole transfer map as consumed.

#ifdef DEBUG

  SCInput::getPair(headerPos.peek(), &tag, &data);

  MOZ_ASSERT(tag == SCTAG_TRANSFER_MAP_HEADER);

  MOZ_ASSERT(TransferableMapHeader(data) != SCTAG_TM_TRANSFERRED);

#endif

  headerPos.write(

      PairToUInt64(SCTAG_TRANSFER_MAP_HEADER, SCTAG_TM_TRANSFERRED));

  return true;

JSObject* JSStructuredCloneReader::readSavedFrameHeader(

    uint32_t principalsTag) {

  Rooted<SavedFrame*> savedFrame(context(), SavedFrame::create(context()));

  if (!savedFrame) {

    return nullptr;

  JSPrincipals* principals;

  if (principalsTag == SCTAG_JSPRINCIPALS) {

    if (!context()->runtime()->readPrincipals) {

      JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                JSMSG_SC_UNSUPPORTED_TYPE);

      return nullptr;

    if (!context()->runtime()->readPrincipals(context(), this, &principals)) {

      return nullptr;

  } else if (principalsTag ==

             SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_SYSTEM) {

    principals = &ReconstructedSavedFramePrincipals::IsSystem;

    principals->refcount++;

  } else if (principalsTag ==

             SCTAG_RECONSTRUCTED_SAVED_FRAME_PRINCIPALS_IS_NOT_SYSTEM) {

    principals = &ReconstructedSavedFramePrincipals::IsNotSystem;

    principals->refcount++;

  } else if (principalsTag == SCTAG_NULL_JSPRINCIPALS) {

    principals = nullptr;

  } else {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "bad SavedFrame principals");

    return nullptr;

  RootedValue mutedErrors(context());

  RootedValue source(context());

    // Read a |mutedErrors| boolean followed by a |source| string.

    // The |mutedErrors| boolean is present in all new structured-clone data,

    // but in older data it will be absent and only the |source| string will be

    // found.

    if (!startRead(&mutedErrors, AtomizeStrings)) {

      return nullptr;

    if (mutedErrors.isBoolean()) {

      if (!startRead(&source, AtomizeStrings) || !source.isString()) {

        return nullptr;

    } else if (mutedErrors.isString()) {

      // Backwards compatibility: Handle missing |mutedErrors| boolean,

      // this is actually just a |source| string.

      source = mutedErrors;

      mutedErrors.setBoolean(true);  // Safe default value.

    } else {

      // Invalid type.

      return nullptr;

  savedFrame->initPrincipalsAlreadyHeldAndMutedErrors(principals,

                                                      mutedErrors.toBoolean());

  savedFrame->initSource(&source.toString()->asAtom());

  uint32_t line;

  if (!readUint32(&line)) {

    return nullptr;

  savedFrame->initLine(line);

  JS::TaggedColumnNumberOneOrigin column;

  if (!readUint32(column.addressOfValueForTranscode())) {

    return nullptr;

  savedFrame->initColumn(column);

  // Don't specify a source ID when reading a cloned saved frame, as these IDs

  // are only valid within a specific process.

  savedFrame->initSourceId(0);

  RootedValue name(context());

  if (!startRead(&name, AtomizeStrings)) {

    return nullptr;

  if (!(name.isString() || name.isNull())) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid saved frame cause");

    return nullptr;

  JSAtom* atomName = nullptr;

  if (name.isString()) {

    atomName = &name.toString()->asAtom();

  savedFrame->initFunctionDisplayName(atomName);

  RootedValue cause(context());

  if (!startRead(&cause, AtomizeStrings)) {

    return nullptr;

  if (!(cause.isString() || cause.isNull())) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid saved frame cause");

    return nullptr;

  JSAtom* atomCause = nullptr;

  if (cause.isString()) {

    atomCause = &cause.toString()->asAtom();

  savedFrame->initAsyncCause(atomCause);

  return savedFrame;

// SavedFrame object: there is one child value, the parent SavedFrame,

// which is either null or another SavedFrame object.

bool JSStructuredCloneReader::readSavedFrameFields(Handle<SavedFrame*> frameObj,

                                                   HandleValue parent,

                                                   bool* state) {

  if (*state) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "multiple SavedFrame parents");

    return false;

  SavedFrame* parentFrame;

  if (parent.isNull()) {

    parentFrame = nullptr;

  } else if (parent.isObject() && parent.toObject().is<SavedFrame>()) {

    parentFrame = &parent.toObject().as<SavedFrame>();

  } else {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid SavedFrame parent");

    return false;

  frameObj->initParent(parentFrame);

  *state = true;

  return true;

JSObject* JSStructuredCloneReader::readErrorHeader(uint32_t type) {

  JSContext* cx = context();

  switch (type) {

    case JSEXN_ERR:

    case JSEXN_EVALERR:

    case JSEXN_RANGEERR:

    case JSEXN_REFERENCEERR:

    case JSEXN_SYNTAXERR:

    case JSEXN_TYPEERR:

    case JSEXN_URIERR:

    case JSEXN_AGGREGATEERR:

      break;

    default:

      JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                JSMSG_SC_BAD_SERIALIZED_DATA,

                                "invalid error type");

      return nullptr;

  RootedString message(cx);

    RootedValue messageVal(cx);

    if (!startRead(&messageVal)) {

      return nullptr;

    if (messageVal.isString()) {

      message = messageVal.toString();

    } else if (!messageVal.isNull()) {

      JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                                JSMSG_SC_BAD_SERIALIZED_DATA,

                                "invalid 'message' field for Error object");

      return nullptr;

  // We have to set |cause| to something if it exists, otherwise the shape

  // would be wrong. The actual value will be overwritten later.

  RootedValue val(cx);

  if (!startRead(&val)) {

    return nullptr;

  bool hasCause = ToBoolean(val);

  Rooted<Maybe<Value>> cause(cx, mozilla::Nothing());

  if (hasCause) {

    cause = mozilla::Some(BooleanValue(true));

  if (!startRead(&val)) {

    return nullptr;

  if (!val.isString()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid 'fileName' field for Error object");

    return nullptr;

  RootedString fileName(cx, val.toString());

  uint32_t lineNumber;

  JS::ColumnNumberOneOrigin columnNumber;

  if (!readUint32(&lineNumber) ||

      !readUint32(columnNumber.addressOfValueForTranscode())) {

    return nullptr;

  // The |cause| and |stack| slots of the objects might be overwritten later.

  // For AggregateErrors the |errors| property will be added.

  RootedObject errorObj(

      cx, ErrorObject::create(cx, static_cast<JSExnType>(type), nullptr,

                              fileName, 0, lineNumber, columnNumber, nullptr,

                              message, cause));

  if (!errorObj) {

    return nullptr;

  return errorObj;

// Error objects have 3 fields, some or all of them null: cause,

// errors, and stack.

bool JSStructuredCloneReader::readErrorFields(Handle<ErrorObject*> errorObj,

                                              HandleValue cause, bool* state) {

  JSContext* cx = context();

  if (*state) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "unexpected child value seen for Error object");

    return false;

  RootedValue errors(cx);

  RootedValue stack(cx);

  if (!startRead(&errors) || !startRead(&stack)) {

    return false;

  bool hasCause = errorObj->getCause().isSome();

  if (hasCause) {

    errorObj->setCauseSlot(cause);

  } else if (!cause.isNull()) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid 'cause' field for Error object");

    return false;

  if (errorObj->type() == JSEXN_AGGREGATEERR) {

    if (!DefineDataProperty(context(), errorObj, cx->names().errors, errors,

                            0)) {

      return false;

  } else if (!errors.isNull()) {

    JS_ReportErrorNumberASCII(

        cx, GetErrorMessage, nullptr, JSMSG_SC_BAD_SERIALIZED_DATA,

        "unexpected 'errors' field seen for non-AggregateError");

    return false;

  if (stack.isObject()) {

    RootedObject stackObj(cx, &stack.toObject());

    if (!stackObj->is<SavedFrame>()) {

      JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                                JSMSG_SC_BAD_SERIALIZED_DATA,

                                "invalid 'stack' field for Error object");

      return false;

    errorObj->setStackSlot(stack);

  } else if (!stack.isNull()) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "invalid 'stack' field for Error object");

    return false;

  *state = true;

  return true;

// Read a value and treat as a key,value pair.

bool JSStructuredCloneReader::readMapField(Handle<MapObject*> mapObj,

                                           HandleValue key) {

  RootedValue val(context());

  if (!startRead(&val)) {

    return false;

  return MapObject::set(context(), mapObj, key, val);

// Read a value and treat as a key,value pair. Interpret as a plain property

// value.

bool JSStructuredCloneReader::readObjectField(HandleObject obj,

                                              HandleValue key) {

  if (!key.isString() && !key.isInt32()) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "property key expected");

    return false;

  RootedValue val(context());

  if (!startRead(&val)) {

    return false;

  RootedId id(context());

  if (!PrimitiveValueToId<CanGC>(context(), key, &id)) {

    return false;

  // Fast path for adding a new property to a plain object. The property names

  // we see here should be unique, but we check for duplicates to guard against

  // corrupt or malicious data.

  if (id.isString() && obj->is<PlainObject>() &&

      MOZ_LIKELY(!obj->as<PlainObject>().contains(context(), id))) {

    return AddDataPropertyToPlainObject(context(), obj.as<PlainObject>(), id,

                                        val);

  // Fast path for adding an array element. The index shouldn't exceed the

  // array's length, but we check for this in `addDenseElementNoLengthChange` to

  // guard against corrupt or malicious data.

  if (id.isInt() && obj->is<ArrayObject>()) {

    ArrayObject* arr = &obj->as<ArrayObject>();

    switch (arr->addDenseElementNoLengthChange(context(), id.toInt(), val)) {

      case DenseElementResult::Failure:

        return false;

      case DenseElementResult::Success:

        return true;

      case DenseElementResult::Incomplete:

        // Fall-through to slow path.

        break;

  return DefineDataProperty(context(), obj, id, val);

// Perform the whole recursive reading procedure.

bool JSStructuredCloneReader::read(MutableHandleValue vp, size_t nbytes) {

  auto startTime = mozilla::TimeStamp::Now();

  if (!readHeader()) {

    return false;

  if (!readTransferMap()) {

    return false;

  MOZ_ASSERT(objs.length() == 0);

  MOZ_ASSERT(objState.length() == 1);

  // Start out by reading in the main object and pushing it onto the 'objs'

  // stack. The data related to this object and its descendants extends from

  // here to the SCTAG_END_OF_KEYS at the end of the stream.

  if (!startRead(vp)) {

    return false;

  // Stop when the stack shows that all objects have been read.

  while (objs.length() != 0) {

    // What happens depends on the top obj on the objs stack.

    RootedObject obj(context(), &objs.back().toObject());

    uint32_t tag, data;

    if (!in.getPair(&tag, &data)) {

      return false;

    if (tag == SCTAG_END_OF_KEYS) {

      // Pop the current obj off the stack, since we are done with it and

      // its children.

      MOZ_ALWAYS_TRUE(in.readPair(&tag, &data));

      objs.popBack();

      if (objState.back().first == obj) {

        objState.popBack();

      continue;

    // Remember the index of the current top of the state stack, which will

    // correspond to the state for `obj` iff `obj` is a type that uses state.

    // startRead() may push additional entries before the state is accessed and

    // updated while filling in the object's data.

    size_t objStateIdx = objState.length() - 1;

    // The input stream contains a sequence of "child" values, whose

    // interpretation depends on the type of obj. These values can be

    // anything, and startRead() will push onto 'objs' for any non-leaf

    // value (i.e., anything that may contain children).

//

    // startRead() will allocate the (empty) object, but note that when

    // startRead() returns, 'key' is not yet initialized with any of its

    // properties. Those will be filled in by returning to the head of this

    // loop, processing the first child obj, and continuing until all

    // children have been fully created.

//

    // Note that this means the ordering in the stream is a little funky for

    // things like Map. See the comment above traverseMap() for an example.

    bool expectKeyValuePairs =

        !(obj->is<MapObject>() || obj->is<SetObject>() ||

          obj->is<SavedFrame>() || obj->is<ErrorObject>());

    RootedValue key(context());

    ShouldAtomizeStrings atomize =

        expectKeyValuePairs ? AtomizeStrings : DontAtomizeStrings;

    if (!startRead(&key, atomize)) {

      return false;

    if (key.isNull() && expectKeyValuePairs) {

      // Backwards compatibility: Null formerly indicated the end of

      // object properties.

      // No legacy objects used the state stack.

      MOZ_ASSERT(objState[objStateIdx].first() != obj);

      objs.popBack();

      continue;

    context()->check(key);

    if (obj->is<SetObject>()) {

      // Set object: the values between obj header (from startRead()) and

      // SCTAG_END_OF_KEYS are all interpreted as values to add to the set.

      if (!SetObject::add(context(), obj, key)) {

        return false;

    } else if (obj->is<MapObject>()) {

      Rooted<MapObject*> mapObj(context(), &obj->as<MapObject>());

      if (!readMapField(mapObj, key)) {

        return false;

    } else if (obj->is<SavedFrame>()) {

      Rooted<SavedFrame*> frameObj(context(), &obj->as<SavedFrame>());

      MOZ_ASSERT(objState[objStateIdx].first() == obj);

      bool state = objState[objStateIdx].second();

      if (!readSavedFrameFields(frameObj, key, &state)) {

        return false;

      objState[objStateIdx].second() = state;

    } else if (obj->is<ErrorObject>()) {

      Rooted<ErrorObject*> errorObj(context(), &obj->as<ErrorObject>());

      MOZ_ASSERT(objState[objStateIdx].first() == obj);

      bool state = objState[objStateIdx].second();

      if (!readErrorFields(errorObj, key, &state)) {

        return false;

      objState[objStateIdx].second() = state;

    } else {

      MOZ_ASSERT(expectKeyValuePairs);

      // Everything else uses a series of key,value,key,value,... Value

      // objects.

      if (!readObjectField(obj, key)) {

        return false;

  allObjs.clear();

  // For fuzzing, it is convenient to allow extra data at the end

  // of the input buffer so that more possible inputs are considered

  // valid.

#ifndef FUZZING

  bool extraData;

  if (tailStartPos.isSome()) {

    // in.tell() is the end of the main data. If "tail" data was consumed,

    // then check whether there's any data between the main data and the

    // beginning of the tail, or after the last read point in the tail.

    extraData = (in.tell() != *tailStartPos || !tailEndPos->done());

  } else {

    extraData = !in.tell().done();

  if (extraData) {

    JS_ReportErrorNumberASCII(context(), GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_SERIALIZED_DATA,

                              "extra data after end");

    return false;

#endif

  JSRuntime* rt = context()->runtime();

  rt->metrics().DESERIALIZE_BYTES(nbytes);

  rt->metrics().DESERIALIZE_ITEMS(numItemsRead);

  mozilla::TimeDuration elapsed = mozilla::TimeStamp::Now() - startTime;

  rt->metrics().DESERIALIZE_US(elapsed);

  return true;

JS_PUBLIC_API bool JS_ReadStructuredClone(

    JSContext* cx, const JSStructuredCloneData& buf, uint32_t version,

    JS::StructuredCloneScope scope, MutableHandleValue vp,

    const JS::CloneDataPolicy& cloneDataPolicy,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure) {

  AssertHeapIsIdle();

  CHECK_THREAD(cx);

  if (version > JS_STRUCTURED_CLONE_VERSION) {

    JS_ReportErrorNumberASCII(cx, GetErrorMessage, nullptr,

                              JSMSG_SC_BAD_CLONE_VERSION);

    return false;

  const JSStructuredCloneCallbacks* callbacks = optionalCallbacks;

  return ReadStructuredClone(cx, buf, scope, vp, cloneDataPolicy, callbacks,

                             closure);

JS_PUBLIC_API bool JS_WriteStructuredClone(

    JSContext* cx, HandleValue value, JSStructuredCloneData* bufp,

    JS::StructuredCloneScope scope, const JS::CloneDataPolicy& cloneDataPolicy,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure,

    HandleValue transferable) {

  AssertHeapIsIdle();

  CHECK_THREAD(cx);

  cx->check(value);

  const JSStructuredCloneCallbacks* callbacks = optionalCallbacks;

  return WriteStructuredClone(cx, value, bufp, scope, cloneDataPolicy,

                              callbacks, closure, transferable);

JS_PUBLIC_API bool JS_StructuredCloneHasTransferables(

    JSStructuredCloneData& data, bool* hasTransferable) {

  *hasTransferable = StructuredCloneHasTransferObjects(data);

  return true;

JS_PUBLIC_API bool JS_StructuredClone(

    JSContext* cx, HandleValue value, MutableHandleValue vp,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure) {

  AssertHeapIsIdle();

  CHECK_THREAD(cx);

  // Strings are associated with zones, not compartments,

  // so we copy the string by wrapping it.

  if (value.isString()) {

    RootedString strValue(cx, value.toString());

    if (!cx->compartment()->wrap(cx, &strValue)) {

      return false;

    vp.setString(strValue);

    return true;

  const JSStructuredCloneCallbacks* callbacks = optionalCallbacks;

  JSAutoStructuredCloneBuffer buf(JS::StructuredCloneScope::SameProcess,

                                  callbacks, closure);

    if (value.isObject()) {

      RootedObject obj(cx, &value.toObject());

      obj = CheckedUnwrapStatic(obj);

      if (!obj) {

        ReportAccessDenied(cx);

        return false;

      AutoRealm ar(cx, obj);

      RootedValue unwrappedVal(cx, ObjectValue(*obj));

      if (!buf.write(cx, unwrappedVal, callbacks, closure)) {

        return false;

    } else {

      if (!buf.write(cx, value, callbacks, closure)) {

        return false;

  return buf.read(cx, vp, JS::CloneDataPolicy(), callbacks, closure);

JSAutoStructuredCloneBuffer::JSAutoStructuredCloneBuffer(

    JSAutoStructuredCloneBuffer&& other)

    : data_(other.scope()) {

  version_ = other.version_;

  other.giveTo(&data_);

JSAutoStructuredCloneBuffer& JSAutoStructuredCloneBuffer::operator=(

    JSAutoStructuredCloneBuffer&& other) {

  MOZ_ASSERT(&other != this);

  MOZ_ASSERT(scope() == other.scope());

  clear();

  version_ = other.version_;

  other.giveTo(&data_);

  return *this;

void JSAutoStructuredCloneBuffer::clear() {

  data_.discardTransferables();

  data_.ownTransferables_ = OwnTransferablePolicy::NoTransferables;

  data_.refsHeld_.releaseAll();

  data_.Clear();

  version_ = 0;

void JSAutoStructuredCloneBuffer::adopt(

    JSStructuredCloneData&& data, uint32_t version,

    const JSStructuredCloneCallbacks* callbacks, void* closure) {

  clear();

  data_ = std::move(data);

  version_ = version;

  data_.setCallbacks(callbacks, closure,

                     OwnTransferablePolicy::OwnsTransferablesIfAny);

void JSAutoStructuredCloneBuffer::giveTo(JSStructuredCloneData* data) {

  *data = std::move(data_);

  version_ = 0;

  data_.setCallbacks(nullptr, nullptr, OwnTransferablePolicy::NoTransferables);

  data_.Clear();

bool JSAutoStructuredCloneBuffer::read(

    JSContext* cx, MutableHandleValue vp,

    const JS::CloneDataPolicy& cloneDataPolicy,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure) {

  MOZ_ASSERT(cx);

  return !!JS_ReadStructuredClone(

      cx, data_, version_, data_.scope(), vp, cloneDataPolicy,

      optionalCallbacks ? optionalCallbacks : data_.callbacks_,

      optionalCallbacks ? closure : data_.closure_);

bool JSAutoStructuredCloneBuffer::write(

    JSContext* cx, HandleValue value,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure) {

  HandleValue transferable = UndefinedHandleValue;

  return write(cx, value, transferable, JS::CloneDataPolicy(),

               optionalCallbacks ? optionalCallbacks : data_.callbacks_,

               optionalCallbacks ? closure : data_.closure_);

bool JSAutoStructuredCloneBuffer::write(

    JSContext* cx, HandleValue value, HandleValue transferable,

    const JS::CloneDataPolicy& cloneDataPolicy,

    const JSStructuredCloneCallbacks* optionalCallbacks, void* closure) {

  clear();

  bool ok = JS_WriteStructuredClone(

      cx, value, &data_, data_.scopeForInternalWriting(), cloneDataPolicy,

      optionalCallbacks ? optionalCallbacks : data_.callbacks_,

      optionalCallbacks ? closure : data_.closure_, transferable);

  if (!ok) {

    version_ = JS_STRUCTURED_CLONE_VERSION;

  return ok;

JS_PUBLIC_API bool JS_ReadUint32Pair(JSStructuredCloneReader* r, uint32_t* p1,

                                     uint32_t* p2) {

  return r->input().readPair((uint32_t*)p1, (uint32_t*)p2);

JS_PUBLIC_API bool JS_ReadBytes(JSStructuredCloneReader* r, void* p,

                                size_t len) {

  return r->input().readBytes(p, len);

JS_PUBLIC_API bool JS_ReadString(JSStructuredCloneReader* r,

                                 MutableHandleString str) {

  uint32_t tag, data;

  if (!r->input().readPair(&tag, &data)) {

    return false;

  if (tag == SCTAG_STRING) {

    if (JSString* s =

            r->readString(data, JSStructuredCloneReader::DontAtomizeStrings)) {

      str.set(s);

      return true;

    return false;

  JS_ReportErrorNumberASCII(r->context(), GetErrorMessage, nullptr,

                            JSMSG_SC_BAD_SERIALIZED_DATA, "expected string");

  return false;

JS_PUBLIC_API bool JS_ReadDouble(JSStructuredCloneReader* r, double* v) {

  return r->input().readDouble(v);

JS_PUBLIC_API bool JS_ReadTypedArray(JSStructuredCloneReader* r,

                                     MutableHandleValue vp) {

  uint32_t tag, data;

  if (!r->input().readPair(&tag, &data)) {

    return false;

  if (tag >= SCTAG_TYPED_ARRAY_V1_MIN && tag <= SCTAG_TYPED_ARRAY_V1_MAX) {

    return r->readTypedArray(TagToV1ArrayType(tag), data, vp, true);

  if (tag == SCTAG_TYPED_ARRAY_OBJECT_V2) {

    // V2 stores the length (nelems) in |data| and the arrayType separately.

    uint64_t arrayType;

    if (!r->input().read(&arrayType)) {

      return false;

    uint64_t nelems = data;

    return r->readTypedArray(arrayType, nelems, vp);

  if (tag == SCTAG_TYPED_ARRAY_OBJECT) {

    // The current version stores the array type in |data| and the length

    // (nelems) separately to support large TypedArrays.

    uint32_t arrayType = data;

    uint64_t nelems;

    if (!r->input().read(&nelems)) {

      return false;

    return r->readTypedArray(arrayType, nelems, vp);

  JS_ReportErrorNumberASCII(r->context(), GetErrorMessage, nullptr,

                            JSMSG_SC_BAD_SERIALIZED_DATA,

                            "expected type array");

  return false;

JS_PUBLIC_API bool JS_WriteUint32Pair(JSStructuredCloneWriter* w, uint32_t tag,

                                      uint32_t data) {

  return w->output().writePair(tag, data);

JS_PUBLIC_API bool JS_WriteBytes(JSStructuredCloneWriter* w, const void* p,

                                 size_t len) {

  return w->output().writeBytes(p, len);

JS_PUBLIC_API bool JS_WriteString(JSStructuredCloneWriter* w,

                                  HandleString str) {

  return w->writeString(SCTAG_STRING, str);

JS_PUBLIC_API bool JS_WriteDouble(JSStructuredCloneWriter* w, double v) {

  return w->output().writeDouble(v);

JS_PUBLIC_API bool JS_WriteTypedArray(JSStructuredCloneWriter* w,

                                      HandleValue v) {

  MOZ_ASSERT(v.isObject());

  w->context()->check(v);

  RootedObject obj(w->context(), &v.toObject());

  // startWrite can write everything, thus we should check here

  // and report error if the user passes a wrong type.

  if (!obj->canUnwrapAs<TypedArrayObject>()) {

    ReportAccessDenied(w->context());

    return false;

  // We should use startWrite instead of writeTypedArray, because

  // typed array is an object, we should add it to the |memory|

  // (allObjs) list. Directly calling writeTypedArray won't add it.

  return w->startWrite(v);

JS_PUBLIC_API bool JS_ObjectNotWritten(JSStructuredCloneWriter* w,

                                       HandleObject obj) {

  w->memory.remove(w->memory.lookup(obj));

  return true;

JS_PUBLIC_API JS::StructuredCloneScope JS_GetStructuredCloneScope(

    JSStructuredCloneWriter* w) {

  return w->output().scope();