mirror of
https://github.com/mozilla/gecko-dev.git
synced 2025-02-27 04:38:02 +00:00

This patch begins re-implements ReadableStreams using WebIDL and DOM technology (vs the existing JS streams implementation). Some more background is [here](https://docs.google.com/document/d/1MWRkF32KV60ngOY-Ip4PnKbCMvl6VK_Y9QLED8MJJxg/edit#) This is guarded under a configure flag `--enable-dom-streams` 1. ByteStreams and ReadableStream.tee will come in future patches. 2. I intentionally crash in other parts of the DOM that require streams (Fetch, Response, Blob), until the integration work is done in future patches. My current plan for that integration doesn't involve re-creating the alternative 'external streams' API from SpiderMonkey's implementation, but I have yet to do enough development to verify that will work. Differential Revision: https://phabricator.services.mozilla.com/D122643
503 lines
16 KiB
C++
503 lines
16 KiB
C++
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
|
|
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
|
|
/* This Source Code Form is subject to the terms of the Mozilla Public
|
|
* License, v. 2.0. If a copy of the MPL was not distributed with this
|
|
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
|
|
|
|
#include "Response.h"
|
|
|
|
#include "nsISupportsImpl.h"
|
|
#include "nsIURI.h"
|
|
#include "nsNetUtil.h"
|
|
#include "nsPIDOMWindow.h"
|
|
#include "mozilla/BasePrincipal.h"
|
|
#include "mozilla/ErrorResult.h"
|
|
#include "mozilla/HoldDropJSObjects.h"
|
|
#include "mozilla/dom/BodyStream.h"
|
|
#include "mozilla/dom/Document.h"
|
|
#include "mozilla/dom/FetchBinding.h"
|
|
#include "mozilla/dom/ResponseBinding.h"
|
|
#include "mozilla/dom/Headers.h"
|
|
#include "mozilla/dom/Promise.h"
|
|
#include "mozilla/dom/URL.h"
|
|
#include "mozilla/dom/WorkerPrivate.h"
|
|
|
|
#include "nsDOMString.h"
|
|
|
|
#include "BodyExtractor.h"
|
|
#include "FetchStreamReader.h"
|
|
#include "InternalResponse.h"
|
|
|
|
namespace mozilla::dom {
|
|
|
|
NS_IMPL_ADDREF_INHERITED(Response, FetchBody<Response>)
|
|
NS_IMPL_RELEASE_INHERITED(Response, FetchBody<Response>)
|
|
|
|
NS_IMPL_CYCLE_COLLECTION_CLASS(Response)
|
|
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK_BEGIN_INHERITED(Response, FetchBody<Response>)
|
|
AbortFollower::Unlink(static_cast<AbortFollower*>(tmp));
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK(mOwner)
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK(mHeaders)
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK(mSignalImpl)
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK(mFetchStreamReader)
|
|
|
|
tmp->mReadableStreamBody = nullptr;
|
|
tmp->mReadableStreamReader = nullptr;
|
|
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK_PRESERVED_WRAPPER
|
|
NS_IMPL_CYCLE_COLLECTION_UNLINK_END
|
|
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE_BEGIN_INHERITED(Response, FetchBody<Response>)
|
|
AbortFollower::Traverse(static_cast<AbortFollower*>(tmp), cb);
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mOwner)
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mHeaders)
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mSignalImpl)
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mFetchStreamReader)
|
|
NS_IMPL_CYCLE_COLLECTION_TRAVERSE_END
|
|
|
|
NS_IMPL_CYCLE_COLLECTION_TRACE_BEGIN_INHERITED(Response, FetchBody<Response>)
|
|
NS_IMPL_CYCLE_COLLECTION_TRACE_JS_MEMBER_CALLBACK(mReadableStreamBody)
|
|
NS_IMPL_CYCLE_COLLECTION_TRACE_JS_MEMBER_CALLBACK(mReadableStreamReader)
|
|
NS_IMPL_CYCLE_COLLECTION_TRACE_PRESERVED_WRAPPER
|
|
NS_IMPL_CYCLE_COLLECTION_TRACE_END
|
|
|
|
NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(Response)
|
|
NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
|
|
NS_INTERFACE_MAP_END_INHERITING(FetchBody<Response>)
|
|
|
|
Response::Response(nsIGlobalObject* aGlobal,
|
|
InternalResponse* aInternalResponse,
|
|
AbortSignalImpl* aSignalImpl)
|
|
: FetchBody<Response>(aGlobal),
|
|
mInternalResponse(aInternalResponse),
|
|
mSignalImpl(aSignalImpl) {
|
|
MOZ_ASSERT(
|
|
aInternalResponse->Headers()->Guard() == HeadersGuardEnum::Immutable ||
|
|
aInternalResponse->Headers()->Guard() == HeadersGuardEnum::Response);
|
|
|
|
mozilla::HoldJSObjects(this);
|
|
}
|
|
|
|
Response::~Response() { mozilla::DropJSObjects(this); }
|
|
|
|
/* static */
|
|
already_AddRefed<Response> Response::Error(const GlobalObject& aGlobal) {
|
|
nsCOMPtr<nsIGlobalObject> global = do_QueryInterface(aGlobal.GetAsSupports());
|
|
RefPtr<InternalResponse> error =
|
|
InternalResponse::NetworkError(NS_ERROR_FAILURE);
|
|
RefPtr<Response> r = new Response(global, error, nullptr);
|
|
return r.forget();
|
|
}
|
|
|
|
/* static */
|
|
already_AddRefed<Response> Response::Redirect(const GlobalObject& aGlobal,
|
|
const nsAString& aUrl,
|
|
uint16_t aStatus,
|
|
ErrorResult& aRv) {
|
|
nsAutoString parsedURL;
|
|
|
|
if (NS_IsMainThread()) {
|
|
nsIURI* baseURI = nullptr;
|
|
nsCOMPtr<nsPIDOMWindowInner> inner(
|
|
do_QueryInterface(aGlobal.GetAsSupports()));
|
|
Document* doc = inner ? inner->GetExtantDoc() : nullptr;
|
|
if (doc) {
|
|
baseURI = doc->GetBaseURI();
|
|
}
|
|
// Don't use NS_ConvertUTF16toUTF8 because that doesn't let us handle OOM.
|
|
nsAutoCString url;
|
|
if (!AppendUTF16toUTF8(aUrl, url, fallible)) {
|
|
aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
|
|
return nullptr;
|
|
}
|
|
|
|
nsCOMPtr<nsIURI> resolvedURI;
|
|
nsresult rv = NS_NewURI(getter_AddRefs(resolvedURI), url, nullptr, baseURI);
|
|
if (NS_WARN_IF(NS_FAILED(rv))) {
|
|
aRv.ThrowTypeError<MSG_INVALID_URL>(url);
|
|
return nullptr;
|
|
}
|
|
|
|
nsAutoCString spec;
|
|
rv = resolvedURI->GetSpec(spec);
|
|
if (NS_WARN_IF(NS_FAILED(rv))) {
|
|
aRv.ThrowTypeError<MSG_INVALID_URL>(url);
|
|
return nullptr;
|
|
}
|
|
|
|
CopyUTF8toUTF16(spec, parsedURL);
|
|
} else {
|
|
WorkerPrivate* worker = GetCurrentThreadWorkerPrivate();
|
|
MOZ_ASSERT(worker);
|
|
worker->AssertIsOnWorkerThread();
|
|
|
|
NS_ConvertUTF8toUTF16 baseURL(worker->GetLocationInfo().mHref);
|
|
RefPtr<URL> url =
|
|
URL::Constructor(aGlobal.GetAsSupports(), aUrl, baseURL, aRv);
|
|
if (aRv.Failed()) {
|
|
return nullptr;
|
|
}
|
|
|
|
url->GetHref(parsedURL);
|
|
}
|
|
|
|
if (aStatus != 301 && aStatus != 302 && aStatus != 303 && aStatus != 307 &&
|
|
aStatus != 308) {
|
|
aRv.ThrowRangeError("Invalid redirect status code.");
|
|
return nullptr;
|
|
}
|
|
|
|
// We can't just pass nullptr for our null-valued Nullable, because the
|
|
// fetch::ResponseBodyInit is a non-temporary type due to the MOZ_RAII
|
|
// annotations on some of its members.
|
|
Nullable<fetch::ResponseBodyInit> body;
|
|
ResponseInit init;
|
|
init.mStatus = aStatus;
|
|
init.mStatusText.AssignASCII("");
|
|
RefPtr<Response> r = Response::Constructor(aGlobal, body, init, aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
|
|
r->GetInternalHeaders()->Set("Location"_ns, NS_ConvertUTF16toUTF8(parsedURL),
|
|
aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
r->GetInternalHeaders()->SetGuard(HeadersGuardEnum::Immutable, aRv);
|
|
MOZ_ASSERT(!aRv.Failed());
|
|
|
|
return r.forget();
|
|
}
|
|
|
|
/*static*/
|
|
already_AddRefed<Response> Response::Constructor(
|
|
const GlobalObject& aGlobal, const Nullable<fetch::ResponseBodyInit>& aBody,
|
|
const ResponseInit& aInit, ErrorResult& aRv) {
|
|
nsCOMPtr<nsIGlobalObject> global = do_QueryInterface(aGlobal.GetAsSupports());
|
|
|
|
if (NS_WARN_IF(!global)) {
|
|
aRv.Throw(NS_ERROR_DOM_INVALID_STATE_ERR);
|
|
return nullptr;
|
|
}
|
|
|
|
if (aInit.mStatus < 200 || aInit.mStatus > 599) {
|
|
aRv.ThrowRangeError("Invalid response status code.");
|
|
return nullptr;
|
|
}
|
|
|
|
// Check if the status text contains illegal characters
|
|
nsACString::const_iterator start, end;
|
|
aInit.mStatusText.BeginReading(start);
|
|
aInit.mStatusText.EndReading(end);
|
|
if (FindCharInReadable('\r', start, end)) {
|
|
aRv.ThrowTypeError<MSG_RESPONSE_INVALID_STATUSTEXT_ERROR>();
|
|
return nullptr;
|
|
}
|
|
// Reset iterator since FindCharInReadable advances it.
|
|
aInit.mStatusText.BeginReading(start);
|
|
if (FindCharInReadable('\n', start, end)) {
|
|
aRv.ThrowTypeError<MSG_RESPONSE_INVALID_STATUSTEXT_ERROR>();
|
|
return nullptr;
|
|
}
|
|
|
|
RefPtr<InternalResponse> internalResponse =
|
|
new InternalResponse(aInit.mStatus, aInit.mStatusText);
|
|
|
|
UniquePtr<mozilla::ipc::PrincipalInfo> principalInfo;
|
|
|
|
// Grab a valid channel info from the global so this response is 'valid' for
|
|
// interception.
|
|
if (NS_IsMainThread()) {
|
|
ChannelInfo info;
|
|
nsCOMPtr<nsPIDOMWindowInner> window = do_QueryInterface(global);
|
|
if (window) {
|
|
Document* doc = window->GetExtantDoc();
|
|
MOZ_ASSERT(doc);
|
|
info.InitFromDocument(doc);
|
|
|
|
principalInfo.reset(new mozilla::ipc::PrincipalInfo());
|
|
nsresult rv =
|
|
PrincipalToPrincipalInfo(doc->NodePrincipal(), principalInfo.get());
|
|
if (NS_WARN_IF(NS_FAILED(rv))) {
|
|
aRv.ThrowTypeError<MSG_FETCH_BODY_CONSUMED_ERROR>();
|
|
return nullptr;
|
|
}
|
|
|
|
internalResponse->InitChannelInfo(info);
|
|
} else if (global->PrincipalOrNull()->IsSystemPrincipal()) {
|
|
info.InitFromChromeGlobal(global);
|
|
|
|
internalResponse->InitChannelInfo(info);
|
|
}
|
|
|
|
/**
|
|
* The channel info is left uninitialized if neither the above `if` nor
|
|
* `else if` statements are executed; this could be because we're in a
|
|
* WebExtensions content script, where the global (i.e. `global`) is a
|
|
* wrapper, and the principal is an expanded principal. In this case,
|
|
* as far as I can tell, there's no way to get the security info, but we'd
|
|
* like the `Response` to be successfully constructed.
|
|
*/
|
|
} else {
|
|
WorkerPrivate* worker = GetCurrentThreadWorkerPrivate();
|
|
MOZ_ASSERT(worker);
|
|
internalResponse->InitChannelInfo(worker->GetChannelInfo());
|
|
principalInfo =
|
|
MakeUnique<mozilla::ipc::PrincipalInfo>(worker->GetPrincipalInfo());
|
|
}
|
|
|
|
internalResponse->SetPrincipalInfo(std::move(principalInfo));
|
|
|
|
RefPtr<Response> r = new Response(global, internalResponse, nullptr);
|
|
|
|
if (aInit.mHeaders.WasPassed()) {
|
|
internalResponse->Headers()->Clear();
|
|
|
|
// Instead of using Fill, create an object to allow the constructor to
|
|
// unwrap the HeadersInit.
|
|
RefPtr<Headers> headers =
|
|
Headers::Create(global, aInit.mHeaders.Value(), aRv);
|
|
if (aRv.Failed()) {
|
|
return nullptr;
|
|
}
|
|
|
|
internalResponse->Headers()->Fill(*headers->GetInternalHeaders(), aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
}
|
|
|
|
if (!aBody.IsNull()) {
|
|
if (aInit.mStatus == 204 || aInit.mStatus == 205 || aInit.mStatus == 304) {
|
|
aRv.ThrowTypeError("Response body is given with a null body status.");
|
|
return nullptr;
|
|
}
|
|
|
|
nsCString contentTypeWithCharset;
|
|
nsCOMPtr<nsIInputStream> bodyStream;
|
|
int64_t bodySize = InternalResponse::UNKNOWN_BODY_SIZE;
|
|
|
|
const fetch::ResponseBodyInit& body = aBody.Value();
|
|
if (body.IsReadableStream()) {
|
|
#ifdef MOZ_DOM_STREAMS
|
|
MOZ_CRASH("MOZ_DOM_STREAMS:NYI");
|
|
#else
|
|
aRv.MightThrowJSException();
|
|
|
|
JSContext* cx = aGlobal.Context();
|
|
const ReadableStream& readableStream = body.GetAsReadableStream();
|
|
|
|
JS::Rooted<JSObject*> readableStreamObj(cx, readableStream.Obj());
|
|
|
|
bool disturbed;
|
|
bool locked;
|
|
if (!JS::ReadableStreamIsDisturbed(cx, readableStreamObj, &disturbed) ||
|
|
!JS::ReadableStreamIsLocked(cx, readableStreamObj, &locked)) {
|
|
aRv.StealExceptionFromJSContext(cx);
|
|
return nullptr;
|
|
}
|
|
if (disturbed || locked) {
|
|
aRv.ThrowTypeError<MSG_FETCH_BODY_CONSUMED_ERROR>();
|
|
return nullptr;
|
|
}
|
|
|
|
r->SetReadableStreamBody(cx, readableStreamObj);
|
|
|
|
JS::ReadableStreamMode streamMode;
|
|
if (!JS::ReadableStreamGetMode(cx, readableStreamObj, &streamMode)) {
|
|
aRv.StealExceptionFromJSContext(cx);
|
|
return nullptr;
|
|
}
|
|
if (streamMode == JS::ReadableStreamMode::ExternalSource) {
|
|
// If this is a DOM generated ReadableStream, we can extract the
|
|
// inputStream directly.
|
|
JS::ReadableStreamUnderlyingSource* underlyingSource = nullptr;
|
|
if (!JS::ReadableStreamGetExternalUnderlyingSource(
|
|
cx, readableStreamObj, &underlyingSource)) {
|
|
aRv.StealExceptionFromJSContext(cx);
|
|
return nullptr;
|
|
}
|
|
|
|
MOZ_ASSERT(underlyingSource);
|
|
|
|
aRv = BodyStream::RetrieveInputStream(underlyingSource,
|
|
getter_AddRefs(bodyStream));
|
|
|
|
// The releasing of the external source is needed in order to avoid an
|
|
// extra stream lock.
|
|
if (!JS::ReadableStreamReleaseExternalUnderlyingSource(
|
|
cx, readableStreamObj)) {
|
|
aRv.StealExceptionFromJSContext(cx);
|
|
return nullptr;
|
|
}
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
} else {
|
|
// If this is a JS-created ReadableStream, let's create a
|
|
// FetchStreamReader.
|
|
aRv = FetchStreamReader::Create(aGlobal.Context(), global,
|
|
getter_AddRefs(r->mFetchStreamReader),
|
|
getter_AddRefs(bodyStream));
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
}
|
|
#endif
|
|
} else {
|
|
uint64_t size = 0;
|
|
aRv = ExtractByteStreamFromBody(body, getter_AddRefs(bodyStream),
|
|
contentTypeWithCharset, size);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
|
|
bodySize = size;
|
|
}
|
|
|
|
internalResponse->SetBody(bodyStream, bodySize);
|
|
|
|
if (!contentTypeWithCharset.IsVoid() &&
|
|
!internalResponse->Headers()->Has("Content-Type"_ns, aRv)) {
|
|
// Ignore Append() failing here.
|
|
ErrorResult error;
|
|
internalResponse->Headers()->Append("Content-Type"_ns,
|
|
contentTypeWithCharset, error);
|
|
error.SuppressException();
|
|
}
|
|
|
|
if (aRv.Failed()) {
|
|
return nullptr;
|
|
}
|
|
}
|
|
|
|
return r.forget();
|
|
}
|
|
|
|
already_AddRefed<Response> Response::Clone(JSContext* aCx, ErrorResult& aRv) {
|
|
bool bodyUsed = GetBodyUsed(aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
|
|
if (!bodyUsed && mReadableStreamBody) {
|
|
aRv.MightThrowJSException();
|
|
|
|
AutoJSAPI jsapi;
|
|
if (!jsapi.Init(mOwner)) {
|
|
aRv.Throw(NS_ERROR_FAILURE);
|
|
return nullptr;
|
|
}
|
|
|
|
JSContext* cx = jsapi.cx();
|
|
JS::Rooted<JSObject*> body(cx, mReadableStreamBody);
|
|
bool locked;
|
|
// We just need to check the 'locked' state because GetBodyUsed() already
|
|
// checked the 'disturbed' state.
|
|
if (!JS::ReadableStreamIsLocked(cx, body, &locked)) {
|
|
aRv.StealExceptionFromJSContext(cx);
|
|
return nullptr;
|
|
}
|
|
|
|
bodyUsed = locked;
|
|
}
|
|
|
|
if (bodyUsed) {
|
|
aRv.ThrowTypeError<MSG_FETCH_BODY_CONSUMED_ERROR>();
|
|
return nullptr;
|
|
}
|
|
|
|
RefPtr<FetchStreamReader> streamReader;
|
|
nsCOMPtr<nsIInputStream> inputStream;
|
|
|
|
JS::Rooted<JSObject*> body(aCx);
|
|
MaybeTeeReadableStreamBody(aCx, &body, getter_AddRefs(streamReader),
|
|
getter_AddRefs(inputStream), aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
|
|
MOZ_ASSERT_IF(body, streamReader);
|
|
MOZ_ASSERT_IF(body, inputStream);
|
|
|
|
RefPtr<InternalResponse> ir =
|
|
mInternalResponse->Clone(body ? InternalResponse::eDontCloneInputStream
|
|
: InternalResponse::eCloneInputStream);
|
|
|
|
RefPtr<Response> response = new Response(mOwner, ir, GetSignalImpl());
|
|
|
|
if (body) {
|
|
// Maybe we have a body, but we receive null from MaybeTeeReadableStreamBody
|
|
// if this body is a native stream. In this case the InternalResponse will
|
|
// have a clone of the native body and the ReadableStream will be created
|
|
// lazily if needed.
|
|
response->SetReadableStreamBody(aCx, body);
|
|
response->mFetchStreamReader = streamReader;
|
|
ir->SetBody(inputStream, InternalResponse::UNKNOWN_BODY_SIZE);
|
|
}
|
|
|
|
return response.forget();
|
|
}
|
|
|
|
already_AddRefed<Response> Response::CloneUnfiltered(JSContext* aCx,
|
|
ErrorResult& aRv) {
|
|
if (GetBodyUsed(aRv)) {
|
|
aRv.ThrowTypeError<MSG_FETCH_BODY_CONSUMED_ERROR>();
|
|
return nullptr;
|
|
}
|
|
|
|
RefPtr<FetchStreamReader> streamReader;
|
|
nsCOMPtr<nsIInputStream> inputStream;
|
|
|
|
JS::Rooted<JSObject*> body(aCx);
|
|
MaybeTeeReadableStreamBody(aCx, &body, getter_AddRefs(streamReader),
|
|
getter_AddRefs(inputStream), aRv);
|
|
if (NS_WARN_IF(aRv.Failed())) {
|
|
return nullptr;
|
|
}
|
|
|
|
MOZ_ASSERT_IF(body, streamReader);
|
|
MOZ_ASSERT_IF(body, inputStream);
|
|
|
|
RefPtr<InternalResponse> clone =
|
|
mInternalResponse->Clone(body ? InternalResponse::eDontCloneInputStream
|
|
: InternalResponse::eCloneInputStream);
|
|
|
|
RefPtr<InternalResponse> ir = clone->Unfiltered();
|
|
RefPtr<Response> ref = new Response(mOwner, ir, GetSignalImpl());
|
|
|
|
if (body) {
|
|
// Maybe we have a body, but we receive null from MaybeTeeReadableStreamBody
|
|
// if this body is a native stream. In this case the InternalResponse will
|
|
// have a clone of the native body and the ReadableStream will be created
|
|
// lazily if needed.
|
|
ref->SetReadableStreamBody(aCx, body);
|
|
ref->mFetchStreamReader = streamReader;
|
|
ir->SetBody(inputStream, InternalResponse::UNKNOWN_BODY_SIZE);
|
|
}
|
|
|
|
return ref.forget();
|
|
}
|
|
|
|
void Response::SetBody(nsIInputStream* aBody, int64_t aBodySize) {
|
|
MOZ_ASSERT(!CheckBodyUsed());
|
|
mInternalResponse->SetBody(aBody, aBodySize);
|
|
}
|
|
|
|
already_AddRefed<InternalResponse> Response::GetInternalResponse() const {
|
|
RefPtr<InternalResponse> ref = mInternalResponse;
|
|
return ref.forget();
|
|
}
|
|
|
|
Headers* Response::Headers_() {
|
|
if (!mHeaders) {
|
|
mHeaders = new Headers(mOwner, mInternalResponse->Headers());
|
|
}
|
|
|
|
return mHeaders;
|
|
}
|
|
|
|
} // namespace mozilla::dom
|