gecko-dev/dom/fetch/FetchUtil.cpp

824 lines
27 KiB
C++
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/* -*- 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 "FetchUtil.h"
#include "zlib.h"
#include "js/friend/ErrorMessages.h" // JSMSG_*
#include "nsCRT.h"
#include "nsError.h"
#include "nsIAsyncInputStream.h"
#include "nsICloneableInputStream.h"
#include "nsIHttpChannel.h"
#include "nsNetUtil.h"
#include "nsStreamUtils.h"
#include "nsString.h"
#include "js/BuildId.h"
#include "mozilla/dom/Document.h"
#include "mozilla/ClearOnShutdown.h"
#include "mozilla/dom/DOMException.h"
#include "mozilla/dom/InternalRequest.h"
#include "mozilla/dom/Response.h"
#include "mozilla/dom/ReferrerInfo.h"
#include "mozilla/dom/WorkerRef.h"
namespace mozilla::dom {
// static
nsresult FetchUtil::GetValidRequestMethod(const nsACString& aMethod,
nsCString& outMethod) {
nsAutoCString upperCaseMethod(aMethod);
ToUpperCase(upperCaseMethod);
if (!NS_IsValidHTTPToken(aMethod)) {
outMethod.SetIsVoid(true);
return NS_ERROR_DOM_SYNTAX_ERR;
}
if (upperCaseMethod.EqualsLiteral("CONNECT") ||
upperCaseMethod.EqualsLiteral("TRACE") ||
upperCaseMethod.EqualsLiteral("TRACK")) {
outMethod.SetIsVoid(true);
return NS_ERROR_DOM_SECURITY_ERR;
}
if (upperCaseMethod.EqualsLiteral("DELETE") ||
upperCaseMethod.EqualsLiteral("GET") ||
upperCaseMethod.EqualsLiteral("HEAD") ||
upperCaseMethod.EqualsLiteral("OPTIONS") ||
upperCaseMethod.EqualsLiteral("POST") ||
upperCaseMethod.EqualsLiteral("PUT")) {
outMethod = upperCaseMethod;
} else {
outMethod = aMethod; // Case unchanged for non-standard methods
}
return NS_OK;
}
static bool FindCRLF(nsACString::const_iterator& aStart,
nsACString::const_iterator& aEnd) {
nsACString::const_iterator end(aEnd);
return FindInReadable("\r\n"_ns, aStart, end);
}
// Reads over a CRLF and positions start after it.
static bool PushOverLine(nsACString::const_iterator& aStart,
const nsACString::const_iterator& aEnd) {
if (*aStart == nsCRT::CR && (aEnd - aStart > 1) && *(++aStart) == nsCRT::LF) {
++aStart; // advance to after CRLF
return true;
}
return false;
}
// static
bool FetchUtil::IncrementPendingKeepaliveRequestSize(
nsILoadGroup* aLoadGroup, const uint64_t aBodyLength) {
uint64_t pendingKeepaliveRequestSize = 0;
MOZ_ASSERT(aLoadGroup);
aLoadGroup->GetTotalKeepAliveBytes(&pendingKeepaliveRequestSize);
pendingKeepaliveRequestSize += aBodyLength;
if (pendingKeepaliveRequestSize > FETCH_KEEPALIVE_MAX_SIZE) {
return false;
}
aLoadGroup->SetTotalKeepAliveBytes(pendingKeepaliveRequestSize);
return true;
}
// static
void FetchUtil::DecrementPendingKeepaliveRequestSize(
nsILoadGroup* aLoadGroup, const uint64_t aBodyLength) {
MOZ_ASSERT(aLoadGroup);
uint64_t pendingKeepaliveRequestSize = 0;
aLoadGroup->GetTotalKeepAliveBytes(&pendingKeepaliveRequestSize);
MOZ_ASSERT(pendingKeepaliveRequestSize >= aBodyLength);
pendingKeepaliveRequestSize -= aBodyLength;
aLoadGroup->SetTotalKeepAliveBytes(pendingKeepaliveRequestSize);
}
// static
nsCOMPtr<nsILoadGroup> FetchUtil::GetLoadGroupFromGlobal(
nsIGlobalObject* aGlobalObject) {
MOZ_ASSERT(NS_IsMainThread());
nsCOMPtr<nsILoadGroup> loadGroup = nullptr;
auto* innerWindow = aGlobalObject->GetAsInnerWindow();
if (innerWindow) {
Document* doc = innerWindow->GetExtantDoc();
if (doc) {
loadGroup = doc->GetDocumentLoadGroup();
}
}
return loadGroup;
}
// static
bool FetchUtil::ExtractHeader(nsACString::const_iterator& aStart,
nsACString::const_iterator& aEnd,
nsCString& aHeaderName, nsCString& aHeaderValue,
bool* aWasEmptyHeader) {
MOZ_ASSERT(aWasEmptyHeader);
// Set it to a valid value here so we don't forget later.
*aWasEmptyHeader = false;
const char* beginning = aStart.get();
nsACString::const_iterator end(aEnd);
if (!FindCRLF(aStart, end)) {
return false;
}
if (aStart.get() == beginning) {
*aWasEmptyHeader = true;
return true;
}
nsAutoCString header(beginning, aStart.get() - beginning);
nsACString::const_iterator headerStart, iter, headerEnd;
header.BeginReading(headerStart);
header.EndReading(headerEnd);
iter = headerStart;
if (!FindCharInReadable(':', iter, headerEnd)) {
return false;
}
aHeaderName.Assign(StringHead(header, iter - headerStart));
aHeaderName.CompressWhitespace();
if (!NS_IsValidHTTPToken(aHeaderName)) {
return false;
}
aHeaderValue.Assign(Substring(++iter, headerEnd));
if (!NS_IsReasonableHTTPHeaderValue(aHeaderValue)) {
return false;
}
aHeaderValue.CompressWhitespace();
return PushOverLine(aStart, aEnd);
}
// static
nsresult FetchUtil::SetRequestReferrer(nsIPrincipal* aPrincipal, Document* aDoc,
nsIHttpChannel* aChannel,
InternalRequest& aRequest) {
MOZ_ASSERT(NS_IsMainThread());
nsresult rv = NS_OK;
nsAutoCString referrer;
aRequest.GetReferrer(referrer);
ReferrerPolicy policy = aRequest.ReferrerPolicy_();
nsCOMPtr<nsIReferrerInfo> referrerInfo;
if (referrer.IsEmpty()) {
// This is the case requests referrer is "no-referrer"
referrerInfo = new ReferrerInfo(nullptr, ReferrerPolicy::No_referrer);
} else if (referrer.EqualsLiteral(kFETCH_CLIENT_REFERRER_STR)) {
referrerInfo = ReferrerInfo::CreateForFetch(aPrincipal, aDoc);
// In the first step, we should use referrer info from requetInit
referrerInfo = static_cast<ReferrerInfo*>(referrerInfo.get())
->CloneWithNewPolicy(policy);
} else {
// From "Determine request's Referrer" step 3
// "If request's referrer is a URL, let referrerSource be request's
// referrer."
nsCOMPtr<nsIURI> referrerURI;
rv = NS_NewURI(getter_AddRefs(referrerURI), referrer);
NS_ENSURE_SUCCESS(rv, rv);
referrerInfo = new ReferrerInfo(referrerURI, policy);
}
rv = aChannel->SetReferrerInfoWithoutClone(referrerInfo);
NS_ENSURE_SUCCESS(rv, rv);
nsAutoCString computedReferrerSpec;
referrerInfo = aChannel->GetReferrerInfo();
if (referrerInfo) {
Unused << referrerInfo->GetComputedReferrerSpec(computedReferrerSpec);
}
// Step 8 https://fetch.spec.whatwg.org/#main-fetch
// If requests referrer is not "no-referrer", set requests referrer to
// the result of invoking determine requests referrer.
aRequest.SetReferrer(computedReferrerSpec);
return NS_OK;
}
class StoreOptimizedEncodingRunnable final : public Runnable {
nsMainThreadPtrHandle<nsICacheInfoChannel> mCache;
Vector<uint8_t> mBytes;
public:
StoreOptimizedEncodingRunnable(
nsMainThreadPtrHandle<nsICacheInfoChannel>&& aCache,
Vector<uint8_t>&& aBytes)
: Runnable("StoreOptimizedEncodingRunnable"),
mCache(std::move(aCache)),
mBytes(std::move(aBytes)) {}
NS_IMETHOD Run() override {
nsresult rv;
nsCOMPtr<nsIAsyncOutputStream> stream;
rv = mCache->OpenAlternativeOutputStream(FetchUtil::WasmAltDataType,
int64_t(mBytes.length()),
getter_AddRefs(stream));
if (NS_FAILED(rv)) {
return rv;
}
auto closeStream = MakeScopeExit([&]() { stream->CloseWithStatus(rv); });
uint32_t written;
rv = stream->Write((char*)mBytes.begin(), mBytes.length(), &written);
if (NS_FAILED(rv)) {
return rv;
}
MOZ_RELEASE_ASSERT(mBytes.length() == written);
return NS_OK;
};
};
class WindowStreamOwner final : public GlobalTeardownObserver {
private:
// Read from any thread but only set/cleared on the main thread. The lifecycle
// of WindowStreamOwner prevents concurrent read/clear.
nsCOMPtr<nsIAsyncInputStream> mStream;
~WindowStreamOwner() { MOZ_ASSERT(NS_IsMainThread()); }
public:
NS_DECL_ISUPPORTS
WindowStreamOwner(nsIAsyncInputStream* aStream, nsIGlobalObject* aGlobal)
: GlobalTeardownObserver(aGlobal), mStream(aStream) {
MOZ_DIAGNOSTIC_ASSERT(aGlobal);
MOZ_ASSERT(NS_IsMainThread());
}
// GlobalTeardownObserver:
void DisconnectFromOwner() override {
MOZ_ASSERT(NS_IsMainThread());
if (!mStream) {
return;
}
// mStream->Close() will call JSStreamConsumer::OnInputStreamReady which may
// then destory itself, but GTO should be strongly grabbing us right as it's
// calling DisconnectFromOwner.
mStream->Close();
mStream = nullptr;
GlobalTeardownObserver::DisconnectFromOwner();
}
};
NS_IMPL_ISUPPORTS0(WindowStreamOwner)
inline nsISupports* ToSupports(WindowStreamOwner* aObj) {
return static_cast<GlobalTeardownObserver*>(aObj);
}
class WorkerStreamOwner final {
public:
NS_INLINE_DECL_REFCOUNTING(WorkerStreamOwner)
explicit WorkerStreamOwner(nsIAsyncInputStream* aStream,
nsCOMPtr<nsIEventTarget>&& target)
: mStream(aStream), mOwningEventTarget(std::move(target)) {}
static already_AddRefed<WorkerStreamOwner> Create(
nsIAsyncInputStream* aStream, WorkerPrivate* aWorker,
nsCOMPtr<nsIEventTarget>&& target) {
RefPtr<WorkerStreamOwner> self =
new WorkerStreamOwner(aStream, std::move(target));
self->mWorkerRef =
StrongWorkerRef::Create(aWorker, "JSStreamConsumer", [self]() {
if (self->mStream) {
// If this Close() calls JSStreamConsumer::OnInputStreamReady and
// drops the last reference to the JSStreamConsumer, 'this' will not
// be destroyed since ~JSStreamConsumer() only enqueues a release
// proxy.
self->mStream->Close();
self->mStream = nullptr;
}
});
if (!self->mWorkerRef) {
return nullptr;
}
return self.forget();
}
static void ProxyRelease(already_AddRefed<WorkerStreamOwner> aDoomed) {
RefPtr<WorkerStreamOwner> doomed = aDoomed;
nsIEventTarget* target = doomed->mOwningEventTarget;
NS_ProxyRelease("WorkerStreamOwner", target, doomed.forget(),
/* aAlwaysProxy = */ true);
}
private:
~WorkerStreamOwner() = default;
// Read from any thread but only set/cleared on the worker thread. The
// lifecycle of WorkerStreamOwner prevents concurrent read/clear.
nsCOMPtr<nsIAsyncInputStream> mStream;
RefPtr<StrongWorkerRef> mWorkerRef;
nsCOMPtr<nsIEventTarget> mOwningEventTarget;
};
class JSStreamConsumer final : public nsIInputStreamCallback,
public JS::OptimizedEncodingListener {
// A LengthPrefixType is stored at the start of the compressed optimized
// encoding, allowing the decompressed buffer to be allocated to exactly
// the right size.
using LengthPrefixType = uint32_t;
static const unsigned PrefixBytes = sizeof(LengthPrefixType);
RefPtr<WindowStreamOwner> mWindowStreamOwner;
RefPtr<WorkerStreamOwner> mWorkerStreamOwner;
nsMainThreadPtrHandle<nsICacheInfoChannel> mCache;
const bool mOptimizedEncoding;
z_stream mZStream;
bool mZStreamInitialized;
Vector<uint8_t> mOptimizedEncodingBytes;
JS::StreamConsumer* mConsumer;
bool mConsumerAborted;
JSStreamConsumer(already_AddRefed<WindowStreamOwner> aWindowStreamOwner,
nsIGlobalObject* aGlobal, JS::StreamConsumer* aConsumer,
nsMainThreadPtrHandle<nsICacheInfoChannel>&& aCache,
bool aOptimizedEncoding)
: mWindowStreamOwner(aWindowStreamOwner),
mCache(std::move(aCache)),
mOptimizedEncoding(aOptimizedEncoding),
mZStreamInitialized(false),
mConsumer(aConsumer),
mConsumerAborted(false) {
MOZ_DIAGNOSTIC_ASSERT(mWindowStreamOwner);
MOZ_DIAGNOSTIC_ASSERT(mConsumer);
}
JSStreamConsumer(RefPtr<WorkerStreamOwner> aWorkerStreamOwner,
nsIGlobalObject* aGlobal, JS::StreamConsumer* aConsumer,
nsMainThreadPtrHandle<nsICacheInfoChannel>&& aCache,
bool aOptimizedEncoding)
: mWorkerStreamOwner(std::move(aWorkerStreamOwner)),
mCache(std::move(aCache)),
mOptimizedEncoding(aOptimizedEncoding),
mZStreamInitialized(false),
mConsumer(aConsumer),
mConsumerAborted(false) {
MOZ_DIAGNOSTIC_ASSERT(mWorkerStreamOwner);
MOZ_DIAGNOSTIC_ASSERT(mConsumer);
}
~JSStreamConsumer() {
if (mZStreamInitialized) {
inflateEnd(&mZStream);
}
// Both WindowStreamOwner and WorkerStreamOwner need to be destroyed on
// their global's event target thread.
if (mWindowStreamOwner) {
MOZ_DIAGNOSTIC_ASSERT(!mWorkerStreamOwner);
NS_ReleaseOnMainThread("JSStreamConsumer::mWindowStreamOwner",
mWindowStreamOwner.forget(),
/* aAlwaysProxy = */ true);
} else {
MOZ_DIAGNOSTIC_ASSERT(mWorkerStreamOwner);
WorkerStreamOwner::ProxyRelease(mWorkerStreamOwner.forget());
}
// Bug 1733674: these annotations currently do nothing, because they are
// member variables and the annotation mechanism only applies to locals. But
// the analysis could be extended so that these could replace the big-hammer
// ~JSStreamConsumer annotation and thus the analysis could check that
// nothing is added that might GC for a different reason.
JS_HAZ_VALUE_IS_GC_SAFE(mWindowStreamOwner);
JS_HAZ_VALUE_IS_GC_SAFE(mWorkerStreamOwner);
}
static nsresult WriteSegment(nsIInputStream* aStream, void* aClosure,
const char* aFromSegment, uint32_t aToOffset,
uint32_t aCount, uint32_t* aWriteCount) {
JSStreamConsumer* self = reinterpret_cast<JSStreamConsumer*>(aClosure);
MOZ_DIAGNOSTIC_ASSERT(!self->mConsumerAborted);
if (self->mOptimizedEncoding) {
if (!self->mZStreamInitialized) {
// mOptimizedEncodingBytes is used as temporary storage until we have
// the full prefix.
MOZ_ASSERT(self->mOptimizedEncodingBytes.length() < PrefixBytes);
uint32_t remain = PrefixBytes - self->mOptimizedEncodingBytes.length();
uint32_t consume = std::min(remain, aCount);
if (!self->mOptimizedEncodingBytes.append(aFromSegment, consume)) {
return NS_ERROR_UNEXPECTED;
}
if (consume == remain) {
// Initialize zlib once all prefix bytes are loaded.
LengthPrefixType length;
memcpy(&length, self->mOptimizedEncodingBytes.begin(), PrefixBytes);
if (!self->mOptimizedEncodingBytes.resizeUninitialized(length)) {
return NS_ERROR_UNEXPECTED;
}
memset(&self->mZStream, 0, sizeof(self->mZStream));
self->mZStream.avail_out = length;
self->mZStream.next_out = self->mOptimizedEncodingBytes.begin();
if (inflateInit(&self->mZStream) != Z_OK) {
return NS_ERROR_UNEXPECTED;
}
self->mZStreamInitialized = true;
}
*aWriteCount = consume;
return NS_OK;
}
// Zlib is initialized, overwrite the prefix with the inflated data.
MOZ_DIAGNOSTIC_ASSERT(aCount > 0);
self->mZStream.avail_in = aCount;
self->mZStream.next_in = (uint8_t*)aFromSegment;
int ret = inflate(&self->mZStream, Z_NO_FLUSH);
MOZ_DIAGNOSTIC_ASSERT(ret == Z_OK || ret == Z_STREAM_END,
"corrupt optimized wasm cache file: data");
MOZ_DIAGNOSTIC_ASSERT(self->mZStream.avail_in == 0,
"corrupt optimized wasm cache file: input");
MOZ_DIAGNOSTIC_ASSERT_IF(ret == Z_STREAM_END,
self->mZStream.avail_out == 0);
// Gracefully handle corruption in release.
bool ok =
(ret == Z_OK || ret == Z_STREAM_END) && self->mZStream.avail_in == 0;
if (!ok) {
return NS_ERROR_UNEXPECTED;
}
} else {
// This callback can be called on any thread which is explicitly allowed
// by this particular JS API call.
if (!self->mConsumer->consumeChunk((const uint8_t*)aFromSegment,
aCount)) {
self->mConsumerAborted = true;
return NS_ERROR_UNEXPECTED;
}
}
*aWriteCount = aCount;
return NS_OK;
}
public:
NS_DECL_THREADSAFE_ISUPPORTS
static bool Start(nsCOMPtr<nsIInputStream> aStream, nsIGlobalObject* aGlobal,
WorkerPrivate* aMaybeWorker, JS::StreamConsumer* aConsumer,
nsMainThreadPtrHandle<nsICacheInfoChannel>&& aCache,
bool aOptimizedEncoding) {
nsCOMPtr<nsIAsyncInputStream> asyncStream;
nsresult rv = NS_MakeAsyncNonBlockingInputStream(
aStream.forget(), getter_AddRefs(asyncStream));
if (NS_WARN_IF(NS_FAILED(rv))) {
return false;
}
RefPtr<JSStreamConsumer> consumer;
if (aMaybeWorker) {
RefPtr<WorkerStreamOwner> owner = WorkerStreamOwner::Create(
asyncStream, aMaybeWorker, aGlobal->SerialEventTarget());
if (!owner) {
return false;
}
consumer = new JSStreamConsumer(std::move(owner), aGlobal, aConsumer,
std::move(aCache), aOptimizedEncoding);
} else {
RefPtr<WindowStreamOwner> owner =
new WindowStreamOwner(asyncStream, aGlobal);
if (!owner) {
return false;
}
consumer = new JSStreamConsumer(owner.forget(), aGlobal, aConsumer,
std::move(aCache), aOptimizedEncoding);
}
// This AsyncWait() creates a ref-cycle between asyncStream and consumer:
//
// asyncStream -> consumer -> (Window|Worker)StreamOwner -> asyncStream
//
// The cycle is broken when the stream completes or errors out and
// asyncStream drops its reference to consumer.
return NS_SUCCEEDED(asyncStream->AsyncWait(consumer, 0, 0, nullptr));
}
// nsIInputStreamCallback:
NS_IMETHOD
OnInputStreamReady(nsIAsyncInputStream* aStream) override {
// Can be called on any stream. The JS API calls made below explicitly
// support being called from any thread.
MOZ_DIAGNOSTIC_ASSERT(!mConsumerAborted);
nsresult rv;
uint64_t available = 0;
rv = aStream->Available(&available);
if (NS_SUCCEEDED(rv) && available == 0) {
rv = NS_BASE_STREAM_CLOSED;
}
if (rv == NS_BASE_STREAM_CLOSED) {
if (mOptimizedEncoding) {
// Gracefully handle corruption of compressed data stream in release.
// From on investigations in bug 1738987, the incomplete data cases
// mostly happen during shutdown. Some corruptions in the cache entry
// can still happen and will be handled in the WriteSegment above.
bool ok = mZStreamInitialized && mZStream.avail_out == 0;
if (!ok) {
mConsumer->streamError(size_t(NS_ERROR_UNEXPECTED));
return NS_OK;
}
mConsumer->consumeOptimizedEncoding(mOptimizedEncodingBytes.begin(),
mOptimizedEncodingBytes.length());
} else {
// If there is cache entry associated with this stream, then listen for
// an optimized encoding so we can store it in the alt data. By JS API
// contract, the compilation process will hold a refcount to 'this'
// until it's done, optionally calling storeOptimizedEncoding().
mConsumer->streamEnd(mCache ? this : nullptr);
}
return NS_OK;
}
if (NS_FAILED(rv)) {
mConsumer->streamError(size_t(rv));
return NS_OK;
}
// Check mConsumerAborted before NS_FAILED to avoid calling streamError()
// if consumeChunk() returned false per JS API contract.
uint32_t written = 0;
rv = aStream->ReadSegments(WriteSegment, this, available, &written);
if (mConsumerAborted) {
return NS_OK;
}
if (NS_WARN_IF(NS_FAILED(rv))) {
mConsumer->streamError(size_t(rv));
return NS_OK;
}
rv = aStream->AsyncWait(this, 0, 0, nullptr);
if (NS_WARN_IF(NS_FAILED(rv))) {
mConsumer->streamError(size_t(rv));
return NS_OK;
}
return NS_OK;
}
// JS::OptimizedEncodingListener
void storeOptimizedEncoding(const uint8_t* aSrcBytes,
size_t aSrcLength) override {
MOZ_ASSERT(mCache, "we only listen if there's a cache entry");
z_stream zstream;
memset(&zstream, 0, sizeof(zstream));
zstream.avail_in = aSrcLength;
zstream.next_in = (uint8_t*)aSrcBytes;
// The wins from increasing compression levels are tiny, while the time
// to compress increases drastically. For example, for a 148mb alt-data
// produced by a 40mb .wasm file, the level 2 takes 2.5s to get a 3.7x size
// reduction while level 9 takes 22.5s to get a 4x size reduction. Read-time
// wins from smaller compressed cache files are not found to be
// significant, thus the fastest compression level is used. (On test
// workloads, level 2 actually was faster *and* smaller than level 1.)
const int COMPRESSION = 2;
if (deflateInit(&zstream, COMPRESSION) != Z_OK) {
return;
}
auto autoDestroy = MakeScopeExit([&]() { deflateEnd(&zstream); });
Vector<uint8_t> dstBytes;
if (!dstBytes.resizeUninitialized(PrefixBytes +
deflateBound(&zstream, aSrcLength))) {
return;
}
MOZ_RELEASE_ASSERT(LengthPrefixType(aSrcLength) == aSrcLength);
LengthPrefixType srcLength = aSrcLength;
memcpy(dstBytes.begin(), &srcLength, PrefixBytes);
uint8_t* compressBegin = dstBytes.begin() + PrefixBytes;
zstream.next_out = compressBegin;
zstream.avail_out = dstBytes.length() - PrefixBytes;
int ret = deflate(&zstream, Z_FINISH);
if (ret == Z_MEM_ERROR) {
return;
}
MOZ_RELEASE_ASSERT(ret == Z_STREAM_END);
dstBytes.shrinkTo(zstream.next_out - dstBytes.begin());
NS_DispatchToMainThread(new StoreOptimizedEncodingRunnable(
std::move(mCache), std::move(dstBytes)));
}
};
NS_IMPL_ISUPPORTS(JSStreamConsumer, nsIInputStreamCallback)
// static
const nsCString FetchUtil::WasmAltDataType;
// static
void FetchUtil::InitWasmAltDataType() {
nsCString& type = const_cast<nsCString&>(WasmAltDataType);
MOZ_ASSERT(type.IsEmpty());
RunOnShutdown([]() {
// Avoid StringBuffer leak tests failures.
const_cast<nsCString&>(WasmAltDataType).Truncate();
});
type.Append(nsLiteralCString("wasm-"));
JS::BuildIdCharVector buildId;
if (!JS::GetOptimizedEncodingBuildId(&buildId)) {
MOZ_CRASH("build id oom");
}
type.Append(buildId.begin(), buildId.length());
}
static bool ThrowException(JSContext* aCx, unsigned errorNumber) {
JS_ReportErrorNumberASCII(aCx, js::GetErrorMessage, nullptr, errorNumber);
return false;
}
// static
bool FetchUtil::StreamResponseToJS(JSContext* aCx, JS::Handle<JSObject*> aObj,
JS::MimeType aMimeType,
JS::StreamConsumer* aConsumer,
WorkerPrivate* aMaybeWorker) {
MOZ_ASSERT(!WasmAltDataType.IsEmpty());
MOZ_ASSERT(!aMaybeWorker == NS_IsMainThread());
RefPtr<Response> response;
nsresult rv = UNWRAP_OBJECT(Response, aObj, response);
if (NS_FAILED(rv)) {
return ThrowException(aCx, JSMSG_WASM_BAD_RESPONSE_VALUE);
}
const char* requiredMimeType = nullptr;
switch (aMimeType) {
case JS::MimeType::Wasm:
requiredMimeType = WASM_CONTENT_TYPE;
break;
}
nsAutoCString mimeType;
nsAutoCString mixedCaseMimeType; // unused
response->GetMimeType(mimeType, mixedCaseMimeType);
if (!mimeType.EqualsASCII(requiredMimeType)) {
JS_ReportErrorNumberASCII(aCx, js::GetErrorMessage, nullptr,
JSMSG_WASM_BAD_RESPONSE_MIME_TYPE, mimeType.get(),
requiredMimeType);
return false;
}
if (response->Type() != ResponseType::Basic &&
response->Type() != ResponseType::Cors &&
response->Type() != ResponseType::Default) {
return ThrowException(aCx, JSMSG_WASM_BAD_RESPONSE_CORS_SAME_ORIGIN);
}
if (!response->Ok()) {
return ThrowException(aCx, JSMSG_WASM_BAD_RESPONSE_STATUS);
}
if (response->BodyUsed()) {
return ThrowException(aCx, JSMSG_WASM_RESPONSE_ALREADY_CONSUMED);
}
switch (aMimeType) {
case JS::MimeType::Wasm:
nsAutoCString url;
response->GetUrl(url);
IgnoredErrorResult result;
nsCString sourceMapUrl;
response->GetInternalHeaders()->Get("SourceMap"_ns, sourceMapUrl, result);
if (NS_WARN_IF(result.Failed())) {
return ThrowException(aCx, JSMSG_WASM_ERROR_CONSUMING_RESPONSE);
}
aConsumer->noteResponseURLs(
url.get(), sourceMapUrl.IsVoid() ? nullptr : sourceMapUrl.get());
break;
}
SafeRefPtr<InternalResponse> ir = response->GetInternalResponse();
if (NS_WARN_IF(!ir)) {
return ThrowException(aCx, JSMSG_OUT_OF_MEMORY);
}
nsCOMPtr<nsIInputStream> stream;
nsMainThreadPtrHandle<nsICacheInfoChannel> cache;
bool optimizedEncoding = false;
if (ir->HasCacheInfoChannel()) {
cache = ir->TakeCacheInfoChannel();
nsAutoCString altDataType;
if (NS_SUCCEEDED(cache->GetAlternativeDataType(altDataType)) &&
WasmAltDataType.Equals(altDataType)) {
optimizedEncoding = true;
rv = cache->GetAlternativeDataInputStream(getter_AddRefs(stream));
if (NS_WARN_IF(NS_FAILED(rv))) {
return ThrowException(aCx, JSMSG_OUT_OF_MEMORY);
}
if (ir->HasBeenCloned()) {
// If `Response` is cloned, clone alternative data stream instance.
// The cache entry does not clone automatically, and multiple
// JSStreamConsumer instances will collide during read if not cloned.
nsCOMPtr<nsICloneableInputStream> original = do_QueryInterface(stream);
if (NS_WARN_IF(!original)) {
return ThrowException(aCx, JSMSG_OUT_OF_MEMORY);
}
rv = original->Clone(getter_AddRefs(stream));
if (NS_WARN_IF(NS_FAILED(rv))) {
return ThrowException(aCx, JSMSG_OUT_OF_MEMORY);
}
}
}
}
if (!optimizedEncoding) {
ir->GetUnfilteredBody(getter_AddRefs(stream));
if (!stream) {
aConsumer->streamEnd();
return true;
}
}
MOZ_ASSERT(stream);
IgnoredErrorResult error;
response->SetBodyUsed(aCx, error);
if (NS_WARN_IF(error.Failed())) {
return ThrowException(aCx, JSMSG_WASM_ERROR_CONSUMING_RESPONSE);
}
nsIGlobalObject* global = xpc::NativeGlobal(js::UncheckedUnwrap(aObj));
if (!JSStreamConsumer::Start(stream, global, aMaybeWorker, aConsumer,
std::move(cache), optimizedEncoding)) {
return ThrowException(aCx, JSMSG_OUT_OF_MEMORY);
}
return true;
}
// static
void FetchUtil::ReportJSStreamError(JSContext* aCx, size_t aErrorCode) {
// For now, convert *all* errors into AbortError.
RefPtr<DOMException> e = DOMException::Create(NS_ERROR_DOM_ABORT_ERR);
JS::Rooted<JS::Value> value(aCx);
if (!GetOrCreateDOMReflector(aCx, e, &value)) {
return;
}
JS_SetPendingException(aCx, value);
}
} // namespace mozilla::dom