diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 01:47:29 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 01:47:29 +0000 |
commit | 0ebf5bdf043a27fd3dfb7f92e0cb63d88954c44d (patch) | |
tree | a31f07c9bcca9d56ce61e9a1ffd30ef350d513aa /dom/script/ScriptLoadHandler.cpp | |
parent | Initial commit. (diff) | |
download | firefox-esr-upstream/115.8.0esr.tar.xz firefox-esr-upstream/115.8.0esr.zip |
Adding upstream version 115.8.0esr.upstream/115.8.0esr
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'dom/script/ScriptLoadHandler.cpp')
-rw-r--r-- | dom/script/ScriptLoadHandler.cpp | 474 |
1 files changed, 474 insertions, 0 deletions
diff --git a/dom/script/ScriptLoadHandler.cpp b/dom/script/ScriptLoadHandler.cpp new file mode 100644 index 0000000000..03ff5301c5 --- /dev/null +++ b/dom/script/ScriptLoadHandler.cpp @@ -0,0 +1,474 @@ +/* -*- 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 "ScriptLoadHandler.h" + +#include <stdlib.h> +#include <utility> +#include "ScriptCompression.h" +#include "ScriptLoader.h" +#include "ScriptTrace.h" +#include "js/Transcoding.h" +#include "js/loader/ScriptLoadRequest.h" +#include "mozilla/Assertions.h" +#include "mozilla/CheckedInt.h" +#include "mozilla/DebugOnly.h" +#include "mozilla/Encoding.h" +#include "mozilla/Logging.h" +#include "mozilla/NotNull.h" +#include "mozilla/PerfStats.h" +#include "mozilla/ScopeExit.h" +#include "mozilla/StaticPrefs_dom.h" +#include "mozilla/Utf8.h" +#include "mozilla/Vector.h" +#include "mozilla/dom/Document.h" +#include "mozilla/dom/SRICheck.h" +#include "mozilla/dom/ScriptDecoding.h" +#include "nsCOMPtr.h" +#include "nsContentUtils.h" +#include "nsDebug.h" +#include "nsICacheInfoChannel.h" +#include "nsIChannel.h" +#include "nsIHttpChannel.h" +#include "nsIRequest.h" +#include "nsIScriptElement.h" +#include "nsIURI.h" +#include "nsJSUtils.h" +#include "nsMimeTypes.h" +#include "nsString.h" +#include "nsTArray.h" +#include "zlib.h" + +namespace mozilla::dom { + +#undef LOG +#define LOG(args) \ + MOZ_LOG(ScriptLoader::gScriptLoaderLog, mozilla::LogLevel::Debug, args) + +#define LOG_ENABLED() \ + MOZ_LOG_TEST(ScriptLoader::gScriptLoaderLog, mozilla::LogLevel::Debug) + +ScriptDecoder::ScriptDecoder(const Encoding* aEncoding, + ScriptDecoder::BOMHandling handleBOM) { + if (handleBOM == BOMHandling::Ignore) { + mDecoder = aEncoding->NewDecoderWithoutBOMHandling(); + } else { + mDecoder = aEncoding->NewDecoderWithBOMRemoval(); + } + MOZ_ASSERT(mDecoder); +} + +template <typename Unit> +nsresult ScriptDecoder::DecodeRawDataHelper( + JS::loader::ScriptLoadRequest* aRequest, const uint8_t* aData, + uint32_t aDataLength, bool aEndOfStream) { + CheckedInt<size_t> needed = + ScriptDecoding<Unit>::MaxBufferLength(mDecoder, aDataLength); + if (!needed.isValid()) { + return NS_ERROR_OUT_OF_MEMORY; + } + + // Reference to the script source buffer which we will update. + JS::loader::ScriptLoadRequest::ScriptTextBuffer<Unit>& scriptText = + aRequest->ScriptText<Unit>(); + + uint32_t haveRead = scriptText.length(); + + CheckedInt<uint32_t> capacity = haveRead; + capacity += needed.value(); + + if (!capacity.isValid() || !scriptText.resize(capacity.value())) { + return NS_ERROR_OUT_OF_MEMORY; + } + + size_t written = ScriptDecoding<Unit>::DecodeInto( + mDecoder, Span(aData, aDataLength), + Span(scriptText.begin() + haveRead, needed.value()), aEndOfStream); + MOZ_ASSERT(written <= needed.value()); + + haveRead += written; + MOZ_ASSERT(haveRead <= capacity.value(), + "mDecoder produced more data than expected"); + MOZ_ALWAYS_TRUE(scriptText.resize(haveRead)); + aRequest->mScriptTextLength = scriptText.length(); + + return NS_OK; +} + +nsresult ScriptDecoder::DecodeRawData(JS::loader::ScriptLoadRequest* aRequest, + const uint8_t* aData, + uint32_t aDataLength, bool aEndOfStream) { + if (aRequest->IsUTF16Text()) { + return DecodeRawDataHelper<char16_t>(aRequest, aData, aDataLength, + aEndOfStream); + } + + return DecodeRawDataHelper<Utf8Unit>(aRequest, aData, aDataLength, + aEndOfStream); +} + +ScriptLoadHandler::ScriptLoadHandler( + ScriptLoader* aScriptLoader, JS::loader::ScriptLoadRequest* aRequest, + UniquePtr<SRICheckDataVerifier>&& aSRIDataVerifier) + : mScriptLoader(aScriptLoader), + mRequest(aRequest), + mSRIDataVerifier(std::move(aSRIDataVerifier)), + mSRIStatus(NS_OK) { + MOZ_ASSERT(aRequest->IsUnknownDataType()); + MOZ_ASSERT(aRequest->IsFetching()); +} + +ScriptLoadHandler::~ScriptLoadHandler() = default; + +NS_IMPL_ISUPPORTS(ScriptLoadHandler, nsIIncrementalStreamLoaderObserver) + +NS_IMETHODIMP +ScriptLoadHandler::OnIncrementalData(nsIIncrementalStreamLoader* aLoader, + nsISupports* aContext, + uint32_t aDataLength, const uint8_t* aData, + uint32_t* aConsumedLength) { + nsCOMPtr<nsIRequest> channelRequest; + aLoader->GetRequest(getter_AddRefs(channelRequest)); + + auto firstTime = !mPreloadStartNotified; + if (!mPreloadStartNotified) { + mPreloadStartNotified = true; + mRequest->GetScriptLoadContext()->NotifyStart(channelRequest); + } + + if (mRequest->IsCanceled()) { + // If request cancelled, ignore any incoming data. + *aConsumedLength = aDataLength; + return NS_OK; + } + + nsresult rv = NS_OK; + if (mRequest->IsUnknownDataType()) { + rv = EnsureKnownDataType(aLoader); + NS_ENSURE_SUCCESS(rv, rv); + } + + if (mRequest->IsBytecode() && firstTime) { + PerfStats::RecordMeasurementStart(PerfStats::Metric::JSBC_IO_Read); + } + + if (mRequest->IsTextSource()) { + if (!EnsureDecoder(aLoader, aData, aDataLength, + /* aEndOfStream = */ false)) { + return NS_OK; + } + + // Below we will/shall consume entire data chunk. + *aConsumedLength = aDataLength; + + // Decoder has already been initialized. -- trying to decode all loaded + // bytes. + rv = mDecoder->DecodeRawData(mRequest, aData, aDataLength, + /* aEndOfStream = */ false); + NS_ENSURE_SUCCESS(rv, rv); + + // If SRI is required for this load, appending new bytes to the hash. + if (mSRIDataVerifier && NS_SUCCEEDED(mSRIStatus)) { + mSRIStatus = mSRIDataVerifier->Update(aDataLength, aData); + } + } else { + MOZ_ASSERT(mRequest->IsBytecode()); + if (!mRequest->mScriptBytecode.append(aData, aDataLength)) { + return NS_ERROR_OUT_OF_MEMORY; + } + + *aConsumedLength = aDataLength; + uint32_t sriLength = 0; + rv = MaybeDecodeSRI(&sriLength); + if (NS_FAILED(rv)) { + return channelRequest->Cancel(mScriptLoader->RestartLoad(mRequest)); + } + if (sriLength) { + mRequest->mBytecodeOffset = JS::AlignTranscodingBytecodeOffset(sriLength); + } + } + + return rv; +} + +bool ScriptLoadHandler::TrySetDecoder(nsIIncrementalStreamLoader* aLoader, + const uint8_t* aData, + uint32_t aDataLength, bool aEndOfStream) { + MOZ_ASSERT(mDecoder == nullptr, + "can't have a decoder already if we're trying to set one"); + + // JavaScript modules are always UTF-8. + if (mRequest->IsModuleRequest()) { + mDecoder = MakeUnique<ScriptDecoder>(UTF_8_ENCODING, + ScriptDecoder::BOMHandling::Remove); + return true; + } + + // Determine if BOM check should be done. This occurs either + // if end-of-stream has been reached, or at least 3 bytes have + // been read from input. + if (!aEndOfStream && (aDataLength < 3)) { + return false; + } + + // Do BOM detection. + const Encoding* encoding; + std::tie(encoding, std::ignore) = Encoding::ForBOM(Span(aData, aDataLength)); + if (encoding) { + mDecoder = + MakeUnique<ScriptDecoder>(encoding, ScriptDecoder::BOMHandling::Remove); + return true; + } + + // BOM detection failed, check content stream for charset. + nsCOMPtr<nsIRequest> req; + nsresult rv = aLoader->GetRequest(getter_AddRefs(req)); + NS_ASSERTION(req, "StreamLoader's request went away prematurely"); + NS_ENSURE_SUCCESS(rv, false); + + nsCOMPtr<nsIChannel> channel = do_QueryInterface(req); + + if (channel) { + nsAutoCString label; + if (NS_SUCCEEDED(channel->GetContentCharset(label)) && + (encoding = Encoding::ForLabel(label))) { + mDecoder = MakeUnique<ScriptDecoder>(encoding, + ScriptDecoder::BOMHandling::Ignore); + return true; + } + } + + // Check the hint charset from the script element or preload + // request. + nsAutoString hintCharset; + if (!mRequest->GetScriptLoadContext()->IsPreload()) { + mRequest->GetScriptLoadContext()->GetScriptElement()->GetScriptCharset( + hintCharset); + } else { + nsTArray<ScriptLoader::PreloadInfo>::index_type i = + mScriptLoader->mPreloads.IndexOf( + mRequest, 0, ScriptLoader::PreloadRequestComparator()); + + NS_ASSERTION(i != mScriptLoader->mPreloads.NoIndex, + "Incorrect preload bookkeeping"); + hintCharset = mScriptLoader->mPreloads[i].mCharset; + } + + if ((encoding = Encoding::ForLabel(hintCharset))) { + mDecoder = + MakeUnique<ScriptDecoder>(encoding, ScriptDecoder::BOMHandling::Ignore); + return true; + } + + // Get the charset from the charset of the document. + if (mScriptLoader->mDocument) { + encoding = mScriptLoader->mDocument->GetDocumentCharacterSet(); + mDecoder = + MakeUnique<ScriptDecoder>(encoding, ScriptDecoder::BOMHandling::Ignore); + return true; + } + + // Curiously, there are various callers that don't pass aDocument. The + // fallback in the old code was ISO-8859-1, which behaved like + // windows-1252. + mDecoder = MakeUnique<ScriptDecoder>(WINDOWS_1252_ENCODING, + ScriptDecoder::BOMHandling::Ignore); + return true; +} + +nsresult ScriptLoadHandler::MaybeDecodeSRI(uint32_t* sriLength) { + *sriLength = 0; + + if (!mSRIDataVerifier || mSRIDataVerifier->IsComplete() || + NS_FAILED(mSRIStatus)) { + return NS_OK; + } + + // Skip until the content is large enough to be decoded. + if (mRequest->mScriptBytecode.length() <= + mSRIDataVerifier->DataSummaryLength()) { + return NS_OK; + } + + mSRIStatus = mSRIDataVerifier->ImportDataSummary( + mRequest->mScriptBytecode.length(), mRequest->mScriptBytecode.begin()); + + if (NS_FAILED(mSRIStatus)) { + // We are unable to decode the hash contained in the alternate data which + // contains the bytecode, or it does not use the same algorithm. + LOG( + ("ScriptLoadHandler::MaybeDecodeSRI, failed to decode SRI, restart " + "request")); + return mSRIStatus; + } + + *sriLength = mSRIDataVerifier->DataSummaryLength(); + MOZ_ASSERT(*sriLength > 0); + return NS_OK; +} + +nsresult ScriptLoadHandler::EnsureKnownDataType( + nsIIncrementalStreamLoader* aLoader) { + MOZ_ASSERT(mRequest->IsUnknownDataType()); + MOZ_ASSERT(mRequest->IsFetching()); + + nsCOMPtr<nsIRequest> req; + nsresult rv = aLoader->GetRequest(getter_AddRefs(req)); + MOZ_ASSERT(req, "StreamLoader's request went away prematurely"); + NS_ENSURE_SUCCESS(rv, rv); + + if (mRequest->mFetchSourceOnly) { + mRequest->SetTextSource(); + TRACE_FOR_TEST(mRequest->GetScriptLoadContext()->GetScriptElement(), + "scriptloader_load_source"); + return NS_OK; + } + + nsCOMPtr<nsICacheInfoChannel> cic(do_QueryInterface(req)); + if (cic) { + nsAutoCString altDataType; + cic->GetAlternativeDataType(altDataType); + if (altDataType.Equals(ScriptLoader::BytecodeMimeTypeFor(mRequest))) { + mRequest->SetBytecode(); + TRACE_FOR_TEST(mRequest->GetScriptLoadContext()->GetScriptElement(), + "scriptloader_load_bytecode"); + return NS_OK; + } + MOZ_ASSERT(altDataType.IsEmpty()); + } + + mRequest->SetTextSource(); + TRACE_FOR_TEST(mRequest->GetScriptLoadContext()->GetScriptElement(), + "scriptloader_load_source"); + + MOZ_ASSERT(!mRequest->IsUnknownDataType()); + MOZ_ASSERT(mRequest->IsFetching()); + return NS_OK; +} + +NS_IMETHODIMP +ScriptLoadHandler::OnStreamComplete(nsIIncrementalStreamLoader* aLoader, + nsISupports* aContext, nsresult aStatus, + uint32_t aDataLength, + const uint8_t* aData) { + nsresult rv = NS_OK; + if (LOG_ENABLED()) { + nsAutoCString url; + mRequest->mURI->GetAsciiSpec(url); + LOG(("ScriptLoadRequest (%p): Stream complete (url = %s)", mRequest.get(), + url.get())); + } + + nsCOMPtr<nsIRequest> channelRequest; + aLoader->GetRequest(getter_AddRefs(channelRequest)); + + auto firstMessage = !mPreloadStartNotified; + if (!mPreloadStartNotified) { + mPreloadStartNotified = true; + mRequest->GetScriptLoadContext()->NotifyStart(channelRequest); + } + + auto notifyStop = MakeScopeExit([&] { + mRequest->GetScriptLoadContext()->NotifyStop(channelRequest, rv); + }); + + if (!mRequest->IsCanceled()) { + if (mRequest->IsUnknownDataType()) { + rv = EnsureKnownDataType(aLoader); + NS_ENSURE_SUCCESS(rv, rv); + } + + if (mRequest->IsBytecode() && !firstMessage) { + // if firstMessage, then entire stream is in aData, and PerfStats would + // measure 0 time + PerfStats::RecordMeasurementEnd(PerfStats::Metric::JSBC_IO_Read); + } + + if (mRequest->IsTextSource()) { + DebugOnly<bool> encoderSet = + EnsureDecoder(aLoader, aData, aDataLength, /* aEndOfStream = */ true); + MOZ_ASSERT(encoderSet); + rv = mDecoder->DecodeRawData(mRequest, aData, aDataLength, + /* aEndOfStream = */ true); + NS_ENSURE_SUCCESS(rv, rv); + + LOG(("ScriptLoadRequest (%p): Source length in code units = %u", + mRequest.get(), unsigned(mRequest->ScriptTextLength()))); + + // If SRI is required for this load, appending new bytes to the hash. + if (mSRIDataVerifier && NS_SUCCEEDED(mSRIStatus)) { + mSRIStatus = mSRIDataVerifier->Update(aDataLength, aData); + } + } else { + MOZ_ASSERT(mRequest->IsBytecode()); + if (!mRequest->mScriptBytecode.append(aData, aDataLength)) { + return NS_ERROR_OUT_OF_MEMORY; + } + + LOG(("ScriptLoadRequest (%p): Bytecode length = %u", mRequest.get(), + unsigned(mRequest->mScriptBytecode.length()))); + + // If we abort while decoding the SRI, we fallback on explictly requesting + // the source. Thus, we should not continue in + // ScriptLoader::OnStreamComplete, which removes the request from the + // waiting lists. + // + // We calculate the SRI length below. + uint32_t unused; + rv = MaybeDecodeSRI(&unused); + if (NS_FAILED(rv)) { + return channelRequest->Cancel(mScriptLoader->RestartLoad(mRequest)); + } + + // The bytecode cache always starts with the SRI hash, thus even if there + // is no SRI data verifier instance, we still want to skip the hash. + uint32_t sriLength; + rv = SRICheckDataVerifier::DataSummaryLength( + mRequest->mScriptBytecode.length(), mRequest->mScriptBytecode.begin(), + &sriLength); + if (NS_FAILED(rv)) { + return channelRequest->Cancel(mScriptLoader->RestartLoad(mRequest)); + } + + mRequest->mBytecodeOffset = JS::AlignTranscodingBytecodeOffset(sriLength); + + Vector<uint8_t> compressedBytecode; + // mRequest has the compressed bytecode, but will be filled with the + // uncompressed bytecode + compressedBytecode.swap(mRequest->mScriptBytecode); + if (!JS::loader::ScriptBytecodeDecompress(compressedBytecode, + mRequest->mBytecodeOffset, + mRequest->mScriptBytecode)) { + return NS_ERROR_UNEXPECTED; + } + } + } + + // Everything went well, keep the CacheInfoChannel alive such that we can + // later save the bytecode on the cache entry. + if (NS_SUCCEEDED(rv) && mRequest->IsSource() && + StaticPrefs::dom_script_loader_bytecode_cache_enabled()) { + mRequest->mCacheInfo = do_QueryInterface(channelRequest); + LOG(("ScriptLoadRequest (%p): nsICacheInfoChannel = %p", mRequest.get(), + mRequest->mCacheInfo.get())); + } + + // we have to mediate and use mRequest. + rv = mScriptLoader->OnStreamComplete(aLoader, mRequest, aStatus, mSRIStatus, + mSRIDataVerifier.get()); + + // In case of failure, clear the mCacheInfoChannel to avoid keeping it alive. + if (NS_FAILED(rv)) { + mRequest->mCacheInfo = nullptr; + } + + return rv; +} + +#undef LOG_ENABLED +#undef LOG + +} // namespace mozilla::dom |