diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 00:47:55 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 00:47:55 +0000 |
commit | 26a029d407be480d791972afb5975cf62c9360a6 (patch) | |
tree | f435a8308119effd964b339f76abb83a57c29483 /netwerk/streamconv/converters/nsMultiMixedConv.cpp | |
parent | Initial commit. (diff) | |
download | firefox-26a029d407be480d791972afb5975cf62c9360a6.tar.xz firefox-26a029d407be480d791972afb5975cf62c9360a6.zip |
Adding upstream version 124.0.1.upstream/124.0.1
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'netwerk/streamconv/converters/nsMultiMixedConv.cpp')
-rw-r--r-- | netwerk/streamconv/converters/nsMultiMixedConv.cpp | 1051 |
1 files changed, 1051 insertions, 0 deletions
diff --git a/netwerk/streamconv/converters/nsMultiMixedConv.cpp b/netwerk/streamconv/converters/nsMultiMixedConv.cpp new file mode 100644 index 0000000000..fb21d6673f --- /dev/null +++ b/netwerk/streamconv/converters/nsMultiMixedConv.cpp @@ -0,0 +1,1051 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* 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 "nsMultiMixedConv.h" +#include "nsIHttpChannel.h" +#include "nsIThreadRetargetableStreamListener.h" +#include "nsNetCID.h" +#include "nsMimeTypes.h" +#include "nsIStringStream.h" +#include "nsCRT.h" +#include "nsIHttpChannelInternal.h" +#include "nsURLHelper.h" +#include "nsIStreamConverterService.h" +#include <algorithm> +#include "nsContentSecurityManager.h" +#include "nsHttp.h" +#include "nsNetUtil.h" +#include "nsIURI.h" +#include "nsHttpHeaderArray.h" +#include "mozilla/AutoRestore.h" +#include "mozilla/Tokenizer.h" +#include "nsComponentManagerUtils.h" +#include "mozilla/StaticPrefs_network.h" + +using namespace mozilla; + +nsPartChannel::nsPartChannel(nsIChannel* aMultipartChannel, uint32_t aPartID, + bool aIsFirstPart, nsIStreamListener* aListener) + : mMultipartChannel(aMultipartChannel), + mListener(aListener), + mPartID(aPartID), + mIsFirstPart(aIsFirstPart) { + // Inherit the load flags from the original channel... + mMultipartChannel->GetLoadFlags(&mLoadFlags); + + mMultipartChannel->GetLoadGroup(getter_AddRefs(mLoadGroup)); +} + +void nsPartChannel::InitializeByteRange(int64_t aStart, int64_t aEnd) { + mIsByteRangeRequest = true; + + mByteRangeStart = aStart; + mByteRangeEnd = aEnd; +} + +nsresult nsPartChannel::SendOnStartRequest(nsISupports* aContext) { + return mListener->OnStartRequest(this); +} + +nsresult nsPartChannel::SendOnDataAvailable(nsISupports* aContext, + nsIInputStream* aStream, + uint64_t aOffset, uint32_t aLen) { + return mListener->OnDataAvailable(this, aStream, aOffset, aLen); +} + +nsresult nsPartChannel::SendOnStopRequest(nsISupports* aContext, + nsresult aStatus) { + // Drop the listener + nsCOMPtr<nsIStreamListener> listener; + listener.swap(mListener); + return listener->OnStopRequest(this, aStatus); +} + +void nsPartChannel::SetContentDisposition( + const nsACString& aContentDispositionHeader) { + mContentDispositionHeader = aContentDispositionHeader; + nsCOMPtr<nsIURI> uri; + GetURI(getter_AddRefs(uri)); + NS_GetFilenameFromDisposition(mContentDispositionFilename, + mContentDispositionHeader); + mContentDisposition = + NS_GetContentDispositionFromHeader(mContentDispositionHeader, this); +} + +// +// nsISupports implementation... +// + +NS_IMPL_ADDREF(nsPartChannel) +NS_IMPL_RELEASE(nsPartChannel) + +NS_INTERFACE_MAP_BEGIN(nsPartChannel) + NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIChannel) + NS_INTERFACE_MAP_ENTRY(nsIRequest) + NS_INTERFACE_MAP_ENTRY(nsIChannel) + NS_INTERFACE_MAP_ENTRY(nsIByteRangeRequest) + NS_INTERFACE_MAP_ENTRY(nsIMultiPartChannel) +NS_INTERFACE_MAP_END + +// +// nsIRequest implementation... +// + +NS_IMETHODIMP +nsPartChannel::GetName(nsACString& aResult) { + return mMultipartChannel->GetName(aResult); +} + +NS_IMETHODIMP +nsPartChannel::IsPending(bool* aResult) { + // For now, consider the active lifetime of each part the same as + // the underlying multipart channel... This is not exactly right, + // but it is good enough :-) + return mMultipartChannel->IsPending(aResult); +} + +NS_IMETHODIMP +nsPartChannel::GetStatus(nsresult* aResult) { + nsresult rv = NS_OK; + + if (NS_FAILED(mStatus)) { + *aResult = mStatus; + } else { + rv = mMultipartChannel->GetStatus(aResult); + } + + return rv; +} + +NS_IMETHODIMP nsPartChannel::SetCanceledReason(const nsACString& aReason) { + return SetCanceledReasonImpl(aReason); +} + +NS_IMETHODIMP nsPartChannel::GetCanceledReason(nsACString& aReason) { + return GetCanceledReasonImpl(aReason); +} + +NS_IMETHODIMP nsPartChannel::CancelWithReason(nsresult aStatus, + const nsACString& aReason) { + return CancelWithReasonImpl(aStatus, aReason); +} + +NS_IMETHODIMP +nsPartChannel::Cancel(nsresult aStatus) { + // Cancelling an individual part must not cancel the underlying + // multipart channel... + // XXX but we should stop sending data for _this_ part channel! + mStatus = aStatus; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetCanceled(bool* aCanceled) { + *aCanceled = NS_FAILED(mStatus); + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::Suspend(void) { + // Suspending an individual part must not suspend the underlying + // multipart channel... + // XXX why not? + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::Resume(void) { + // Resuming an individual part must not resume the underlying + // multipart channel... + // XXX why not? + return NS_OK; +} + +// +// nsIChannel implementation +// + +NS_IMETHODIMP +nsPartChannel::GetOriginalURI(nsIURI** aURI) { + return mMultipartChannel->GetOriginalURI(aURI); +} + +NS_IMETHODIMP +nsPartChannel::SetOriginalURI(nsIURI* aURI) { + return mMultipartChannel->SetOriginalURI(aURI); +} + +NS_IMETHODIMP +nsPartChannel::GetURI(nsIURI** aURI) { return mMultipartChannel->GetURI(aURI); } + +NS_IMETHODIMP +nsPartChannel::Open(nsIInputStream** aStream) { + nsCOMPtr<nsIStreamListener> listener; + nsresult rv = + nsContentSecurityManager::doContentSecurityCheck(this, listener); + NS_ENSURE_SUCCESS(rv, rv); + + // This channel cannot be opened! + return NS_ERROR_FAILURE; +} + +NS_IMETHODIMP +nsPartChannel::AsyncOpen(nsIStreamListener* aListener) { + nsCOMPtr<nsIStreamListener> listener = aListener; + nsresult rv = + nsContentSecurityManager::doContentSecurityCheck(this, listener); + NS_ENSURE_SUCCESS(rv, rv); + + // This channel cannot be opened! + return NS_ERROR_FAILURE; +} + +NS_IMETHODIMP +nsPartChannel::GetLoadFlags(nsLoadFlags* aLoadFlags) { + *aLoadFlags = mLoadFlags; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetLoadFlags(nsLoadFlags aLoadFlags) { + mLoadFlags = aLoadFlags; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetTRRMode(nsIRequest::TRRMode* aTRRMode) { + return GetTRRModeImpl(aTRRMode); +} + +NS_IMETHODIMP +nsPartChannel::SetTRRMode(nsIRequest::TRRMode aTRRMode) { + return SetTRRModeImpl(aTRRMode); +} + +NS_IMETHODIMP +nsPartChannel::GetIsDocument(bool* aIsDocument) { + return NS_GetIsDocumentChannel(this, aIsDocument); +} + +NS_IMETHODIMP +nsPartChannel::GetLoadGroup(nsILoadGroup** aLoadGroup) { + *aLoadGroup = do_AddRef(mLoadGroup).take(); + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetLoadGroup(nsILoadGroup* aLoadGroup) { + mLoadGroup = aLoadGroup; + + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetOwner(nsISupports** aOwner) { + return mMultipartChannel->GetOwner(aOwner); +} + +NS_IMETHODIMP +nsPartChannel::SetOwner(nsISupports* aOwner) { + return mMultipartChannel->SetOwner(aOwner); +} + +NS_IMETHODIMP +nsPartChannel::GetLoadInfo(nsILoadInfo** aLoadInfo) { + return mMultipartChannel->GetLoadInfo(aLoadInfo); +} + +NS_IMETHODIMP +nsPartChannel::SetLoadInfo(nsILoadInfo* aLoadInfo) { + MOZ_RELEASE_ASSERT(aLoadInfo, "loadinfo can't be null"); + return mMultipartChannel->SetLoadInfo(aLoadInfo); +} + +NS_IMETHODIMP +nsPartChannel::GetNotificationCallbacks(nsIInterfaceRequestor** aCallbacks) { + return mMultipartChannel->GetNotificationCallbacks(aCallbacks); +} + +NS_IMETHODIMP +nsPartChannel::SetNotificationCallbacks(nsIInterfaceRequestor* aCallbacks) { + return mMultipartChannel->SetNotificationCallbacks(aCallbacks); +} + +NS_IMETHODIMP +nsPartChannel::GetSecurityInfo(nsITransportSecurityInfo** aSecurityInfo) { + return mMultipartChannel->GetSecurityInfo(aSecurityInfo); +} + +NS_IMETHODIMP +nsPartChannel::GetContentType(nsACString& aContentType) { + aContentType = mContentType; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetContentType(const nsACString& aContentType) { + bool dummy; + net_ParseContentType(aContentType, mContentType, mContentCharset, &dummy); + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetContentCharset(nsACString& aContentCharset) { + aContentCharset = mContentCharset; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetContentCharset(const nsACString& aContentCharset) { + mContentCharset = aContentCharset; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetContentLength(int64_t* aContentLength) { + *aContentLength = mContentLength; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetContentLength(int64_t aContentLength) { + mContentLength = aContentLength; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetContentDisposition(uint32_t* aContentDisposition) { + if (mContentDispositionHeader.IsEmpty()) return NS_ERROR_NOT_AVAILABLE; + + *aContentDisposition = mContentDisposition; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetContentDisposition(uint32_t aContentDisposition) { + return NS_ERROR_NOT_AVAILABLE; +} + +NS_IMETHODIMP +nsPartChannel::GetContentDispositionFilename( + nsAString& aContentDispositionFilename) { + if (mContentDispositionFilename.IsEmpty()) return NS_ERROR_NOT_AVAILABLE; + + aContentDispositionFilename = mContentDispositionFilename; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::SetContentDispositionFilename( + const nsAString& aContentDispositionFilename) { + return NS_ERROR_NOT_AVAILABLE; +} + +NS_IMETHODIMP +nsPartChannel::GetContentDispositionHeader( + nsACString& aContentDispositionHeader) { + if (mContentDispositionHeader.IsEmpty()) return NS_ERROR_NOT_AVAILABLE; + + aContentDispositionHeader = mContentDispositionHeader; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetPartID(uint32_t* aPartID) { + *aPartID = mPartID; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetIsFirstPart(bool* aIsFirstPart) { + *aIsFirstPart = mIsFirstPart; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetIsLastPart(bool* aIsLastPart) { + *aIsLastPart = mIsLastPart; + return NS_OK; +} + +// +// nsIByteRangeRequest implementation... +// + +NS_IMETHODIMP +nsPartChannel::GetIsByteRangeRequest(bool* aIsByteRangeRequest) { + *aIsByteRangeRequest = mIsByteRangeRequest; + + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetStartRange(int64_t* aStartRange) { + *aStartRange = mByteRangeStart; + + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetEndRange(int64_t* aEndRange) { + *aEndRange = mByteRangeEnd; + return NS_OK; +} + +NS_IMETHODIMP +nsPartChannel::GetBaseChannel(nsIChannel** aReturn) { + NS_ENSURE_ARG_POINTER(aReturn); + + *aReturn = do_AddRef(mMultipartChannel).take(); + return NS_OK; +} + +// nsISupports implementation +NS_IMPL_ISUPPORTS(nsMultiMixedConv, nsIStreamConverter, nsIStreamListener, + nsIThreadRetargetableStreamListener, nsIRequestObserver) + +// nsIStreamConverter implementation + +// No syncronous conversion at this time. +NS_IMETHODIMP +nsMultiMixedConv::Convert(nsIInputStream* aFromStream, const char* aFromType, + const char* aToType, nsISupports* aCtxt, + nsIInputStream** _retval) { + return NS_ERROR_NOT_IMPLEMENTED; +} + +// Stream converter service calls this to initialize the actual stream converter +// (us). +NS_IMETHODIMP +nsMultiMixedConv::AsyncConvertData(const char* aFromType, const char* aToType, + nsIStreamListener* aListener, + nsISupports* aCtxt) { + NS_ASSERTION(aListener && aFromType && aToType, + "null pointer passed into multi mixed converter"); + + // hook up our final listener. this guy gets the various On*() calls we want + // to throw at him. + // + // WARNING: this listener must be able to handle multiple OnStartRequest, + // OnDataAvail() and OnStopRequest() call combinations. We call of series + // of these for each sub-part in the raw stream. + mFinalListener = aListener; + + return NS_OK; +} + +NS_IMETHODIMP +nsMultiMixedConv::GetConvertedType(const nsACString& aFromType, + nsIChannel* aChannel, nsACString& aToType) { + return NS_ERROR_NOT_IMPLEMENTED; +} + +// nsIRequestObserver implementation +NS_IMETHODIMP +nsMultiMixedConv::OnStartRequest(nsIRequest* request) { + // we're assuming the content-type is available at this stage + NS_ASSERTION(mBoundary.IsEmpty(), "a second on start???"); + + nsresult rv; + + mTotalSent = 0; + mChannel = do_QueryInterface(request, &rv); + if (NS_FAILED(rv)) return rv; + + nsAutoCString contentType; + + // ask the HTTP channel for the content-type and extract the boundary from it. + nsCOMPtr<nsIHttpChannel> httpChannel = do_QueryInterface(mChannel, &rv); + if (NS_SUCCEEDED(rv)) { + rv = httpChannel->GetResponseHeader("content-type"_ns, contentType); + if (NS_FAILED(rv)) { + return rv; + } + nsCString csp; + rv = httpChannel->GetResponseHeader("content-security-policy"_ns, csp); + if (NS_SUCCEEDED(rv)) { + mRootContentSecurityPolicy = csp; + } + } else { + // try asking the channel directly + rv = mChannel->GetContentType(contentType); + if (NS_FAILED(rv)) { + return NS_ERROR_FAILURE; + } + } + + Tokenizer p(contentType); + p.SkipUntil(Token::Char(';')); + if (!p.CheckChar(';')) { + return NS_ERROR_CORRUPTED_CONTENT; + } + p.SkipWhites(); + if (!p.CheckWord("boundary")) { + return NS_ERROR_CORRUPTED_CONTENT; + } + p.SkipWhites(); + if (!p.CheckChar('=')) { + return NS_ERROR_CORRUPTED_CONTENT; + } + p.SkipWhites(); + Unused << p.ReadUntil(Token::Char(';'), mBoundary); + mBoundary.Trim( + " \""); // ignoring potential quoted string formatting violations + if (mBoundary.IsEmpty()) { + return NS_ERROR_CORRUPTED_CONTENT; + } + + mHeaderTokens[HEADER_CONTENT_TYPE] = mTokenizer.AddCustomToken( + "content-type", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_CONTENT_LENGTH] = mTokenizer.AddCustomToken( + "content-length", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_CONTENT_DISPOSITION] = mTokenizer.AddCustomToken( + "content-disposition", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_SET_COOKIE] = mTokenizer.AddCustomToken( + "set-cookie", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_CONTENT_RANGE] = mTokenizer.AddCustomToken( + "content-range", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_RANGE] = + mTokenizer.AddCustomToken("range", mTokenizer.CASE_INSENSITIVE, false); + mHeaderTokens[HEADER_CONTENT_SECURITY_POLICY] = mTokenizer.AddCustomToken( + "content-security-policy", mTokenizer.CASE_INSENSITIVE, false); + + mLFToken = mTokenizer.AddCustomToken("\n", mTokenizer.CASE_SENSITIVE, false); + mCRLFToken = + mTokenizer.AddCustomToken("\r\n", mTokenizer.CASE_SENSITIVE, false); + + SwitchToControlParsing(); + + mBoundaryToken = + mTokenizer.AddCustomToken(mBoundary, mTokenizer.CASE_SENSITIVE); + mBoundaryTokenWithDashes = + mTokenizer.AddCustomToken("--"_ns + mBoundary, mTokenizer.CASE_SENSITIVE); + + return NS_OK; +} + +// nsIStreamListener implementation +NS_IMETHODIMP +nsMultiMixedConv::OnDataAvailable(nsIRequest* request, nsIInputStream* inStr, + uint64_t sourceOffset, uint32_t count) { + // Failing these assertions may indicate that some of the target listeners of + // this converter is looping the thead queue, which is harmful to how we + // collect the raw (content) data. + MOZ_DIAGNOSTIC_ASSERT(!mInOnDataAvailable, + "nsMultiMixedConv::OnDataAvailable reentered!"); + MOZ_DIAGNOSTIC_ASSERT( + !mRawData, "There are unsent data from the previous tokenizer feed!"); + + if (mInOnDataAvailable) { + // The multipart logic is incapable of being reentered. + return NS_ERROR_UNEXPECTED; + } + + mozilla::AutoRestore<bool> restore(mInOnDataAvailable); + mInOnDataAvailable = true; + + nsresult rv_feed = mTokenizer.FeedInput(inStr, count); + // We must do this every time. Regardless if something has failed during the + // parsing process. Otherwise the raw data reference would not be thrown + // away. + nsresult rv_send = SendData(); + + return NS_FAILED(rv_send) ? rv_send : rv_feed; +} + +NS_IMETHODIMP +nsMultiMixedConv::OnDataFinished(nsresult aStatus) { return NS_OK; } + +NS_IMETHODIMP +nsMultiMixedConv::CheckListenerChain() { return NS_ERROR_NOT_IMPLEMENTED; } + +NS_IMETHODIMP +nsMultiMixedConv::OnStopRequest(nsIRequest* request, nsresult aStatus) { + nsresult rv; + + if (mPartChannel) { + mPartChannel->SetIsLastPart(); + + MOZ_DIAGNOSTIC_ASSERT( + !mRawData, "There are unsent data from the previous tokenizer feed!"); + + rv = mTokenizer.FinishInput(); + if (NS_SUCCEEDED(aStatus)) { + aStatus = rv; + } + rv = SendData(); + if (NS_SUCCEEDED(aStatus)) { + aStatus = rv; + } + + (void)SendStop(aStatus); + } else if (NS_FAILED(aStatus) && !mRequestListenerNotified) { + // underlying data production problem. we should not be in + // the middle of sending data. if we were, mPartChannel, + // above, would have been non-null. + + (void)mFinalListener->OnStartRequest(request); + (void)mFinalListener->OnStopRequest(request, aStatus); + } + + nsCOMPtr<nsIMultiPartChannelListener> multiListener = + do_QueryInterface(mFinalListener); + if (multiListener) { + multiListener->OnAfterLastPart(aStatus); + } + + return NS_OK; +} + +nsresult nsMultiMixedConv::ConsumeToken(Token const& token) { + nsresult rv; + + switch (mParserState) { + case PREAMBLE: + if (token.Equals(mBoundaryTokenWithDashes)) { + // The server first used boundary '--boundary'. Hence, we no longer + // accept plain 'boundary' token as a delimiter. + mTokenizer.RemoveCustomToken(mBoundaryToken); + mParserState = BOUNDARY_CRLF; + break; + } + if (token.Equals(mBoundaryToken)) { + // And here the opposite from the just above block... + mTokenizer.RemoveCustomToken(mBoundaryTokenWithDashes); + mParserState = BOUNDARY_CRLF; + break; + } + + // This is a preamble, just ignore it and wait for the boundary. + break; + + case BOUNDARY_CRLF: + if (token.Equals(Token::NewLine())) { + mParserState = HEADER_NAME; + mResponseHeader = HEADER_UNKNOWN; + HeadersToDefault(); + SetHeaderTokensEnabled(true); + break; + } + return NS_ERROR_CORRUPTED_CONTENT; + + case HEADER_NAME: + SetHeaderTokensEnabled(false); + if (token.Equals(Token::NewLine())) { + mParserState = BODY_INIT; + SwitchToBodyParsing(); + break; + } + for (uint32_t h = HEADER_CONTENT_TYPE; h < HEADER_UNKNOWN; ++h) { + if (token.Equals(mHeaderTokens[h])) { + mResponseHeader = static_cast<EHeader>(h); + break; + } + } + mParserState = HEADER_SEP; + break; + + case HEADER_SEP: + if (token.Equals(Token::Char(':'))) { + mParserState = HEADER_VALUE; + mResponseHeaderValue.Truncate(); + break; + } + if (mResponseHeader == HEADER_UNKNOWN) { + // If the header is not of any we understand, just pass everything till + // ':' + break; + } + if (token.Equals(Token::Whitespace())) { + // Accept only header-name traling whitespaces after known headers + break; + } + return NS_ERROR_CORRUPTED_CONTENT; + + case HEADER_VALUE: + if (token.Equals(Token::Whitespace()) && mResponseHeaderValue.IsEmpty()) { + // Eat leading whitespaces + break; + } + if (token.Equals(Token::NewLine())) { + nsresult rv = ProcessHeader(); + if (NS_FAILED(rv)) { + return rv; + } + mParserState = HEADER_NAME; + mResponseHeader = HEADER_UNKNOWN; + SetHeaderTokensEnabled(true); + } else { + mResponseHeaderValue.Append(token.Fragment()); + } + break; + + case BODY_INIT: + rv = SendStart(); + if (NS_FAILED(rv)) { + return rv; + } + mParserState = BODY; + [[fallthrough]]; + + case BODY: { + if (!token.Equals(mLFToken) && !token.Equals(mCRLFToken)) { + if (token.Equals(mBoundaryTokenWithDashes) || + token.Equals(mBoundaryToken)) { + // Allow CRLF to NOT be part of the boundary as well + SwitchToControlParsing(); + mParserState = TRAIL_DASH1; + break; + } + AccumulateData(token); + break; + } + + // After CRLF we must explicitly check for boundary. If found, + // that CRLF is part of the boundary and must not be send to the + // data listener. + Token token2; + if (!mTokenizer.Next(token2)) { + // Note: this will give us the CRLF token again when more data + // or OnStopRequest arrive. I.e. we will enter BODY case in + // the very same state as we are now and start this block over. + mTokenizer.NeedMoreInput(); + break; + } + if (token2.Equals(mBoundaryTokenWithDashes) || + token2.Equals(mBoundaryToken)) { + SwitchToControlParsing(); + mParserState = TRAIL_DASH1; + break; + } + + AccumulateData(token); + AccumulateData(token2); + break; + } + + case TRAIL_DASH1: + if (token.Equals(Token::NewLine())) { + rv = SendStop(NS_OK); + if (NS_FAILED(rv)) { + return rv; + } + mParserState = BOUNDARY_CRLF; + mTokenizer.Rollback(); + break; + } + if (token.Equals(Token::Char('-'))) { + mParserState = TRAIL_DASH2; + break; + } + return NS_ERROR_CORRUPTED_CONTENT; + + case TRAIL_DASH2: + if (token.Equals(Token::Char('-'))) { + mPartChannel->SetIsLastPart(); + // SendStop calls SendData first. + rv = SendStop(NS_OK); + if (NS_FAILED(rv)) { + return rv; + } + mParserState = EPILOGUE; + break; + } + return NS_ERROR_CORRUPTED_CONTENT; + + case EPILOGUE: + // Just ignore + break; + + default: + MOZ_ASSERT(false, "Missing parser state handling branch"); + break; + } // switch + + return NS_OK; +} + +void nsMultiMixedConv::SetHeaderTokensEnabled(bool aEnable) { + for (uint32_t h = HEADER_FIRST; h < HEADER_UNKNOWN; ++h) { + mTokenizer.EnableCustomToken(mHeaderTokens[h], aEnable); + } +} + +void nsMultiMixedConv::SwitchToBodyParsing() { + mTokenizer.SetTokenizingMode(Tokenizer::Mode::CUSTOM_ONLY); + mTokenizer.EnableCustomToken(mLFToken, true); + mTokenizer.EnableCustomToken(mCRLFToken, true); + mTokenizer.EnableCustomToken(mBoundaryTokenWithDashes, true); + mTokenizer.EnableCustomToken(mBoundaryToken, true); +} + +void nsMultiMixedConv::SwitchToControlParsing() { + mTokenizer.SetTokenizingMode(Tokenizer::Mode::FULL); + mTokenizer.EnableCustomToken(mLFToken, false); + mTokenizer.EnableCustomToken(mCRLFToken, false); + mTokenizer.EnableCustomToken(mBoundaryTokenWithDashes, false); + mTokenizer.EnableCustomToken(mBoundaryToken, false); +} + +// nsMultiMixedConv methods +nsMultiMixedConv::nsMultiMixedConv() + // XXX: This is a hack to bypass the raw pointer to refcounted object in + // lambda analysis. It should be removed and replaced when the + // IncrementalTokenizer API is improved to avoid the need for such + // workarounds. + // + // This is safe because `mTokenizer` will not outlive `this`, meaning + // that this std::bind object will be destroyed before `this` dies. + : mTokenizer(std::bind(&nsMultiMixedConv::ConsumeToken, this, + std::placeholders::_1)) {} + +nsresult nsMultiMixedConv::SendStart() { + nsresult rv = NS_OK; + + nsCOMPtr<nsIStreamListener> partListener(mFinalListener); + if (mContentType.IsEmpty()) { + mContentType.AssignLiteral(UNKNOWN_CONTENT_TYPE); + nsCOMPtr<nsIStreamConverterService> serv = + do_GetService(NS_STREAMCONVERTERSERVICE_CONTRACTID, &rv); + if (NS_SUCCEEDED(rv)) { + nsCOMPtr<nsIStreamListener> converter; + rv = serv->AsyncConvertData(UNKNOWN_CONTENT_TYPE, "*/*", mFinalListener, + mContext, getter_AddRefs(converter)); + if (NS_SUCCEEDED(rv)) { + partListener = converter; + } + } + } + + // if we already have an mPartChannel, that means we never sent a Stop() + // before starting up another "part." that would be bad. + MOZ_ASSERT(!mPartChannel, "tisk tisk, shouldn't be overwriting a channel"); + + nsPartChannel* newChannel; + newChannel = new nsPartChannel(mChannel, mCurrentPartID, mCurrentPartID == 0, + partListener); + + ++mCurrentPartID; + + if (mIsByteRangeRequest) { + newChannel->InitializeByteRange(mByteRangeStart, mByteRangeEnd); + } + + mTotalSent = 0; + + // Set up the new part channel... + mPartChannel = newChannel; + + rv = mPartChannel->SetContentType(mContentType); + if (NS_FAILED(rv)) return rv; + + rv = mPartChannel->SetContentLength(mContentLength); + if (NS_FAILED(rv)) return rv; + + mPartChannel->SetContentDisposition(mContentDisposition); + + // Each part of a multipart/replace response can be used + // for the top level document. We must inform upper layers + // about this by setting the LOAD_REPLACE flag so that certain + // state assertions are evaluated as positive. + nsLoadFlags loadFlags = 0; + mPartChannel->GetLoadFlags(&loadFlags); + loadFlags |= nsIChannel::LOAD_REPLACE; + mPartChannel->SetLoadFlags(loadFlags); + + nsCOMPtr<nsILoadGroup> loadGroup; + (void)mPartChannel->GetLoadGroup(getter_AddRefs(loadGroup)); + + // Add the new channel to the load group (if any) + if (loadGroup) { + rv = loadGroup->AddRequest(mPartChannel, nullptr); + if (NS_FAILED(rv)) return rv; + } + + // This prevents artificial call to OnStart/StopRequest when the root + // channel fails. Since now it's ensured to keep with the nsIStreamListener + // contract every time. + mRequestListenerNotified = true; + + // Let's start off the load. NOTE: we don't forward on the channel passed + // into our OnDataAvailable() as it's the root channel for the raw stream. + return mPartChannel->SendOnStartRequest(mContext); +} + +nsresult nsMultiMixedConv::SendStop(nsresult aStatus) { + // Make sure we send out all accumulcated data prior call to OnStopRequest. + // If there is no data, this is a no-op. + nsresult rv = SendData(); + if (NS_SUCCEEDED(aStatus)) { + aStatus = rv; + } + if (mPartChannel) { + rv = mPartChannel->SendOnStopRequest(mContext, aStatus); + // don't check for failure here, we need to remove the channel from + // the loadgroup. + + // Remove the channel from its load group (if any) + nsCOMPtr<nsILoadGroup> loadGroup; + (void)mPartChannel->GetLoadGroup(getter_AddRefs(loadGroup)); + if (loadGroup) { + (void)loadGroup->RemoveRequest(mPartChannel, mContext, aStatus); + } + } + + mPartChannel = nullptr; + return rv; +} + +void nsMultiMixedConv::AccumulateData(Token const& aToken) { + if (!mRawData) { + // This is the first read of raw data during this FeedInput loop + // of the incremental tokenizer. All 'raw' tokens are coming from + // the same linear buffer, hence begining of this loop raw data + // is begining of the first raw token. Length of this loop raw + // data is just sum of all 'raw' tokens we collect during this loop. + // + // It's ensured we flush (send to to the listener via OnDataAvailable) + // and nullify the collected raw data right after FeedInput call. + // Hence, the reference can't outlive the actual buffer. + mRawData = aToken.Fragment().BeginReading(); + mRawDataLength = 0; + } + + mRawDataLength += aToken.Fragment().Length(); +} + +nsresult nsMultiMixedConv::SendData() { + nsresult rv; + + if (!mRawData) { + return NS_OK; + } + + nsACString::const_char_iterator rawData = mRawData; + mRawData = nullptr; + + if (!mPartChannel) { + return NS_ERROR_FAILURE; // something went wrong w/ processing + } + + if (mContentLength != UINT64_MAX) { + // make sure that we don't send more than the mContentLength + // XXX why? perhaps the Content-Length header was actually wrong!! + if ((uint64_t(mRawDataLength) + mTotalSent) > mContentLength) { + mRawDataLength = static_cast<uint32_t>(mContentLength - mTotalSent); + } + + if (mRawDataLength == 0) return NS_OK; + } + + uint64_t offset = mTotalSent; + mTotalSent += mRawDataLength; + + nsCOMPtr<nsIStringInputStream> ss( + do_CreateInstance("@mozilla.org/io/string-input-stream;1", &rv)); + if (NS_FAILED(rv)) return rv; + + rv = ss->ShareData(rawData, mRawDataLength); + mRawData = nullptr; + if (NS_FAILED(rv)) return rv; + + return mPartChannel->SendOnDataAvailable(mContext, ss, offset, + mRawDataLength); +} + +void nsMultiMixedConv::HeadersToDefault() { + mContentLength = UINT64_MAX; + mContentType.Truncate(); + mContentDisposition.Truncate(); + mContentSecurityPolicy.Truncate(); + mIsByteRangeRequest = false; +} + +nsresult nsMultiMixedConv::ProcessHeader() { + mozilla::Tokenizer p(mResponseHeaderValue); + + switch (mResponseHeader) { + case HEADER_CONTENT_TYPE: + mContentType = mResponseHeaderValue; + mContentType.CompressWhitespace(); + break; + case HEADER_CONTENT_LENGTH: + p.SkipWhites(); + if (!p.ReadInteger(&mContentLength)) { + return NS_ERROR_CORRUPTED_CONTENT; + } + break; + case HEADER_CONTENT_DISPOSITION: + mContentDisposition = mResponseHeaderValue; + mContentDisposition.CompressWhitespace(); + break; + case HEADER_SET_COOKIE: { + nsCOMPtr<nsIHttpChannelInternal> httpInternal = + do_QueryInterface(mChannel); + mResponseHeaderValue.CompressWhitespace(); + if (!StaticPrefs::network_cookie_prevent_set_cookie_from_multipart() && + httpInternal) { + DebugOnly<nsresult> rv = httpInternal->SetCookie(mResponseHeaderValue); + MOZ_ASSERT(NS_SUCCEEDED(rv)); + } + break; + } + case HEADER_RANGE: + case HEADER_CONTENT_RANGE: { + if (!p.CheckWord("bytes") || !p.CheckWhite()) { + return NS_ERROR_CORRUPTED_CONTENT; + } + p.SkipWhites(); + if (p.CheckChar('*')) { + mByteRangeStart = mByteRangeEnd = 0; + } else if (!p.ReadInteger(&mByteRangeStart) || !p.CheckChar('-') || + !p.ReadInteger(&mByteRangeEnd)) { + return NS_ERROR_CORRUPTED_CONTENT; + } + mIsByteRangeRequest = true; + if (mContentLength == UINT64_MAX) { + mContentLength = uint64_t(mByteRangeEnd - mByteRangeStart + 1); + } + break; + } + case HEADER_CONTENT_SECURITY_POLICY: { + mContentSecurityPolicy = mResponseHeaderValue; + mContentSecurityPolicy.CompressWhitespace(); + nsCOMPtr<nsIHttpChannel> httpChannel = do_QueryInterface(mChannel); + if (httpChannel) { + nsCString resultCSP = mRootContentSecurityPolicy; + if (!mContentSecurityPolicy.IsEmpty()) { + // We are updating the root channel CSP header respectively for + // each part as: CSP-root + CSP-partN, where N is the part number. + // Here we append current part's CSP to root CSP and reset CSP + // header for each part. + if (!resultCSP.IsEmpty()) { + resultCSP.Append(";"); + } + resultCSP.Append(mContentSecurityPolicy); + } + nsresult rv = httpChannel->SetResponseHeader( + "Content-Security-Policy"_ns, resultCSP, false); + if (NS_FAILED(rv)) { + return NS_ERROR_CORRUPTED_CONTENT; + } + } + break; + } + case HEADER_UNKNOWN: + // We ignore anything else... + break; + } + + return NS_OK; +} + +nsresult NS_NewMultiMixedConv(nsMultiMixedConv** aMultiMixedConv) { + MOZ_ASSERT(aMultiMixedConv != nullptr, "null ptr"); + + RefPtr<nsMultiMixedConv> conv = new nsMultiMixedConv(); + conv.forget(aMultiMixedConv); + return NS_OK; +} |