Move multipart resource handling to core/fetch (1/2)

third_party/WebKit/Source/core/fetch/MultipartImageResourceParser.cpp

+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "core/fetch/MultipartImageResourceParser.h"
+
+#include "public/platform/Platform.h"
+#include "public/platform/WebURLResponse.h"
+#include "wtf/NotFound.h"
+#include "wtf/StdLibExtras.h"
+#include "wtf/text/WTFString.h"
+
+namespace blink {
+
+namespace {
+
+// The list of response headers that we do not copy from the original
+// response when generating a WebURLResponse for a MIME payload.
+const char* kReplaceHeaders[] = {
+    "content-type",
+    "content-length",
+    "content-disposition",
+    "content-range",
+    "range",
+    "set-cookie"
+};
+
+size_t find(const char* s, size_t sSize, const char* pattern, size_t patternSize)

[hiroshige, 2016/02/25 18:10:24]

I think it's better to use std::search() rather than re-implementing here.

[yhirano, 2016/02/25 18:35:13]

Oh, I didn't know the function. Thank you!

+{
+    // Naive search
+    for (size_t i = 0; i + patternSize <= sSize; ++i) {
+        bool found = true;
+        for (size_t j = 0; found && j < patternSize; ++j)
+            found = (s[i + j] == pattern[j]);
+        if (found)
+            return i;
+    }
+    return kNotFound;
+}
+
+} // namespace
+
+MultipartImageResourceParser::MultipartImageResourceParser(const ResourceResponse& response, const Vector<char>& boundary, Client* client)
+    : m_originalResponse(response)
+    , m_boundary(boundary)
+    , m_client(client)
+{
+    // Some servers report a boundary prefixed with "--".  See https://crbug.com/5786.
+    if (m_boundary.size() < 2 || m_boundary[0] != '-' || m_boundary[1] != '-')
+        m_boundary.prepend("--", 2);
+}
+
+void MultipartImageResourceParser::addData(const char* bytes, size_t size)
+{
+    // m_sawLastBoundary means that we've already received the final boundary
+    // token. The server should stop sending us data at this point, but if it
+    // does, we just throw it away.
+    if (m_sawLastBoundary)
+        return;
+    m_data.append(bytes, size);
+
+    if (m_isParsingTop) {
+        // Eat leading \r\n
+        int pos = pushOverLine(m_data, 0);
+        if (pos)
+            m_data.remove(0, pos);
+
+        if (m_data.size() < m_boundary.size() + 2) {
+            // We don't have enough data yet to make a boundary token.  Just
+            // wait until the next chunk of data arrives.
+            return;
+        }
+
+        // Some servers don't send a boundary token before the first chunk of
+        // data.  We handle this case anyway (Gecko does too).
+        if (0 != memcmp(m_data.data(), m_boundary.data(), m_boundary.size())) {
+            m_data.prepend("\n", 1);
+            const auto& boundary = m_boundary;

[hiroshige, 2016/02/25 18:10:24]

nit: Why don't we do |m_data.prepend(m_boundary);|?

[yhirano, 2016/02/25 18:35:13]

It causes a strange compile error.

[hiroshige, 2016/02/25 21:46:31]

I see. Probably this is WTF::Vector's problem.

+            m_data.prepend(boundary);
+        }
+        m_isParsingTop = false;
+    }
+    ASSERT(!m_isParsingTop);
+
+    // Headers
+    if (m_isParsingHeaders) {
+        // Eat leading \r\n
+        int pos = pushOverLine(m_data, 0);
+        if (pos)
+            m_data.remove(0, pos);
+
+        if (parseHeaders()) {
+            // Successfully parsed headers.
+            m_isParsingHeaders = false;
+        } else {
+            // Get more data before trying again.
+            return;
+        }
+        if (isCancelled())

[hiroshige, 2016/02/25 18:10:24]

Should we place isCancelled() check here?
In the case of isCancelled() in Line 117, it seems to check whether this is
cancelled inside didReceiveData() in Line 116.

In the case of here Line 99, there doesn't seem such things that lead
isCancelled() to be true. If isCancelled() is true here, is it also true from
the beginning of this method?

[hiroshige, 2016/02/25 18:31:05]

Er, I understood, parseHeaders() calls didReceiveResponse().
Then, I think we should rename parseHeaders(), e.g. to
parseHeadersAndNotifyClient(), because parseHeaders() looks like it just parse
headers.

[yhirano, 2016/02/25 18:35:13]

This "parser" class parses incoming data and notifies the client, so I think it
is consistent that a parsing function notifies the client.

+            return;
+    }
+    ASSERT(!m_isParsingHeaders);
+
+    size_t boundaryPosition;
+    while ((boundaryPosition = findBoundary(m_data, &m_boundary)) != kNotFound) {
+        // Strip out trailing \r\n characters in the buffer preceding the
+        // boundary on the same lines as Firefox.
+        size_t dataSize = boundaryPosition;
+        if (boundaryPosition > 0 && m_data[boundaryPosition - 1] == '\n') {
+            dataSize--;
+            if (boundaryPosition > 1 && m_data[boundaryPosition - 2] == '\r') {
+                dataSize--;
+            }
+        }
+        if (dataSize > 0) {
+            m_client->didReceiveData(m_data.data(), dataSize);
+            if (isCancelled())
+                return;
+        }
+        size_t boundaryEndPosition = boundaryPosition + m_boundary.size();
+        if (boundaryEndPosition < m_data.size() && '-' == m_data[boundaryEndPosition]) {
+            // This was the last boundary so we can stop processing.
+            m_sawLastBoundary = true;
+            m_data.clear();
+            return;
+        }
+
+        // We can now throw out data up through the boundary
+        int offset = pushOverLine(m_data, boundaryEndPosition);
+        m_data.remove(0, boundaryEndPosition + offset);
+
+        // Ok, back to parsing headers
+        if (!parseHeaders()) {
+            m_isParsingHeaders = true;
+            break;
+        }
+    }
+
+    // At this point, we should send over any data we have, but keep enough data
+    // buffered to handle a boundary that may have been truncated.
+    if (!m_isParsingHeaders && m_data.size() > m_boundary.size()) {
+        // If the last character is a new line character, go ahead and just send
+        // everything we have buffered.  This matches an optimization in Gecko.
+        size_t sendLength = m_data.size() - m_boundary.size();
+        if (m_data.last() == '\n')
+            sendLength = m_data.size();
+        m_client->didReceiveData(m_data.data(), sendLength);
+        m_data.remove(0, sendLength);
+    }
+}
+
+void MultipartImageResourceParser::finish()
+{
+    ASSERT(!isCancelled());
+    // If we have any pending data and we're not in a header, go ahead and send
+    // it to WebCore.

[hiroshige, 2016/02/25 18:10:24]

nit: s/send it to WebCore/send it to Client/ (or something)?

[yhirano, 2016/02/25 18:35:13]

Done.

+    if (!m_isParsingHeaders && !m_data.isEmpty() && !m_sawLastBoundary)
+        m_client->didReceiveData(m_data.data(), m_data.size());
+    m_data.clear();
+    m_sawLastBoundary = true;
+}
+
+int MultipartImageResourceParser::pushOverLine(const Vector<char>& data, size_t pos)
+{
+    int offset = 0;
+    if (pos < data.size() && (data[pos] == '\r' || data[pos] == '\n')) {
+        ++offset;
+        if (pos + 1 < data.size() && data[pos + 1] == '\n')
+            ++offset;
+    }
+    return offset;
+}
+
+bool MultipartImageResourceParser::parseHeaders()
+{
+    // Create a WebURLResponse based on the original set of headers + the
+    // replacement headers. We only replace the same few headers that gecko
+    // does. See netwerk/streamconv/converters/nsMultiMixedConv.cpp.
+    WebURLResponse response(m_originalResponse.url());
+
+    size_t end = 0;
+    if (!Platform::current()->parseAdditionalHeaders(m_data.data(), m_data.size(), &response, &end))
+        return false;
+    m_data.remove(0, end);
+
+    // Copy the response headers from the original response.
+    for (const auto& header : m_originalResponse.httpHeaderFields()) {
+        bool forbidden = false;
+        for (size_t i = 0; !forbidden && i < WTF_ARRAY_LENGTH(kReplaceHeaders); ++i)
+            forbidden = equalIgnoringCase(header.key, kReplaceHeaders[i]);
+        if (!forbidden)
+            response.addHTTPHeaderField(header.key, header.value);
+    }
+
+    // To avoid recording every multipart load as a separate visit in
+    // the history database, we want to keep track of whether the response
+    // is part of a multipart payload.  We do want to record the first visit,
+    // so we only set isMultipartPayload to true after the first visit.
+    response.setIsMultipartPayload(!m_isFirstPart);
+    m_isFirstPart = false;
+    // Send the response!
+    m_client->didReceiveResponse(response.toResourceResponse());
+
+    return true;
+}
+
+// Boundaries are supposed to be preceeded with --, but it looks like gecko
+// doesn't require the dashes to exist.  See nsMultiMixedConv::FindToken.
+size_t MultipartImageResourceParser::findBoundary(const Vector<char>& data, Vector<char>* boundary)
+{
+    size_t boundaryPosition = find(data.data(), data.size(), boundary->data(), boundary->size());
+    if (boundaryPosition != kNotFound) {
+        // Back up over -- for backwards compat
+        // TODO(tc): Don't we only want to do this once?  Gecko code doesn't
+        // seem to care.
+        if (boundaryPosition >= 2) {
+            if ('-' == data[boundaryPosition - 1] && '-' == data[boundaryPosition - 2]) {
+                boundaryPosition -= 2;
+                Vector<char> v(2, '-');
+                v.appendVector(*boundary);
+                *boundary = v;
+            }
+        }
+    }
+    return boundaryPosition;
+}
+
+DEFINE_TRACE(MultipartImageResourceParser)
+{
+    visitor->trace(m_client);
+}
+
+} // namespace blink