gecko-dev/browser/components/feeds/nsFeedSniffer.cpp

364 строки
11 KiB
C++

/* -*- Mode: C++; tab-width: 8; 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 "nsFeedSniffer.h"
#include "nsNetCID.h"
#include "nsXPCOM.h"
#include "nsCOMPtr.h"
#include "nsStringStream.h"
#include "nsBrowserCompsCID.h"
#include "nsICategoryManager.h"
#include "nsIServiceManager.h"
#include "nsComponentManagerUtils.h"
#include "nsServiceManagerUtils.h"
#include "nsIStreamConverterService.h"
#include "nsIStreamConverter.h"
#include "nsIStreamListener.h"
#include "nsIHttpChannel.h"
#include "nsIMIMEHeaderParam.h"
#include "nsMimeTypes.h"
#include "nsIURI.h"
#include <algorithm>
#define TYPE_ATOM "application/atom+xml"
#define TYPE_RSS "application/rss+xml"
#define TYPE_MAYBE_FEED "application/vnd.mozilla.maybe.feed"
#define NS_RDF "http://www.w3.org/1999/02/22-rdf-syntax-ns#"
#define NS_RSS "http://purl.org/rss/1.0/"
#define MAX_BYTES 512u
NS_IMPL_ISUPPORTS(nsFeedSniffer,
nsIContentSniffer,
nsIStreamListener,
nsIRequestObserver)
nsresult
nsFeedSniffer::ConvertEncodedData(nsIRequest* request,
const uint8_t* data,
uint32_t length)
{
nsresult rv = NS_OK;
mDecodedData = "";
nsCOMPtr<nsIHttpChannel> httpChannel(do_QueryInterface(request));
if (!httpChannel)
return NS_ERROR_NO_INTERFACE;
nsAutoCString contentEncoding;
httpChannel->GetResponseHeader(NS_LITERAL_CSTRING("Content-Encoding"),
contentEncoding);
if (!contentEncoding.IsEmpty()) {
nsCOMPtr<nsIStreamConverterService> converterService(do_GetService(NS_STREAMCONVERTERSERVICE_CONTRACTID));
if (converterService) {
ToLowerCase(contentEncoding);
nsCOMPtr<nsIStreamListener> converter;
rv = converterService->AsyncConvertData(contentEncoding.get(),
"uncompressed", this, nullptr,
getter_AddRefs(converter));
NS_ENSURE_SUCCESS(rv, rv);
converter->OnStartRequest(request, nullptr);
nsCOMPtr<nsIStringInputStream> rawStream =
do_CreateInstance(NS_STRINGINPUTSTREAM_CONTRACTID);
if (!rawStream)
return NS_ERROR_FAILURE;
rv = rawStream->SetData((const char*)data, length);
NS_ENSURE_SUCCESS(rv, rv);
rv = converter->OnDataAvailable(request, nullptr, rawStream, 0, length);
NS_ENSURE_SUCCESS(rv, rv);
converter->OnStopRequest(request, nullptr, NS_OK);
}
}
return rv;
}
template<int N>
static bool
StringBeginsWithLowercaseLiteral(nsAString& aString,
const char (&aSubstring)[N])
{
return StringHead(aString, N).LowerCaseEqualsLiteral(aSubstring);
}
bool
HasAttachmentDisposition(nsIHttpChannel* httpChannel)
{
if (!httpChannel)
return false;
uint32_t disp;
nsresult rv = httpChannel->GetContentDisposition(&disp);
if (NS_SUCCEEDED(rv) && disp == nsIChannel::DISPOSITION_ATTACHMENT)
return true;
return false;
}
/**
* @return the first occurrence of a character within a string buffer,
* or nullptr if not found
*/
static const char*
FindChar(char c, const char *begin, const char *end)
{
for (; begin < end; ++begin) {
if (*begin == c)
return begin;
}
return nullptr;
}
/**
*
* Determine if a substring is the "documentElement" in the document.
*
* All of our sniffed substrings: <rss, <feed, <rdf:RDF must be the "document"
* element within the XML DOM, i.e. the root container element. Otherwise,
* it's possible that someone embedded one of these tags inside a document of
* another type, e.g. a HTML document, and we don't want to show the preview
* page if the document isn't actually a feed.
*
* @param start
* The beginning of the data being sniffed
* @param end
* The end of the data being sniffed, right before the substring that
* was found.
* @returns true if the found substring is the documentElement, false
* otherwise.
*/
static bool
IsDocumentElement(const char *start, const char* end)
{
// For every tag in the buffer, check to see if it's a PI, Doctype or
// comment, our desired substring or something invalid.
while ( (start = FindChar('<', start, end)) ) {
++start;
if (start >= end)
return false;
// Check to see if the character following the '<' is either '?' or '!'
// (processing instruction or doctype or comment)... these are valid nodes
// to have in the prologue.
if (*start != '?' && *start != '!')
return false;
// Now advance the iterator until the '>' (We do this because we don't want
// to sniff indicator substrings that are embedded within other nodes, e.g.
// comments: <!-- <rdf:RDF .. > -->
start = FindChar('>', start, end);
if (!start)
return false;
++start;
}
return true;
}
/**
* Determines whether or not a string exists as the root element in an XML data
* string buffer.
* @param dataString
* The data being sniffed
* @param substring
* The substring being tested for existence and root-ness.
* @returns true if the substring exists and is the documentElement, false
* otherwise.
*/
static bool
ContainsTopLevelSubstring(nsACString& dataString, const char *substring)
{
int32_t offset = dataString.Find(substring);
if (offset == -1)
return false;
const char *begin = dataString.BeginReading();
// Only do the validation when we find the substring.
return IsDocumentElement(begin, begin + offset);
}
NS_IMETHODIMP
nsFeedSniffer::GetMIMETypeFromContent(nsIRequest* request,
const uint8_t* data,
uint32_t length,
nsACString& sniffedType)
{
nsCOMPtr<nsIHttpChannel> channel(do_QueryInterface(request));
if (!channel)
return NS_ERROR_NO_INTERFACE;
// Check that this is a GET request, since you can't subscribe to a POST...
nsAutoCString method;
channel->GetRequestMethod(method);
if (!method.EqualsLiteral("GET")) {
sniffedType.Truncate();
return NS_OK;
}
// We need to find out if this is a load of a view-source document. In this
// case we do not want to override the content type, since the source display
// does not need to be converted from feed format to XUL. More importantly,
// we don't want to change the content type from something
// nsContentDLF::CreateInstance knows about (e.g. application/xml, text/html
// etc) to something that only the application fe knows about (maybe.feed)
// thus deactivating syntax highlighting.
nsCOMPtr<nsIURI> originalURI;
channel->GetOriginalURI(getter_AddRefs(originalURI));
nsAutoCString scheme;
originalURI->GetScheme(scheme);
if (scheme.EqualsLiteral("view-source")) {
sniffedType.Truncate();
return NS_OK;
}
// Check the Content-Type to see if it is set correctly. If it is set to
// something specific that we think is a reliable indication of a feed, don't
// bother sniffing since we assume the site maintainer knows what they're
// doing.
nsAutoCString contentType;
channel->GetContentType(contentType);
bool noSniff = contentType.EqualsLiteral(TYPE_RSS) ||
contentType.EqualsLiteral(TYPE_ATOM);
// Check to see if this was a feed request from the location bar or from
// the feed: protocol. This is also a reliable indication.
// The value of the header doesn't matter.
if (!noSniff) {
nsAutoCString sniffHeader;
nsresult foundHeader =
channel->GetRequestHeader(NS_LITERAL_CSTRING("X-Moz-Is-Feed"),
sniffHeader);
noSniff = NS_SUCCEEDED(foundHeader);
}
if (noSniff) {
// check for an attachment after we have a likely feed.
if(HasAttachmentDisposition(channel)) {
sniffedType.Truncate();
return NS_OK;
}
// set the feed header as a response header, since we have good metadata
// telling us that the feed is supposed to be RSS or Atom
channel->SetResponseHeader(NS_LITERAL_CSTRING("X-Moz-Is-Feed"),
NS_LITERAL_CSTRING("1"), false);
sniffedType.AssignLiteral(TYPE_MAYBE_FEED);
return NS_OK;
}
// Don't sniff arbitrary types. Limit sniffing to situations that
// we think can reasonably arise.
if (!contentType.EqualsLiteral(TEXT_HTML) &&
!contentType.EqualsLiteral(APPLICATION_OCTET_STREAM) &&
// Same criterion as XMLHttpRequest. Should we be checking for "+xml"
// and check for text/xml and application/xml by hand instead?
contentType.Find("xml") == -1) {
sniffedType.Truncate();
return NS_OK;
}
// Now we need to potentially decompress data served with
// Content-Encoding: gzip
nsresult rv = ConvertEncodedData(request, data, length);
if (NS_FAILED(rv))
return rv;
// We cap the number of bytes to scan at MAX_BYTES to prevent picking up
// false positives by accidentally reading document content, e.g. a "how to
// make a feed" page.
const char* testData;
if (mDecodedData.IsEmpty()) {
testData = (const char*)data;
length = std::min(length, MAX_BYTES);
} else {
testData = mDecodedData.get();
length = std::min(mDecodedData.Length(), MAX_BYTES);
}
// The strategy here is based on that described in:
// http://blogs.msdn.com/rssteam/articles/PublishersGuide.aspx
// for interoperarbility purposes.
// Thus begins the actual sniffing.
nsDependentCSubstring dataString((const char*)testData, length);
bool isFeed = false;
// RSS 0.91/0.92/2.0
isFeed = ContainsTopLevelSubstring(dataString, "<rss");
// Atom 1.0
if (!isFeed)
isFeed = ContainsTopLevelSubstring(dataString, "<feed");
// RSS 1.0
if (!isFeed) {
isFeed = ContainsTopLevelSubstring(dataString, "<rdf:RDF") &&
dataString.Find(NS_RDF) != -1 &&
dataString.Find(NS_RSS) != -1;
}
// If we sniffed a feed, coerce our internal type
if (isFeed && !HasAttachmentDisposition(channel))
sniffedType.AssignLiteral(TYPE_MAYBE_FEED);
else
sniffedType.Truncate();
return NS_OK;
}
NS_IMETHODIMP
nsFeedSniffer::OnStartRequest(nsIRequest* request, nsISupports* context)
{
return NS_OK;
}
NS_METHOD
nsFeedSniffer::AppendSegmentToString(nsIInputStream* inputStream,
void* closure,
const char* rawSegment,
uint32_t toOffset,
uint32_t count,
uint32_t* writeCount)
{
nsCString* decodedData = static_cast<nsCString*>(closure);
decodedData->Append(rawSegment, count);
*writeCount = count;
return NS_OK;
}
NS_IMETHODIMP
nsFeedSniffer::OnDataAvailable(nsIRequest* request, nsISupports* context,
nsIInputStream* stream, uint64_t offset,
uint32_t count)
{
uint32_t read;
return stream->ReadSegments(AppendSegmentToString, &mDecodedData, count,
&read);
}
NS_IMETHODIMP
nsFeedSniffer::OnStopRequest(nsIRequest* request, nsISupports* context,
nsresult status)
{
return NS_OK;
}