[utils] Rework URL path munging for ., .. components

* move processing to YoutubeDLHandler * also process `Location` header for redirect * use tests from https://github.com/yt-dlp/yt-dlp/pull/7662
[utils] Rework decoding of Content-Encodings
2024-12-20 23:52:07 +00:00 · 2023-07-29 14:27:26 +01:00 · 2023-07-29 14:27:26 +01:00 · 2023-07-29 14:27:26 +01:00
6 changed files with 203 additions and 85 deletions
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@ -278,7 +278,7 @@ jobs:
    #-------- Jython ------
    - name: Set up Java 8
      if: ${{ matrix.python-impl == 'jython' }}
-      uses: actions/setup-java@v2
+      uses: actions/setup-java@v3
      with:
        java-version: 8
        distribution: 'zulu'
--- a/test/test_http.py
+++ b/test/test_http.py
@ -180,6 +180,12 @@ class HTTPTestRequestHandler(compat_http_server.BaseHTTPRequestHandler):
            respond()
        elif self.path == '/%c7%9f':
            respond()
        elif self.path == '/redirect_dotsegments':
            self.send_response(301)
            # redirect to /headers but with dot segments before
            self.send_header('Location', '/a/b/./../../headers')
            self.send_header('Content-Length', '0')
            self.end_headers()
        elif self.path.startswith('/redirect_'):
            self._redirect()
        elif self.path.startswith('/method'):
@ -461,33 +467,23 @@ class TestHTTP(unittest.TestCase):
                sanitized_Request(
                    self._test_url('content-encoding'),
                    headers={'ytdl-encoding': encoding}))
-            self.assertEqual(res.headers.get('Content-Encoding'), encoding)
+            # decoded encodings are removed: only check for valid decompressed data
            self.assertEqual(res.read(), b'<html><video src="/vid.mp4" /></html>')
    @unittest.skipUnless(brotli, 'brotli support is not installed')
    @unittest.expectedFailure
    def test_brotli(self):
        self.__test_compression('br')
    @unittest.expectedFailure
    def test_deflate(self):
        self.__test_compression('deflate')
    @unittest.expectedFailure
    def test_gzip(self):
        self.__test_compression('gzip')
    @unittest.expectedFailure  # not yet implemented
    def test_multiple_encodings(self):
        # https://www.rfc-editor.org/rfc/rfc9110.html#section-8.4
        with FakeYDL() as ydl:
        for pair in ('gzip,deflate', 'deflate, gzip', 'gzip, gzip', 'deflate, deflate'):
-                res = ydl.urlopen(
+            self.__test_compression(pair)
                    sanitized_Request(
                        self._test_url('content-encoding'),
                        headers={'ytdl-encoding': pair}))
                self.assertEqual(res.headers.get('Content-Encoding'), pair)
                self.assertEqual(res.read(), b'<html><video src="/vid.mp4" /></html>')
    def test_unsupported_encoding(self):
        # it should return the raw content
@ -499,6 +495,14 @@ class TestHTTP(unittest.TestCase):
            self.assertEqual(res.headers.get('Content-Encoding'), 'unsupported')
            self.assertEqual(res.read(), b'raw')
    def test_remove_dot_segments(self):
        with FakeYDL() as ydl:
            res = ydl.urlopen(sanitized_Request(self._test_url('a/b/./../../headers')))
            self.assertEqual(compat_urllib_parse.urlparse(res.geturl()).path, '/headers')
            res = ydl.urlopen(sanitized_Request(self._test_url('redirect_dotsegments')))
            self.assertEqual(compat_urllib_parse.urlparse(res.geturl()).path, '/headers')
 def _build_proxy_handler(name):
    class HTTPTestRequestHandler(compat_http_server.BaseHTTPRequestHandler):
--- a/test/test_utils.py
+++ b/test/test_utils.py
@ -64,6 +64,7 @@ from youtube_dl.utils import (
    parse_age_limit,
    parse_duration,
    parse_filesize,
    parse_codecs,
    parse_count,
    parse_iso8601,
    parse_resolution,
@ -114,7 +115,7 @@ from youtube_dl.utils import (
    cli_option,
    cli_valueless_option,
    cli_bool_option,
-    parse_codecs,
+    YoutubeDLHandler,
 )
 from youtube_dl.compat import (
    compat_chr,
@ -905,6 +906,32 @@ class TestUtil(unittest.TestCase):
        )
        self.assertEqual(escape_url('http://vimeo.com/56015672#at=0'), 'http://vimeo.com/56015672#at=0')
    def test_remove_dot_segments(self):
        def remove_dot_segments(p):
            q = '' if p.startswith('/') else '/'
            p = 'http://example.com' + q + p
            p = compat_urlparse.urlsplit(YoutubeDLHandler._fix_path(p)).path
            return p[1:] if q else p
        self.assertEqual(remove_dot_segments('/a/b/c/./../../g'), '/a/g')
        self.assertEqual(remove_dot_segments('mid/content=5/../6'), 'mid/6')
        self.assertEqual(remove_dot_segments('/ad/../cd'), '/cd')
        self.assertEqual(remove_dot_segments('/ad/../cd/'), '/cd/')
        self.assertEqual(remove_dot_segments('/..'), '/')
        self.assertEqual(remove_dot_segments('/./'), '/')
        self.assertEqual(remove_dot_segments('/./a'), '/a')
        self.assertEqual(remove_dot_segments('/abc/./.././d/././e/.././f/./../../ghi'), '/ghi')
        self.assertEqual(remove_dot_segments('/'), '/')
        self.assertEqual(remove_dot_segments('/t'), '/t')
        self.assertEqual(remove_dot_segments('t'), 't')
        self.assertEqual(remove_dot_segments(''), '')
        self.assertEqual(remove_dot_segments('/../a/b/c'), '/a/b/c')
        self.assertEqual(remove_dot_segments('../a'), 'a')
        self.assertEqual(remove_dot_segments('./a'), 'a')
        self.assertEqual(remove_dot_segments('.'), '')
        self.assertEqual(remove_dot_segments('////'), '////')
    def test_js_to_json_vars_strings(self):
        self.assertDictEqual(
            json.loads(js_to_json(
--- a/youtube_dl/YoutubeDL.py
+++ b/youtube_dl/YoutubeDL.py
@ -71,7 +71,6 @@ from .utils import (
    format_bytes,
    formatSeconds,
    GeoRestrictedError,
    HEADRequest,
    int_or_none,
    ISO3166Utils,
    join_nonempty,
@ -88,7 +87,6 @@ from .utils import (
    preferredencoding,
    prepend_extension,
    process_communicate_or_kill,
    PUTRequest,
    register_socks_protocols,
    render_table,
    replace_extension,
@ -2460,27 +2458,6 @@ class YoutubeDL(object):
        """ Start an HTTP download """
        if isinstance(req, compat_basestring):
            req = sanitized_Request(req)
        # an embedded /../ sequence is not automatically handled by urllib2
        # see https://github.com/yt-dlp/yt-dlp/issues/3355
        url = req.get_full_url()
        parts = url.partition('/../')
        if parts[1]:
            url = compat_urllib_parse.urljoin(parts[0] + parts[1][:1], parts[1][1:] + parts[2])
        if url:
            # worse, URL path may have initial /../ against RFCs: work-around
            # by stripping such prefixes, like eg Firefox
            parts = compat_urllib_parse.urlsplit(url)
            path = parts.path
            while path.startswith('/../'):
                path = path[3:]
            url = parts._replace(path=path).geturl()
            # get a new Request with the munged URL
            if url != req.get_full_url():
                req_type = {'HEAD': HEADRequest, 'PUT': PUTRequest}.get(
                    req.get_method(), compat_urllib_request.Request)
                req = req_type(
                    url, data=req.data, headers=dict(req.header_items()),
                    origin_req_host=req.origin_req_host, unverifiable=req.unverifiable)
        return self._opener.open(req, timeout=self._socket_timeout)
    def print_debug_header(self):
--- a/youtube_dl/compat.py
+++ b/youtube_dl/compat.py
@ -3200,6 +3200,18 @@ except AttributeError:
    def compat_datetime_timedelta_total_seconds(td):
        return (td.microseconds + (td.seconds + td.days * 24 * 3600) * 10**6) / 10**6
 # optional decompression packages
 # PyPi brotli package implements 'br' Content-Encoding
 try:
    import brotli as compat_brotli
 except ImportError:
    compat_brotli = None
 # PyPi ncompress package implements 'compress' Content-Encoding
 try:
    import ncompress as compat_ncompress
 except ImportError:
    compat_ncompress = None
 legacy = [
    'compat_HTMLParseError',
@ -3234,6 +3246,7 @@ __all__ = [
    'compat_Struct',
    'compat_base64_b64decode',
    'compat_basestring',
    'compat_brotli',
    'compat_casefold',
    'compat_chr',
    'compat_collections_abc',
@ -3259,6 +3272,7 @@ __all__ = [
    'compat_itertools_zip_longest',
    'compat_kwargs',
    'compat_map',
    'compat_ncompress',
    'compat_numeric_types',
    'compat_open',
    'compat_ord',
--- a/youtube_dl/utils.py
+++ b/youtube_dl/utils.py
@ -15,7 +15,6 @@ import email.utils
 import email.header
 import errno
 import functools
 import gzip
 import inspect
 import io
 import itertools
@ -42,6 +41,7 @@ from .compat import (
    compat_HTMLParseError,
    compat_HTMLParser,
    compat_basestring,
    compat_brotli as brotli,
    compat_casefold,
    compat_chr,
    compat_collections_abc,
@ -55,6 +55,7 @@ from .compat import (
    compat_http_client,
    compat_integer_types,
    compat_kwargs,
    compat_ncompress as ncompress,
    compat_os_name,
    compat_re_Match,
    compat_re_Pattern,
@ -2638,23 +2639,91 @@ class YoutubeDLHandler(compat_urllib_request.HTTPHandler):
            req)
    @staticmethod
-    def deflate(data):
+    def deflate_gz(data):
        try:
-            return zlib.decompress(data, -zlib.MAX_WBITS)
+            # format:zlib,gzip + windowsize:32768
            return data and zlib.decompress(data, 32 + zlib.MAX_WBITS)
        except zlib.error:
-            return zlib.decompress(data)
+            # raw zlib * windowsize:32768 (RFC 9110: "non-conformant")
            return zlib.decompress(data, -zlib.MAX_WBITS)
    @staticmethod
    def gzip(data):
        from gzip import GzipFile
        def _gzip(data):
            with io.BytesIO(data) as data_buf:
                gz = GzipFile(fileobj=data_buf, mode='rb')
                return gz.read()
        try:
            return _gzip(data)
        except IOError as original_ioerror:
            # There may be junk at the end of the file
            # See http://stackoverflow.com/q/4928560/35070 for details
            for i in range(1, 1024):
                try:
                    return _gzip(data[:-i])
                except IOError:
                    continue
            else:
                raise original_ioerror
    @staticmethod
    def brotli(data):
        return data and brotli.decompress(data)
    @staticmethod
    def compress(data):
        return data and ncompress.decompress(data)
    @staticmethod
    def _fix_path(url):
        # an embedded /../ or /./ sequence is not automatically handled by urllib2
        # see https://github.com/yt-dlp/yt-dlp/issues/3355
        parsed_url = compat_urllib_parse.urlsplit(url)
        path = parsed_url.path
        if not path.endswith('/'):
            path += '/'
        parts = path.partition('/./')
        if not parts[1]:
            parts = path.partition('/../')
        if parts[1]:
            path = compat_urllib_parse.urljoin(
                parts[0] + parts[1][:1],
                parts[1][1:] + (parts[2] if parsed_url.path.endswith('/') else parts[2][:-1]))
            url = parsed_url._replace(path=path).geturl()
        if '/.' in url:
            # worse, URL path may have initial /../ against RFCs: work-around
            # by stripping such prefixes, like eg Firefox
            path = parsed_url.path + '/'
            while path.startswith('/.'):
                if path.startswith('/../'):
                    path = path[3:]
                elif path.startswith('/./'):
                    path = path[2:]
                else:
                    break
            path = path[:-1]
            if not path.startswith('/') and parsed_url.path.startswith('/'):
                path = '/' + path
            url = parsed_url._replace(path=path).geturl()
        return url
    def http_request(self, req):
-        # According to RFC 3986, URLs can not contain non-ASCII characters, however this is not
+        url = req.get_full_url()
-        # always respected by websites, some tend to give out URLs with non percent-encoded
+        # resolve embedded . and ..
        url_fixed = self._fix_path(url)
        # According to RFC 3986, URLs can not contain non-ASCII characters; however this is not
        # always respected by websites: some tend to give out URLs with non percent-encoded
        # non-ASCII characters (see telemb.py, ard.py [#3412])
        # urllib chokes on URLs with non-ASCII characters (see http://bugs.python.org/issue3991)
        # To work around aforementioned issue we will replace request's original URL with
        # percent-encoded one
        # Since redirects are also affected (e.g. http://www.southpark.de/alle-episoden/s18e09)
        # the code of this workaround has been moved here from YoutubeDL.urlopen()
-        url = req.get_full_url()
+        url_escaped = escape_url(url_fixed)
        url_escaped = escape_url(url)
        # Substitute URL if any change after escaping
        if url != url_escaped:
@ -2668,7 +2737,10 @@ class YoutubeDLHandler(compat_urllib_request.HTTPHandler):
        req.headers = handle_youtubedl_headers(req.headers)
-        if sys.version_info < (2, 7) and '#' in req.get_full_url():
+        if sys.version_info < (2, 7):
            # avoid possible race where __r_type may be unset
            req.get_type()
            if '#' in req.get_full_url():
                # Python 2.6 is brain-dead when it comes to fragments
                req._Request__original = req._Request__original.partition('#')[0]
                req._Request__r_type = req._Request__r_type.partition('#')[0]
@ -2679,33 +2751,59 @@ class YoutubeDLHandler(compat_urllib_request.HTTPHandler):
    def http_response(self, req, resp):
        old_resp = resp
-        # gzip
+
-        if resp.headers.get('Content-encoding', '') == 'gzip':
+        # Content-Encoding header lists the encodings in order that they were applied [1].
-            content = resp.read()
+        # To decompress, we simply do the reverse.
-            gz = gzip.GzipFile(fileobj=io.BytesIO(content), mode='rb')
+        # [1]: https://datatracker.ietf.org/doc/html/rfc9110#name-content-encoding
        decoded_response = None
        decoders = {
            'gzip': self.deflate_gz,
            'deflate': self.deflate_gz,
        }
        if brotli:
            decoders['br'] = self.brotli
        if ncompress:
            decoders['compress'] = self.compress
        if sys.platform.startswith('java'):
            # Jython zlib implementation misses gzip
            decoders['gzip'] = self.gzip
        def encodings(hdrs):
            # A header field that allows multiple values can have multiple instances [2].
            # [2]: https://datatracker.ietf.org/doc/html/rfc9110#name-fields
            for e in reversed(','.join(hdrs).split(',')):
                if e:
                    yield e.strip()
        encodings_left = []
        try:
-                uncompressed = io.BytesIO(gz.read())
+            resp.headers.get_all
-            except IOError as original_ioerror:
+            hdrs = resp.headers
-                # There may be junk at the end of the file
+        except AttributeError:
-                # See http://stackoverflow.com/q/4928560/35070 for details
+            # Py2 has no get_all() method: headers are rfc822.Message
-                for i in range(1, 1024):
+            from email.message import Message
-                    try:
+            hdrs = Message()
-                        gz = gzip.GzipFile(fileobj=io.BytesIO(content[:-i]), mode='rb')
+            for k, v in resp.headers.items():
-                        uncompressed = io.BytesIO(gz.read())
+                hdrs[k] = v
-                    except IOError:
+
        decoder, decoded_response = True, None
        for encoding in encodings(hdrs.get_all('Content-Encoding', [])):
            # "SHOULD consider" x-compress, x-gzip as compress, gzip
            decoder = decoder and decoders.get(remove_start(encoding, 'x-'))
            if not decoder:
                encodings_left.insert(0, encoding)
                continue
-                    break
+            decoded_response = decoder(decoded_response or resp.read())
-                else:
+        if decoded_response is not None:
-                    raise original_ioerror
+            resp = compat_urllib_request.addinfourl(
-            resp = compat_urllib_request.addinfourl(uncompressed, old_resp.headers, old_resp.url, old_resp.code)
+                io.BytesIO(decoded_response), old_resp.headers, old_resp.url, old_resp.code)
            resp.msg = old_resp.msg
-            del resp.headers['Content-encoding']
+            del resp.headers['Content-Length']
-        # deflate
+            resp.headers['Content-Length'] = '%d' % len(decoded_response)
-        if resp.headers.get('Content-encoding', '') == 'deflate':
+        del resp.headers['Content-Encoding']
-            gz = io.BytesIO(self.deflate(resp.read()))
+        if encodings_left:
-            resp = compat_urllib_request.addinfourl(gz, old_resp.headers, old_resp.url, old_resp.code)
+            resp.headers['Content-Encoding'] = ', '.join(encodings_left)
-            resp.msg = old_resp.msg
+
            del resp.headers['Content-encoding']
        # Percent-encode redirect URL of Location HTTP header to satisfy RFC 3986 (see
        # https://github.com/ytdl-org/youtube-dl/issues/6457).
        if 300 <= resp.code < 400:
@ -2715,10 +2813,13 @@ class YoutubeDLHandler(compat_urllib_request.HTTPHandler):
                if sys.version_info >= (3, 0):
                    location = location.encode('iso-8859-1')
                location = location.decode('utf-8')
-                location_escaped = escape_url(location)
+                # resolve embedded . and ..
                location_fixed = self._fix_path(location)
                location_escaped = escape_url(location_fixed)
                if location != location_escaped:
                    del resp.headers['Location']
-                    if sys.version_info < (3, 0):
+                    # if sys.version_info < (3, 0):
                    if not isinstance(location_escaped, str):
                        location_escaped = location_escaped.encode('utf-8')
                    resp.headers['Location'] = location_escaped
        return resp
@ -4188,13 +4289,8 @@ def update_Request(req, url=None, data=None, headers={}, query={}):
    req_headers.update(headers)
    req_data = data if data is not None else req.data
    req_url = update_url_query(url or req.get_full_url(), query)
-    req_get_method = req.get_method()
+    req_type = {'HEAD': HEADRequest, 'PUT': PUTRequest}.get(
-    if req_get_method == 'HEAD':
+        req.get_method(), compat_urllib_request.Request)
        req_type = HEADRequest
    elif req_get_method == 'PUT':
        req_type = PUTRequest
    else:
        req_type = compat_urllib_request.Request
    new_req = req_type(
        req_url, data=req_data, headers=req_headers,
        origin_req_host=req.origin_req_host, unverifiable=req.unverifiable)