yt_dlp/networking/_requests.py

   1 import contextlib
   2 import functools
   3 import http.client
   4 import logging
   5 import re
   6 import socket
   7 import warnings
   8
   9 from ..dependencies import brotli, requests, urllib3
  10 from ..utils import bug_reports_message, int_or_none, variadic
  11 from ..utils.networking import normalize_url
  12
  13 if requests is None:
  14     raise ImportError('requests module is not installed')
  15
  16 if urllib3 is None:
  17     raise ImportError('urllib3 module is not installed')
  18
  19 urllib3_version = tuple(int_or_none(x, default=0) for x in urllib3.__version__.split('.'))
  20
  21 if urllib3_version < (1, 26, 17):
  22     raise ImportError('Only urllib3 >= 1.26.17 is supported')
  23
  24 if requests.__build__ < 0x023100:
  25     raise ImportError('Only requests >= 2.31.0 is supported')
  26
  27 import requests.adapters
  28 import requests.utils
  29 import urllib3.connection
  30 import urllib3.exceptions
  31 import urllib3.util
  32
  33 from ._helper import (
  34     InstanceStoreMixin,
  35     add_accept_encoding_header,
  36     create_connection,
  37     create_socks_proxy_socket,
  38     get_redirect_method,
  39     make_socks_proxy_opts,
  40     select_proxy,
  41 )
  42 from .common import (
  43     Features,
  44     RequestHandler,
  45     Response,
  46     register_preference,
  47     register_rh,
  48 )
  49 from .exceptions import (
  50     CertificateVerifyError,
  51     HTTPError,
  52     IncompleteRead,
  53     ProxyError,
  54     RequestError,
  55     SSLError,
  56     TransportError,
  57 )
  58 from ..socks import ProxyError as SocksProxyError
  59
  60 SUPPORTED_ENCODINGS = [
  61     'gzip', 'deflate'
  62 ]
  63
  64 if brotli is not None:
  65     SUPPORTED_ENCODINGS.append('br')
  66
  67 """
  68 Override urllib3's behavior to not convert lower-case percent-encoded characters
  69 to upper-case during url normalization process.
  70
  71 RFC3986 defines that the lower or upper case percent-encoded hexidecimal characters are equivalent
  72 and normalizers should convert them to uppercase for consistency [1].
  73
  74 However, some sites may have an incorrect implementation where they provide
  75 a percent-encoded url that is then compared case-sensitively.[2]
  76
  77 While this is a very rare case, since urllib does not do this normalization step, it
  78 is best to avoid it in requests too for compatability reasons.
  79
  80 1: https://tools.ietf.org/html/rfc3986#section-2.1
  81 2: https://github.com/streamlink/streamlink/pull/4003
  82 """
  83
  84
  85 class Urllib3PercentREOverride:
  86     def __init__(self, r: re.Pattern):
  87         self.re = r
  88
  89     # pass through all other attribute calls to the original re
  90     def __getattr__(self, item):
  91         return self.re.__getattribute__(item)
  92
  93     def subn(self, repl, string, *args, **kwargs):
  94         return string, self.re.subn(repl, string, *args, **kwargs)[1]
  95
  96
  97 # urllib3 >= 1.25.8 uses subn:
  98 # https://github.com/urllib3/urllib3/commit/a2697e7c6b275f05879b60f593c5854a816489f0
  99 import urllib3.util.url  # noqa: E305
 100
 101 if hasattr(urllib3.util.url, 'PERCENT_RE'):
 102     urllib3.util.url.PERCENT_RE = Urllib3PercentREOverride(urllib3.util.url.PERCENT_RE)
 103 elif hasattr(urllib3.util.url, '_PERCENT_RE'):  # urllib3 >= 2.0.0
 104     urllib3.util.url._PERCENT_RE = Urllib3PercentREOverride(urllib3.util.url._PERCENT_RE)
 105 else:
 106     warnings.warn('Failed to patch PERCENT_RE in urllib3 (does the attribute exist?)' + bug_reports_message())
 107
 108 """
 109 Workaround for issue in urllib.util.ssl_.py: ssl_wrap_context does not pass
 110 server_hostname to SSLContext.wrap_socket if server_hostname is an IP,
 111 however this is an issue because we set check_hostname to True in our SSLContext.
 112
 113 Monkey-patching IS_SECURETRANSPORT forces ssl_wrap_context to pass server_hostname regardless.
 114
 115 This has been fixed in urllib3 2.0+.
 116 See: https://github.com/urllib3/urllib3/issues/517
 117 """
 118
 119 if urllib3_version < (2, 0, 0):
 120     with contextlib.suppress(Exception):
 121         urllib3.util.IS_SECURETRANSPORT = urllib3.util.ssl_.IS_SECURETRANSPORT = True
 122
 123
 124 # Requests will not automatically handle no_proxy by default
 125 # due to buggy no_proxy handling with proxy dict [1].
 126 # 1. https://github.com/psf/requests/issues/5000
 127 requests.adapters.select_proxy = select_proxy
 128
 129
 130 class RequestsResponseAdapter(Response):
 131     def __init__(self, res: requests.models.Response):
 132         super().__init__(
 133             fp=res.raw, headers=res.headers, url=res.url,
 134             status=res.status_code, reason=res.reason)
 135
 136         self._requests_response = res
 137
 138     def read(self, amt: int = None):
 139         try:
 140             # Interact with urllib3 response directly.
 141             return self.fp.read(amt, decode_content=True)
 142
 143         # See urllib3.response.HTTPResponse.read() for exceptions raised on read
 144         except urllib3.exceptions.SSLError as e:
 145             raise SSLError(cause=e) from e
 146
 147         except urllib3.exceptions.ProtocolError as e:
 148             # IncompleteRead is always contained within ProtocolError
 149             # See urllib3.response.HTTPResponse._error_catcher()
 150             ir_err = next(
 151                 (err for err in (e.__context__, e.__cause__, *variadic(e.args))
 152                  if isinstance(err, http.client.IncompleteRead)), None)
 153             if ir_err is not None:
 154                 # `urllib3.exceptions.IncompleteRead` is subclass of `http.client.IncompleteRead`
 155                 # but uses an `int` for its `partial` property.
 156                 partial = ir_err.partial if isinstance(ir_err.partial, int) else len(ir_err.partial)
 157                 raise IncompleteRead(partial=partial, expected=ir_err.expected) from e
 158             raise TransportError(cause=e) from e
 159
 160         except urllib3.exceptions.HTTPError as e:
 161             # catch-all for any other urllib3 response exceptions
 162             raise TransportError(cause=e) from e
 163
 164
 165 class RequestsHTTPAdapter(requests.adapters.HTTPAdapter):
 166     def __init__(self, ssl_context=None, proxy_ssl_context=None, source_address=None, **kwargs):
 167         self._pm_args = {}
 168         if ssl_context:
 169             self._pm_args['ssl_context'] = ssl_context
 170         if source_address:
 171             self._pm_args['source_address'] = (source_address, 0)
 172         self._proxy_ssl_context = proxy_ssl_context or ssl_context
 173         super().__init__(**kwargs)
 174
 175     def init_poolmanager(self, *args, **kwargs):
 176         return super().init_poolmanager(*args, **kwargs, **self._pm_args)
 177
 178     def proxy_manager_for(self, proxy, **proxy_kwargs):
 179         extra_kwargs = {}
 180         if not proxy.lower().startswith('socks') and self._proxy_ssl_context:
 181             extra_kwargs['proxy_ssl_context'] = self._proxy_ssl_context
 182         return super().proxy_manager_for(proxy, **proxy_kwargs, **self._pm_args, **extra_kwargs)
 183
 184     # Skip `requests` internal verification; we use our own SSLContext
 185     # requests 2.31.0+
 186     def cert_verify(*args, **kwargs):
 187         pass
 188
 189     # requests 2.31.0-2.32.1
 190     def _get_connection(self, request, *_, proxies=None, **__):
 191         return self.get_connection(request.url, proxies)
 192
 193     # requests 2.32.2+: Reimplementation without `_urllib3_request_context`
 194     def get_connection_with_tls_context(self, request, verify, proxies=None, cert=None):
 195         url = urllib3.util.parse_url(request.url).url
 196
 197         manager = self.poolmanager
 198         if proxy := select_proxy(url, proxies):
 199             manager = self.proxy_manager_for(proxy)
 200
 201         return manager.connection_from_url(url)
 202
 203
 204 class RequestsSession(requests.sessions.Session):
 205     """
 206     Ensure unified redirect method handling with our urllib redirect handler.
 207     """
 208
 209     def rebuild_method(self, prepared_request, response):
 210         new_method = get_redirect_method(prepared_request.method, response.status_code)
 211
 212         # HACK: requests removes headers/body on redirect unless code was a 307/308.
 213         if new_method == prepared_request.method:
 214             response._real_status_code = response.status_code
 215             response.status_code = 308
 216
 217         prepared_request.method = new_method
 218
 219         # Requests fails to resolve dot segments on absolute redirect locations
 220         # See: https://github.com/yt-dlp/yt-dlp/issues/9020
 221         prepared_request.url = normalize_url(prepared_request.url)
 222
 223     def rebuild_auth(self, prepared_request, response):
 224         # HACK: undo status code change from rebuild_method, if applicable.
 225         # rebuild_auth runs after requests would remove headers/body based on status code
 226         if hasattr(response, '_real_status_code'):
 227             response.status_code = response._real_status_code
 228             del response._real_status_code
 229         return super().rebuild_auth(prepared_request, response)
 230
 231
 232 class Urllib3LoggingFilter(logging.Filter):
 233
 234     def filter(self, record):
 235         # Ignore HTTP request messages since HTTPConnection prints those
 236         if record.msg == '%s://%s:%s "%s %s %s" %s %s':
 237             return False
 238         return True
 239
 240
 241 class Urllib3LoggingHandler(logging.Handler):
 242     """Redirect urllib3 logs to our logger"""
 243
 244     def __init__(self, logger, *args, **kwargs):
 245         super().__init__(*args, **kwargs)
 246         self._logger = logger
 247
 248     def emit(self, record):
 249         try:
 250             msg = self.format(record)
 251             if record.levelno >= logging.ERROR:
 252                 self._logger.error(msg)
 253             else:
 254                 self._logger.stdout(msg)
 255
 256         except Exception:
 257             self.handleError(record)
 258
 259
 260 @register_rh
 261 class RequestsRH(RequestHandler, InstanceStoreMixin):
 262
 263     """Requests RequestHandler
 264     https://github.com/psf/requests
 265     """
 266     _SUPPORTED_URL_SCHEMES = ('http', 'https')
 267     _SUPPORTED_ENCODINGS = tuple(SUPPORTED_ENCODINGS)
 268     _SUPPORTED_PROXY_SCHEMES = ('http', 'https', 'socks4', 'socks4a', 'socks5', 'socks5h')
 269     _SUPPORTED_FEATURES = (Features.NO_PROXY, Features.ALL_PROXY)
 270     RH_NAME = 'requests'
 271
 272     def __init__(self, *args, **kwargs):
 273         super().__init__(*args, **kwargs)
 274
 275         # Forward urllib3 debug messages to our logger
 276         logger = logging.getLogger('urllib3')
 277         self.__logging_handler = Urllib3LoggingHandler(logger=self._logger)
 278         self.__logging_handler.setFormatter(logging.Formatter('requests: %(message)s'))
 279         self.__logging_handler.addFilter(Urllib3LoggingFilter())
 280         logger.addHandler(self.__logging_handler)
 281         # TODO: Use a logger filter to suppress pool reuse warning instead
 282         logger.setLevel(logging.ERROR)
 283
 284         if self.verbose:
 285             # Setting this globally is not ideal, but is easier than hacking with urllib3.
 286             # It could technically be problematic for scripts embedding yt-dlp.
 287             # However, it is unlikely debug traffic is used in that context in a way this will cause problems.
 288             urllib3.connection.HTTPConnection.debuglevel = 1
 289             logger.setLevel(logging.DEBUG)
 290         # this is expected if we are using --no-check-certificate
 291         urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
 292
 293     def close(self):
 294         self._clear_instances()
 295         # Remove the logging handler that contains a reference to our logger
 296         # See: https://github.com/yt-dlp/yt-dlp/issues/8922
 297         logging.getLogger('urllib3').removeHandler(self.__logging_handler)
 298
 299     def _check_extensions(self, extensions):
 300         super()._check_extensions(extensions)
 301         extensions.pop('cookiejar', None)
 302         extensions.pop('timeout', None)
 303
 304     def _create_instance(self, cookiejar):
 305         session = RequestsSession()
 306         http_adapter = RequestsHTTPAdapter(
 307             ssl_context=self._make_sslcontext(),
 308             source_address=self.source_address,
 309             max_retries=urllib3.util.retry.Retry(False),
 310         )
 311         session.adapters.clear()
 312         session.headers = requests.models.CaseInsensitiveDict({'Connection': 'keep-alive'})
 313         session.mount('https://', http_adapter)
 314         session.mount('http://', http_adapter)
 315         session.cookies = cookiejar
 316         session.trust_env = False  # no need, we already load proxies from env
 317         return session
 318
 319     def _send(self, request):
 320
 321         headers = self._merge_headers(request.headers)
 322         add_accept_encoding_header(headers, SUPPORTED_ENCODINGS)
 323
 324         max_redirects_exceeded = False
 325
 326         session = self._get_instance(cookiejar=self._get_cookiejar(request))
 327
 328         try:
 329             requests_res = session.request(
 330                 method=request.method,
 331                 url=request.url,
 332                 data=request.data,
 333                 headers=headers,
 334                 timeout=self._calculate_timeout(request),
 335                 proxies=self._get_proxies(request),
 336                 allow_redirects=True,
 337                 stream=True
 338             )
 339
 340         except requests.exceptions.TooManyRedirects as e:
 341             max_redirects_exceeded = True
 342             requests_res = e.response
 343
 344         except requests.exceptions.SSLError as e:
 345             if 'CERTIFICATE_VERIFY_FAILED' in str(e):
 346                 raise CertificateVerifyError(cause=e) from e
 347             raise SSLError(cause=e) from e
 348
 349         except requests.exceptions.ProxyError as e:
 350             raise ProxyError(cause=e) from e
 351
 352         except (requests.exceptions.ConnectionError, requests.exceptions.Timeout) as e:
 353             raise TransportError(cause=e) from e
 354
 355         except urllib3.exceptions.HTTPError as e:
 356             # Catch any urllib3 exceptions that may leak through
 357             raise TransportError(cause=e) from e
 358
 359         except requests.exceptions.RequestException as e:
 360             # Miscellaneous Requests exceptions. May not necessary be network related e.g. InvalidURL
 361             raise RequestError(cause=e) from e
 362
 363         res = RequestsResponseAdapter(requests_res)
 364
 365         if not 200 <= res.status < 300:
 366             raise HTTPError(res, redirect_loop=max_redirects_exceeded)
 367
 368         return res
 369
 370
 371 @register_preference(RequestsRH)
 372 def requests_preference(rh, request):
 373     return 100
 374
 375
 376 # Use our socks proxy implementation with requests to avoid an extra dependency.
 377 class SocksHTTPConnection(urllib3.connection.HTTPConnection):
 378     def __init__(self, _socks_options, *args, **kwargs):  # must use _socks_options to pass PoolKey checks
 379         self._proxy_args = _socks_options
 380         super().__init__(*args, **kwargs)
 381
 382     def _new_conn(self):
 383         try:
 384             return create_connection(
 385                 address=(self._proxy_args['addr'], self._proxy_args['port']),
 386                 timeout=self.timeout,
 387                 source_address=self.source_address,
 388                 _create_socket_func=functools.partial(
 389                     create_socks_proxy_socket, (self.host, self.port), self._proxy_args))
 390         except (socket.timeout, TimeoutError) as e:
 391             raise urllib3.exceptions.ConnectTimeoutError(
 392                 self, f'Connection to {self.host} timed out. (connect timeout={self.timeout})') from e
 393         except SocksProxyError as e:
 394             raise urllib3.exceptions.ProxyError(str(e), e) from e
 395         except OSError as e:
 396             raise urllib3.exceptions.NewConnectionError(
 397                 self, f'Failed to establish a new connection: {e}') from e
 398
 399
 400 class SocksHTTPSConnection(SocksHTTPConnection, urllib3.connection.HTTPSConnection):
 401     pass
 402
 403
 404 class SocksHTTPConnectionPool(urllib3.HTTPConnectionPool):
 405     ConnectionCls = SocksHTTPConnection
 406
 407
 408 class SocksHTTPSConnectionPool(urllib3.HTTPSConnectionPool):
 409     ConnectionCls = SocksHTTPSConnection
 410
 411
 412 class SocksProxyManager(urllib3.PoolManager):
 413
 414     def __init__(self, socks_proxy, username=None, password=None, num_pools=10, headers=None, **connection_pool_kw):
 415         connection_pool_kw['_socks_options'] = make_socks_proxy_opts(socks_proxy)
 416         super().__init__(num_pools, headers, **connection_pool_kw)
 417         self.pool_classes_by_scheme = {
 418             'http': SocksHTTPConnectionPool,
 419             'https': SocksHTTPSConnectionPool
 420         }
 421
 422
 423 requests.adapters.SOCKSProxyManager = SocksProxyManager