Release 2024.03.10
[yt-dlp.git] / yt_dlp / networking / _requests.py
blob6545028c814379c00ea5e75a9495f72575a1ea1d
1 import contextlib
2 import functools
3 import http.client
4 import logging
5 import re
6 import socket
7 import warnings
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
13 if requests is None:
14 raise ImportError('requests module is not installed')
16 if urllib3 is None:
17 raise ImportError('urllib3 module is not installed')
19 urllib3_version = tuple(int_or_none(x, default=0) for x in urllib3.__version__.split('.'))
21 if urllib3_version < (1, 26, 17):
22 raise ImportError('Only urllib3 >= 1.26.17 is supported')
24 if requests.__build__ < 0x023100:
25 raise ImportError('Only requests >= 2.31.0 is supported')
27 import requests.adapters
28 import requests.utils
29 import urllib3.connection
30 import urllib3.exceptions
32 from ._helper import (
33 InstanceStoreMixin,
34 add_accept_encoding_header,
35 create_connection,
36 create_socks_proxy_socket,
37 get_redirect_method,
38 make_socks_proxy_opts,
39 select_proxy,
41 from .common import (
42 Features,
43 RequestHandler,
44 Response,
45 register_preference,
46 register_rh,
48 from .exceptions import (
49 CertificateVerifyError,
50 HTTPError,
51 IncompleteRead,
52 ProxyError,
53 RequestError,
54 SSLError,
55 TransportError,
57 from ..socks import ProxyError as SocksProxyError
59 SUPPORTED_ENCODINGS = [
60 'gzip', 'deflate'
63 if brotli is not None:
64 SUPPORTED_ENCODINGS.append('br')
66 """
67 Override urllib3's behavior to not convert lower-case percent-encoded characters
68 to upper-case during url normalization process.
70 RFC3986 defines that the lower or upper case percent-encoded hexidecimal characters are equivalent
71 and normalizers should convert them to uppercase for consistency [1].
73 However, some sites may have an incorrect implementation where they provide
74 a percent-encoded url that is then compared case-sensitively.[2]
76 While this is a very rare case, since urllib does not do this normalization step, it
77 is best to avoid it in requests too for compatability reasons.
79 1: https://tools.ietf.org/html/rfc3986#section-2.1
80 2: https://github.com/streamlink/streamlink/pull/4003
81 """
84 class Urllib3PercentREOverride:
85 def __init__(self, r: re.Pattern):
86 self.re = r
88 # pass through all other attribute calls to the original re
89 def __getattr__(self, item):
90 return self.re.__getattribute__(item)
92 def subn(self, repl, string, *args, **kwargs):
93 return string, self.re.subn(repl, string, *args, **kwargs)[1]
96 # urllib3 >= 1.25.8 uses subn:
97 # https://github.com/urllib3/urllib3/commit/a2697e7c6b275f05879b60f593c5854a816489f0
98 import urllib3.util.url # noqa: E305
100 if hasattr(urllib3.util.url, 'PERCENT_RE'):
101 urllib3.util.url.PERCENT_RE = Urllib3PercentREOverride(urllib3.util.url.PERCENT_RE)
102 elif hasattr(urllib3.util.url, '_PERCENT_RE'): # urllib3 >= 2.0.0
103 urllib3.util.url._PERCENT_RE = Urllib3PercentREOverride(urllib3.util.url._PERCENT_RE)
104 else:
105 warnings.warn('Failed to patch PERCENT_RE in urllib3 (does the attribute exist?)' + bug_reports_message())
108 Workaround for issue in urllib.util.ssl_.py: ssl_wrap_context does not pass
109 server_hostname to SSLContext.wrap_socket if server_hostname is an IP,
110 however this is an issue because we set check_hostname to True in our SSLContext.
112 Monkey-patching IS_SECURETRANSPORT forces ssl_wrap_context to pass server_hostname regardless.
114 This has been fixed in urllib3 2.0+.
115 See: https://github.com/urllib3/urllib3/issues/517
118 if urllib3_version < (2, 0, 0):
119 with contextlib.suppress(Exception):
120 urllib3.util.IS_SECURETRANSPORT = urllib3.util.ssl_.IS_SECURETRANSPORT = True
123 # Requests will not automatically handle no_proxy by default
124 # due to buggy no_proxy handling with proxy dict [1].
125 # 1. https://github.com/psf/requests/issues/5000
126 requests.adapters.select_proxy = select_proxy
129 class RequestsResponseAdapter(Response):
130 def __init__(self, res: requests.models.Response):
131 super().__init__(
132 fp=res.raw, headers=res.headers, url=res.url,
133 status=res.status_code, reason=res.reason)
135 self._requests_response = res
137 def read(self, amt: int = None):
138 try:
139 # Interact with urllib3 response directly.
140 return self.fp.read(amt, decode_content=True)
142 # See urllib3.response.HTTPResponse.read() for exceptions raised on read
143 except urllib3.exceptions.SSLError as e:
144 raise SSLError(cause=e) from e
146 except urllib3.exceptions.ProtocolError as e:
147 # IncompleteRead is always contained within ProtocolError
148 # See urllib3.response.HTTPResponse._error_catcher()
149 ir_err = next(
150 (err for err in (e.__context__, e.__cause__, *variadic(e.args))
151 if isinstance(err, http.client.IncompleteRead)), None)
152 if ir_err is not None:
153 # `urllib3.exceptions.IncompleteRead` is subclass of `http.client.IncompleteRead`
154 # but uses an `int` for its `partial` property.
155 partial = ir_err.partial if isinstance(ir_err.partial, int) else len(ir_err.partial)
156 raise IncompleteRead(partial=partial, expected=ir_err.expected) from e
157 raise TransportError(cause=e) from e
159 except urllib3.exceptions.HTTPError as e:
160 # catch-all for any other urllib3 response exceptions
161 raise TransportError(cause=e) from e
164 class RequestsHTTPAdapter(requests.adapters.HTTPAdapter):
165 def __init__(self, ssl_context=None, proxy_ssl_context=None, source_address=None, **kwargs):
166 self._pm_args = {}
167 if ssl_context:
168 self._pm_args['ssl_context'] = ssl_context
169 if source_address:
170 self._pm_args['source_address'] = (source_address, 0)
171 self._proxy_ssl_context = proxy_ssl_context or ssl_context
172 super().__init__(**kwargs)
174 def init_poolmanager(self, *args, **kwargs):
175 return super().init_poolmanager(*args, **kwargs, **self._pm_args)
177 def proxy_manager_for(self, proxy, **proxy_kwargs):
178 extra_kwargs = {}
179 if not proxy.lower().startswith('socks') and self._proxy_ssl_context:
180 extra_kwargs['proxy_ssl_context'] = self._proxy_ssl_context
181 return super().proxy_manager_for(proxy, **proxy_kwargs, **self._pm_args, **extra_kwargs)
183 def cert_verify(*args, **kwargs):
184 # lean on SSLContext for cert verification
185 pass
188 class RequestsSession(requests.sessions.Session):
190 Ensure unified redirect method handling with our urllib redirect handler.
193 def rebuild_method(self, prepared_request, response):
194 new_method = get_redirect_method(prepared_request.method, response.status_code)
196 # HACK: requests removes headers/body on redirect unless code was a 307/308.
197 if new_method == prepared_request.method:
198 response._real_status_code = response.status_code
199 response.status_code = 308
201 prepared_request.method = new_method
203 # Requests fails to resolve dot segments on absolute redirect locations
204 # See: https://github.com/yt-dlp/yt-dlp/issues/9020
205 prepared_request.url = normalize_url(prepared_request.url)
207 def rebuild_auth(self, prepared_request, response):
208 # HACK: undo status code change from rebuild_method, if applicable.
209 # rebuild_auth runs after requests would remove headers/body based on status code
210 if hasattr(response, '_real_status_code'):
211 response.status_code = response._real_status_code
212 del response._real_status_code
213 return super().rebuild_auth(prepared_request, response)
216 class Urllib3LoggingFilter(logging.Filter):
218 def filter(self, record):
219 # Ignore HTTP request messages since HTTPConnection prints those
220 if record.msg == '%s://%s:%s "%s %s %s" %s %s':
221 return False
222 return True
225 class Urllib3LoggingHandler(logging.Handler):
226 """Redirect urllib3 logs to our logger"""
228 def __init__(self, logger, *args, **kwargs):
229 super().__init__(*args, **kwargs)
230 self._logger = logger
232 def emit(self, record):
233 try:
234 msg = self.format(record)
235 if record.levelno >= logging.ERROR:
236 self._logger.error(msg)
237 else:
238 self._logger.stdout(msg)
240 except Exception:
241 self.handleError(record)
244 @register_rh
245 class RequestsRH(RequestHandler, InstanceStoreMixin):
247 """Requests RequestHandler
248 https://github.com/psf/requests
250 _SUPPORTED_URL_SCHEMES = ('http', 'https')
251 _SUPPORTED_ENCODINGS = tuple(SUPPORTED_ENCODINGS)
252 _SUPPORTED_PROXY_SCHEMES = ('http', 'https', 'socks4', 'socks4a', 'socks5', 'socks5h')
253 _SUPPORTED_FEATURES = (Features.NO_PROXY, Features.ALL_PROXY)
254 RH_NAME = 'requests'
256 def __init__(self, *args, **kwargs):
257 super().__init__(*args, **kwargs)
259 # Forward urllib3 debug messages to our logger
260 logger = logging.getLogger('urllib3')
261 self.__logging_handler = Urllib3LoggingHandler(logger=self._logger)
262 self.__logging_handler.setFormatter(logging.Formatter('requests: %(message)s'))
263 self.__logging_handler.addFilter(Urllib3LoggingFilter())
264 logger.addHandler(self.__logging_handler)
265 # TODO: Use a logger filter to suppress pool reuse warning instead
266 logger.setLevel(logging.ERROR)
268 if self.verbose:
269 # Setting this globally is not ideal, but is easier than hacking with urllib3.
270 # It could technically be problematic for scripts embedding yt-dlp.
271 # However, it is unlikely debug traffic is used in that context in a way this will cause problems.
272 urllib3.connection.HTTPConnection.debuglevel = 1
273 logger.setLevel(logging.DEBUG)
274 # this is expected if we are using --no-check-certificate
275 urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
277 def close(self):
278 self._clear_instances()
279 # Remove the logging handler that contains a reference to our logger
280 # See: https://github.com/yt-dlp/yt-dlp/issues/8922
281 logging.getLogger('urllib3').removeHandler(self.__logging_handler)
283 def _check_extensions(self, extensions):
284 super()._check_extensions(extensions)
285 extensions.pop('cookiejar', None)
286 extensions.pop('timeout', None)
288 def _create_instance(self, cookiejar):
289 session = RequestsSession()
290 http_adapter = RequestsHTTPAdapter(
291 ssl_context=self._make_sslcontext(),
292 source_address=self.source_address,
293 max_retries=urllib3.util.retry.Retry(False),
295 session.adapters.clear()
296 session.headers = requests.models.CaseInsensitiveDict({'Connection': 'keep-alive'})
297 session.mount('https://', http_adapter)
298 session.mount('http://', http_adapter)
299 session.cookies = cookiejar
300 session.trust_env = False # no need, we already load proxies from env
301 return session
303 def _send(self, request):
305 headers = self._merge_headers(request.headers)
306 add_accept_encoding_header(headers, SUPPORTED_ENCODINGS)
308 max_redirects_exceeded = False
310 session = self._get_instance(
311 cookiejar=request.extensions.get('cookiejar') or self.cookiejar)
313 try:
314 requests_res = session.request(
315 method=request.method,
316 url=request.url,
317 data=request.data,
318 headers=headers,
319 timeout=float(request.extensions.get('timeout') or self.timeout),
320 proxies=request.proxies or self.proxies,
321 allow_redirects=True,
322 stream=True
325 except requests.exceptions.TooManyRedirects as e:
326 max_redirects_exceeded = True
327 requests_res = e.response
329 except requests.exceptions.SSLError as e:
330 if 'CERTIFICATE_VERIFY_FAILED' in str(e):
331 raise CertificateVerifyError(cause=e) from e
332 raise SSLError(cause=e) from e
334 except requests.exceptions.ProxyError as e:
335 raise ProxyError(cause=e) from e
337 except (requests.exceptions.ConnectionError, requests.exceptions.Timeout) as e:
338 raise TransportError(cause=e) from e
340 except urllib3.exceptions.HTTPError as e:
341 # Catch any urllib3 exceptions that may leak through
342 raise TransportError(cause=e) from e
344 except requests.exceptions.RequestException as e:
345 # Miscellaneous Requests exceptions. May not necessary be network related e.g. InvalidURL
346 raise RequestError(cause=e) from e
348 res = RequestsResponseAdapter(requests_res)
350 if not 200 <= res.status < 300:
351 raise HTTPError(res, redirect_loop=max_redirects_exceeded)
353 return res
356 @register_preference(RequestsRH)
357 def requests_preference(rh, request):
358 return 100
361 # Use our socks proxy implementation with requests to avoid an extra dependency.
362 class SocksHTTPConnection(urllib3.connection.HTTPConnection):
363 def __init__(self, _socks_options, *args, **kwargs): # must use _socks_options to pass PoolKey checks
364 self._proxy_args = _socks_options
365 super().__init__(*args, **kwargs)
367 def _new_conn(self):
368 try:
369 return create_connection(
370 address=(self._proxy_args['addr'], self._proxy_args['port']),
371 timeout=self.timeout,
372 source_address=self.source_address,
373 _create_socket_func=functools.partial(
374 create_socks_proxy_socket, (self.host, self.port), self._proxy_args))
375 except (socket.timeout, TimeoutError) as e:
376 raise urllib3.exceptions.ConnectTimeoutError(
377 self, f'Connection to {self.host} timed out. (connect timeout={self.timeout})') from e
378 except SocksProxyError as e:
379 raise urllib3.exceptions.ProxyError(str(e), e) from e
380 except OSError as e:
381 raise urllib3.exceptions.NewConnectionError(
382 self, f'Failed to establish a new connection: {e}') from e
385 class SocksHTTPSConnection(SocksHTTPConnection, urllib3.connection.HTTPSConnection):
386 pass
389 class SocksHTTPConnectionPool(urllib3.HTTPConnectionPool):
390 ConnectionCls = SocksHTTPConnection
393 class SocksHTTPSConnectionPool(urllib3.HTTPSConnectionPool):
394 ConnectionCls = SocksHTTPSConnection
397 class SocksProxyManager(urllib3.PoolManager):
399 def __init__(self, socks_proxy, username=None, password=None, num_pools=10, headers=None, **connection_pool_kw):
400 connection_pool_kw['_socks_options'] = make_socks_proxy_opts(socks_proxy)
401 super().__init__(num_pools, headers, **connection_pool_kw)
402 self.pool_classes_by_scheme = {
403 'http': SocksHTTPConnectionPool,
404 'https': SocksHTTPSConnectionPool
408 requests.adapters.SOCKSProxyManager = SocksProxyManager