jasonfan's picture
Add files using upload-large-folder tool
71687cf verified
# Copyright (C) 2012 Anaconda, Inc
# SPDX-License-Identifier: BSD-3-Clause
"""Requests session configured with all accepted scheme adapters."""
from __future__ import annotations
from fnmatch import fnmatch
from functools import cache
from logging import getLogger
from threading import local
from typing import TYPE_CHECKING
from ... import CondaError
from ...auxlib.ish import dals
from ...base.constants import CONDA_HOMEPAGE_URL
from ...base.context import context
from ...common.url import (
add_username_and_password,
get_proxy_username_and_pass,
split_anaconda_token,
urlparse,
)
from ...exceptions import OfflineError, PluginError, ProxyError
from ...models.channel import Channel
from ..anaconda_client import read_binstar_tokens
from . import (
AuthBase,
BaseAdapter,
CaseInsensitiveDict,
Retry,
Session,
_basic_auth_str,
extract_cookies_to_jar,
get_auth_from_url,
get_netrc_auth,
)
from .adapters.ftp import FTPAdapter
from .adapters.http import HTTPAdapter
from .adapters.localfs import LocalFSAdapter
from .adapters.s3 import S3Adapter
if TYPE_CHECKING:
from requests.models import PreparedRequest, Request
log = getLogger(__name__)
RETRIES = 3
CONDA_SESSION_SCHEMES = frozenset(
(
"http",
"https",
"ftp",
"s3",
"file",
)
)
# Forbidden headers, which should never be set by plugins. Based on
# https://web.archive.org/web/20251124174612/https://developer.mozilla.org/en-US/docs/Glossary/Forbidden_request_header
# (last updated as of November 17, 2025). Set-Cookie is a response header and not a request header, but
# it's semantically invalid, so we include it here.
FORBIDDEN_HEADERS = frozenset(
[
"accept-charset",
"accept-encoding",
"access-control-request-headers",
"access-control-request-method",
"connection",
"content-length",
"cookie",
"date",
"dnt",
"expect",
"host",
"keep-alive",
"origin",
"referer",
"set-cookie",
"te",
"trailer",
"transfer-encoding",
"upgrade",
"via",
]
)
# Headers that override HTTP methods, which are forbidden when they contain the following.
FORBIDDEN_METHOD_OVERRIDE_HEADERS = frozenset(
[
"x-http-method",
"x-http-method-override",
"x-method-override",
]
)
FORBIDDEN_HTTP_METHODS = frozenset(["connect", "trace", "track"])
class EnforceUnusedAdapter(BaseAdapter):
def send(self, request: Request, *args, **kwargs):
raise OfflineError(
f"EnforceUnusedAdapter called with url {request.url}.\n"
"This command is using a remote connection in offline mode."
)
def close(self):
raise NotImplementedError()
def get_channel_name_from_url(url: str) -> str | None:
"""
Given a URL, determine the channel it belongs to and return its name.
"""
return Channel.from_url(url).canonical_name
def _validate_plugin_headers(headers: dict) -> None:
"""
Validate plugin-provided headers and raise PluginError if any forbidden
headers are detected. This prevents plugins from compromising conda's
core network stack.
:param headers: Dictionary of headers to validate
:raises PluginError: If any forbidden headers are detected
"""
for key, value in headers.items():
key_lower = key.lower()
if key_lower in FORBIDDEN_HEADERS:
raise PluginError(
f"Plugin attempted to set forbidden header '{key}'. "
"This header is not allowed as it could compromise conda's network stack. "
"See https://developer.mozilla.org/en-US/docs/Glossary/Forbidden_request_header"
)
if key_lower.startswith("proxy-") or key_lower.startswith("sec-"):
raise PluginError(
f"Plugin attempted to set forbidden header '{key}'. "
"Headers starting with 'Proxy-' or 'Sec-' are not allowed as they could "
"compromise conda's network stack. "
"See https://developer.mozilla.org/en-US/docs/Glossary/Forbidden_request_header"
)
if key_lower in FORBIDDEN_METHOD_OVERRIDE_HEADERS:
if isinstance(value, str) and value.lower() in FORBIDDEN_HTTP_METHODS:
raise PluginError(
f"Plugin attempted to set header '{key}' with forbidden method '{value}'. "
"This method override is not allowed as it could compromise conda's network stack. "
"See https://developer.mozilla.org/en-US/docs/Glossary/Forbidden_request_header"
)
@cache
def get_session(url: str):
"""
Function that determines the correct Session object to be returned
based on the URL that is passed in.
"""
channel_name = get_channel_name_from_url(url)
# If for whatever reason a channel name can't be determined, (should be unlikely)
# we just return the default session object.
if channel_name is None:
return CondaSession()
# We ensure here if there are duplicates defined, we choose the last one
channel_settings = {}
for settings in context.channel_settings:
channel = settings.get("channel", "")
if channel == channel_name:
# First we check for exact match
channel_settings = settings
continue
# If we don't have an exact match, we attempt to match a URL pattern
parsed_url = urlparse(url)
parsed_setting = urlparse(channel)
# We require that the schemes must be identical to prevent downgrade attacks.
# This includes the case of a scheme-less pattern like "*", which is not allowed.
if parsed_setting.scheme != parsed_url.scheme:
continue
url_without_schema = parsed_url.netloc + parsed_url.path
pattern = parsed_setting.netloc + parsed_setting.path
if fnmatch(url_without_schema, pattern):
channel_settings = settings
auth_handler = channel_settings.get("auth", "").strip() or None
# Return default session object
if auth_handler is None:
return CondaSession()
auth_handler_cls = context.plugin_manager.get_auth_handler(auth_handler)
if not auth_handler_cls:
return CondaSession()
return CondaSession(auth=auth_handler_cls(channel_name))
def get_session_storage_key(auth) -> str:
"""
Function that determines which storage key to use for our CondaSession object caching
"""
if auth is None:
return "default"
if isinstance(auth, tuple):
return hash(auth)
auth_type = type(auth)
return f"{auth_type.__module__}.{auth_type.__qualname__}::{auth.channel_name}"
class CondaSessionType(type):
"""
Takes advice from https://github.com/requests/requests/issues/1871#issuecomment-33327847
and creates one Session instance per thread.
"""
def __new__(mcs, name, bases, dct):
dct["_thread_local"] = local()
return super().__new__(mcs, name, bases, dct)
def __call__(cls, **kwargs):
storage_key = get_session_storage_key(kwargs.get("auth"))
try:
return cls._thread_local.sessions[storage_key]
except AttributeError:
session = super().__call__(**kwargs)
cls._thread_local.sessions = {storage_key: session}
except KeyError:
session = cls._thread_local.sessions[storage_key] = super().__call__(
**kwargs
)
return session
class CondaSession(Session, metaclass=CondaSessionType):
def __init__(
self,
auth: AuthBase | tuple[str, str] | None = None,
):
"""
:param auth: Optionally provide ``requests.AuthBase`` compliant objects
"""
super().__init__()
self.auth = auth or CondaHttpAuth()
self.proxies.update(context.proxy_servers)
ssl_context = None
if context.ssl_verify == "truststore":
try:
import ssl
import truststore
ssl_context = truststore.SSLContext(ssl.PROTOCOL_TLS_CLIENT)
except ImportError:
raise CondaError(
"The `ssl_verify: truststore` setting is only supported on"
"Python 3.10 or later."
)
self.verify = True
else:
self.verify = context.ssl_verify
if context.offline:
unused_adapter = EnforceUnusedAdapter()
self.mount("http://", unused_adapter)
self.mount("https://", unused_adapter)
self.mount("ftp://", unused_adapter)
self.mount("s3://", unused_adapter)
else:
# Configure retries
retry = Retry(
total=context.remote_max_retries,
backoff_factor=context.remote_backoff_factor,
status_forcelist=[413, 429, 500, 503],
raise_on_status=False,
respect_retry_after_header=False,
)
http_adapter = HTTPAdapter(max_retries=retry, ssl_context=ssl_context)
self.mount("http://", http_adapter)
self.mount("https://", http_adapter)
self.mount("ftp://", FTPAdapter())
self.mount("s3://", S3Adapter())
self.mount("file://", LocalFSAdapter())
self.headers["User-Agent"] = context.user_agent
if context.client_ssl_cert_key:
self.cert = (context.client_ssl_cert, context.client_ssl_cert_key)
elif context.client_ssl_cert:
self.cert = context.client_ssl_cert
def prepare_request(self, request: Request) -> PreparedRequest:
# inject headers from plugins if this is a https/http request
url = urlparse(request.url)
if url.scheme in ("https", "http"):
session_headers = context.plugin_manager.get_cached_session_headers(
host=url.netloc
)
request_headers = context.plugin_manager.get_cached_request_headers(
host=url.netloc, path=url.path
)
# Validate plugin headers - raises PluginError if forbidden headers detected
_validate_plugin_headers(session_headers)
_validate_plugin_headers(request_headers)
request.headers = CaseInsensitiveDict(
{
# hardcoded session headers (self.headers) are injected in super().prepare_request
**session_headers,
**request_headers,
**(request.headers or {}),
}
)
return super().prepare_request(request)
@classmethod
def cache_clear(cls):
try:
cls._thread_local.sessions.clear()
except AttributeError:
# AttributeError: thread's session cache has not been initialized
pass
class CondaHttpAuth(AuthBase):
# TODO: make this class thread-safe by adding some of the requests.auth.HTTPDigestAuth() code
def __call__(self, request):
request.url = CondaHttpAuth.add_binstar_token(request.url)
self._apply_basic_auth(request)
request.register_hook("response", self.handle_407)
return request
@staticmethod
def _apply_basic_auth(request):
# this logic duplicated from Session.prepare_request and PreparedRequest.prepare_auth
url_auth = get_auth_from_url(request.url)
auth = url_auth if any(url_auth) else None
if auth is None:
# look for auth information in a .netrc file
auth = get_netrc_auth(request.url)
if isinstance(auth, tuple) and len(auth) == 2:
request.headers["Authorization"] = _basic_auth_str(*auth)
return request
@staticmethod
def add_binstar_token(url):
clean_url, token = split_anaconda_token(url)
if not token and context.add_anaconda_token:
for binstar_url, token in read_binstar_tokens().items():
if clean_url.startswith(binstar_url):
log.debug("Adding anaconda token for url <%s>", clean_url)
from ...models.channel import Channel
channel = Channel(clean_url)
channel.token = token
return channel.url(with_credentials=True)
return url
@staticmethod
def handle_407(response, **kwargs): # pragma: no cover
"""
Prompts the user for the proxy username and password and modifies the
proxy in the session object to include it.
This method is modeled after
* requests.auth.HTTPDigestAuth.handle_401()
* requests.auth.HTTPProxyAuth
* the previous conda.fetch.handle_proxy_407()
It both adds 'username:password' to the proxy URL, as well as adding a
'Proxy-Authorization' header. If any of this is incorrect, please file an issue.
"""
# kwargs = {'verify': True, 'cert': None, 'proxies': {}, 'stream': False,
# 'timeout': (3.05, 60)}
if response.status_code != 407:
return response
# Consume content and release the original connection
# to allow our new request to reuse the same one.
response.content
response.close()
proxies = kwargs.pop("proxies")
proxy_scheme = urlparse(response.url).scheme
if proxy_scheme not in proxies:
raise ProxyError(
dals(
f"""
Could not find a proxy for {proxy_scheme!r}. See
{CONDA_HOMEPAGE_URL}/docs/html#configure-conda-for-use-behind-a-proxy-server
for more information on how to configure proxies.
"""
)
)
# fix-up proxy_url with username & password
proxy_url = proxies[proxy_scheme]
username, password = get_proxy_username_and_pass(proxy_scheme)
proxy_url = add_username_and_password(proxy_url, username, password)
proxy_authorization_header = _basic_auth_str(username, password)
proxies[proxy_scheme] = proxy_url
kwargs["proxies"] = proxies
prep = response.request.copy()
extract_cookies_to_jar(prep._cookies, response.request, response.raw)
prep.prepare_cookies(prep._cookies)
prep.headers["Proxy-Authorization"] = proxy_authorization_header
_response = response.connection.send(prep, **kwargs)
_response.history.append(response)
_response.request = prep
return _response