sanic/sanic/config.py
L. Kärkkäinen 1e4b1c4d1a Forwarded headers and otherwise improved proxy handling (#1638)
* Added support for HTTP Forwarded header and combined parsing of other proxy headers.

- Accessible via request.forwarded that tries parse_forwarded and then parse_xforwarded
- parse_forwarded uses the Forwarded header, if config.FORWARDED_SECRET is provided and a matching header field is found
- parse_xforwarded uses X-Real-IP and X-Forwarded-* much alike the existing implementation
- This commit does not change existing request properties that still use the old code and won't make use of Forwarded headers.

* Use req.forwarded in req properties server_name, server_port, scheme and remote_addr.

X-Scheme handling moved to parse_xforwarded.

* Cleanup and fix req.server_port; no longer reports socket port if any forwards headers are used.

* Update docstrings to incidate that forwarded header is used first.

* Remove testing function.

* Fix tests and linting.

- One test removed due to change of semantics - no socket port will be used if any forwarded headers are in effect.
- Other tests augmented with X-Forwarded-For, to allow the header being tested take effect (shouldn't affect old implementation).

* Try to workaround buggy tools complaining about incorrect ordering of imports.

* Cleanup forwarded processing, add comments. secret is now also returned.

* Added tests, fixed quoted string handling, cleanup.

* Further tests for full coverage.

* Try'n make linter happy.

* Add support for multiple Forwarded headers. Unify parse_forwarded parameters with parse_xforwarded.

* Implement multiple headers support for X-Forwarded-For.

- Previously only the first header was used, so this BUGFIX may affect functionality.

* Bugfix for request.server_name: strip port and other parts.

- request.server_name docs claim that it returns the hostname only (no port).
- config.SERVER_NAME may be full URL, so strip scheme, port and path
- HTTP Host and consequently forwarded Host may include port number, so
  strip that also for forwarded hosts (previously only done for HTTP Host).
- Possible performance benefit of limiting to one split.

* Fallback to app.url_for and let it handle SERVER_NAME if defined (until a proper solution is implemented).

* Revise previous commit. Only fallback for full URL SERVER_NAMEs; allows host to be defined and proxied information still being used.

* Heil lintnazi.

* Modify testcase not to use underscores in URLs. Use hyphens which the spec allows for.

* Forwarded and Host header parsing improved.

- request.forwarded lowercases hosts, separates host:port into their own fields and lowercases addresses
- forwarded.parse_host helper function added and used for parsing all host-style headers (IPv6 cannot be simply split(":")).
- more tests fixed not to use underscores in hosts as those are no longer accepted and lead to the field being rejected

* Fixed typo in docstring.

* Added IPv6 address tests for Host header.

* Fix regex.

* Further tests and stricter forwarded handling.

* Fix merge commit

* Linter

* Linter

* Linter

* Add  to avoid re-using the  variable. Make a few raw strings non-raw.

* Remove unnecessary or

* Updated docs (work in progress).

* Enable REAL_IP_HEADER parsing irregardless of PROXIES_COUNT setting.

- Also cleanup and added comments

* New defaults for PROXIES_COUNT and REAL_IP_HEADER, updated tests.

* Remove support for PROXIES_COUNT=-1.

* Linter errors.

- This is getting ridiculous: cannot fit an URL on one line, linter requires
  splitting the string literal!

* Add support for by=_proxySecret, updated docs, updated tests.

* Forwarded headers' semantics tuning.

- Forwarded host is now preserved in original format
- request.host now returns a forwarded host if available, else the Host header
- Forwarded options are preserved in original order, and later keys override earlier ones
- Forwarded path is automatically URL-unquoted
- Forwarded 'by' and 'for' are omitted if their value is unknown
- Tests modified accordingly
- Cleanup and improved documentation

* Add ASGI test.

* Linter

* Linter #2
2019-09-02 08:50:56 -05:00

158 lines
5.0 KiB
Python

import os
import types
from sanic.exceptions import PyFileError
from sanic.helpers import import_string
SANIC_PREFIX = "SANIC_"
BASE_LOGO = """
Sanic
Build Fast. Run Fast.
"""
DEFAULT_CONFIG = {
"REQUEST_MAX_SIZE": 100000000, # 100 megabytes
"REQUEST_BUFFER_QUEUE_SIZE": 100,
"REQUEST_TIMEOUT": 60, # 60 seconds
"RESPONSE_TIMEOUT": 60, # 60 seconds
"KEEP_ALIVE": True,
"KEEP_ALIVE_TIMEOUT": 5, # 5 seconds
"WEBSOCKET_MAX_SIZE": 2 ** 20, # 1 megabytes
"WEBSOCKET_MAX_QUEUE": 32,
"WEBSOCKET_READ_LIMIT": 2 ** 16,
"WEBSOCKET_WRITE_LIMIT": 2 ** 16,
"GRACEFUL_SHUTDOWN_TIMEOUT": 15.0, # 15 sec
"ACCESS_LOG": True,
"FORWARDED_SECRET": None,
"REAL_IP_HEADER": None,
"PROXIES_COUNT": None,
"FORWARDED_FOR_HEADER": "X-Forwarded-For",
}
class Config(dict):
def __init__(self, defaults=None, load_env=True, keep_alive=None):
defaults = defaults or {}
super().__init__({**DEFAULT_CONFIG, **defaults})
self.LOGO = BASE_LOGO
if keep_alive is not None:
self.KEEP_ALIVE = keep_alive
if load_env:
prefix = SANIC_PREFIX if load_env is True else load_env
self.load_environment_vars(prefix=prefix)
def __getattr__(self, attr):
try:
return self[attr]
except KeyError as ke:
raise AttributeError("Config has no '{}'".format(ke.args[0]))
def __setattr__(self, attr, value):
self[attr] = value
def from_envvar(self, variable_name):
"""Load a configuration from an environment variable pointing to
a configuration file.
:param variable_name: name of the environment variable
:return: bool. ``True`` if able to load config, ``False`` otherwise.
"""
config_file = os.environ.get(variable_name)
if not config_file:
raise RuntimeError(
"The environment variable %r is not set and "
"thus configuration could not be loaded." % variable_name
)
return self.from_pyfile(config_file)
def from_pyfile(self, filename):
"""Update the values in the config from a Python file.
Only the uppercase variables in that module are stored in the config.
:param filename: an absolute path to the config file
"""
module = types.ModuleType("config")
module.__file__ = filename
try:
with open(filename) as config_file:
exec( # nosec
compile(config_file.read(), filename, "exec"),
module.__dict__,
)
except IOError as e:
e.strerror = "Unable to load configuration file (%s)" % e.strerror
raise
except Exception as e:
raise PyFileError(filename) from e
self.from_object(module)
return True
def from_object(self, obj):
"""Update the values from the given object.
Objects are usually either modules or classes.
Just the uppercase variables in that object are stored in the config.
Example usage::
from yourapplication import default_config
app.config.from_object(default_config)
or also:
app.config.from_object('myproject.config.MyConfigClass')
You should not use this function to load the actual configuration but
rather configuration defaults. The actual config should be loaded
with :meth:`from_pyfile` and ideally from a location not within the
package because the package might be installed system wide.
:param obj: an object holding the configuration
"""
if isinstance(obj, str):
obj = import_string(obj)
for key in dir(obj):
if key.isupper():
self[key] = getattr(obj, key)
def load_environment_vars(self, prefix=SANIC_PREFIX):
"""
Looks for prefixed environment variables and applies
them to the configuration if present.
"""
for k, v in os.environ.items():
if k.startswith(prefix):
_, config_key = k.split(prefix, 1)
try:
self[config_key] = int(v)
except ValueError:
try:
self[config_key] = float(v)
except ValueError:
try:
self[config_key] = strtobool(v)
except ValueError:
self[config_key] = v
def strtobool(val):
"""
This function was borrowed from distutils.utils. While distutils
is part of stdlib, it feels odd to use distutils in main application code.
The function was modified to walk its talk and actually return bool
and not int.
"""
val = val.lower()
if val in ("y", "yes", "t", "true", "on", "1"):
return True
elif val in ("n", "no", "f", "false", "off", "0"):
return False
else:
raise ValueError("invalid truth value %r" % (val,))