advocate 0.6.2

Advocate

Advocate is a set of tools based around the requests library for safely making HTTP requests on behalf of a third party. Specifically, it aims to prevent common techniques that enable SSRF attacks.

Advocate was inspired by fin1te’s SafeCurl project.

Installation

pip install advocate

Advocate is officially supported on CPython 2.7+, CPython 3.4+ and PyPy 2. PyPy 3 may work as well, but you’ll need a copy of the ipaddress module from elsewhere.

See it in action

If you want to try out Advocate to see what kind of things it catches, there’s a test site up on advocate.saynotolinux.com.

Examples

Advocate is more-or-less a drop-in replacement for requests. In most cases you can just replace “requests” with “advocate” where necessary and be good to go:

>>> import advocate
>>> print advocate.get("http://google.com/")
<Response [200]>

Advocate also provides a subclassed requests.Session with sane defaults for validation already set up:

>>> import advocate
>>> sess = advocate.Session()
>>> print sess.get("http://google.com/")
<Response [200]>
>>> print sess.get("http://localhost/")
advocate.exceptions.UnacceptableAddressException: ('localhost', 80)

All of the wrapped request functions accept a validator kwarg where you can set additional rules:

>>> import advocate
>>> validator = advocate.AddrValidator(hostname_blacklist={"*.museum",})
>>> print advocate.get("http://educational.MUSEUM/", validator=validator)
advocate.exceptions.UnacceptableAddressException: educational.MUSEUM

If you require more advanced rules than the defaults, but don’t want to have to pass the validator kwarg everywhere, there’s RequestsAPIWrapper . You can define a wrapper in a common file and import it instead of advocate:

>>> from advocate import AddrValidator, RequestsAPIWrapper
>>> from advocate.packages import ipaddress
>>> dougs_advocate = RequestsAPIWrapper(AddrValidator(ip_blacklist={
...     # Contains data incomprehensible to mere mortals
...     ipaddress.ip_network("42.42.42.42/32")
... }))
>>> print dougs_advocate.get("http://42.42.42.42/")
advocate.exceptions.UnacceptableAddressException: ('42.42.42.42', 80)

Other than that, you can do just about everything with Advocate that you can with an unwrapped requests. Advocate passes requests’ test suite with the exception of tests that require Session.mount().

Conditionally bypassing protection

If you want to allow certain users to bypass Advocate’s restrictions, just use plain ‘ol requests by doing something like:

if user == "mr_skeltal":
    requests_module = requests
else:
    requests_module = advocate
resp = requests_module.get("http://example.com/doot_doot")

requests-futures support

A thin wrapper around requests-futures is provided to ease writing async-friendly code:

>>> from advocate.futures import FuturesSession
>>> sess = FuturesSession()
>>> fut = sess.get("http://example.com/")
>>> fut
<Future at 0x10c717f28 state=finished returned Response>
>>> fut.result()
<Response [200]>

You can do basically everything you can do with regular FuturesSession s and advocate.Session s:

>>> from advocate import AddrValidator
>>> from advocate.futures import FuturesSession
>>> sess = FuturesSession(max_workers=20, validator=AddrValidator(hostname_blacklist={"*.museum"}))
>>> fut = sess.get("http://anice.museum/")
>>> fut
<Future at 0x10c696668 state=running>
>>> fut.result()
Traceback (most recent call last):
# [...]
advocate.exceptions.UnacceptableAddressException: anice.museum

When should I use Advocate?

Any time you’re fetching resources over HTTP for / from someone you don’t trust!

When should I not use Advocate?

That’s a tough one. There are a few cases I can think of where I wouldn’t:

• When good, safe support for IPv6 is important

• When internal hosts use globally routable addresses and you can’t guess their prefix to blacklist it ahead of time

• You already have a good handle on network security within your network

Actually, if you’re comfortable enough with Squid and network security, you should set up a secured Squid instance on a segregated subnet and proxy through that instead. Advocate attempts to guess whether an address references an internal host and block access, but it’s definitely preferable to proxy through a host can’t access anything internal in the first place!

Of course, if you’re writing an app / library that’s meant to be usable OOTB on other people’s networks, Advocate + a user-configurable blacklist is probably the safer bet.

This seems like it’s been done before

There’ve been a few similar projects, but in my opinion Advocate’s approach is the best because:

def connect_socket(host, port):
    for res in socket.getaddrinfo(host, port):
        # where `res` will be a tuple containing the IP for the host
        if not is_blacklisted(res):
            # ... connect the socket using `res`

TODO

Proper IPv6 Support?

Advocate’s IPv6 support is still a work-in-progress, since I’m not that familiar with the spec, and there are so many ways to tunnel IPv4 over IPv6, as well as other non-obvious gotchas. IPv6 records are ignored by default for now, but you can enable by using an AddrValidator with allow_ipv6=True.

It should mostly work as expected, but Advocate’s approach might not even make sense with most IPv6 deployments, see Issue #3 for more info.

If you can think of any improvements to the IPv6 handling, please submit an issue or PR!

Caveats

• This is beta-quality software, the API might change without warning!

• mount() ing other adapters is disallowed to prevent Advocate’s validating adapters from being clobbered.

• Advocate does not, and might never support the use of HTTP proxies.

• Proper IPv6 support is still a WIP as noted above.

Acknowledgements

• https://github.com/fin1te/safecurl for inspiration

• https://github.com/kennethreitz/requests for the lovely requests module

• https://bitbucket.org/kwi/py2-ipaddress for the backport of ipaddress

• https://github.com/hakobe/paranoidhttp a similar project targeting golang

• https://github.com/uber-common/paranoid-request a similar project targeting Node

• http://search.cpan.org/~tsibley/LWP-UserAgent-Paranoid/ a similar project targeting Perl 5