How do I safely get the user's ip address in Flask that has a proxy?

Question

I am using Flask and need to get the user's IP address. This is usually done through request.remote_addr but since this app is hosted at a 3rd party (and using cloudflare) it just returns the localhost.

Flask suggests getting the X-Forwarded-Host but then they immediately say it is a security risk. Is there a safe way to get the client's real ip?

Answer 1

The Problem

The issue here is not that the ProxyFix itself will cause the user to get access to your system, but rather the fact that the ProxyFix will take what was once mostly reliable information and replace it instead with potentially unreliable information.

For starters, when you don't use ProxyFix, the REMOTE_ADDR attribute is most likely retrieved from the source IP address in the TCP packets. While not impossible, the source IP address in TCP packets are tough to spoof . Therefore, if you need a reliable way to retrieve the user's IP address, REMOTE_ADDR is a good way to do it; in most cases, you can rely on it to provide you something that is accurate when you do request.remote_addr .

The problem is, of course, in a reverse-proxy situation the TCP connection is not coming from the end user; instead, the end user makes a TCP connection with the reverse proxy, and the reverse proxy then makes a second TCP connection with your web app. Therefore, the request.remote_addr in your app will have the IP address of the reverse proxy rather than the original user.

A Potential Solution

ProxyFix is supposed to solve this problem so that you can make request.remote_addr have the user's IP address rather than the proxy. It does this by looking at the typical HTTP header that remote proxies (like Apache and Nginx) add into the HTTP header ( X-Forwarded-For ) and use the user's IP address it finds there. Note that Cloudflare uses a different HTTP Header, so ProxyFix probably won't help you; you'll need to write your own implementation of this middleware to get request.remote_addr to use the original client's IP address. However, in the rest of this answer I will continue to refer to that fix as "ProxyFix".

This solution, however, is problematic. The problem is that while the TCP header is mostly reliable, the HTTP headers are not; if a user can bypass your reverse proxy and send data right to the server, they can put whatever they want in the HTTP header. For example, they can make the IP address in the HTTP header the IP address of someone else! If you use the IP address for authentication, the user can spoof that authentication mechanism. If you store the IP address in your database and then display it in your application to another user in HTML, the user could inject SQL or Javascript into the header, potentially causing SQL injection or XSS vulnerabilities.

So, to summarize; ProxyFix takes a known mostly-safe solution to retrieve the user's IP address from a TCP packet and switches it to using the not-very-safe-by-itself solution of parsing an easily-spoofed HTTP header.

Therefore, the recomendation to use ProxyFix ONLY in reverse proxy situations means just that: don't use this if you accept connections from places that are NOT the proxy. This is often means have the reverse proxy (like Nginx or Apache) handle all your incoming traffic and have your application that actually uses ProxyFix safe behind a firewall.

You should also read this post which explains how ProxyFix was broken in the past ( although is now fixed ). This will also explains how ProxyFix works, and give you ideas on how to set your num_proxies argument .

A Better Solution

Let's say your user is at point A, they send the request to Cloudflare (B) which eventually sends the request to your final application (point C). Cloudflare will send the IP address of A in the CF-Connecting-IP header .

As explained above, if the user finds the IP address to point C, they could send a specially crafted HTTP request directly to point C which includes any header info they want. ProxyFix will use its logic to determine what the IP address is from the HTTP header, which of course is problematic if you rely on that value for, well, mostly anything.

Therefore, you might want to look at using something like mod_cloudflare , which allows you to do these proxy fixes directly in the Apache mod, but only when the HTTP connection comes from Cloudflare IP addresses (as defined by the TCP IP source). You can also have it only accept connections from Cloudflare. See How do I restore original visitor IP to my server logs for more info on this and help doing this with other servers (like Nginx).

This should give you a start. However, keep in mind that you're still not "safe": you've only shut down one possible attack vector, and that attack vector assumed that the attacker knew the IP address of your actual application. In that case, the malicious user could try to do a TCP attack with a spoofed Cloudflare IP address, although this would be extremely difficult . More likely, if they wanted to cause havoc, they would just DDOS your source server since they've bypassed Cloudflare. So, there are plenty more things to think about in securing, your application. Hopefully this helps you with understanding how to make one part slightly safer.