(λ (x) (create x) '(knowledge))

Simple Iptables Firewalls

It doesn't have to be scary · July 23rd, 2022

A while back I talked about documenting some of the things I've been learning from my LFCE studying, I think if I've gotten to a point where I feel comfortable writing about a topic then I'm likely prepared to test on that subject too. So let's talk a little bit about iptables!

What is it?

Iptables is simply a firewall software, and unfortunately it gets a reputation for being complicated and confusing. It's definitely not a point and click solution like you get with UFW or Firewalld, but it powers both of those solutions, so why shouldn't you learn it? Even something like Alpine's Awall is powered by iptables, and while I have a personal affinity for Awall, iptables is still the root; if you understand how it works it doesn't much matter what you're dealing with. You can easily figure out how a UFW, Firewalld, Awall, or any other iptables backed firewall software works, and that includes plenty of the off the shelf enterprise solutions out there. Mikrotik's for example work this way, and their custom tooling follows very closely alongside iptables.

Additionally there are some really neat features you can leverage with iptables, such as rate-limiting by local user, much like you'd do inside of something like a Fortigate's NGFW. Nifty, and free!

My Droid's Firewall

Here's an example of a really simple workstation firewall. Characteristically its operation is simple, it allows any sort of outbound traffic, and only allows certain types of inbound traffic. I'd say this is likely the simplest and most relatable configuration to frame iptables with. Any laptop, desktop, or even something esoteric like the droid can be considered a workstation if you're working on it regularly. Typically you want anything you do on that system to be sanctioned outwards, but you want to more granularly control which ports are open and what can access the resources on your workstation. Unlike a server the expectation is that traffic originates outwards to multiple points, and inbound traffic is rare and should meet expected parameters.


*filter
:INPUT DROP [0:0]
:FORWARD DROP [0:0]
:OUTPUT ACCEPT [0:0]
#Route established and related traffic
-A INPUT -m state --state RELATED,ESTABLISHED -j ACCEPT
-A INPUT -i lo -j ACCEPT
#Allow SSH
-A INPUT -i wlan0 -p tcp -m state --state NEW -m tcp --dport 22 -j ACCEPT
-A INPUT -i wlan0 -p udp -m state --state NEW -m udp --dport 60000:61000 -j ACCEPT
#Allow Lapis Dev
-A INPUT -s 192.168.88.0/24 -i wlan0 -p tcp -m state --state NEW -m tcp --dport 8080 -j ACCEPT
#Allow fserv
-A INPUT -i wlan0 -p tcp -m state --state NEW -m tcp --dport 8090 -j ACCEPT
#Allow PNTP
-A INPUT -i wlan0 -p tcp -m state --state NEW -m tcp --dport 8091 -j ACCEPT
#Drop other unlisted input, drop forwards, accept output
-A INPUT -j DROP
-A FORWARD -j DROP
-A OUTPUT -j ACCEPT
COMMIT
 

Nice and short, maybe not as easy to grok as the pretty UFW output, but I promise it's not that bad either. These rules are in the format that iptables-save expects, you can pretty much append "sudo iptables" to any of the -A CHAIN rules there and it'll add that specific rule temporarily to your iptables ruleset in the specified chain.


*filter
:INPUT DROP [0:0]
:FORWARD DROP [0:0]
:OUTPUT ACCEPT [0:0]
 

At the very front we define our filter table it contains three chains by default, these chains essentially store our rules and let us think about our firewall in a consistent way. The default chains are pretty straight forward to work with, INPUT is anything coming into the firewall, FORWARD is anything that is going through our firewall, and OUTPUT is anything leaving the firewall. A quick glance at the full ruleset and you'll note that we use all three chains. Lets look at just the top and bottom of our ruleset to see those in action.


#Route established and related traffic
-A INPUT -m state --state RELATED,ESTABLISHED -j ACCEPT
-A INPUT -i lo -j ACCEPT
#Drop other unlisted input, drop forwards, accept output
-A INPUT -j DROP
-A FORWARD -j DROP
-A OUTPUT -j ACCEPT
COMMIT
 

Since iptables rules are processed from the top down (unless a JUMP to a specific chain is defined) it's easy to build out traffic exclusions. The tail end of the droid's ruleset is precisely this, you can read it as follows:

  • Accept any traffic that is already ESTABLISHED or RELATED to existing traffic
  • Accept any sort of input from our lo interface
  • Drop any INPUT traffic that doesn't match accepted INPUT rules above
  • Drop any FORWARD traffic that doesn't match accepted FORWARD rules above
  • Accept ALL OUTPUT traffic coming from the system

So if our ruleset only defined these items it would ACCEPT any sort of OUTBOUND traffic, anything that uses the interface lo, and DROP any INBOUND or FORWARD packets, effectively blocking the outside world but allow our own traffic to tentatively find its way into the wild unknown. This is actually a solid baseline for a simple but effective firewall. But we can't just shut ourselves off from the world right? If you're like me you really need to be able to SSH into every system you own, or maybe you need to expose an HTTP port for testing a project.


#Allow SSH
-A INPUT -i wlan0 -p tcp -m state --state NEW -m tcp --dport 22 -j ACCEPT
 

Fortunately those use cases are easy to define, the above example can be read as follows. Append to the INPUT chain to allow any NEW traffic coming into interface wlan0 of packet type TCP bound for port 22. We specifically bind this to the NEW state because we allow ESTABLISHED and RELATED traffic at the top, so it's redundant to look for anything else to allow new SSH connections.


#Allow MOSH
-A INPUT -i wlan0 -p udp -m state --state NEW -m udp --dport 60000:61000 -j ACCEPT
 

Some applications require multiple ports to function, such as Mobile Shell or Mosh for short. That application looks for a UDP port inbetween 60000-61000, so we give out --dport arg a range of min:max to work with, but otherwise the rule is exactly the same as a the simpler SSH rule.


#Allow Lapis Dev
-A INPUT -s 192.168.88.0/24 -i wlan0 -p tcp -m state --state NEW -m tcp --dport 8080 -j ACCEPT
 

No perhaps we want to filter based on where traffic is coming from, for instance I like to run the Lapis application for my blog when I'm traveling. I can always access it on 127.0.0.1 because we allow all traffic from -i lo, but I don't need random strangers on a public wifi network to see my in-dev work. Adding a -s 192.168.88.0/24 restricts the INPUT to any addresses in that subnet. So anything on my home LAN can access that port, but nothing else. Obviously that's not perfect design, there easily could be a public wifi network that uses that subnet, as it's Mikrtoik's default DHCP address range. You should also consider your firewall a single layer in a multi-layer defence!

The syntax is a little weird, but if you break each argument of the rule down it starts to make sense. Here's how I read these things.


-A INPUT
Append to chain INPUT

-s 192.168.88.0/24
Any traffic with source IP of subnet 192.168.88.0/24

-i wlan0
Inbound from interface wlan0

-p tcp
That is TCP protocol traffic

-m state --state NEW
And is NEW traffic

-m tcp --dport 8080
Which is TCP traffic destined to port 8080

-j ACCEPT
If all of that checks out, jump to the ACCEPT chain, and push the traffic through.
  

Phew, yeah there's a reason that people look at iptables and think "I can't make heds or tails of this" it's extremely verbose, and once you move out of simple usecases like this and into something like a full blown NATing firewall configuration it can be a little scary. But that verbosity is your friend! These rules state very explicitly what they do and do not do. And each flag can be read as a specific check that's performed on the traffic. I'm honestly very happy to have moved away from UFW for my systems and now maintain iptables rules for all of the systems in my homelab. The flat file configuration nature works perfectly for simple provisioning, and the full rulesets can be revisioned in git for long term maintenance.

I'll revisit this topic sometime in the future so we can work through desigining a NATing firewall with iptables. There's a lot of dependent systems there too, so that will give us a chance to dig into DHCPD, and BIND at very least. I've got an idea in mind, it just needs to be fleshed out before I actually bring it to the blog.

Bio

(defparameter *Will_Sinatra* '((Age . 31) (Occupation . DevOps Engineer) (FOSS-Dev . true) (Locale . Maine) (Languages . ("Lisp" "Fennel" "Lua" "Go" "Nim")) (Certs . ("LFCS"))))

"Very little indeed is needed to live a happy life." - Aurelius