gotosocial/docs/advanced/security/firewall.md
Daenney a15415b1a2
[docs] Describe how to block IPs using a firewall (#2459)
* [docs] Describe how to block IPs using a firewall

This adds some documentation on how to block one, multiple or whole IP
ranges using the firewall. This can be helpful to protect from DDoS
attacks or block certain parties from being able to communicate with
your instance at all.

* [docs] Fix indenting of nft rule

* [docs] Split iptables and nftables into sections
2024-01-03 10:14:38 +00:00

9.4 KiB

Firewall

You should deploy a firewall on your instance to close off any open ports and give you a mechanism to ban potentially misbehaving clients. Many firewall frontends will also automatically install some rules that block obvious malicious packets.

It can be helpful to deploy tools that monitor your log files for certain patterns and automatically ban clients exhibiting certain behaviour. This can be use to monitor your SSH and web server access logs for things like SSH brute-force attacks.

Ports

For GoToSocial, you'll want to ensure port 443 remains open. Without it, nobody will be able to reach your instance. Federation will fail and client apps won't be able to work at all.

If you provision TLS certificates using ACME or GoToSocial's built-in Lets Encrypt support, you'll also need port 80 to be open.

In order to access your instance over SSH, you'll need to keep the port your SSH daemon is bound on open too. By default this is port 22.

ICMP

Internet Control Message Protocol are exchanged between machines in order to detect certain network conditions or troubleshoot things. Many firewalls have a tendency of blocking ICMP entirely but this is undesirable. A few ICMP types should be allowed and you can use your firewall to configure rate limiting for them instead.

IPv4

In order for things to work reliably, your firewall must allow:

  • ICMP Type 3: "Destination Unreachable" and also aids in Path-MTU Discovery
  • ICMP Type 4: "Source Quench"

If you want to be able to ping things or be pinged, you should also allow:

  • ICMP Type 0: "Echo Reply"
  • ICMP Type 8: "Echo Request"

For traceroute to work, it can be helpful to also allow:

  • ICMP Type 11: "Time Exceeded"

IPv6

ICMP is heavily relied on by all parts of the IPv6 stack and things will break in exciting and hard to debug ways if you block it. RFC 4890 was specifically written to address this and is worthwhile to review.

Roughly speaking, you must always allow:

  • ICMP Type 1: "Destination Unreachable"
  • ICMP Type 2: "Packet Too Big"
  • ICMP Type 3, code 0: "Time Exceeded"
  • ICMP Type 4, code 1, 2: "Parameter Problem"

For ping, you should allow:

  • ICMP Type 128: "Echo Request"
  • ICMP Type 129: "Echo Response"

Firewall configuration

On Linux, firewalling is typically done using either iptables or the more modern and faster nftables as the backend. Most distributions are switching to nftables and many firewall frontends can be configured to use nftables instead. You'll need to refer to your distribution's documentation on the matter, but typically there will be an iptables or nftables service your init-system can start with a predefined location to load firewall rules from.

Doing this by hand using raw iptables or nftables rules offers the most control but can be challenging if you're not familiar with these systems. In order to help with that, a number of configuration frontends exist that you can use.

On the Debian and Ubuntu as well as openSUSE family of distributions, UFW is commonly used. It's a simple firewall front-end and many tutorials targeting those distributions will be using it.

For the Red Hat/CentOS family of distributions, firewalld is typically used. It's a much more advanced firewall configuration utility which also has a desktop GUI and Cockpit integration.

Despite distribution preferences, you can use UFW, firewalld or something else entirely with any Linux distribution.

Brute-force protection

fail2ban and SSHGuard can be set up to monitor your log files for attempts to brute-force logins and other malicious behaviour. They can be configured to automatically insert firewall rules to block malicious IP addresses, either for a specific period of time or even indefinitely.

SSHGuard was initially designed just for SSH, but nowadays supports a variety of services. Fail2ban tends to support anything you can generate consistent log lines for, whereas SSHGuard's signature approach can catch more sophisticated or stealthy attacks as it computes an attack score over time.

Both SSHGuard and fail2ban ship with "backends" that can target iptables and nftables directly, or work with your frontend of choice like UFW or firewalld on Linux or pf on *BSD. Make sure you review their documentation on how to correctly configure it.

For fail2ban, you can use the following regex, which triggers fail2ban on failed logins and not another 'Unauthorized' errors (API for example):

statusCode=401 path=/auth/sign_in clientIP=<HOST> .* msg=\"Unauthorized:

IP blocking

GoToSocial implements rate-limiting in order to try and protect your instance from one party taking up all your processing capacity. However, if you know this traffic isn't legitimate or coming from an instance you don't wish to federate with anyway, you can block the IP(s) the traffic is originating from instead and spare GoToSocial from having to do any work.

Linux

Blocking IPs is done with iptables or nftables. If you're using a firewall frontend like UFW or firewalld, use their facilities to block an IP.

In iptables, people tend to add a DROP rule for an IP in the filter table on the INPUT chain. On nftables, it's often done on a table with a chain with the ip or ip6 address family. In both those cases the kernel has already done a lot of unnecessary processing of the incoming traffic, just for it to then be blocked by an IP match.

When using iptables, this can be done more effectively using the mangle table and the PREROUTING chain. You can check this blog post on how that works in iptables. For nftables, you want to block using the netdev family instead.

iptables

An example of blocking an IP using iptables:

iptables -t mangle -A PREROUTING -s 1.0.0.0/8 -j DROP
ip6tables -t mangle -A PREROUTING -s fc00::/7 -j DROP

When using iptables, adding many rules slows things down significantly, including reloading the firewall when adding/removing rules. Since you may wish to block many IP addresses, use the ipset module and add a single block rule for the set instead.

Start by creating your sets and adding some IPs to them:

ipset create baddiesv4 hash:ip family inet
ipset create baddiesv6 hash:ip family inet6

ipset add baddiesv4 1.0.0.0/8
ipset add baddiesv6 fc00::/7

Then, update your iptables rules to target the set instead:

iptables -t mangle -A PREROUTING -m set --match-set baddiesv4 src -j DROP
ip6tables -t mangle -A PREROUTING -m set --match-set baddiesv6 src -j DROP

nftables

For nftables, you can use something like:

table netdev filter {
    chain ingress {
        set baddiesv4 {
            type ipv4_addr
            flags interval
            elements = { \
                1.0.0.0/8, \
                2.2.2.2/32 \
            }
        }
        set baddiesv6 {
            type ipv6_addr
            flags interval
            elements = { \
                2620:4f:8000::/48, \
                fc00::/7 \
            }
        }

        type filter hook ingress device <interface name> priority -500;
        ip saddr @baddiesv4 drop
        ip6 saddr @baddiesv6 drop
    }
}

BSDs

When using pf, you can create a persistent table, typically named <badhosts>, to which you add the IP addresses you want to block. Tables can also read from other files, so it's possible to keep the list of IPs outside of your main pf.conf.

An example of how to do this can be found in the pf manual.