donatas abraitis

Filter out routes by type of blackhole in FRRouting

SYN

Basically, there are two purposes of using blackhole routes:

  • To prevent from DDoS attacks
  • To inject fake routes into the routing table with longer network mask (e.g.: for BGP advertisements)

The former can also be implemented using well-known iptables. But it always depends on what is better iptables, iproute2 using blackhole, nftables, eBPF, XDP or finally XDP offload on the hardware when talking about good enough.

If you have thousands of rules in iptables then the kernel has a lot of work to do by touching every rule in the list. In other words, when you modify a single rule, iptables has to fetch the whole basket of rules from the kernel (of course using locking because it’s a linked-list and synchronization between CPUs), do the change and put all rules back. So, avoid having a plethora of rules under iptables - otherwise it will take half a minute to reload.

I’m pretty sure that 99% of servers (if not using the route on the host) has a routing table only with a default route, which means that dropping packets to specific destinations is much cheaper than using iptables.

ip route add blackhole 192.168.0.1/32
ip -6 route add blackhole 2a02:4780:dead::beef/128

vs.

iptables -A INPUT -d 192.168.0.1/32 -j DROP
ip6tables -A INPUT -d 2a02:4780:dead::beef/128 -j DROP

The difference in processing the packets is that blackhole doesn’t send any packets back to the source while netfilter generates ICMP responses.

Let’s say you want to filter out blackhole routes from being advertised to BGP neighbors, what you can do? If it’s /32 or /128 then it can be filtered using prefix-list, otherwise can’t.

With my latest patch to FRRouting it can be implemented just by matching the type of the next-hop like:

route-map bh deny 10
  match ip next-hop type blackhole
route-map bh deny 20
  match ipv6 next-hop type blackhole

FIN