The IPv6 threat model: what changes
Identify the attack surface that IPv6 adds, including on networks that believe they are IPv4-only. Map common IPv4 security controls to their IPv6 equivalents and spot the gaps. Prioritise IPv6 risks for a dual-stack organisation.
Take the quiz · pass mark 70% · 9 questions
Explanation
IPv6 is not more or less secure than IPv4. It is different, and most incidents come from the differences: controls that were built for IPv4 and never extended, and IPv6 features that are on by default and nobody watches. This level teaches the defences; this lesson maps the ground.
Lab ethics. Every attack in Level 5 is run only inside your own lab. Running these tools against networks you do not own or have written permission to test is illegal in most countries, including under South Africa's Cybercrimes Act.
Your "IPv4-only" network already runs IPv6
Every modern operating system enables IPv6 by default and prefers it when it works (L3.04). RFC 7123 describes what that means for a network whose operators think it is IPv4-only:
- Link-local IPv6 is always on. Hosts form
fe80::addresses, answer Neighbor Discovery and listen for Router Advertisements, whether or not any router sends them. - A rogue RA turns IPv6 on. One attacker host sending an RA with a prefix (and optionally RDNSS) gives every host a global address and a default router: the attacker. Because hosts prefer IPv6, traffic to dual-stack destinations now flows through the attacker, and an attacker running NAT64/DNS64 can intercept IPv4-only destinations too. Your IPv4 firewall, IDS and proxy see none of it.
- Tunnels bypass the perimeter. Automatic tunnels (6to4, ISATAP and Teredo, which runs over UDP port 3544) carry IPv6 inside IPv4 straight through IPv4-only filters. Modern systems have mostly disabled them, but a policy should block IPv4 protocol 41 and outbound UDP 3544 unless needed.
- Security tools may ignore IPv6. Logs that store addresses in 15-character fields, SIEM rules keyed on IPv4, IDS sensors without IPv6 decoding, and asset inventories without IPv6 addresses all leave blind spots.
The fix is not to "disable IPv6" (Microsoft, for example, does not support Windows with IPv6 disabled, and things break), but to manage it: deploy it properly, or at least defend the first hop and monitor for it.
What is new, what is the same
| Area | IPv4 | IPv6 | What changes for defenders |
|---|---|---|---|
| Address resolution | ARP | ND (ICMPv6) | ND spoofing replaces ARP spoofing; RA spoofing is new (L5.03) |
| Router discovery | DHCP option 3 | RAs from any host | Rogue RAs; needs RA-Guard (L5.03) |
| Address assignment | DHCP | SLAAC, DHCPv6 | Rogue DHCPv6; DHCPv6-Shield (L5.03) |
| ICMP | Often blocked | Required for ND and PMTUD | Filter selectively (L5.02) |
| Options | Rarely used IPv4 options | Extension headers, fragmentation rules | New evasion tricks (L5.04) |
| Perimeter | NAT as accidental firewall | No NAT, global addresses | Explicit stateful policy (L5.05) |
| Scanning | /24 sweep in seconds | /64 sweep impossible | Reconnaissance moves to DNS, logs, patterns (L5.06) |
| Logging | Fixed-width addresses | Many text forms, temporary addresses | Canonical form, aggregation by /64 (L5.08) |
Some things are the same: application vulnerabilities, phishing, weak passwords and unpatched servers do not care which IP version carries them. And IPsec is not a reason to call IPv6 secure (L5.07).
Neighbor Discovery trust
ND (RFC 4861) assumes every node on the link is trustworthy: any host can send an RA, answer an NS for someone else's address, or send a Redirect. RFC 3756 analyses these threats. SEND (Secure Neighbor Discovery) was designed to sign ND messages, but it was never widely implemented, so in practice the first hop is protected by the switch (RA-Guard, DHCPv6-Shield, SAVI) and by monitoring.
Prioritising
RFC 9099 is the operator's catalogue of IPv6 security practice. For a typical dual-stack organisation, a sensible order is:
- First-hop security on every access network, including "IPv4-only" ones: RA-Guard and DHCPv6-Shield (L5.03).
- Equivalent perimeter policy for IPv6, built from the same intent as IPv4, with correct ICMPv6 handling (L5.02, L5.05).
- Visibility: IPv6 in logs, flows, SIEM and the asset inventory (L5.08).
- Hardening of exposed hosts: services listening on
::, stable-privacy addresses for clients (L5.06). - Extension-header and fragment policy at the edge, without breaking legitimate traffic (L5.04).
ictNetworks scenario
ictNetworks' security team uses this threat model for its own network and for managed business customers. Its standard for every customer LAN, even the few that are still IPv4-only, is RA-Guard and DHCPv6-Shield on the access switches and a rogue-RA alert in the SOC. A recent incident at a customer, where a misconfigured Windows laptop with internet connection sharing sent RAs and broke the office's Internet access, is the case study used in this level.
Drawing diagram…
Worked examples
Example 1: Where the victim's traffic goes
A victim on an "IPv4-only" LAN receives the attacker's RA for 2001:db8:bad::/64 with an RDNSS option pointing at the attacker. It forms an address with SLAAC and a default route via the attacker's link-local address. When it resolves a dual-stack site, it gets a AAAA record (from the attacker's resolver if it chooses), prefers IPv6 and sends the traffic to the attacker.
Example 2: Intercepting neighbour resolution
To answer resolution for the victim 2001:db8:3001:1::25, an attacker must receive the NS, which is sent to the victim's solicited-node group ff02::1:ff00:25. On a switch that floods multicast (no MLD snooping), every host receives it, so the attacker can race the real owner with a forged NA.
Example 3: Blocking tunnels at an IPv4 perimeter
6to4 and ISATAP use IPv4 protocol 41. Teredo clients contact servers on UDP port 3544. Blocking both (outbound and inbound) at the IPv4 perimeter stops the common automatic tunnels, but not tunnels on other ports or VPNs, so monitoring is still needed.
Lab
Goal: Take over traffic on an "IPv4-only" LAN with a rogue RA, and document the impact.
You need: a Linux bridge in a network namespace lab (or a Proxmox VLAN) with an IPv4-only router (IPv4 address and NAT to a test server, no IPv6 configured), two victim hosts (Linux, and Windows if available) and an attacker host with Scapy (or THC-IPv6). The test server is dual-stack, reachable at 203.0.113.80 and 2001:db8:f000::80, with a AAAA record in the lab DNS.
-
Show that the victims have only link-local IPv6 addresses and no IPv6 default route, and that
curl http://www.lab.example/goes over IPv4 (capture on the router). -
Capture for two minutes and list every IPv6 packet the "IPv4-only" LAN already carries (DAD, MLD reports, RS, mDNS). Explain each.
-
From the attacker, send a rogue RA with Scapy:
from scapy.all import * ra = (Ether() / IPv6(src="fe80::bad", dst="ff02::1") / ICMPv6ND_RA(routerlifetime=1800, prf=1) / ICMPv6NDOptPrefixInfo(prefix="2001:db8:bad::", prefixlen=64) / ICMPv6NDOptSrcLLAddr(lladdr=get_if_hwaddr("eth1"))) sendp(ra, iface="eth1", loop=1, inter=10)Enable forwarding on the attacker with a route towards the test server. Show the victims' new address, default route and the
curltraffic now crossing the attacker over IPv6. -
Add an RDNSS option (
ICMPv6NDOptRDNSS(dns=["2001:db8:bad::53"])) and record which victims use it. -
Stop the attack. Record how long each victim keeps the rogue router and address, and how to clean up (
ip -6 route flush, waiting for lifetimes, or an RA with lifetime 0 from the same source). -
Write an impact statement (what an attacker could read or change) and the three controls you would deploy first.
What to submit: captures, before and after outputs from each victim, your Scapy script, the timing results and the impact statement.
Summary
- "IPv4-only" networks run IPv6 on the link, and a single rogue RA activates it with the attacker as router.
- Tunnels (protocol 41, Teredo on UDP 3544) can bypass IPv4-only perimeters.
- Map every IPv4 control to IPv6: first hop, perimeter, ICMP, logging, scanning assumptions.
- ND trusts every node on the link; SEND is not deployed, so the switch and monitoring must protect it.
- Priorities: first-hop security, equivalent perimeter policy, visibility, host hardening, extension-header policy.
- Next lesson (L5.02): filtering ICMPv6 correctly.
Quiz imported · 9 questions · pass mark 70%. Lab questions are not auto-marked.