IPv6 at Scale

IPv6 at Scale

Synopsis #

This chapter describes operating IPv6 at scale on OpenBSD: interface addressing, Router Advertisements (RA), Neighbor Discovery (ND), upstream integration, and production guardrails. It uses the in-base router advertisement daemon rad(8) with configuration in rad.conf(5) , client autoconfiguration with slaacd(8) , ND inspection via ndp(8) , interface management with ifconfig(8) , and packet filtering in pf.conf(5) applied by pfctl(8) . Use these patterns for campus segments, data-center pods, and ISP edges where you operate multiple /64s or larger aggregates.

Design Considerations #

  • Prefix plan. Allocate one /64 per L2 segment. For campuses and pods, obtain at least a /56 (256 × /64) or /48 (65,536 × /64) from your provider. Use hierarchical addressing that mirrors your failure domains and routing boundaries.
  • Forwarding model. Route between VLANs; avoid L2 extension across failure domains. IPv6 first-hop redundancy can reuse the CARP pattern from the HA chapter.
  • Router Advertisements. RAs convey on-link prefixes, default gateway, DNS (RDNSS), and search domains (DNSSL). Keep RA timers consistent and advertise exactly one /64 per L2.
  • ND and ICMPv6. Do not block ICMPv6. Neighbor Discovery, Path MTU Discovery, and error handling depend on it.
  • Prefix Delegation (PD). Upstreams may delegate a /56 or /60 dynamically. Plan how you subdivide and advertise the delegated space to downstream L2s. Review the release notes for OpenBSD 7.8 and slaacd(8) for current PD client capabilities.
  • NPTv6. Prefer proper routing. Only consider Network Prefix Translation (NPTv6) for constrained renumbering scenarios, and test thoroughly. See pf.conf(5) for current support and caveats.
  • Operations at scale. Standardize RA defaults, centralize logging, and instrument ND tables and error counters per segment. Use change windows to test DAD, RA reception, and PMTU.

Configuration #

The examples assume:

  • Router with interfaces em1 (LAN-A VLAN 10), em2 (LAN-B VLAN 20), em0 (WAN).
  • Assigned aggregate 2001:db8:10::/56, where 2001:db8:10:0010::/64 is LAN-A and 2001:db8:10:0020::/64 is LAN-B.

1) System forwarding and interface addressing (router) #

Enable IPv4/IPv6 forwarding and assign IPv6 addresses to the router’s LAN interfaces.

# printf '%s\n' \
    'net.inet.ip.forwarding=1' \
    'net.inet6.ip6.forwarding=1' \
    >> /etc/sysctl.conf
  # Persist L3 forwarding

# sysctl net.inet.ip.forwarding=1 net.inet6.ip6.forwarding=1
  # Apply immediately

# cat > /etc/hostname.em1 <<'EOF'
inet6 2001:db8:10:10::1 64
up
EOF
  # LAN-A: router address and prefix length

# cat > /etc/hostname.em2 <<'EOF'
inet6 2001:db8:10:20::1 64
up
EOF
  # LAN-B: router address and prefix length

# sh /etc/netstart em1 em2
  # Bring the interfaces up now

2) Router Advertisements with rad(8) #

Advertise one /64 per L2. Include RDNSS and DNSSL so hosts learn DNS settings via RA.

## /etc/rad.conf — advertise /64s and DNS on two interfaces

# LAN-A (VLAN 10)
interface em1
    prefix 2001:db8:10:10::/64
    rdnss 2001:db8:10:10::53
    dnssl example.internal

# LAN-B (VLAN 20)
interface em2
    prefix 2001:db8:10:20::/64
    rdnss 2001:db8:10:20::53
    dnssl example.internal

# rcctl enable rad
  # Start at boot
# rcctl start rad
  # Launch now; begin sending RAs

Keep one advertising router per segment during initial rollout. If you deploy redundant routers with CARP, ensure both run rad(8) and that the virtual default gateway address appears in DNS and documentation.

3) Client autoconfiguration (hosts) #

OpenBSD hosts use slaacd(8) ) to form addresses from RA. Ensure the interface requests autoconfiguration.

## /etc/hostname.em1 on a host
inet6 autoconf
up

# rcctl enable slaacd
# rcctl start slaacd
  # The host will configure a global address and default route from RA

4) PF allowances for IPv6 control-plane and data #

Permit essential ICMPv6, allow LAN traffic, and apply a conservative scrub for PMTU stability. Adjust to your security model as needed.

## /etc/pf.conf — minimal IPv6 allowances

set skip on lo

lan = "{ em1, em2 }"
wan = "em0"

block all

# ICMPv6 is mandatory for ND and PMTU; allow it on all interfaces
pass inet6 proto icmp6 keep state

# Permit LAN IPv6; constrain further per policy
pass in on $lan inet6 from { 2001:db8:10:10::/64, 2001:db8:10:20::/64 } to any keep state

# Outbound to WAN
pass out on $wan inet6 from any to any keep state

# Conservative scrub to avoid PMTU black holes during rollout
scrub in on $wan inet6

# pfctl -f /etc/pf.conf
# pfctl -sr | grep icmp6
  # Verify ICMPv6 allowance is active

5) Upstream integration and Prefix Delegation (overview) #

When an ISP delegates a prefix (for example, a /56), subdivide it deterministically (for example, VLAN-ID ↔ 0xNN in 2001:db8:10:NN::/64) and advertise per-L2 via rad(8). Depending on release and provider requirements, prefix delegation may be delivered dynamically. Review OpenBSD 7.8 documentation for slaacd(8) and plan operational procedures (lease monitoring, re-advertisement on change). Where PD is not available, use static assignment from your aggregate.

6) Notes on NPTv6 #

Where renumbering pressure is high and routing is not an option, NPTv6 can translate one global /64 to another while preserving host bits. Prefer native routing; reserve NPTv6 for edge-cases and lab validation. For syntax and constraints, consult pf.conf(5) for your OpenBSD 7.8 release and test in an isolated segment before production.

Verification #

  • Addresses and routes: confirm per-L2 /64s and the default route.
$ ifconfig em1 em2 | grep inet6
  # Verify router and host global addresses
$ route -n show -inet6
  # Confirm default route via the on-link router on hosts; confirm connected /64s on the router
  • Neighbor Discovery: inspect neighbor caches and router entries with ndp(8) .
$ ndp -an
  # Neighbor and router cache; look for stale or incomplete entries
  • RAs and ND on the wire: capture ICMPv6 on a LAN with tcpdump(8) .
# tcpdump -ni em1 icmp6
  # Expect Router Advertisements, Neighbor Solicitations/Advertisements
  • DNS via RDNSS: verify that hosts learned DNS servers via RA.
$ getent hosts www.example.internal
  # Name resolution using learned RDNSS

Troubleshooting #

  • Hosts do not get IPv6 addresses. Confirm rad(8) is running and advertising on the correct interface, net.inet6.ip6.forwarding=1 on routers, and that PF is not blocking ICMPv6. Use tcpdump -ni emX icmp6 to observe RAs.
  • Duplicate Address Detection (DAD) failures. If an address remains tentative or is marked duplicate, check for stray RA senders or misconfigured anycast. Inspect ndp -an for conflicting MACs.
  • Intermittent reachability or stalls. Packet Too Big messages may be blocked. Ensure ICMPv6 is permitted end-to-end and verify PMTU with large pings (ping -s).
  • Unexpected router selection. Ensure only the intended routers advertise on each L2. During migrations, disable RA on the retiring router first, then withdraw prefixes.
  • Prefix changes with PD. When the delegated prefix changes, update interface addresses and rad.conf(5) promptly. Automate downstream updates or use deterministic mapping from the delegated aggregate.
  • Firewall denies IPv6. Verify that your policy contains explicit inet6 rules. Mixed inet/inet6 environments often mistakenly permit only IPv4.

See Also #

Report a bug