Skip to content
Community

IPv6 for an IPv4 person: what's the difference between fe80:: link-local and a global address, and is SLAAC replacing DHCP?

22maya2n2 · posted · 53 views

Coming from IPv4 where every interface just had one address I configured (or got from DHCP). In the NetForge lab I enabled IPv6 on a router interface and suddenly show ipv6 interface lists two addresses: an fe80:: one I never typed, plus the 2001:db8:: global I set.

A few things I can't map back to IPv4:

  1. What is the fe80:: address for and why does it appear automatically?
  2. When would a host use link-local vs the global address?
  3. Everyone says "SLAAC" for auto-addressing — is that just DHCPv6 with a new name, or a totally different mechanism? When do I still need DHCPv6?

Trying to build the right mental model before I go further. Thanks!

  • One clarifier on the mental model: SLAAC's prefix is always a `/64` — the interface ID is a full 64 bits, so IPv6 subnets basically stop being variable-length inside a LAN. Don't try to SLAAC a `/120` and wonder why it's silent. — ariel.lahav22,

2 Answers

Accepted

Good instinct separating these out. Map it like this:

Link-local (fe80::/10) — every IPv6 interface auto-generates one the moment IPv6 comes up; you never configure it. It's only valid on that single link (routers never forward it off-segment). Its job is the plumbing IPv4 hid from you: Neighbor Discovery (NS/NA — the ICMPv6 replacement for ARP), Router Advertisements, and it's the next-hop for your routing table. Look at show ipv6 route — next hops are fe80:: addresses, not globals.

Global unicast (2000::/3, docs use 2001:db8::/32) — internet-routable, the equivalent of your public/routable IPv4. Hosts use link-local to talk to on-link neighbors and the router, and the global to reach anything off-link. So both coexist on one interface by design.

SLAAC vs DHCPv6 — genuinely different, not a rename:

  • SLAAC = StateLess Address AutoConfig. The router sends a Router Advertisement containing a /64 prefix; the host builds its own address (prefix + a self-generated 64-bit interface ID, EUI-64 or a random/privacy ID). Router keeps zero per-host state.
  • DHCPv6 = stateful, a server hands out and tracks specific addresses (like IPv4 DHCP).

The RA carries two flags that decide which:

  • M (Managed) = get your address from DHCPv6.
  • O (Other) = address via SLAAC, but fetch extras (DNS) from DHCPv6 — this is stateless DHCPv6.

Classic reason you still want DHCPv6: plain SLAAC historically didn't hand out DNS well, and you need it when you want address tracking/logging. In the lab, enable IPv6 on two router interfaces, debug ipv6 nd and watch the RA go out — you'll see the prefix the host uses to SLAAC itself.

ben31833 · answered
  • The "next-hop in the routing table is the fe80:: address" bit is what finally clicked for me — I kept wondering why routes didn't point at globals. Thanks! — 22maya2n2,
  • Worth adding a design note: because SLAAC keeps no per-host state, it scales beautifully but gives you no address-to-user record. If you need auditing/accountability, that's the real reason to reach for stateful DHCPv6 over the O-flag setup. — ariel.lahav22,

Ben covered the theory; here's the troubleshooting angle that makes it stick.

Remember: a host can ping and route with only a link-local address. So "IPv6 is up" and "IPv6 has a usable global address" are two separate checks. When something's broken, split them:

  1. show ipv6 interface brief — is there an fe80:: at all? No link-local = IPv6 isn't really enabled on that interface.
  2. Is there a global? If not, and you expected SLAAC, the host isn't hearing an RA. On the router the interface needs ipv6 address ... /64 and IPv6 unicast-routing on so it actually sends RAs.
  3. Ping the neighbor's fe80:: — but you must add the exit interface (e.g. ping fe80::1 won't work without specifying which link), because link-local isn't globally unique.

That last point trips up every IPv4 person: link-local addresses are only unique per link, so they always need a zone/interface scope. In the NetForge canvas, wire two routers, enable IPv6 on both ends, and watch the Neighbor Solicitation resolve on the cable — it's the exact same story as watching ARP resolve, just ICMPv6 instead.

Shira Sivroni · answered

Your answer

Posting questions and answers is included with every paid plan, from $15/mo. On the Free plan you can read, like, comment, connect and message. Compare plans →