In my previous article, IPv6 Tunneling over IPv4 Networks, I discussed how to connect a LAN to the IPv6 public Internet. This article continues the theme, discussing the issues surrounding the LAN’s configuration for IPv6.

Looking at how LAN’s are configured for IPv4 proves to be instructive. Many of the techniques, and all of the patterns turn out to have parallels in IPv6. During the design of IPv6 engineers took the opportunity for refactoring using lessons learned from IPv4. Initial host configuration was an area that received some attention, and so, we will explore the new methods provided by IPv6 as well as the traditional techniques carried forward from IPv4.

Network Autoconfiguration

The goal here is to identify how hosts joining an IPv6 network can come to discover and then interoperate on that network. The network may also have IPv4, but it should not be required. Specifically, the goal is to show how hosts can, with emphasis towards automation, configure themselves on an IPv6 only network. The same (pre)configuration should also work if the network happens to support IPv4, and the hosts should then configure and join the IPv4 network as well.

There are two use cases that are used to vet the proposed methods. The first is the “small lab” use case. Think of a small, less than 100 hosts, lab environment, or SOHO office. In this sort of environment some of the hosts are very dynamic, and some are more static with specific functions assigned to them; for instance, a mail, or IRC server. The second use case is the prototype environment for a distributed application. Often such environments are created with VM’s using tools such as Vagrant inside a developer’s laptop. Such environments can be paused and restarted, relocated, and the hosts are often destroyed and recreated for a “fresh install”. In both cases, formal rigor in configuring and maintaining the network are not as important as a degree of automation combined with flexibility to manually adjust things as needed for the task at hand.

In an IPv4 world, DHCP combined with DNS are the go-to tools for such configuration. As luck has it, both are available in the IPv6 world as well. DNS remains unchanged, except for the addition of the AAAA record that holds IPv6 addresses; AAAA records are retrievable via IPv4 as well. DHCP functions the same, but the specifics are adapted for IPv6 and to disambiguate, it is known as DHCPv6; there is a separate RFC: RFC-3315.

IPv6 Autoconfiguration

During the design of IPv6, then codenamed ‘IPng’, the design of “autoconfiguration” for IP was refactored to support a more automated method by which hosts could join an IPv6 network. In the base IPv4 protocol there is no explicit support for autoconfiguration. In IPv6 autoconfiguration support was initially integrated into ICMPv6. This comes in the form of Neighbor Discovery Protocol (NDP) and includes the ability for a host to automatically discover:

  • Neighbor Discovery and Advertisement, replacing IPv4’s ARP.
  • Network address prefix value.
    • Note: network address prefix length is fixed in IPv6 to /64.
  • Link (L2) parameter discovery, such as MTU.
  • Next-hop routing determination through Router Solicitation and Advertisement (RA).
  • Duplicate address detection, thus allowing hosts to generate a host address and determine if it is a duplicate. (If so, they will pick another).

The above tools that made up the original NDP allow a host to perform what is formally called Stateless address autoconfiguration (SLAAC). Additionally, NDP, through Router Solicitation and Router Advertisement (RA), allows a host to define itself a unique host address, determine its network address, and router(s). These are all the necessary details a host needs to start sending and receiving IP packets. The IPng designers believed they had successfully refactored the base IP protocol, including ICMP, to handle autoconfiguration.

In hindsight, the refactoring was not agile enough. While hosts can configure the IP layer with SLAAC, it turns out that DHCPv4 is used provide added functionality beyond simply configuring the IP layer of a host. The almost universally used feature of DHCP, beyond the IP layer, is communicating the DNS server’s IP address. SLAAC did not initially have support for this and thus DHCPv6 was born. The ICMPv6 Router Advertisement was extended in RFC-6106 to allow the inclusion of DNS configuration information in the RA message.

RFC-6106, which allows DNS details in router advertisements, is not the end of the story. Firstly, its adoption has been slow, and secondly, there remain additional configuration details not covered by the RFC that are desired when autoconfiguring hosts on an IP network. In fact, IANA cites well over 150 option parameters registered for DHCP [1].

Stateful and Stateless DHCPv6

The ability for a host to autoconfigure, through SLAAC, it’s IPv6 details gives rise to a new mode of use for DHCP - “Stateless DHCPv6”. In this mode, the host first uses SLAAC to configure its IPv6 details and then uses DHCPv6 to request additional details like DNS, NTP, etc. In this mode the host indicates it is operating “stateless” and the server does not perform address assignment.

In addition to the stateless mode, DHCPv6 can also be used in a “stateful” mode where an IPv6 address is assigned from the server’s pool of addresses. IPv6 allows and most often requires hosts to have multiple IPv6 addresses assigned to a single interface; the link-local address is an excellent example of this. Because hosts must support multiple addresses per interface it is possible to use both SLAAC and stateful DHCPv6 to configure a host. If both methods are used then there will be multiple IPv6 addresses assigned to the interface. The link-local address is also required, so using both methods will ensure the interface has at least three addresses.

The late arrival of DNS configuration as part of SLAAC caused some vendors, Microsoft Windows most notably, to pursue client IPv6 autoconfiguration which required DHCPv6. This makes the use of DHCPv6 a near requirement in any network that has liberal requirements for operating system support.

Host Registration in DNS

Providing DNS configuration to an autoconfiguring host does not imply registering that host in the local DNS tables. In fact, neither DHCPv4, nor DHCPv6 address the issue of DNS registration for newly configured hosts. In many environments, including lab and prototype use case environments, registering configured hosts ranges from very helpful to required. Some clients will perform this task, but this behavior is not common enough to rely on it.

There are a number of different tools to support registering hosts in DNS as they join a network, but one project is more appealing than the others for our given use cases: Dnsmasq. The Dnsmasq project combines DHCP and DNS in a single daemon and supports both IPv4 and IPv6. Dnsmasq also supports portions of IPv6 autoconfiguration, including router advertisement (RA). The feature that places Dnsmasq in the most appealing position is that its DHCP and DNS integration includes automatically registering DHCP leases in the DNS tables when a hostname is provided in the DHCP request; solving exactly the problem not explicitly addressed in DHCP or DNS specifications.

Through experimentation it was determined that Windows and Mac OSX based systems consistently provide the hostname as part of their DHCP request. Unfortunately, the most popular, and widely used, DHCP client, from ISC, either does not, or is rarely configured to send the hostname. Fortunately, an alternative DHCP client does: the dhcpcd client. Investigating the dhcpcd client also revealed that it is trivial to swap the ISC client for dhcpcd on most Linux hosts.

Router and Client Configuration

The following is the Dnsmasq configuration applied on the router that allows for the most effective autoconfiguration of IPv6:

# dnsmasq configuration for router "appliance"





The configuration is broken down as follows:

no-resolve:Disable using /etc/resolv.conf as a basis for configuring the dnsmasq server. This was done to make this example clearly explicit.
 Configure the server that dnsmasq forwards all DNS requests it can not handle locally to. Note that the forwarder does not need to be a IPv6 address, although it could be. Multiple forwarders are configurable if desired.
local=/cloud1/:Declare the domain “cloud1” as the domain being served locally.
domain=cloud1:Declare the domain “cloud1” as the domain for all DHCP requests
dhcp-fqdn:Force all DHCP clients to be placed in the “cloud1” domain, regardless of what domain they specify in the DHCP request.
enable-ra:Perform IPv6 Router Advertisement as part of Dnsmasq’s operation. Other router advertisement daemons should not be run. If the host Dnsmasq is running on is not the router then disable this.
 Explicitly configure and ensure the dns-server option is sent in the DHCP reply. The address listed is an address assigned to the ‘em1’ interface on this host.
 Send ‘cloud1’ as the assigned domain to all clients performing DHCP requests.
dhcp-range…:Issue IPv6 addresses between ::100 and ::1ff in response to DHCP requests. The clause, “constructor:em1” directs the configuration to use the network prefix of the ‘em1’ interface as the network prefix for the leased addresses. The actual address returned will be [em1 prefix]::[100-1ff].

Enabling the log-dhcp or log-queries parameters in Dnsmasq will enable verbose reporting of either DHCP or DNS is debugging is required.

Client DHCP Configuration

No explicit configuration is required on the client side except replacing the ISC DHCP client with the dhcpcd client. For Debian derived Linux installations:

sudo apt-get remove isc-dhcp-client isc-dhcp-common
sudo apt-get install dhcpcd5

No configuration files need to be modified.


By utilizing the Dnsmasq and dhcpcd projects a very simple configuration can be constructed that supports autoconfiguration of an IPv6 network. This pattern can be utilized in both a heterogeneous lab and prototype environments, including Vagrant based setups on developer laptops.


comments powered by Disqus