IPv6 Static Addresses using Tokens

From Gentoo Wiki
Jump to: navigation, search

Network admins may require static, simple and stable IP addresses for both IPv4 and IPv6. The goal of this article is to assign a static custom IPv6 suffix to an interface without additional software while keeping full RA functionality from the router.

When combined with a fix ULA Prefix advertised from the router, IPv6 Addresses can then be simple, organized, static and easily be handled in DNS.

Default kernel IPv6 address creation mechanism

In a IPv6 network, addresses are often self-assigned (constructed) by the kernel using both PREFIX: information provided by a router (DHCPv6 server) and :SUFFIX information extracted from hardware (MAC address). The router mechanism for this functionality is called "Router Advertisement" (RA). A machine usually has multiple IPv6 addresses for global, link or network scope. If the IPv6 Prefix can change (for example on a dialup connection), static addresses can not be assigned as easily as on IPv4.

Example: default MAC construction

Take a look at a simple example configuration with IPv6 addresses completely self-constructed with prefix information received by a router. Note that the IPv4 configuration is already static:

FILE ~/etc/conf.d/net
rc_keyword="-stop"
config_eth0="192.168.0.35/24" 
routes_eth0="default via 192.168.0.2"

when the interface is started, the address configuration could like this:

root #ip addr show eth0
1: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
        link/ether 9a:02:79:45:ce:d2 brd ff:ff:ff:ff:ff:ff
        inet 192.168.0.35/24 brd 192.168.0.255 scope global eth0
           valid_lft forever preferred_lft forever
        inet6 2a02:810a:8240:a2c:9802:79ff:fe45:ced2/64 scope global
        inet6 fd00::9802:79ff:fe45:ced2/64 scope global
        inet6 fe80::9802:79ff:fe45:ced2/64 scope link
           valid_lft forever preferred_lft forever

Note that the Suffix part 9802:79ff:fe45:ced2 is constructed out of the MAC address.

Drawbacks on MAC-constructed addresses

There a couple drawbacks of MAC-constructed addresses, for example

  • complicated to read
  • change when the MAC address changes (hardware, VM setup etc)
  • not associated with their IPv4 counterpart

IPv6 tokenized interface identifiers

Static IPv6 address problems

The word "static IPv6 addresses" is a bit misleading. When on dialup, the Prefix may change and so does the global IPv6 address of the interface. If the machine needs access to the internet without an additional router or NAT, it needs a public IPv6 address with the same prefix as the IPv6 gateway.

Within a closed network, for example a LAN, that's not important. For DNS and connectivity within a closed network, the Unique Local Address (or ULA) is important. This address can be fully fixed.

Since the Prefix can change, static addresses cannot be assigned for all scopes. When using RA to get IPv6 configured, an additional IPv6 address could be entered in /etc/conf.d/net. Drawback: The kernel still constructs and adds another address using it's constructions defaults. And this kernel-auto-created address becomes route default.

Solution: Tokens

An easy solution is called IPv6 Tokenized Interface Identifiers: tell the kernel which SUFFIX to use for a particular interface.

When combined with a fixed ULA Prefix (Unique Local Address Prefix) announced from the router, the Local Address (for example fd00:) will stay the same while the Global Address gets updated by the kernel as soon as the router announces a prefix change. And thanks to the token, the Global Address also keeps it's suffix!

Unfortunately, IPV6 tokens are not supported by netifrc yet, which is Gentoo's default framework for configuring network when using OpenRC. Further more, tokens can not be configured using sysctl. Tokens can only be set using the ip (or ifconfig) command.

Example: Assign the simple suffix ::35 to interface eth0, in order to match the IPv4 counterpart 192.168.0.35:

root #ip token set ::35 dev eth0

Restart the interface, the changes apply:

root #rc-service net.eth0 restart
root #ip addr show eth0
1: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
        link/ether 9a:02:79:45:ce:d2 brd ff:ff:ff:ff:ff:ff
        inet 192.168.0.35/24 brd 192.168.0.255 scope global eth0
           valid_lft forever preferred_lft forever
        inet6 2a02:810a:8240:a2c::35/64 scope global
        inet6 fd00::35/64 scope global
        inet6 fe80::9802:79ff:fe45:ced2/64 scope link
           valid_lft forever preferred_lft forever

Notice the new ::35 suffix on all IPV6 addresses, except for the Link Local address fe80:. That's by design, but can be changed too if desired by changing the address after its creation:

root #ip addr flush scope link dev eth0
root #ip addr add fe80::35/64 dev eth0
root #rc-service net.eth0 restart
root #ip addr show eth0
1: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP group default qlen 1000
        link/ether 9a:02:79:45:ce:d2 brd ff:ff:ff:ff:ff:ff
        inet 192.168.0.35/24 brd 192.168.0.255 scope global eth0
           valid_lft forever preferred_lft forever
        inet6 2a02:810a:8240:a2c::35/64 scope global
        inet6 fd00::35/64 scope global
        inet6 fe80::35/64 scope link
           valid_lft forever preferred_lft forever

Now all IPv6 addresses on eth0 have the simple suffix of ::35.

In this example, the Router additionally announces a fixed ULA Prefix (Unique Local Address Prefix) of fd00:. The full Unique Local Address (ULA) is now simply fd00::35 - this address is used for all traffic within the local network (LAN). The global (Public, WAN side) address of this interface is now 2a02:810a:8240:a2c::35.

fd00::35 can now be registered in local DNS - it will never change (unless the ULA Prefix would change).

The interface now has fully functional, easy to read and manage IPv6 addresses while keeping full RA functionality from the router. This means that the kernel will instantly re-configure the global IPV6 address when for example the Prefix changes.

Putting it together: On boot

Since netifrc does not support tokens yet out of the box, a parameter cannot be added to /etc/conf.d/net for configuration.

A simple workaround is to call an ip command before launching an interface. This can easily be achieved by implementing hook functions directly in /etc/conf.d/net:

FILE /etc/conf.d/net
rc_keyword="-stop"
config_eth0="192.168.0.35/24" 
routes_eth0="default via 192.168.0.1"

# set IPv6 interface token
preup() {
  ip token set ::35 dev eth0
  return 0
}

# optional: assign the token ::35 address to fe80:
postup() {
  ip addr flush scope link dev eth0
  ip addr add fe80::35/64 dev eth0
  return 0
}
Note
The setup listed above will not work on bridges due to a design limitation of netifrc: At the point in time when the preup() function is called, the bridge interface does not exist yet.

A workaround for bridges is to only use the postup() function with a slight modification. The drawback: When creating (starting) the bridge, it has more addresses than necessary. But this can be ignored since the auto-generated (non-tokenized old MAC) addresses automatically vanish after their expiration time ("valid_lft").

FILE /etc/conf.d/net
rc_keyword="-stop"
config_br0="192.168.0.35/24" 
routes_br0="default via 192.168.0.1"

# set IPv6 interface token
# and set the IPv6 token to fe80: identifier
postup() {
  ip token set ::35 dev br0
  ip addr flush scope link dev br0
  ip addr add fe80::35/64 dev br0
  return 0
}