5

I am studying up on IPv6 and part of the talk was about how the machines generate or can have multiple IPv6 addresses for internal use and external use.

Does this mean that there is no longer a "public" IP address that all the machines inside a network will use?

The video guide also mentioned that the point of the "privacy" IPv6 was to keep people from accessing the machine directly. Do firewalls no longer exist or was this just a poorly worded explanation?

This was the CBT IPv6 video series.

9

Let's look at a live example. This being from my Linux workstation. (And for simplicity I actually omitted a couple of addresses.) I'll explain each of the addresses in turn:

$ ip a s dev br0
3: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
    link/ether fc:aa:14:25:f1:f1 brd ff:ff:ff:ff:ff:ff
    inet 172.25.50.17/24 brd 172.25.50.255 scope global br0
       valid_lft forever preferred_lft forever

Here is your ordinary RFC 1918 address.

    inet6 2001:db8:8301:424c:71a2:70f2:6b1d:a9af/64 scope global temporary deprecated dynamic 
       valid_lft 2120sec preferred_lft 0sec
    inet6 2001:db8:8301:424c:3caa:9ce6:65ef:1f15/64 scope global temporary deprecated dynamic 
       valid_lft 2120sec preferred_lft 0sec

Next we get to global IPv6 addresses. These are routable on the public Internet, and subject to firewall rules, bidirectional communication is possible. That means that anyone (with IPv6) can initiate a connection to this system. Because NAT is not in use, the connection is direct, without any need for port forwarding hackery. But a host or network firewall may still prohibit the connection.

But these IPv6 addresses are also RFC 4941 privacy addresses. These addresses are created at configurable intervals, and then the previous privacy address is deprecated. Once an address is deprecated it is no longer used for new outgoing connections. After another configurable interval, the deprecated address is automatically removed from the interface, and no communication is possible at all.

Privacy addresses are meant to protect the privacy of the end host, especially as it moves from one network to another. SLAAC addresses configured using modified EUI-64 always have the same last 64 bits, regardless of what network they connect to, so a host (such as a laptop) could be tracked as it moved across networks by correlating these. Privacy addresses eliminate this problem.

    inet6 2001:db8:8301:424c:8e6e:e3b8:ce7e:aff8/64 scope global mngtmpaddr noprefixroute dynamic 
       valid_lft 2120sec preferred_lft 2120sec

This is a fun one. It's also a global address, but it's an RFC 7217 stable privacy address. These are not simply random, as with RFC 4941 addresses, but generated using a PRNG fed a secret key combined with host-specific information and the assigned IPv6 prefix. They are still unpredictable, but once created, they remain the same for any given IPv6 prefix.

Ordinary privacy addresses have the problem that a host using them has no fixed address at which it can be reached when it is on a given network. So incoming connections are impossible. This wasn't really the point, but more of a side effect (RFC 4941 section 2.4 discusses this in detail.). If you want incoming connections, but still want privacy when the host moves around, then stable privacy addresses come to the rescue. The host will get a different interface identifier when it moves to another network, but when it goes back to its original network, it will get the same interface identifier it had previously.

So, on this system, outgoing connections use the RFC 4941 privacy addresses, which are rotated at intervals, while incoming connections use the stable privacy address, which is published in the global DNS.

    inet6 fda8:75f3:eca7:0:3caa:9ce6:65ef:1f15/64 scope global temporary deprecated dynamic 
       valid_lft 442822sec preferred_lft 0sec
    inet6 fda8:75f3:eca7:0:d05b:6572:c01c:107/64 scope global temporary deprecated dynamic 
       valid_lft 357022sec preferred_lft 0sec
    inet6 fda8:75f3:eca7:0:81e4:b3f8:bb92:5fa5/64 scope global temporary deprecated dynamic 
       valid_lft 271224sec preferred_lft 0sec

Here we have unique local addresses, as defined in RFC 4193. These are routable across subnets, and even across wide area networks, but they are not allowed on the public Internet. This makes them usable as private addresses for any network from the smallest home network to the largest global enterprises.

A ULA prefix is always a /48 within fd00::/8 (technically fc00::/7 but the other half is meant to be assigned by IANA) and is meant to be constructed by an algorithm given in RFC 4193, but can be done randomly. It must not be assigned from zero or in any other pattern. If a /48 is not large enough, or two companies merge and need to connect their networks, multiple /48s can be used and the appropriate routes created. The 2^40 space of possible subnets is 256 times the number of possible IPv4 addresses, making collisions extremely unlikely if the ULA prefix is generated properly.

The three addresses shown above are deprecated RFC 4941 privacy addresses.

    inet6 fda8:75f3:eca7:0:71a2:70f2:6b1d:a9af/64 scope global temporary dynamic 
       valid_lft 528621sec preferred_lft 9621sec

Here is the currently valid RFC 4941 privacy address.

    inet6 fda8:75f3:eca7:0:3014:2169:37ba:ebd1/64 scope global mngtmpaddr noprefixroute 
       valid_lft forever preferred_lft forever

Here is a unique local address which is also an RFC 7217 stable privacy address. You can see that its interface identifier is different from the interface identifier of the global stable privacy address.

    inet6 fda8:75f3:eca7::2e4/128 scope global 
       valid_lft forever preferred_lft forever

And this unique local address was assigned via DHCPv6.

    inet6 fe80::6ef0:f0b6:9b0d:26c6/64 scope link 
       valid_lft forever preferred_lft forever

Finally, this is a link-local address. It is not routable, and can be reached only on the same layer 2 network. Because this host is using RFC 7217 stable privacy addresses, it too is a stable privacy address.


On this network, this host can be reached globally at a stable address which also leaks no information about the host's MAC address. It can also be reached via its unique local address by other hosts in several networks on the site to and from which these addresses are routed. And all of these addresses might be used simultaneously.

  • 1
    As always Michael, amazing answer! – Jason Nov 18 '16 at 17:03
3

For the most part, IPv6 hosts will be assigned Global IPv6 addresses, There are far more of those than can even be imagined for the foreseeable future. All the hosts in in the world can have public IPv6 addresses.

One major weakness of IPv4 is that there are not nearly enough IPv4 addresses to assign a unique IPv4 address to every host, and that was one of the IP design foundations. To forestall the IPv4 address shortage until IPv6 becomes ubiquitous, NAT was created, but NAT is a kludge that breaks the IP end-to-end paradigm, and it causes problems for some things. IPv6 restores the end-to-end paradigm by providing enough IPv6 addresses that every host can have a unique IPv6 address.

Yes, there will still be firewalls. IPv6 can use firewalls just like IPv4 can. There is no change that prevents that. Firewalls block or allow traffic as the firewall rules have been configured.

You may be confusing NAT with firewalls, but the two really have nothing to do with each other. In many cases, a firewall is a convenient place to configure NAT and routing, but neither of those require a firewall. Firewalls are what secure your network, and firewalls run perfectly fine without NAT. In many companies, there are lots of firewalls which do not have NAT configured. Also, there are many places NAT is used which are not on a firewall. All NAT does is to replace either the source or destination, or both, IPv4 addresses in packets with different IPv4 addresses. That is particularly useful if all you have are private IPv4 addresses, and you need to connect to the public Internet, but that is not how IPv4 was originally designed, and it is the shortage of IPv4 addresses which forced NAT on us.

While there are RFCs covering IPv4 NAT, the RFC for IPv6 NAT is classified Experimental, and there may never be a Standard RFC for IPv6 NAT; it simply isn't needed, and the Experimental RFC for IPv6 NAT breaks some things in IPv6.

There are IPv6 ULA (Unique Local Addresses). These addresses can never connect to the public Internet. These are kind of like IPv4 Private addresses, but there are some rules around ULA. The first half of the address block is reserved for a global authority to assign them. The second half of the address block can be used by anyone, but there are 40 bits immediately after the ULA prefix that must be randomly chosen; you can't simply start with 0 then increment for each network. The idea is that you end up with a globally unique set of network addresses.

  • 3
    One of the biggest problems that NAT has caused is that a whole generation of developers and system and network engineers think it is normal, rather than the messy hack that it actually is, which leads to confusion when they try to learn how routing actually works. – Michael Hampton Nov 17 '16 at 5:04
  • One addition about ULA: it's not just used by hosts that can never connect to the internet. Another use case is to use it side by side with normal global addresses in case the global addresses aren't necessarily stable, like with some consumer ISPs. ULA can provide stable addresses inside the house even when the global addresses change. – Sander Steffann Nov 17 '16 at 5:41
  • And using multiple addresses on one host is totally fine with IPv6 :) – Sander Steffann Nov 17 '16 at 5:42
  • 1
    @SanderSteffann I suggested an edit that hopefully makes that more clear. – Michael Hampton Nov 17 '16 at 5:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.