IPv6 Basic Services

1
IPv6 Basic Services

• Stig Venaas
• sv_at_ecs.soton.ac.uk
• University of Southampton/UNINETT

2
Overview

• Neighbour discovery
• Autoconfiguration
• PathMTU discovery
• DHCP
• DNS
• NTP

3
Multicast addresses

• 4 bits flags (well known/transient), 4 bits
  scope, 32 bits group ID
• Node local ff01, link local ff02, site local
  ff05, organization local ff08
• Some special multicast addresses
• All nodes addresses ff011, ff021
• All routers addresses ff012, ff022, ff052
• Solicited-node address ff021ffXXXXXX
• For each unicast address a node responds to the
  solicited-node address where the Xes are the
  last 24 bits of the unicast address.

4
Neighbour discovery

• RFC 2461, Replaces IPv4 ARP, no broadcast
• Uses multicast and ICMP
• Address resolution
• Sends solicitation message to the solicited-node
  multicast address of the target address (sender
  includes its link layer address). Target responds
  with an advertisement message containing its link
  layer address
• Individual hosts will receive much less address
  resolution packets than in IPv4
• Using ICMP is more media-independent than ARP and
  allows for IP security mechanisms.

5
Auto configuration(1)

• RFC 2462, stateless autoconfiguration of
  addresses, default router, MTU etc. Not DHCPv6
  (separate protocol)
• Uses multicast and ICMP
• Router advertisement messages
• Sent by routers at regular intervals or when
  prompted
• Info about prefixes, hop limit, MTU, life time
  etc.
• Sent to all-nodes multicast address or a specific
  host
• Router solicitation messages
• Prompts for a router advertisement, advertisement
  is sent to the source address.

6
Auto configuration (2)

• Interface identifier
• 64 bit identifier, required to be unique on the
  link
• One way is IEEE EUI-64, for ethernet EUI-64 gives
  a way of creating 64 bits identifier from 48 bits
  MAC
• Link local address FE80EUI-64
• Check uniqueness by sending neighbor solicitation
• Global and site-local addresses
• Append interface identifier to known prefixes
• Prefixes usually known from router advertisements
• Hosts also use Neighbour Discovery to check that
  addresses are unique on the link (Duplicate
  Address Detection)

7
PathMTU discovery

• IPv6 requires MTU to be at least 1280 bytes
• PathMTU discovery always done
• Minimal implementations can avoid it by using the
  minimal MTU of 1280
• Done similar to IPv4
• Only sender fragments (like IPv4 with DF) and
  usually only receiver reassembles
• If router is about to forward a packet out an
  interface with MTU less than packet size, it will
  send ICMP Packet Too Big to the packets source
  address
• Fragmentation should be avoided, better that the
  application reduces packet size

8
DHCP (1)

• IPv6 has stateless address autoconfiguration but
  DHCPv6 (RFC 3315) is available too
• DHCPv6 can be used both for assigning addresses
  and providing other information like nameserver,
  ntpserver etc
• If not using DHCPv6 for addresses, no state is
  required on server side and only part of the
  protocol is needed. This is called Stateless
  DHCPv6 (RFC 3736)
• Some server and client implementations only do
  Stateless DHCPv6 while others do the full DHCP
  protocol
• The two main approaches are
• Stateless address autoconfiguration with
  stateless DHCPv6 for other information
• Using DHCPv6 for both addresses and other
  information to obtain better control of address
  assignment

9
DHCP (2)

• One possible problem for DHCP is that DHCPv4 only
  provides IPv4 information (addresses for servers
  etc) while DHCPv6 only provides IPv6 information.
  Should a dual-stack host run both or only one
  (which one)?
• Several vendors working on DHCP but only a few
  implementations available at the moment
• DHCPv6 http//dhcpv6.sourceforge.net/
• dibbler http//klub.com.pl/dhcpv6/
• NEC, Lucent etc. are working on their own
  implementations
• Cisco routers have a built-in stateless server
  that provides basic things like nameserver and
  domain name (also SIP server options in image I
  checked).
• DHCP can also be used between routers for prefix
  delegation (RFC 3633). There are several
  implementations. E.g. Cisco routers can act as
  both client and server

10
DNS IPv6 transport

• Most name servers are not accessible over IPv6
• Situation is improving. Several TLD servers are
  accessible over IPv6 (including .com)
• A few root servers are now accessible with IPv6
  but not registered with AAAA in DNS
• I believe the issue is that if you ask for NS for
  . The additional section with addresses should
  fit within a single packet (512 bytes when not
  supporting EDNS0)
• IPv6-only hosts will need help of a dual-stack
  name server for resolving (translation possible
  solution but not recommended)
• This may be as simple as putting the IPv6 address
  of recursive dual-stack nameserver in named.conf
• Only a few stub resolver libraries support IPv6
  transport, glibc on Linux does. One solution is
  using another resolver library or running a
  purely caching nameserver locally
• BIND9 and possibly nsd supports IPv6 transport as
  standard
• Patches available for djbdns (dont think native
  yet)

11
DNS storing IPv6 info

• Practically all up to date nameservers allow you
  to store DNS data
• Almost like IPv4, e.g.
• sverresborg AAAA 2001700e000020475fffee4423
  b
• storhaugen AAAA 2001700e000029027fffe22
  7186
• ORIGIN 0.0.0.0.0.0.0.e.0.0.7.0.1.0.0.2.ip6.arpa
• b.3.2.4.4.e.e.f.f.f.5.7.4.0.2.0 PTR
  sverresborg.uninett.no.
• 6.8.1.7.2.2.e.f.f.f.7.2.0.9.2.0 PTR
  storhaugen.uninett.no.
• ORIGIN as above or a zone per /64 makes things a
  little easier
• Some resolvers still use ip6.int, if you really
  want you can register both. If no ORIGIN in
  zone file you can use same file for ip6.arpa and
  ip6.int

12
DNS named.conf for BIND9

• For IPv6 transport add e.g.
• listen-on-v6 any
• transfer-source-v6 to specify IPv6 source
  address for transfers
• query-source-v6 address port to specify
  IPv6 source for queries
• IPv6 addresses can be used in ACLs
• Rest as normal. With a zone for each /64 you
  might have e.g.
• zone 0.0.0.0.0.0.0.e.0.0.7.0.1.0.0.2.ip6.arpa (
• type master
• file pz/0.0.0.0.0.0.0.e.0.0.7.0.1.0.0
  .2.ip6.arpa
• )
• If want to support ip6.int and no ORIGIN in the
  file, we can also do
• zone 0.0.0.0.0.0.0.e.0.0.7.0.1.0.0.2.ip6.int (
• type master
• file pz/0.0.0.0.0.0.0.e.0.0.7.0.1.0.0
  .2.ip6.arpa
• )

13
NTP

• NTP server and client from http//www.ntp.org/
• IPv6 addresses can be used all places where IPv4
  is used, except for reference clock addresses
  which always are IPv4
• -4 or -6 can specified in front of hostnames to
  force v4 or v6
• Can also use IPv6 multicast. IANA has reserved
  the site-local multicast address ff05101
• The IPv6 enabled RIPE TTM can also be used as an
  IPv6 NTP server
• Also commercial Meinberg NTP server
• See http//eng.hexago.com/services/ntp.shtml for
  list of public IPv6 NTP servers